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/5.4/sys/netinet/tcp_subr.c 145979 2005-05-07 03:58:25Z cperciva $
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/tcp.h>
80 #include <netinet/tcp_fsm.h>
81 #include <netinet/tcp_seq.h>
82 #include <netinet/tcp_timer.h>
83 #include <netinet/tcp_var.h>
84 #ifdef INET6
85 #include <netinet6/tcp6_var.h>
86 #endif
87 #include <netinet/tcpip.h>
88 #ifdef TCPDEBUG
89 #include <netinet/tcp_debug.h>
90 #endif
91 #include <netinet6/ip6protosw.h>
92
93 #ifdef IPSEC
94 #include <netinet6/ipsec.h>
95 #ifdef INET6
96 #include <netinet6/ipsec6.h>
97 #endif
98 #include <netkey/key.h>
99 #endif /*IPSEC*/
100
101 #ifdef FAST_IPSEC
102 #include <netipsec/ipsec.h>
103 #include <netipsec/xform.h>
104 #ifdef INET6
105 #include <netipsec/ipsec6.h>
106 #endif
107 #include <netipsec/key.h>
108 #define IPSEC
109 #endif /*FAST_IPSEC*/
110
111 #include <machine/in_cksum.h>
112 #include <sys/md5.h>
113
114 int tcp_mssdflt = TCP_MSS;
115 SYSCTL_INT(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLFLAG_RW,
116 &tcp_mssdflt , 0, "Default TCP Maximum Segment Size");
117
118 #ifdef INET6
119 int tcp_v6mssdflt = TCP6_MSS;
120 SYSCTL_INT(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
121 CTLFLAG_RW, &tcp_v6mssdflt , 0,
122 "Default TCP Maximum Segment Size for IPv6");
123 #endif
124
125 /*
126 * Minimum MSS we accept and use. This prevents DoS attacks where
127 * we are forced to a ridiculous low MSS like 20 and send hundreds
128 * of packets instead of one. The effect scales with the available
129 * bandwidth and quickly saturates the CPU and network interface
130 * with packet generation and sending. Set to zero to disable MINMSS
131 * checking. This setting prevents us from sending too small packets.
132 */
133 int tcp_minmss = TCP_MINMSS;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
135 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
136 /*
137 * Number of TCP segments per second we accept from remote host
138 * before we start to calculate average segment size. If average
139 * segment size drops below the minimum TCP MSS we assume a DoS
140 * attack and reset+drop the connection. Care has to be taken not to
141 * set this value too small to not kill interactive type connections
142 * (telnet, SSH) which send many small packets.
143 */
144 int tcp_minmssoverload = TCP_MINMSSOVERLOAD;
145 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmssoverload, CTLFLAG_RW,
146 &tcp_minmssoverload , 0, "Number of TCP Segments per Second allowed to"
147 "be under the MINMSS Size");
148
149 #if 0
150 static int tcp_rttdflt = TCPTV_SRTTDFLT / PR_SLOWHZ;
151 SYSCTL_INT(_net_inet_tcp, TCPCTL_RTTDFLT, rttdflt, CTLFLAG_RW,
152 &tcp_rttdflt , 0, "Default maximum TCP Round Trip Time");
153 #endif
154
155 int tcp_do_rfc1323 = 1;
156 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
157 &tcp_do_rfc1323 , 0, "Enable rfc1323 (high performance TCP) extensions");
158
159 int tcp_do_rfc1644 = 0;
160 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
161 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
162
163 static int tcp_tcbhashsize = 0;
164 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
165 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
166
167 static int do_tcpdrain = 1;
168 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
169 "Enable tcp_drain routine for extra help when low on mbufs");
170
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
172 &tcbinfo.ipi_count, 0, "Number of active PCBs");
173
174 static int icmp_may_rst = 1;
175 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
176 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
177
178 static int tcp_isn_reseed_interval = 0;
179 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
180 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
181
182 /*
183 * TCP bandwidth limiting sysctls. Note that the default lower bound of
184 * 1024 exists only for debugging. A good production default would be
185 * something like 6100.
186 */
187 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
188 "TCP inflight data limiting");
189
190 static int tcp_inflight_enable = 1;
191 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
192 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
193
194 static int tcp_inflight_debug = 0;
195 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
196 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
197
198 static int tcp_inflight_min = 6144;
199 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
200 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
201
202 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
203 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
204 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
205
206 static int tcp_inflight_stab = 20;
207 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
208 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
209
210 uma_zone_t sack_hole_zone;
211
212 static struct inpcb *tcp_notify(struct inpcb *, int);
213 static void tcp_discardcb(struct tcpcb *);
214 static void tcp_isn_tick(void *);
215
216 /*
217 * Target size of TCP PCB hash tables. Must be a power of two.
218 *
219 * Note that this can be overridden by the kernel environment
220 * variable net.inet.tcp.tcbhashsize
221 */
222 #ifndef TCBHASHSIZE
223 #define TCBHASHSIZE 512
224 #endif
225
226 /*
227 * XXX
228 * Callouts should be moved into struct tcp directly. They are currently
229 * separate because the tcpcb structure is exported to userland for sysctl
230 * parsing purposes, which do not know about callouts.
231 */
232 struct tcpcb_mem {
233 struct tcpcb tcb;
234 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
235 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
236 };
237
238 static uma_zone_t tcpcb_zone;
239 static uma_zone_t tcptw_zone;
240 struct callout isn_callout;
241
242 /*
243 * Tcp initialization
244 */
245 void
246 tcp_init()
247 {
248 int hashsize = TCBHASHSIZE;
249
250 tcp_ccgen = 1;
251
252 tcp_delacktime = TCPTV_DELACK;
253 tcp_keepinit = TCPTV_KEEP_INIT;
254 tcp_keepidle = TCPTV_KEEP_IDLE;
255 tcp_keepintvl = TCPTV_KEEPINTVL;
256 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
257 tcp_msl = TCPTV_MSL;
258 tcp_rexmit_min = TCPTV_MIN;
259 tcp_rexmit_slop = TCPTV_CPU_VAR;
260
261 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
262 LIST_INIT(&tcb);
263 tcbinfo.listhead = &tcb;
264 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
265 if (!powerof2(hashsize)) {
266 printf("WARNING: TCB hash size not a power of 2\n");
267 hashsize = 512; /* safe default */
268 }
269 tcp_tcbhashsize = hashsize;
270 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
271 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
272 &tcbinfo.porthashmask);
273 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
274 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
275 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
276 #ifdef INET6
277 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
278 #else /* INET6 */
279 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
280 #endif /* INET6 */
281 if (max_protohdr < TCP_MINPROTOHDR)
282 max_protohdr = TCP_MINPROTOHDR;
283 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
284 panic("tcp_init");
285 #undef TCP_MINPROTOHDR
286 /*
287 * These have to be type stable for the benefit of the timers.
288 */
289 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
290 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
291 uma_zone_set_max(tcpcb_zone, maxsockets);
292 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
293 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
294 uma_zone_set_max(tcptw_zone, maxsockets / 5);
295 tcp_timer_init();
296 syncache_init();
297 tcp_hc_init();
298 tcp_reass_init();
299 callout_init(&isn_callout, CALLOUT_MPSAFE);
300 tcp_isn_tick(NULL);
301 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
302 SHUTDOWN_PRI_DEFAULT);
303 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
304 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 }
306
307 void
308 tcp_fini(xtp)
309 void *xtp;
310 {
311 callout_stop(&isn_callout);
312
313 }
314
315 /*
316 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
317 * tcp_template used to store this data in mbufs, but we now recopy it out
318 * of the tcpcb each time to conserve mbufs.
319 */
320 void
321 tcpip_fillheaders(inp, ip_ptr, tcp_ptr)
322 struct inpcb *inp;
323 void *ip_ptr;
324 void *tcp_ptr;
325 {
326 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
327
328 INP_LOCK_ASSERT(inp);
329
330 #ifdef INET6
331 if ((inp->inp_vflag & INP_IPV6) != 0) {
332 struct ip6_hdr *ip6;
333
334 ip6 = (struct ip6_hdr *)ip_ptr;
335 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
336 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
337 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
338 (IPV6_VERSION & IPV6_VERSION_MASK);
339 ip6->ip6_nxt = IPPROTO_TCP;
340 ip6->ip6_plen = sizeof(struct tcphdr);
341 ip6->ip6_src = inp->in6p_laddr;
342 ip6->ip6_dst = inp->in6p_faddr;
343 } else
344 #endif
345 {
346 struct ip *ip;
347
348 ip = (struct ip *)ip_ptr;
349 ip->ip_v = IPVERSION;
350 ip->ip_hl = 5;
351 ip->ip_tos = inp->inp_ip_tos;
352 ip->ip_len = 0;
353 ip->ip_id = 0;
354 ip->ip_off = 0;
355 ip->ip_ttl = inp->inp_ip_ttl;
356 ip->ip_sum = 0;
357 ip->ip_p = IPPROTO_TCP;
358 ip->ip_src = inp->inp_laddr;
359 ip->ip_dst = inp->inp_faddr;
360 }
361 th->th_sport = inp->inp_lport;
362 th->th_dport = inp->inp_fport;
363 th->th_seq = 0;
364 th->th_ack = 0;
365 th->th_x2 = 0;
366 th->th_off = 5;
367 th->th_flags = 0;
368 th->th_win = 0;
369 th->th_urp = 0;
370 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
371 }
372
373 /*
374 * Create template to be used to send tcp packets on a connection.
375 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
376 * use for this function is in keepalives, which use tcp_respond.
377 */
378 struct tcptemp *
379 tcpip_maketemplate(inp)
380 struct inpcb *inp;
381 {
382 struct mbuf *m;
383 struct tcptemp *n;
384
385 m = m_get(M_DONTWAIT, MT_HEADER);
386 if (m == NULL)
387 return (0);
388 m->m_len = sizeof(struct tcptemp);
389 n = mtod(m, struct tcptemp *);
390
391 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
392 return (n);
393 }
394
395 /*
396 * Send a single message to the TCP at address specified by
397 * the given TCP/IP header. If m == NULL, then we make a copy
398 * of the tcpiphdr at ti and send directly to the addressed host.
399 * This is used to force keep alive messages out using the TCP
400 * template for a connection. If flags are given then we send
401 * a message back to the TCP which originated the * segment ti,
402 * and discard the mbuf containing it and any other attached mbufs.
403 *
404 * In any case the ack and sequence number of the transmitted
405 * segment are as specified by the parameters.
406 *
407 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
408 */
409 void
410 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
411 struct tcpcb *tp;
412 void *ipgen;
413 register struct tcphdr *th;
414 register struct mbuf *m;
415 tcp_seq ack, seq;
416 int flags;
417 {
418 register int tlen;
419 int win = 0;
420 struct ip *ip;
421 struct tcphdr *nth;
422 #ifdef INET6
423 struct ip6_hdr *ip6;
424 int isipv6;
425 #endif /* INET6 */
426 int ipflags = 0;
427 struct inpcb *inp;
428
429 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
430
431 #ifdef INET6
432 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
433 ip6 = ipgen;
434 #endif /* INET6 */
435 ip = ipgen;
436
437 if (tp != NULL) {
438 inp = tp->t_inpcb;
439 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
440 INP_INFO_WLOCK_ASSERT(&tcbinfo);
441 INP_LOCK_ASSERT(inp);
442 } else
443 inp = NULL;
444
445 if (tp != NULL) {
446 if (!(flags & TH_RST)) {
447 win = sbspace(&inp->inp_socket->so_rcv);
448 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
449 win = (long)TCP_MAXWIN << tp->rcv_scale;
450 }
451 }
452 if (m == NULL) {
453 m = m_gethdr(M_DONTWAIT, MT_HEADER);
454 if (m == NULL)
455 return;
456 tlen = 0;
457 m->m_data += max_linkhdr;
458 #ifdef INET6
459 if (isipv6) {
460 bcopy((caddr_t)ip6, mtod(m, caddr_t),
461 sizeof(struct ip6_hdr));
462 ip6 = mtod(m, struct ip6_hdr *);
463 nth = (struct tcphdr *)(ip6 + 1);
464 } else
465 #endif /* INET6 */
466 {
467 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
468 ip = mtod(m, struct ip *);
469 nth = (struct tcphdr *)(ip + 1);
470 }
471 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
472 flags = TH_ACK;
473 } else {
474 m_freem(m->m_next);
475 m->m_next = NULL;
476 m->m_data = (caddr_t)ipgen;
477 /* m_len is set later */
478 tlen = 0;
479 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
480 #ifdef INET6
481 if (isipv6) {
482 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
483 nth = (struct tcphdr *)(ip6 + 1);
484 } else
485 #endif /* INET6 */
486 {
487 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
488 nth = (struct tcphdr *)(ip + 1);
489 }
490 if (th != nth) {
491 /*
492 * this is usually a case when an extension header
493 * exists between the IPv6 header and the
494 * TCP header.
495 */
496 nth->th_sport = th->th_sport;
497 nth->th_dport = th->th_dport;
498 }
499 xchg(nth->th_dport, nth->th_sport, n_short);
500 #undef xchg
501 }
502 #ifdef INET6
503 if (isipv6) {
504 ip6->ip6_flow = 0;
505 ip6->ip6_vfc = IPV6_VERSION;
506 ip6->ip6_nxt = IPPROTO_TCP;
507 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
508 tlen));
509 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
510 } else
511 #endif
512 {
513 tlen += sizeof (struct tcpiphdr);
514 ip->ip_len = tlen;
515 ip->ip_ttl = ip_defttl;
516 if (path_mtu_discovery)
517 ip->ip_off |= IP_DF;
518 }
519 m->m_len = tlen;
520 m->m_pkthdr.len = tlen;
521 m->m_pkthdr.rcvif = NULL;
522 #ifdef MAC
523 if (inp != NULL) {
524 /*
525 * Packet is associated with a socket, so allow the
526 * label of the response to reflect the socket label.
527 */
528 INP_LOCK_ASSERT(inp);
529 mac_create_mbuf_from_inpcb(inp, m);
530 } else {
531 /*
532 * Packet is not associated with a socket, so possibly
533 * update the label in place.
534 */
535 mac_reflect_mbuf_tcp(m);
536 }
537 #endif
538 nth->th_seq = htonl(seq);
539 nth->th_ack = htonl(ack);
540 nth->th_x2 = 0;
541 nth->th_off = sizeof (struct tcphdr) >> 2;
542 nth->th_flags = flags;
543 if (tp != NULL)
544 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
545 else
546 nth->th_win = htons((u_short)win);
547 nth->th_urp = 0;
548 #ifdef INET6
549 if (isipv6) {
550 nth->th_sum = 0;
551 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
552 sizeof(struct ip6_hdr),
553 tlen - sizeof(struct ip6_hdr));
554 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
555 NULL, NULL);
556 } else
557 #endif /* INET6 */
558 {
559 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
560 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
561 m->m_pkthdr.csum_flags = CSUM_TCP;
562 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
563 }
564 #ifdef TCPDEBUG
565 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
566 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
567 #endif
568 #ifdef INET6
569 if (isipv6)
570 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
571 else
572 #endif /* INET6 */
573 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
574 }
575
576 /*
577 * Create a new TCP control block, making an
578 * empty reassembly queue and hooking it to the argument
579 * protocol control block. The `inp' parameter must have
580 * come from the zone allocator set up in tcp_init().
581 */
582 struct tcpcb *
583 tcp_newtcpcb(inp)
584 struct inpcb *inp;
585 {
586 struct tcpcb_mem *tm;
587 struct tcpcb *tp;
588 #ifdef INET6
589 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
590 #endif /* INET6 */
591 int callout_flag;
592
593 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
594 if (tm == NULL)
595 return (NULL);
596 tp = &tm->tcb;
597 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
598 tp->t_maxseg = tp->t_maxopd =
599 #ifdef INET6
600 isipv6 ? tcp_v6mssdflt :
601 #endif /* INET6 */
602 tcp_mssdflt;
603
604 /* Set up our timeouts. */
605 callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0;
606 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag);
607 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag);
608 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag);
609 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag);
610 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag);
611
612 if (tcp_do_rfc1323)
613 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
614 if (tcp_do_rfc1644)
615 tp->t_flags |= TF_REQ_CC;
616 tp->sack_enable = tcp_do_sack;
617 tp->t_inpcb = inp; /* XXX */
618 /*
619 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
620 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
621 * reasonable initial retransmit time.
622 */
623 tp->t_srtt = TCPTV_SRTTBASE;
624 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
625 tp->t_rttmin = tcp_rexmit_min;
626 tp->t_rxtcur = TCPTV_RTOBASE;
627 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
628 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
629 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
630 tp->t_rcvtime = ticks;
631 tp->t_bw_rtttime = ticks;
632 /*
633 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
634 * because the socket may be bound to an IPv6 wildcard address,
635 * which may match an IPv4-mapped IPv6 address.
636 */
637 inp->inp_ip_ttl = ip_defttl;
638 inp->inp_ppcb = (caddr_t)tp;
639 return (tp); /* XXX */
640 }
641
642 /*
643 * Drop a TCP connection, reporting
644 * the specified error. If connection is synchronized,
645 * then send a RST to peer.
646 */
647 struct tcpcb *
648 tcp_drop(tp, errno)
649 register struct tcpcb *tp;
650 int errno;
651 {
652 struct socket *so = tp->t_inpcb->inp_socket;
653
654 INP_LOCK_ASSERT(tp->t_inpcb);
655 if (TCPS_HAVERCVDSYN(tp->t_state)) {
656 tp->t_state = TCPS_CLOSED;
657 (void) tcp_output(tp);
658 tcpstat.tcps_drops++;
659 } else
660 tcpstat.tcps_conndrops++;
661 if (errno == ETIMEDOUT && tp->t_softerror)
662 errno = tp->t_softerror;
663 so->so_error = errno;
664 return (tcp_close(tp));
665 }
666
667 static void
668 tcp_discardcb(tp)
669 struct tcpcb *tp;
670 {
671 struct tseg_qent *q;
672 struct inpcb *inp = tp->t_inpcb;
673 struct socket *so = inp->inp_socket;
674 #ifdef INET6
675 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
676 #endif /* INET6 */
677
678 INP_LOCK_ASSERT(inp);
679
680 /*
681 * Make sure that all of our timers are stopped before we
682 * delete the PCB.
683 */
684 callout_stop(tp->tt_rexmt);
685 callout_stop(tp->tt_persist);
686 callout_stop(tp->tt_keep);
687 callout_stop(tp->tt_2msl);
688 callout_stop(tp->tt_delack);
689
690 /*
691 * If we got enough samples through the srtt filter,
692 * save the rtt and rttvar in the routing entry.
693 * 'Enough' is arbitrarily defined as 4 rtt samples.
694 * 4 samples is enough for the srtt filter to converge
695 * to within enough % of the correct value; fewer samples
696 * and we could save a bogus rtt. The danger is not high
697 * as tcp quickly recovers from everything.
698 * XXX: Works very well but needs some more statistics!
699 */
700 if (tp->t_rttupdated >= 4) {
701 struct hc_metrics_lite metrics;
702 u_long ssthresh;
703
704 bzero(&metrics, sizeof(metrics));
705 /*
706 * Update the ssthresh always when the conditions below
707 * are satisfied. This gives us better new start value
708 * for the congestion avoidance for new connections.
709 * ssthresh is only set if packet loss occured on a session.
710 */
711 ssthresh = tp->snd_ssthresh;
712 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
713 /*
714 * convert the limit from user data bytes to
715 * packets then to packet data bytes.
716 */
717 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
718 if (ssthresh < 2)
719 ssthresh = 2;
720 ssthresh *= (u_long)(tp->t_maxseg +
721 #ifdef INET6
722 (isipv6 ? sizeof (struct ip6_hdr) +
723 sizeof (struct tcphdr) :
724 #endif
725 sizeof (struct tcpiphdr)
726 #ifdef INET6
727 )
728 #endif
729 );
730 } else
731 ssthresh = 0;
732 metrics.rmx_ssthresh = ssthresh;
733
734 metrics.rmx_rtt = tp->t_srtt;
735 metrics.rmx_rttvar = tp->t_rttvar;
736 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
737 metrics.rmx_bandwidth = tp->snd_bandwidth;
738 metrics.rmx_cwnd = tp->snd_cwnd;
739 metrics.rmx_sendpipe = 0;
740 metrics.rmx_recvpipe = 0;
741
742 tcp_hc_update(&inp->inp_inc, &metrics);
743 }
744
745 /* free the reassembly queue, if any */
746 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
747 LIST_REMOVE(q, tqe_q);
748 m_freem(q->tqe_m);
749 uma_zfree(tcp_reass_zone, q);
750 tp->t_segqlen--;
751 tcp_reass_qsize--;
752 }
753 tcp_free_sackholes(tp);
754 inp->inp_ppcb = NULL;
755 tp->t_inpcb = NULL;
756 uma_zfree(tcpcb_zone, tp);
757 soisdisconnected(so);
758 }
759
760 /*
761 * Close a TCP control block:
762 * discard all space held by the tcp
763 * discard internet protocol block
764 * wake up any sleepers
765 */
766 struct tcpcb *
767 tcp_close(tp)
768 struct tcpcb *tp;
769 {
770 struct inpcb *inp = tp->t_inpcb;
771 #ifdef INET6
772 struct socket *so = inp->inp_socket;
773 #endif
774
775 INP_LOCK_ASSERT(inp);
776
777 tcp_discardcb(tp);
778 #ifdef INET6
779 if (INP_CHECK_SOCKAF(so, AF_INET6))
780 in6_pcbdetach(inp);
781 else
782 #endif
783 in_pcbdetach(inp);
784 tcpstat.tcps_closed++;
785 return (NULL);
786 }
787
788 void
789 tcp_drain()
790 {
791 if (do_tcpdrain)
792 {
793 struct inpcb *inpb;
794 struct tcpcb *tcpb;
795 struct tseg_qent *te;
796
797 /*
798 * Walk the tcpbs, if existing, and flush the reassembly queue,
799 * if there is one...
800 * XXX: The "Net/3" implementation doesn't imply that the TCP
801 * reassembly queue should be flushed, but in a situation
802 * where we're really low on mbufs, this is potentially
803 * usefull.
804 */
805 INP_INFO_RLOCK(&tcbinfo);
806 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
807 if (inpb->inp_vflag & INP_TIMEWAIT)
808 continue;
809 INP_LOCK(inpb);
810 if ((tcpb = intotcpcb(inpb)) != NULL) {
811 while ((te = LIST_FIRST(&tcpb->t_segq))
812 != NULL) {
813 LIST_REMOVE(te, tqe_q);
814 m_freem(te->tqe_m);
815 uma_zfree(tcp_reass_zone, te);
816 tcpb->t_segqlen--;
817 tcp_reass_qsize--;
818 }
819 }
820 INP_UNLOCK(inpb);
821 }
822 INP_INFO_RUNLOCK(&tcbinfo);
823 }
824 }
825
826 /*
827 * Notify a tcp user of an asynchronous error;
828 * store error as soft error, but wake up user
829 * (for now, won't do anything until can select for soft error).
830 *
831 * Do not wake up user since there currently is no mechanism for
832 * reporting soft errors (yet - a kqueue filter may be added).
833 */
834 static struct inpcb *
835 tcp_notify(inp, error)
836 struct inpcb *inp;
837 int error;
838 {
839 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
840
841 INP_LOCK_ASSERT(inp);
842
843 /*
844 * Ignore some errors if we are hooked up.
845 * If connection hasn't completed, has retransmitted several times,
846 * and receives a second error, give up now. This is better
847 * than waiting a long time to establish a connection that
848 * can never complete.
849 */
850 if (tp->t_state == TCPS_ESTABLISHED &&
851 (error == EHOSTUNREACH || error == ENETUNREACH ||
852 error == EHOSTDOWN)) {
853 return (inp);
854 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
855 tp->t_softerror) {
856 tcp_drop(tp, error);
857 return (struct inpcb *)0;
858 } else {
859 tp->t_softerror = error;
860 return (inp);
861 }
862 #if 0
863 wakeup( &so->so_timeo);
864 sorwakeup(so);
865 sowwakeup(so);
866 #endif
867 }
868
869 static int
870 tcp_pcblist(SYSCTL_HANDLER_ARGS)
871 {
872 int error, i, n, s;
873 struct inpcb *inp, **inp_list;
874 inp_gen_t gencnt;
875 struct xinpgen xig;
876
877 /*
878 * The process of preparing the TCB list is too time-consuming and
879 * resource-intensive to repeat twice on every request.
880 */
881 if (req->oldptr == NULL) {
882 n = tcbinfo.ipi_count;
883 req->oldidx = 2 * (sizeof xig)
884 + (n + n/8) * sizeof(struct xtcpcb);
885 return (0);
886 }
887
888 if (req->newptr != NULL)
889 return (EPERM);
890
891 /*
892 * OK, now we're committed to doing something.
893 */
894 s = splnet();
895 INP_INFO_RLOCK(&tcbinfo);
896 gencnt = tcbinfo.ipi_gencnt;
897 n = tcbinfo.ipi_count;
898 INP_INFO_RUNLOCK(&tcbinfo);
899 splx(s);
900
901 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
902 + n * sizeof(struct xtcpcb));
903 if (error != 0)
904 return (error);
905
906 xig.xig_len = sizeof xig;
907 xig.xig_count = n;
908 xig.xig_gen = gencnt;
909 xig.xig_sogen = so_gencnt;
910 error = SYSCTL_OUT(req, &xig, sizeof xig);
911 if (error)
912 return (error);
913
914 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
915 if (inp_list == NULL)
916 return (ENOMEM);
917
918 s = splnet();
919 INP_INFO_RLOCK(&tcbinfo);
920 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
921 inp = LIST_NEXT(inp, inp_list)) {
922 INP_LOCK(inp);
923 if (inp->inp_gencnt <= gencnt) {
924 /*
925 * XXX: This use of cr_cansee(), introduced with
926 * TCP state changes, is not quite right, but for
927 * now, better than nothing.
928 */
929 if (inp->inp_vflag & INP_TIMEWAIT)
930 error = cr_cansee(req->td->td_ucred,
931 intotw(inp)->tw_cred);
932 else
933 error = cr_canseesocket(req->td->td_ucred,
934 inp->inp_socket);
935 if (error == 0)
936 inp_list[i++] = inp;
937 }
938 INP_UNLOCK(inp);
939 }
940 INP_INFO_RUNLOCK(&tcbinfo);
941 splx(s);
942 n = i;
943
944 error = 0;
945 for (i = 0; i < n; i++) {
946 inp = inp_list[i];
947 if (inp->inp_gencnt <= gencnt) {
948 struct xtcpcb xt;
949 caddr_t inp_ppcb;
950
951 bzero(&xt, sizeof(xt));
952 xt.xt_len = sizeof xt;
953 /* XXX should avoid extra copy */
954 bcopy(inp, &xt.xt_inp, sizeof *inp);
955 inp_ppcb = inp->inp_ppcb;
956 if (inp_ppcb == NULL)
957 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
958 else if (inp->inp_vflag & INP_TIMEWAIT) {
959 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
960 xt.xt_tp.t_state = TCPS_TIME_WAIT;
961 } else
962 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
963 if (inp->inp_socket != NULL)
964 sotoxsocket(inp->inp_socket, &xt.xt_socket);
965 else {
966 bzero(&xt.xt_socket, sizeof xt.xt_socket);
967 xt.xt_socket.xso_protocol = IPPROTO_TCP;
968 }
969 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
970 error = SYSCTL_OUT(req, &xt, sizeof xt);
971 }
972 }
973 if (!error) {
974 /*
975 * Give the user an updated idea of our state.
976 * If the generation differs from what we told
977 * her before, she knows that something happened
978 * while we were processing this request, and it
979 * might be necessary to retry.
980 */
981 s = splnet();
982 INP_INFO_RLOCK(&tcbinfo);
983 xig.xig_gen = tcbinfo.ipi_gencnt;
984 xig.xig_sogen = so_gencnt;
985 xig.xig_count = tcbinfo.ipi_count;
986 INP_INFO_RUNLOCK(&tcbinfo);
987 splx(s);
988 error = SYSCTL_OUT(req, &xig, sizeof xig);
989 }
990 free(inp_list, M_TEMP);
991 return (error);
992 }
993
994 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
995 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
996
997 static int
998 tcp_getcred(SYSCTL_HANDLER_ARGS)
999 {
1000 struct xucred xuc;
1001 struct sockaddr_in addrs[2];
1002 struct inpcb *inp;
1003 int error, s;
1004
1005 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1006 if (error)
1007 return (error);
1008 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1009 if (error)
1010 return (error);
1011 s = splnet();
1012 INP_INFO_RLOCK(&tcbinfo);
1013 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1014 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1015 if (inp == NULL) {
1016 error = ENOENT;
1017 goto outunlocked;
1018 }
1019 INP_LOCK(inp);
1020 if (inp->inp_socket == NULL) {
1021 error = ENOENT;
1022 goto out;
1023 }
1024 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1025 if (error)
1026 goto out;
1027 cru2x(inp->inp_socket->so_cred, &xuc);
1028 out:
1029 INP_UNLOCK(inp);
1030 outunlocked:
1031 INP_INFO_RUNLOCK(&tcbinfo);
1032 splx(s);
1033 if (error == 0)
1034 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1035 return (error);
1036 }
1037
1038 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1039 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1040 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1041
1042 #ifdef INET6
1043 static int
1044 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1045 {
1046 struct xucred xuc;
1047 struct sockaddr_in6 addrs[2];
1048 struct in6_addr a6[2];
1049 struct inpcb *inp;
1050 int error, s, mapped = 0;
1051
1052 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1053 if (error)
1054 return (error);
1055 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1056 if (error)
1057 return (error);
1058 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1059 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1060 mapped = 1;
1061 else
1062 return (EINVAL);
1063 } else {
1064 error = in6_embedscope(&a6[0], &addrs[0], NULL, NULL);
1065 if (error)
1066 return (EINVAL);
1067 error = in6_embedscope(&a6[1], &addrs[1], NULL, NULL);
1068 if (error)
1069 return (EINVAL);
1070 }
1071 s = splnet();
1072 INP_INFO_RLOCK(&tcbinfo);
1073 if (mapped == 1)
1074 inp = in_pcblookup_hash(&tcbinfo,
1075 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1076 addrs[1].sin6_port,
1077 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1078 addrs[0].sin6_port,
1079 0, NULL);
1080 else
1081 inp = in6_pcblookup_hash(&tcbinfo, &a6[1], addrs[1].sin6_port,
1082 &a6[0], addrs[0].sin6_port, 0, NULL);
1083 if (inp == NULL) {
1084 error = ENOENT;
1085 goto outunlocked;
1086 }
1087 INP_LOCK(inp);
1088 if (inp->inp_socket == NULL) {
1089 error = ENOENT;
1090 goto out;
1091 }
1092 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1093 if (error)
1094 goto out;
1095 cru2x(inp->inp_socket->so_cred, &xuc);
1096 out:
1097 INP_UNLOCK(inp);
1098 outunlocked:
1099 INP_INFO_RUNLOCK(&tcbinfo);
1100 splx(s);
1101 if (error == 0)
1102 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1103 return (error);
1104 }
1105
1106 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1107 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1108 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1109 #endif
1110
1111
1112 void
1113 tcp_ctlinput(cmd, sa, vip)
1114 int cmd;
1115 struct sockaddr *sa;
1116 void *vip;
1117 {
1118 struct ip *ip = vip;
1119 struct tcphdr *th;
1120 struct in_addr faddr;
1121 struct inpcb *inp;
1122 struct tcpcb *tp;
1123 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1124 tcp_seq icmp_seq;
1125 int s;
1126
1127 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1128 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1129 return;
1130
1131 if (cmd == PRC_QUENCH)
1132 notify = tcp_quench;
1133 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1134 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1135 notify = tcp_drop_syn_sent;
1136 else if (cmd == PRC_MSGSIZE)
1137 notify = tcp_mtudisc;
1138 /*
1139 * Redirects don't need to be handled up here.
1140 */
1141 else if (PRC_IS_REDIRECT(cmd))
1142 return;
1143 /*
1144 * Hostdead is ugly because it goes linearly through all PCBs.
1145 * XXX: We never get this from ICMP, otherwise it makes an
1146 * excellent DoS attack on machines with many connections.
1147 */
1148 else if (cmd == PRC_HOSTDEAD)
1149 ip = NULL;
1150 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1151 return;
1152 if (ip != NULL) {
1153 s = splnet();
1154 th = (struct tcphdr *)((caddr_t)ip
1155 + (ip->ip_hl << 2));
1156 INP_INFO_WLOCK(&tcbinfo);
1157 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1158 ip->ip_src, th->th_sport, 0, NULL);
1159 if (inp != NULL) {
1160 INP_LOCK(inp);
1161 if (inp->inp_socket != NULL) {
1162 icmp_seq = htonl(th->th_seq);
1163 tp = intotcpcb(inp);
1164 if (SEQ_GEQ(icmp_seq, tp->snd_una) &&
1165 SEQ_LT(icmp_seq, tp->snd_max))
1166 inp = (*notify)(inp, inetctlerrmap[cmd]);
1167 }
1168 if (inp != NULL)
1169 INP_UNLOCK(inp);
1170 } else {
1171 struct in_conninfo inc;
1172
1173 inc.inc_fport = th->th_dport;
1174 inc.inc_lport = th->th_sport;
1175 inc.inc_faddr = faddr;
1176 inc.inc_laddr = ip->ip_src;
1177 #ifdef INET6
1178 inc.inc_isipv6 = 0;
1179 #endif
1180 syncache_unreach(&inc, th);
1181 }
1182 INP_INFO_WUNLOCK(&tcbinfo);
1183 splx(s);
1184 } else
1185 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1186 }
1187
1188 #ifdef INET6
1189 void
1190 tcp6_ctlinput(cmd, sa, d)
1191 int cmd;
1192 struct sockaddr *sa;
1193 void *d;
1194 {
1195 struct tcphdr th;
1196 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1197 struct ip6_hdr *ip6;
1198 struct mbuf *m;
1199 struct ip6ctlparam *ip6cp = NULL;
1200 const struct sockaddr_in6 *sa6_src = NULL;
1201 int off;
1202 struct tcp_portonly {
1203 u_int16_t th_sport;
1204 u_int16_t th_dport;
1205 } *thp;
1206
1207 if (sa->sa_family != AF_INET6 ||
1208 sa->sa_len != sizeof(struct sockaddr_in6))
1209 return;
1210
1211 if (cmd == PRC_QUENCH)
1212 notify = tcp_quench;
1213 else if (cmd == PRC_MSGSIZE)
1214 notify = tcp_mtudisc;
1215 else if (!PRC_IS_REDIRECT(cmd) &&
1216 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1217 return;
1218
1219 /* if the parameter is from icmp6, decode it. */
1220 if (d != NULL) {
1221 ip6cp = (struct ip6ctlparam *)d;
1222 m = ip6cp->ip6c_m;
1223 ip6 = ip6cp->ip6c_ip6;
1224 off = ip6cp->ip6c_off;
1225 sa6_src = ip6cp->ip6c_src;
1226 } else {
1227 m = NULL;
1228 ip6 = NULL;
1229 off = 0; /* fool gcc */
1230 sa6_src = &sa6_any;
1231 }
1232
1233 if (ip6 != NULL) {
1234 struct in_conninfo inc;
1235 /*
1236 * XXX: We assume that when IPV6 is non NULL,
1237 * M and OFF are valid.
1238 */
1239
1240 /* check if we can safely examine src and dst ports */
1241 if (m->m_pkthdr.len < off + sizeof(*thp))
1242 return;
1243
1244 bzero(&th, sizeof(th));
1245 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1246
1247 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1248 (struct sockaddr *)ip6cp->ip6c_src,
1249 th.th_sport, cmd, NULL, notify);
1250
1251 inc.inc_fport = th.th_dport;
1252 inc.inc_lport = th.th_sport;
1253 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1254 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1255 inc.inc_isipv6 = 1;
1256 INP_INFO_WLOCK(&tcbinfo);
1257 syncache_unreach(&inc, &th);
1258 INP_INFO_WUNLOCK(&tcbinfo);
1259 } else
1260 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1261 0, cmd, NULL, notify);
1262 }
1263 #endif /* INET6 */
1264
1265
1266 /*
1267 * Following is where TCP initial sequence number generation occurs.
1268 *
1269 * There are two places where we must use initial sequence numbers:
1270 * 1. In SYN-ACK packets.
1271 * 2. In SYN packets.
1272 *
1273 * All ISNs for SYN-ACK packets are generated by the syncache. See
1274 * tcp_syncache.c for details.
1275 *
1276 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1277 * depends on this property. In addition, these ISNs should be
1278 * unguessable so as to prevent connection hijacking. To satisfy
1279 * the requirements of this situation, the algorithm outlined in
1280 * RFC 1948 is used, with only small modifications.
1281 *
1282 * Implementation details:
1283 *
1284 * Time is based off the system timer, and is corrected so that it
1285 * increases by one megabyte per second. This allows for proper
1286 * recycling on high speed LANs while still leaving over an hour
1287 * before rollover.
1288 *
1289 * As reading the *exact* system time is too expensive to be done
1290 * whenever setting up a TCP connection, we increment the time
1291 * offset in two ways. First, a small random positive increment
1292 * is added to isn_offset for each connection that is set up.
1293 * Second, the function tcp_isn_tick fires once per clock tick
1294 * and increments isn_offset as necessary so that sequence numbers
1295 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1296 * random positive increments serve only to ensure that the same
1297 * exact sequence number is never sent out twice (as could otherwise
1298 * happen when a port is recycled in less than the system tick
1299 * interval.)
1300 *
1301 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1302 * between seeding of isn_secret. This is normally set to zero,
1303 * as reseeding should not be necessary.
1304 *
1305 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1306 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1307 * general, this means holding an exclusive (write) lock.
1308 */
1309
1310 #define ISN_BYTES_PER_SECOND 1048576
1311 #define ISN_STATIC_INCREMENT 4096
1312 #define ISN_RANDOM_INCREMENT (4096 - 1)
1313
1314 static u_char isn_secret[32];
1315 static int isn_last_reseed;
1316 static u_int32_t isn_offset, isn_offset_old;
1317 static MD5_CTX isn_ctx;
1318
1319 tcp_seq
1320 tcp_new_isn(tp)
1321 struct tcpcb *tp;
1322 {
1323 u_int32_t md5_buffer[4];
1324 tcp_seq new_isn;
1325
1326 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1327 INP_LOCK_ASSERT(tp->t_inpcb);
1328
1329 /* Seed if this is the first use, reseed if requested. */
1330 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1331 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1332 < (u_int)ticks))) {
1333 read_random(&isn_secret, sizeof(isn_secret));
1334 isn_last_reseed = ticks;
1335 }
1336
1337 /* Compute the md5 hash and return the ISN. */
1338 MD5Init(&isn_ctx);
1339 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1340 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1341 #ifdef INET6
1342 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1343 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1344 sizeof(struct in6_addr));
1345 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1346 sizeof(struct in6_addr));
1347 } else
1348 #endif
1349 {
1350 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1351 sizeof(struct in_addr));
1352 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1353 sizeof(struct in_addr));
1354 }
1355 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1356 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1357 new_isn = (tcp_seq) md5_buffer[0];
1358 isn_offset += ISN_STATIC_INCREMENT +
1359 (arc4random() & ISN_RANDOM_INCREMENT);
1360 new_isn += isn_offset;
1361 return (new_isn);
1362 }
1363
1364 /*
1365 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1366 * to keep time flowing at a relatively constant rate. If the random
1367 * increments have already pushed us past the projected offset, do nothing.
1368 */
1369 static void
1370 tcp_isn_tick(xtp)
1371 void *xtp;
1372 {
1373 u_int32_t projected_offset;
1374
1375 INP_INFO_WLOCK(&tcbinfo);
1376 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1377
1378 if (projected_offset > isn_offset)
1379 isn_offset = projected_offset;
1380
1381 isn_offset_old = isn_offset;
1382 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1383 INP_INFO_WUNLOCK(&tcbinfo);
1384 }
1385
1386 /*
1387 * When a source quench is received, close congestion window
1388 * to one segment. We will gradually open it again as we proceed.
1389 */
1390 struct inpcb *
1391 tcp_quench(inp, errno)
1392 struct inpcb *inp;
1393 int errno;
1394 {
1395 struct tcpcb *tp = intotcpcb(inp);
1396
1397 INP_LOCK_ASSERT(inp);
1398 if (tp != NULL)
1399 tp->snd_cwnd = tp->t_maxseg;
1400 return (inp);
1401 }
1402
1403 /*
1404 * When a specific ICMP unreachable message is received and the
1405 * connection state is SYN-SENT, drop the connection. This behavior
1406 * is controlled by the icmp_may_rst sysctl.
1407 */
1408 struct inpcb *
1409 tcp_drop_syn_sent(inp, errno)
1410 struct inpcb *inp;
1411 int errno;
1412 {
1413 struct tcpcb *tp = intotcpcb(inp);
1414
1415 INP_LOCK_ASSERT(inp);
1416 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) {
1417 tcp_drop(tp, errno);
1418 return (NULL);
1419 }
1420 return (inp);
1421 }
1422
1423 /*
1424 * When `need fragmentation' ICMP is received, update our idea of the MSS
1425 * based on the new value in the route. Also nudge TCP to send something,
1426 * since we know the packet we just sent was dropped.
1427 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1428 */
1429 struct inpcb *
1430 tcp_mtudisc(inp, errno)
1431 struct inpcb *inp;
1432 int errno;
1433 {
1434 struct tcpcb *tp = intotcpcb(inp);
1435 struct rmxp_tao tao;
1436 struct socket *so = inp->inp_socket;
1437 u_int maxmtu;
1438 u_int romtu;
1439 int mss;
1440 #ifdef INET6
1441 int isipv6;
1442 #endif /* INET6 */
1443 bzero(&tao, sizeof(tao));
1444
1445 INP_LOCK_ASSERT(inp);
1446 if (tp != NULL) {
1447 #ifdef INET6
1448 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1449 #endif
1450 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1451 romtu =
1452 #ifdef INET6
1453 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1454 #endif /* INET6 */
1455 tcp_maxmtu(&inp->inp_inc);
1456 if (!maxmtu)
1457 maxmtu = romtu;
1458 else
1459 maxmtu = min(maxmtu, romtu);
1460 if (!maxmtu) {
1461 tp->t_maxopd = tp->t_maxseg =
1462 #ifdef INET6
1463 isipv6 ? tcp_v6mssdflt :
1464 #endif /* INET6 */
1465 tcp_mssdflt;
1466 return (inp);
1467 }
1468 mss = maxmtu -
1469 #ifdef INET6
1470 (isipv6 ?
1471 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1472 #endif /* INET6 */
1473 sizeof(struct tcpiphdr)
1474 #ifdef INET6
1475 )
1476 #endif /* INET6 */
1477 ;
1478
1479 if (tcp_do_rfc1644) {
1480 tcp_hc_gettao(&inp->inp_inc, &tao);
1481 if (tao.tao_mssopt)
1482 mss = min(mss, tao.tao_mssopt);
1483 }
1484 /*
1485 * XXX - The above conditional probably violates the TCP
1486 * spec. The problem is that, since we don't know the
1487 * other end's MSS, we are supposed to use a conservative
1488 * default. But, if we do that, then MTU discovery will
1489 * never actually take place, because the conservative
1490 * default is much less than the MTUs typically seen
1491 * on the Internet today. For the moment, we'll sweep
1492 * this under the carpet.
1493 *
1494 * The conservative default might not actually be a problem
1495 * if the only case this occurs is when sending an initial
1496 * SYN with options and data to a host we've never talked
1497 * to before. Then, they will reply with an MSS value which
1498 * will get recorded and the new parameters should get
1499 * recomputed. For Further Study.
1500 */
1501 if (tp->t_maxopd <= mss)
1502 return (inp);
1503 tp->t_maxopd = mss;
1504
1505 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1506 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1507 mss -= TCPOLEN_TSTAMP_APPA;
1508 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1509 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1510 mss -= TCPOLEN_CC_APPA;
1511 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1512 if (mss > MCLBYTES)
1513 mss &= ~(MCLBYTES-1);
1514 #else
1515 if (mss > MCLBYTES)
1516 mss = mss / MCLBYTES * MCLBYTES;
1517 #endif
1518 if (so->so_snd.sb_hiwat < mss)
1519 mss = so->so_snd.sb_hiwat;
1520
1521 tp->t_maxseg = mss;
1522
1523 tcpstat.tcps_mturesent++;
1524 tp->t_rtttime = 0;
1525 tp->snd_nxt = tp->snd_una;
1526 tcp_output(tp);
1527 }
1528 return (inp);
1529 }
1530
1531 /*
1532 * Look-up the routing entry to the peer of this inpcb. If no route
1533 * is found and it cannot be allocated, then return NULL. This routine
1534 * is called by TCP routines that access the rmx structure and by tcp_mss
1535 * to get the interface MTU.
1536 */
1537 u_long
1538 tcp_maxmtu(inc)
1539 struct in_conninfo *inc;
1540 {
1541 struct route sro;
1542 struct sockaddr_in *dst;
1543 struct ifnet *ifp;
1544 u_long maxmtu = 0;
1545
1546 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1547
1548 bzero(&sro, sizeof(sro));
1549 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1550 dst = (struct sockaddr_in *)&sro.ro_dst;
1551 dst->sin_family = AF_INET;
1552 dst->sin_len = sizeof(*dst);
1553 dst->sin_addr = inc->inc_faddr;
1554 rtalloc_ign(&sro, RTF_CLONING);
1555 }
1556 if (sro.ro_rt != NULL) {
1557 ifp = sro.ro_rt->rt_ifp;
1558 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1559 maxmtu = ifp->if_mtu;
1560 else
1561 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1562 RTFREE(sro.ro_rt);
1563 }
1564 return (maxmtu);
1565 }
1566
1567 #ifdef INET6
1568 u_long
1569 tcp_maxmtu6(inc)
1570 struct in_conninfo *inc;
1571 {
1572 struct route_in6 sro6;
1573 struct ifnet *ifp;
1574 u_long maxmtu = 0;
1575
1576 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1577
1578 bzero(&sro6, sizeof(sro6));
1579 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1580 sro6.ro_dst.sin6_family = AF_INET6;
1581 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1582 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1583 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1584 }
1585 if (sro6.ro_rt != NULL) {
1586 ifp = sro6.ro_rt->rt_ifp;
1587 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1588 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1589 else
1590 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1591 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1592 RTFREE(sro6.ro_rt);
1593 }
1594
1595 return (maxmtu);
1596 }
1597 #endif /* INET6 */
1598
1599 #ifdef IPSEC
1600 /* compute ESP/AH header size for TCP, including outer IP header. */
1601 size_t
1602 ipsec_hdrsiz_tcp(tp)
1603 struct tcpcb *tp;
1604 {
1605 struct inpcb *inp;
1606 struct mbuf *m;
1607 size_t hdrsiz;
1608 struct ip *ip;
1609 #ifdef INET6
1610 struct ip6_hdr *ip6;
1611 #endif
1612 struct tcphdr *th;
1613
1614 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1615 return (0);
1616 MGETHDR(m, M_DONTWAIT, MT_DATA);
1617 if (!m)
1618 return (0);
1619
1620 #ifdef INET6
1621 if ((inp->inp_vflag & INP_IPV6) != 0) {
1622 ip6 = mtod(m, struct ip6_hdr *);
1623 th = (struct tcphdr *)(ip6 + 1);
1624 m->m_pkthdr.len = m->m_len =
1625 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1626 tcpip_fillheaders(inp, ip6, th);
1627 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1628 } else
1629 #endif /* INET6 */
1630 {
1631 ip = mtod(m, struct ip *);
1632 th = (struct tcphdr *)(ip + 1);
1633 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1634 tcpip_fillheaders(inp, ip, th);
1635 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1636 }
1637
1638 m_free(m);
1639 return (hdrsiz);
1640 }
1641 #endif /*IPSEC*/
1642
1643 /*
1644 * Move a TCP connection into TIME_WAIT state.
1645 * tcbinfo is locked.
1646 * inp is locked, and is unlocked before returning.
1647 */
1648 void
1649 tcp_twstart(tp)
1650 struct tcpcb *tp;
1651 {
1652 struct tcptw *tw;
1653 struct inpcb *inp;
1654 int tw_time, acknow;
1655 struct socket *so;
1656
1657 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1658 INP_LOCK_ASSERT(tp->t_inpcb);
1659
1660 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1661 if (tw == NULL) {
1662 tw = tcp_timer_2msl_tw(1);
1663 if (tw == NULL) {
1664 tcp_close(tp);
1665 return;
1666 }
1667 }
1668 inp = tp->t_inpcb;
1669 tw->tw_inpcb = inp;
1670
1671 /*
1672 * Recover last window size sent.
1673 */
1674 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1675
1676 /*
1677 * Set t_recent if timestamps are used on the connection.
1678 */
1679 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1680 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1681 tw->t_recent = tp->ts_recent;
1682 else
1683 tw->t_recent = 0;
1684
1685 tw->snd_nxt = tp->snd_nxt;
1686 tw->rcv_nxt = tp->rcv_nxt;
1687 tw->iss = tp->iss;
1688 tw->irs = tp->irs;
1689 tw->cc_recv = tp->cc_recv;
1690 tw->cc_send = tp->cc_send;
1691 tw->t_starttime = tp->t_starttime;
1692 tw->tw_time = 0;
1693
1694 /* XXX
1695 * If this code will
1696 * be used for fin-wait-2 state also, then we may need
1697 * a ts_recent from the last segment.
1698 */
1699 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1700 if (tp->cc_recv != 0 && (ticks - tp->t_starttime) < tcp_msl) {
1701 tw_time = tp->t_rxtcur * TCPTV_TWTRUNC;
1702 /* For T/TCP client, force ACK now. */
1703 acknow = 1;
1704 } else {
1705 tw_time = 2 * tcp_msl;
1706 acknow = tp->t_flags & TF_ACKNOW;
1707 }
1708 tcp_discardcb(tp);
1709 so = inp->inp_socket;
1710 ACCEPT_LOCK();
1711 SOCK_LOCK(so);
1712 so->so_pcb = NULL;
1713 tw->tw_cred = crhold(so->so_cred);
1714 tw->tw_so_options = so->so_options;
1715 sotryfree(so);
1716 inp->inp_socket = NULL;
1717 if (acknow)
1718 tcp_twrespond(tw, TH_ACK);
1719 inp->inp_ppcb = (caddr_t)tw;
1720 inp->inp_vflag |= INP_TIMEWAIT;
1721 tcp_timer_2msl_reset(tw, tw_time);
1722 INP_UNLOCK(inp);
1723 }
1724
1725 /*
1726 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1727 * the actual rate is slightly higher due to the addition of
1728 * random positive increments.
1729 *
1730 * Most other new OSes use semi-randomized ISN values, so we
1731 * do not need to worry about them.
1732 */
1733 #define MS_ISN_BYTES_PER_SECOND 250000
1734
1735 /*
1736 * Determine if the ISN we will generate has advanced beyond the last
1737 * sequence number used by the previous connection. If so, indicate
1738 * that it is safe to recycle this tw socket by returning 1.
1739 *
1740 * XXXRW: This function should assert the inpcb lock as it does multiple
1741 * non-atomic reads from the tcptw, but is currently called without it from
1742 * in_pcb.c:in_pcblookup_local().
1743 */
1744 int
1745 tcp_twrecycleable(struct tcptw *tw)
1746 {
1747 tcp_seq new_iss = tw->iss;
1748 tcp_seq new_irs = tw->irs;
1749
1750 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1751 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1752
1753 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1754 return (1);
1755 else
1756 return (0);
1757 }
1758
1759 struct tcptw *
1760 tcp_twclose(struct tcptw *tw, int reuse)
1761 {
1762 struct inpcb *inp;
1763
1764 inp = tw->tw_inpcb;
1765 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1766 INP_LOCK_ASSERT(inp);
1767
1768 tw->tw_inpcb = NULL;
1769 tcp_timer_2msl_stop(tw);
1770 inp->inp_ppcb = NULL;
1771 #ifdef INET6
1772 if (inp->inp_vflag & INP_IPV6PROTO)
1773 in6_pcbdetach(inp);
1774 else
1775 #endif
1776 in_pcbdetach(inp);
1777 tcpstat.tcps_closed++;
1778 crfree(tw->tw_cred);
1779 tw->tw_cred = NULL;
1780 if (reuse)
1781 return (tw);
1782 uma_zfree(tcptw_zone, tw);
1783 return (NULL);
1784 }
1785
1786 int
1787 tcp_twrespond(struct tcptw *tw, int flags)
1788 {
1789 struct inpcb *inp = tw->tw_inpcb;
1790 struct tcphdr *th;
1791 struct mbuf *m;
1792 struct ip *ip = NULL;
1793 u_int8_t *optp;
1794 u_int hdrlen, optlen;
1795 int error;
1796 #ifdef INET6
1797 struct ip6_hdr *ip6 = NULL;
1798 int isipv6 = inp->inp_inc.inc_isipv6;
1799 #endif
1800
1801 INP_LOCK_ASSERT(inp);
1802
1803 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1804 if (m == NULL)
1805 return (ENOBUFS);
1806 m->m_data += max_linkhdr;
1807
1808 #ifdef MAC
1809 mac_create_mbuf_from_inpcb(inp, m);
1810 #endif
1811
1812 #ifdef INET6
1813 if (isipv6) {
1814 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1815 ip6 = mtod(m, struct ip6_hdr *);
1816 th = (struct tcphdr *)(ip6 + 1);
1817 tcpip_fillheaders(inp, ip6, th);
1818 } else
1819 #endif
1820 {
1821 hdrlen = sizeof(struct tcpiphdr);
1822 ip = mtod(m, struct ip *);
1823 th = (struct tcphdr *)(ip + 1);
1824 tcpip_fillheaders(inp, ip, th);
1825 }
1826 optp = (u_int8_t *)(th + 1);
1827
1828 /*
1829 * Send a timestamp and echo-reply if both our side and our peer
1830 * have sent timestamps in our SYN's and this is not a RST.
1831 */
1832 if (tw->t_recent && flags == TH_ACK) {
1833 u_int32_t *lp = (u_int32_t *)optp;
1834
1835 /* Form timestamp option as shown in appendix A of RFC 1323. */
1836 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1837 *lp++ = htonl(ticks);
1838 *lp = htonl(tw->t_recent);
1839 optp += TCPOLEN_TSTAMP_APPA;
1840 }
1841
1842 /*
1843 * Send `CC-family' options if needed, and it's not a RST.
1844 */
1845 if (tw->cc_recv != 0 && flags == TH_ACK) {
1846 u_int32_t *lp = (u_int32_t *)optp;
1847
1848 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1849 *lp = htonl(tw->cc_send);
1850 optp += TCPOLEN_CC_APPA;
1851 }
1852 optlen = optp - (u_int8_t *)(th + 1);
1853
1854 m->m_len = hdrlen + optlen;
1855 m->m_pkthdr.len = m->m_len;
1856
1857 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1858
1859 th->th_seq = htonl(tw->snd_nxt);
1860 th->th_ack = htonl(tw->rcv_nxt);
1861 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1862 th->th_flags = flags;
1863 th->th_win = htons(tw->last_win);
1864
1865 #ifdef INET6
1866 if (isipv6) {
1867 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1868 sizeof(struct tcphdr) + optlen);
1869 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1870 error = ip6_output(m, inp->in6p_outputopts, NULL,
1871 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1872 } else
1873 #endif
1874 {
1875 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1876 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1877 m->m_pkthdr.csum_flags = CSUM_TCP;
1878 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1879 ip->ip_len = m->m_pkthdr.len;
1880 if (path_mtu_discovery)
1881 ip->ip_off |= IP_DF;
1882 error = ip_output(m, inp->inp_options, NULL,
1883 (tw->tw_so_options & SO_DONTROUTE), NULL, inp);
1884 }
1885 if (flags & TH_ACK)
1886 tcpstat.tcps_sndacks++;
1887 else
1888 tcpstat.tcps_sndctrl++;
1889 tcpstat.tcps_sndtotal++;
1890 return (error);
1891 }
1892
1893 /*
1894 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1895 *
1896 * This code attempts to calculate the bandwidth-delay product as a
1897 * means of determining the optimal window size to maximize bandwidth,
1898 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1899 * routers. This code also does a fairly good job keeping RTTs in check
1900 * across slow links like modems. We implement an algorithm which is very
1901 * similar (but not meant to be) TCP/Vegas. The code operates on the
1902 * transmitter side of a TCP connection and so only effects the transmit
1903 * side of the connection.
1904 *
1905 * BACKGROUND: TCP makes no provision for the management of buffer space
1906 * at the end points or at the intermediate routers and switches. A TCP
1907 * stream, whether using NewReno or not, will eventually buffer as
1908 * many packets as it is able and the only reason this typically works is
1909 * due to the fairly small default buffers made available for a connection
1910 * (typicaly 16K or 32K). As machines use larger windows and/or window
1911 * scaling it is now fairly easy for even a single TCP connection to blow-out
1912 * all available buffer space not only on the local interface, but on
1913 * intermediate routers and switches as well. NewReno makes a misguided
1914 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1915 * then backing off, then steadily increasing the window again until another
1916 * failure occurs, ad-infinitum. This results in terrible oscillation that
1917 * is only made worse as network loads increase and the idea of intentionally
1918 * blowing out network buffers is, frankly, a terrible way to manage network
1919 * resources.
1920 *
1921 * It is far better to limit the transmit window prior to the failure
1922 * condition being achieved. There are two general ways to do this: First
1923 * you can 'scan' through different transmit window sizes and locate the
1924 * point where the RTT stops increasing, indicating that you have filled the
1925 * pipe, then scan backwards until you note that RTT stops decreasing, then
1926 * repeat ad-infinitum. This method works in principle but has severe
1927 * implementation issues due to RTT variances, timer granularity, and
1928 * instability in the algorithm which can lead to many false positives and
1929 * create oscillations as well as interact badly with other TCP streams
1930 * implementing the same algorithm.
1931 *
1932 * The second method is to limit the window to the bandwidth delay product
1933 * of the link. This is the method we implement. RTT variances and our
1934 * own manipulation of the congestion window, bwnd, can potentially
1935 * destabilize the algorithm. For this reason we have to stabilize the
1936 * elements used to calculate the window. We do this by using the minimum
1937 * observed RTT, the long term average of the observed bandwidth, and
1938 * by adding two segments worth of slop. It isn't perfect but it is able
1939 * to react to changing conditions and gives us a very stable basis on
1940 * which to extend the algorithm.
1941 */
1942 void
1943 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1944 {
1945 u_long bw;
1946 u_long bwnd;
1947 int save_ticks;
1948
1949 INP_LOCK_ASSERT(tp->t_inpcb);
1950
1951 /*
1952 * If inflight_enable is disabled in the middle of a tcp connection,
1953 * make sure snd_bwnd is effectively disabled.
1954 */
1955 if (tcp_inflight_enable == 0) {
1956 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1957 tp->snd_bandwidth = 0;
1958 return;
1959 }
1960
1961 /*
1962 * Figure out the bandwidth. Due to the tick granularity this
1963 * is a very rough number and it MUST be averaged over a fairly
1964 * long period of time. XXX we need to take into account a link
1965 * that is not using all available bandwidth, but for now our
1966 * slop will ramp us up if this case occurs and the bandwidth later
1967 * increases.
1968 *
1969 * Note: if ticks rollover 'bw' may wind up negative. We must
1970 * effectively reset t_bw_rtttime for this case.
1971 */
1972 save_ticks = ticks;
1973 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1974 return;
1975
1976 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1977 (save_ticks - tp->t_bw_rtttime);
1978 tp->t_bw_rtttime = save_ticks;
1979 tp->t_bw_rtseq = ack_seq;
1980 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1981 return;
1982 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1983
1984 tp->snd_bandwidth = bw;
1985
1986 /*
1987 * Calculate the semi-static bandwidth delay product, plus two maximal
1988 * segments. The additional slop puts us squarely in the sweet
1989 * spot and also handles the bandwidth run-up case and stabilization.
1990 * Without the slop we could be locking ourselves into a lower
1991 * bandwidth.
1992 *
1993 * Situations Handled:
1994 * (1) Prevents over-queueing of packets on LANs, especially on
1995 * high speed LANs, allowing larger TCP buffers to be
1996 * specified, and also does a good job preventing
1997 * over-queueing of packets over choke points like modems
1998 * (at least for the transmit side).
1999 *
2000 * (2) Is able to handle changing network loads (bandwidth
2001 * drops so bwnd drops, bandwidth increases so bwnd
2002 * increases).
2003 *
2004 * (3) Theoretically should stabilize in the face of multiple
2005 * connections implementing the same algorithm (this may need
2006 * a little work).
2007 *
2008 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2009 * be adjusted with a sysctl but typically only needs to be
2010 * on very slow connections. A value no smaller then 5
2011 * should be used, but only reduce this default if you have
2012 * no other choice.
2013 */
2014 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2015 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2016 #undef USERTT
2017
2018 if (tcp_inflight_debug > 0) {
2019 static int ltime;
2020 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2021 ltime = ticks;
2022 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2023 tp,
2024 bw,
2025 tp->t_rttbest,
2026 tp->t_srtt,
2027 bwnd
2028 );
2029 }
2030 }
2031 if ((long)bwnd < tcp_inflight_min)
2032 bwnd = tcp_inflight_min;
2033 if (bwnd > tcp_inflight_max)
2034 bwnd = tcp_inflight_max;
2035 if ((long)bwnd < tp->t_maxseg * 2)
2036 bwnd = tp->t_maxseg * 2;
2037 tp->snd_bwnd = bwnd;
2038 }
2039
2040 #ifdef TCP_SIGNATURE
2041 /*
2042 * Callback function invoked by m_apply() to digest TCP segment data
2043 * contained within an mbuf chain.
2044 */
2045 static int
2046 tcp_signature_apply(void *fstate, void *data, u_int len)
2047 {
2048
2049 MD5Update(fstate, (u_char *)data, len);
2050 return (0);
2051 }
2052
2053 /*
2054 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2055 *
2056 * Parameters:
2057 * m pointer to head of mbuf chain
2058 * off0 offset to TCP header within the mbuf chain
2059 * len length of TCP segment data, excluding options
2060 * optlen length of TCP segment options
2061 * buf pointer to storage for computed MD5 digest
2062 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2063 *
2064 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2065 * When called from tcp_input(), we can be sure that th_sum has been
2066 * zeroed out and verified already.
2067 *
2068 * This function is for IPv4 use only. Calling this function with an
2069 * IPv6 packet in the mbuf chain will yield undefined results.
2070 *
2071 * Return 0 if successful, otherwise return -1.
2072 *
2073 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2074 * search with the destination IP address, and a 'magic SPI' to be
2075 * determined by the application. This is hardcoded elsewhere to 1179
2076 * right now. Another branch of this code exists which uses the SPD to
2077 * specify per-application flows but it is unstable.
2078 */
2079 int
2080 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2081 u_char *buf, u_int direction)
2082 {
2083 union sockaddr_union dst;
2084 struct ippseudo ippseudo;
2085 MD5_CTX ctx;
2086 int doff;
2087 struct ip *ip;
2088 struct ipovly *ipovly;
2089 struct secasvar *sav;
2090 struct tcphdr *th;
2091 u_short savecsum;
2092
2093 KASSERT(m != NULL, ("NULL mbuf chain"));
2094 KASSERT(buf != NULL, ("NULL signature pointer"));
2095
2096 /* Extract the destination from the IP header in the mbuf. */
2097 ip = mtod(m, struct ip *);
2098 bzero(&dst, sizeof(union sockaddr_union));
2099 dst.sa.sa_len = sizeof(struct sockaddr_in);
2100 dst.sa.sa_family = AF_INET;
2101 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2102 ip->ip_src : ip->ip_dst;
2103
2104 /* Look up an SADB entry which matches the address of the peer. */
2105 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2106 if (sav == NULL) {
2107 printf("%s: SADB lookup failed for %s\n", __func__,
2108 inet_ntoa(dst.sin.sin_addr));
2109 return (EINVAL);
2110 }
2111
2112 MD5Init(&ctx);
2113 ipovly = (struct ipovly *)ip;
2114 th = (struct tcphdr *)((u_char *)ip + off0);
2115 doff = off0 + sizeof(struct tcphdr) + optlen;
2116
2117 /*
2118 * Step 1: Update MD5 hash with IP pseudo-header.
2119 *
2120 * XXX The ippseudo header MUST be digested in network byte order,
2121 * or else we'll fail the regression test. Assume all fields we've
2122 * been doing arithmetic on have been in host byte order.
2123 * XXX One cannot depend on ipovly->ih_len here. When called from
2124 * tcp_output(), the underlying ip_len member has not yet been set.
2125 */
2126 ippseudo.ippseudo_src = ipovly->ih_src;
2127 ippseudo.ippseudo_dst = ipovly->ih_dst;
2128 ippseudo.ippseudo_pad = 0;
2129 ippseudo.ippseudo_p = IPPROTO_TCP;
2130 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2131 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2132
2133 /*
2134 * Step 2: Update MD5 hash with TCP header, excluding options.
2135 * The TCP checksum must be set to zero.
2136 */
2137 savecsum = th->th_sum;
2138 th->th_sum = 0;
2139 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2140 th->th_sum = savecsum;
2141
2142 /*
2143 * Step 3: Update MD5 hash with TCP segment data.
2144 * Use m_apply() to avoid an early m_pullup().
2145 */
2146 if (len > 0)
2147 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2148
2149 /*
2150 * Step 4: Update MD5 hash with shared secret.
2151 */
2152 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2153 MD5Final(buf, &ctx);
2154
2155 key_sa_recordxfer(sav, m);
2156 KEY_FREESAV(&sav);
2157 return (0);
2158 }
2159 #endif /* TCP_SIGNATURE */
2160
2161 static int
2162 sysctl_drop(SYSCTL_HANDLER_ARGS)
2163 {
2164 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2165 struct sockaddr_storage addrs[2];
2166 struct inpcb *inp;
2167 struct tcpcb *tp;
2168 struct sockaddr_in *fin, *lin;
2169 #ifdef INET6
2170 struct sockaddr_in6 *fin6, *lin6;
2171 struct in6_addr f6, l6;
2172 #endif
2173 int error;
2174
2175 inp = NULL;
2176 fin = lin = NULL;
2177 #ifdef INET6
2178 fin6 = lin6 = NULL;
2179 #endif
2180 error = 0;
2181
2182 if (req->oldptr != NULL || req->oldlen != 0)
2183 return (EINVAL);
2184 if (req->newptr == NULL)
2185 return (EPERM);
2186 if (req->newlen < sizeof(addrs))
2187 return (ENOMEM);
2188 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2189 if (error)
2190 return (error);
2191
2192 switch (addrs[0].ss_family) {
2193 #ifdef INET6
2194 case AF_INET6:
2195 fin6 = (struct sockaddr_in6 *)&addrs[0];
2196 lin6 = (struct sockaddr_in6 *)&addrs[1];
2197 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2198 lin6->sin6_len != sizeof(struct sockaddr_in6))
2199 return (EINVAL);
2200 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2201 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2202 return (EINVAL);
2203 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2204 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2205 fin = (struct sockaddr_in *)&addrs[0];
2206 lin = (struct sockaddr_in *)&addrs[1];
2207 break;
2208 }
2209 error = in6_embedscope(&f6, fin6, NULL, NULL);
2210 if (error)
2211 return (EINVAL);
2212 error = in6_embedscope(&l6, lin6, NULL, NULL);
2213 if (error)
2214 return (EINVAL);
2215 break;
2216 #endif
2217 case AF_INET:
2218 fin = (struct sockaddr_in *)&addrs[0];
2219 lin = (struct sockaddr_in *)&addrs[1];
2220 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2221 lin->sin_len != sizeof(struct sockaddr_in))
2222 return (EINVAL);
2223 break;
2224 default:
2225 return (EINVAL);
2226 }
2227 INP_INFO_WLOCK(&tcbinfo);
2228 switch (addrs[0].ss_family) {
2229 #ifdef INET6
2230 case AF_INET6:
2231 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2232 &l6, lin6->sin6_port, 0, NULL);
2233 break;
2234 #endif
2235 case AF_INET:
2236 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2237 lin->sin_addr, lin->sin_port, 0, NULL);
2238 break;
2239 }
2240 if (inp != NULL) {
2241 INP_LOCK(inp);
2242 if ((tp = intotcpcb(inp)) &&
2243 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
2244 tp = tcp_drop(tp, ECONNABORTED);
2245 if (tp != NULL)
2246 INP_UNLOCK(inp);
2247 } else
2248 INP_UNLOCK(inp);
2249 } else
2250 error = ESRCH;
2251 INP_INFO_WUNLOCK(&tcbinfo);
2252 return (error);
2253 }
2254
2255 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2256 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2257 0, sysctl_drop, "", "Drop TCP connection");
Cache object: 0408dc8a010403b1e606fb4847b68fda
|