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 * $FreeBSD$
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 int tcp_do_rfc1644 = 0;
161 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1644, rfc1644, CTLFLAG_RW,
162 &tcp_do_rfc1644 , 0, "Enable rfc1644 (TTCP) extensions");
163
164 static int tcp_tcbhashsize = 0;
165 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
166 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
167
168 static int do_tcpdrain = 1;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
170 "Enable tcp_drain routine for extra help when low on mbufs");
171
172 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
173 &tcbinfo.ipi_count, 0, "Number of active PCBs");
174
175 static int icmp_may_rst = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW, &icmp_may_rst, 0,
177 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
178
179 static int tcp_isn_reseed_interval = 0;
180 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
181 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
182
183 /*
184 * TCP bandwidth limiting sysctls. Note that the default lower bound of
185 * 1024 exists only for debugging. A good production default would be
186 * something like 6100.
187 */
188 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
189 "TCP inflight data limiting");
190
191 static int tcp_inflight_enable = 1;
192 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
193 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
194
195 static int tcp_inflight_debug = 0;
196 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
197 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
198
199 static int tcp_inflight_min = 6144;
200 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
201 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
202
203 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
204 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
205 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
206
207 static int tcp_inflight_stab = 20;
208 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
209 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
210
211 uma_zone_t sack_hole_zone;
212
213 static struct inpcb *tcp_notify(struct inpcb *, int);
214 static void tcp_discardcb(struct tcpcb *);
215 static void tcp_isn_tick(void *);
216
217 /*
218 * Target size of TCP PCB hash tables. Must be a power of two.
219 *
220 * Note that this can be overridden by the kernel environment
221 * variable net.inet.tcp.tcbhashsize
222 */
223 #ifndef TCBHASHSIZE
224 #define TCBHASHSIZE 512
225 #endif
226
227 /*
228 * XXX
229 * Callouts should be moved into struct tcp directly. They are currently
230 * separate because the tcpcb structure is exported to userland for sysctl
231 * parsing purposes, which do not know about callouts.
232 */
233 struct tcpcb_mem {
234 struct tcpcb tcb;
235 struct callout tcpcb_mem_rexmt, tcpcb_mem_persist, tcpcb_mem_keep;
236 struct callout tcpcb_mem_2msl, tcpcb_mem_delack;
237 };
238
239 static uma_zone_t tcpcb_zone;
240 static uma_zone_t tcptw_zone;
241 struct callout isn_callout;
242
243 /*
244 * Tcp initialization
245 */
246 void
247 tcp_init()
248 {
249 int hashsize = TCBHASHSIZE;
250
251 tcp_ccgen = 1;
252
253 tcp_delacktime = TCPTV_DELACK;
254 tcp_keepinit = TCPTV_KEEP_INIT;
255 tcp_keepidle = TCPTV_KEEP_IDLE;
256 tcp_keepintvl = TCPTV_KEEPINTVL;
257 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
258 tcp_msl = TCPTV_MSL;
259 tcp_rexmit_min = TCPTV_MIN;
260 tcp_rexmit_slop = TCPTV_CPU_VAR;
261
262 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
263 LIST_INIT(&tcb);
264 tcbinfo.listhead = &tcb;
265 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
266 if (!powerof2(hashsize)) {
267 printf("WARNING: TCB hash size not a power of 2\n");
268 hashsize = 512; /* safe default */
269 }
270 tcp_tcbhashsize = hashsize;
271 tcbinfo.hashbase = hashinit(hashsize, M_PCB, &tcbinfo.hashmask);
272 tcbinfo.porthashbase = hashinit(hashsize, M_PCB,
273 &tcbinfo.porthashmask);
274 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
275 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
276 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
277 #ifdef INET6
278 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
279 #else /* INET6 */
280 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
281 #endif /* INET6 */
282 if (max_protohdr < TCP_MINPROTOHDR)
283 max_protohdr = TCP_MINPROTOHDR;
284 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
285 panic("tcp_init");
286 #undef TCP_MINPROTOHDR
287 /*
288 * These have to be type stable for the benefit of the timers.
289 */
290 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
291 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
292 uma_zone_set_max(tcpcb_zone, maxsockets);
293 tcptw_zone = uma_zcreate("tcptw", sizeof(struct tcptw),
294 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
295 uma_zone_set_max(tcptw_zone, maxsockets / 5);
296 tcp_timer_init();
297 syncache_init();
298 tcp_hc_init();
299 tcp_reass_init();
300 callout_init(&isn_callout, CALLOUT_MPSAFE);
301 tcp_isn_tick(NULL);
302 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
303 SHUTDOWN_PRI_DEFAULT);
304 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
305 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
306 }
307
308 void
309 tcp_fini(xtp)
310 void *xtp;
311 {
312 callout_stop(&isn_callout);
313
314 }
315
316 /*
317 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
318 * tcp_template used to store this data in mbufs, but we now recopy it out
319 * of the tcpcb each time to conserve mbufs.
320 */
321 void
322 tcpip_fillheaders(inp, ip_ptr, tcp_ptr)
323 struct inpcb *inp;
324 void *ip_ptr;
325 void *tcp_ptr;
326 {
327 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
328
329 INP_LOCK_ASSERT(inp);
330
331 #ifdef INET6
332 if ((inp->inp_vflag & INP_IPV6) != 0) {
333 struct ip6_hdr *ip6;
334
335 ip6 = (struct ip6_hdr *)ip_ptr;
336 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
337 (inp->in6p_flowinfo & IPV6_FLOWINFO_MASK);
338 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
339 (IPV6_VERSION & IPV6_VERSION_MASK);
340 ip6->ip6_nxt = IPPROTO_TCP;
341 ip6->ip6_plen = sizeof(struct tcphdr);
342 ip6->ip6_src = inp->in6p_laddr;
343 ip6->ip6_dst = inp->in6p_faddr;
344 } else
345 #endif
346 {
347 struct ip *ip;
348
349 ip = (struct ip *)ip_ptr;
350 ip->ip_v = IPVERSION;
351 ip->ip_hl = 5;
352 ip->ip_tos = inp->inp_ip_tos;
353 ip->ip_len = 0;
354 ip->ip_id = 0;
355 ip->ip_off = 0;
356 ip->ip_ttl = inp->inp_ip_ttl;
357 ip->ip_sum = 0;
358 ip->ip_p = IPPROTO_TCP;
359 ip->ip_src = inp->inp_laddr;
360 ip->ip_dst = inp->inp_faddr;
361 }
362 th->th_sport = inp->inp_lport;
363 th->th_dport = inp->inp_fport;
364 th->th_seq = 0;
365 th->th_ack = 0;
366 th->th_x2 = 0;
367 th->th_off = 5;
368 th->th_flags = 0;
369 th->th_win = 0;
370 th->th_urp = 0;
371 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
372 }
373
374 /*
375 * Create template to be used to send tcp packets on a connection.
376 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
377 * use for this function is in keepalives, which use tcp_respond.
378 */
379 struct tcptemp *
380 tcpip_maketemplate(inp)
381 struct inpcb *inp;
382 {
383 struct mbuf *m;
384 struct tcptemp *n;
385
386 m = m_get(M_DONTWAIT, MT_HEADER);
387 if (m == NULL)
388 return (0);
389 m->m_len = sizeof(struct tcptemp);
390 n = mtod(m, struct tcptemp *);
391
392 tcpip_fillheaders(inp, (void *)&n->tt_ipgen, (void *)&n->tt_t);
393 return (n);
394 }
395
396 /*
397 * Send a single message to the TCP at address specified by
398 * the given TCP/IP header. If m == NULL, then we make a copy
399 * of the tcpiphdr at ti and send directly to the addressed host.
400 * This is used to force keep alive messages out using the TCP
401 * template for a connection. If flags are given then we send
402 * a message back to the TCP which originated the * segment ti,
403 * and discard the mbuf containing it and any other attached mbufs.
404 *
405 * In any case the ack and sequence number of the transmitted
406 * segment are as specified by the parameters.
407 *
408 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
409 */
410 void
411 tcp_respond(tp, ipgen, th, m, ack, seq, flags)
412 struct tcpcb *tp;
413 void *ipgen;
414 register struct tcphdr *th;
415 register struct mbuf *m;
416 tcp_seq ack, seq;
417 int flags;
418 {
419 register int tlen;
420 int win = 0;
421 struct ip *ip;
422 struct tcphdr *nth;
423 #ifdef INET6
424 struct ip6_hdr *ip6;
425 int isipv6;
426 #endif /* INET6 */
427 int ipflags = 0;
428 struct inpcb *inp;
429
430 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
431
432 #ifdef INET6
433 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
434 ip6 = ipgen;
435 #endif /* INET6 */
436 ip = ipgen;
437
438 if (tp != NULL) {
439 inp = tp->t_inpcb;
440 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
441 INP_INFO_WLOCK_ASSERT(&tcbinfo);
442 INP_LOCK_ASSERT(inp);
443 } else
444 inp = NULL;
445
446 if (tp != NULL) {
447 if (!(flags & TH_RST)) {
448 win = sbspace(&inp->inp_socket->so_rcv);
449 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
450 win = (long)TCP_MAXWIN << tp->rcv_scale;
451 }
452 }
453 if (m == NULL) {
454 m = m_gethdr(M_DONTWAIT, MT_HEADER);
455 if (m == NULL)
456 return;
457 tlen = 0;
458 m->m_data += max_linkhdr;
459 #ifdef INET6
460 if (isipv6) {
461 bcopy((caddr_t)ip6, mtod(m, caddr_t),
462 sizeof(struct ip6_hdr));
463 ip6 = mtod(m, struct ip6_hdr *);
464 nth = (struct tcphdr *)(ip6 + 1);
465 } else
466 #endif /* INET6 */
467 {
468 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
469 ip = mtod(m, struct ip *);
470 nth = (struct tcphdr *)(ip + 1);
471 }
472 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
473 flags = TH_ACK;
474 } else {
475 m_freem(m->m_next);
476 m->m_next = NULL;
477 m->m_data = (caddr_t)ipgen;
478 /* m_len is set later */
479 tlen = 0;
480 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
481 #ifdef INET6
482 if (isipv6) {
483 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
484 nth = (struct tcphdr *)(ip6 + 1);
485 } else
486 #endif /* INET6 */
487 {
488 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, n_long);
489 nth = (struct tcphdr *)(ip + 1);
490 }
491 if (th != nth) {
492 /*
493 * this is usually a case when an extension header
494 * exists between the IPv6 header and the
495 * TCP header.
496 */
497 nth->th_sport = th->th_sport;
498 nth->th_dport = th->th_dport;
499 }
500 xchg(nth->th_dport, nth->th_sport, n_short);
501 #undef xchg
502 }
503 #ifdef INET6
504 if (isipv6) {
505 ip6->ip6_flow = 0;
506 ip6->ip6_vfc = IPV6_VERSION;
507 ip6->ip6_nxt = IPPROTO_TCP;
508 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
509 tlen));
510 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
511 } else
512 #endif
513 {
514 tlen += sizeof (struct tcpiphdr);
515 ip->ip_len = tlen;
516 ip->ip_ttl = ip_defttl;
517 if (path_mtu_discovery)
518 ip->ip_off |= IP_DF;
519 }
520 m->m_len = tlen;
521 m->m_pkthdr.len = tlen;
522 m->m_pkthdr.rcvif = NULL;
523 #ifdef MAC
524 if (inp != NULL) {
525 /*
526 * Packet is associated with a socket, so allow the
527 * label of the response to reflect the socket label.
528 */
529 INP_LOCK_ASSERT(inp);
530 mac_create_mbuf_from_inpcb(inp, m);
531 } else {
532 /*
533 * Packet is not associated with a socket, so possibly
534 * update the label in place.
535 */
536 mac_reflect_mbuf_tcp(m);
537 }
538 #endif
539 nth->th_seq = htonl(seq);
540 nth->th_ack = htonl(ack);
541 nth->th_x2 = 0;
542 nth->th_off = sizeof (struct tcphdr) >> 2;
543 nth->th_flags = flags;
544 if (tp != NULL)
545 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
546 else
547 nth->th_win = htons((u_short)win);
548 nth->th_urp = 0;
549 #ifdef INET6
550 if (isipv6) {
551 nth->th_sum = 0;
552 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
553 sizeof(struct ip6_hdr),
554 tlen - sizeof(struct ip6_hdr));
555 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
556 NULL, NULL);
557 } else
558 #endif /* INET6 */
559 {
560 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
561 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
562 m->m_pkthdr.csum_flags = CSUM_TCP;
563 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
564 }
565 #ifdef TCPDEBUG
566 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
567 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
568 #endif
569 #ifdef INET6
570 if (isipv6)
571 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
572 else
573 #endif /* INET6 */
574 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
575 }
576
577 /*
578 * Create a new TCP control block, making an
579 * empty reassembly queue and hooking it to the argument
580 * protocol control block. The `inp' parameter must have
581 * come from the zone allocator set up in tcp_init().
582 */
583 struct tcpcb *
584 tcp_newtcpcb(inp)
585 struct inpcb *inp;
586 {
587 struct tcpcb_mem *tm;
588 struct tcpcb *tp;
589 #ifdef INET6
590 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
591 #endif /* INET6 */
592 int callout_flag;
593
594 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
595 if (tm == NULL)
596 return (NULL);
597 tp = &tm->tcb;
598 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
599 tp->t_maxseg = tp->t_maxopd =
600 #ifdef INET6
601 isipv6 ? tcp_v6mssdflt :
602 #endif /* INET6 */
603 tcp_mssdflt;
604
605 /* Set up our timeouts. */
606 callout_flag = debug_mpsafenet ? CALLOUT_MPSAFE : 0;
607 callout_init(tp->tt_rexmt = &tm->tcpcb_mem_rexmt, callout_flag);
608 callout_init(tp->tt_persist = &tm->tcpcb_mem_persist, callout_flag);
609 callout_init(tp->tt_keep = &tm->tcpcb_mem_keep, callout_flag);
610 callout_init(tp->tt_2msl = &tm->tcpcb_mem_2msl, callout_flag);
611 callout_init(tp->tt_delack = &tm->tcpcb_mem_delack, callout_flag);
612
613 if (tcp_do_rfc1323)
614 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
615 if (tcp_do_rfc1644)
616 tp->t_flags |= TF_REQ_CC;
617 tp->sack_enable = tcp_do_sack;
618 tp->t_inpcb = inp; /* XXX */
619 /*
620 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
621 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
622 * reasonable initial retransmit time.
623 */
624 tp->t_srtt = TCPTV_SRTTBASE;
625 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
626 tp->t_rttmin = tcp_rexmit_min;
627 tp->t_rxtcur = TCPTV_RTOBASE;
628 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
629 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
630 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
631 tp->t_rcvtime = ticks;
632 tp->t_bw_rtttime = ticks;
633 /*
634 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
635 * because the socket may be bound to an IPv6 wildcard address,
636 * which may match an IPv4-mapped IPv6 address.
637 */
638 inp->inp_ip_ttl = ip_defttl;
639 inp->inp_ppcb = (caddr_t)tp;
640 return (tp); /* XXX */
641 }
642
643 /*
644 * Drop a TCP connection, reporting
645 * the specified error. If connection is synchronized,
646 * then send a RST to peer.
647 */
648 struct tcpcb *
649 tcp_drop(tp, errno)
650 register struct tcpcb *tp;
651 int errno;
652 {
653 struct socket *so = tp->t_inpcb->inp_socket;
654
655 INP_INFO_WLOCK_ASSERT(&tcbinfo);
656 INP_LOCK_ASSERT(tp->t_inpcb);
657
658 if (TCPS_HAVERCVDSYN(tp->t_state)) {
659 tp->t_state = TCPS_CLOSED;
660 (void) tcp_output(tp);
661 tcpstat.tcps_drops++;
662 } else
663 tcpstat.tcps_conndrops++;
664 if (errno == ETIMEDOUT && tp->t_softerror)
665 errno = tp->t_softerror;
666 so->so_error = errno;
667 return (tcp_close(tp));
668 }
669
670 static void
671 tcp_discardcb(tp)
672 struct tcpcb *tp;
673 {
674 struct tseg_qent *q;
675 struct inpcb *inp = tp->t_inpcb;
676 struct socket *so = inp->inp_socket;
677 #ifdef INET6
678 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
679 #endif /* INET6 */
680
681 INP_LOCK_ASSERT(inp);
682
683 /*
684 * Make sure that all of our timers are stopped before we
685 * delete the PCB.
686 */
687 callout_stop(tp->tt_rexmt);
688 callout_stop(tp->tt_persist);
689 callout_stop(tp->tt_keep);
690 callout_stop(tp->tt_2msl);
691 callout_stop(tp->tt_delack);
692
693 /*
694 * If we got enough samples through the srtt filter,
695 * save the rtt and rttvar in the routing entry.
696 * 'Enough' is arbitrarily defined as 4 rtt samples.
697 * 4 samples is enough for the srtt filter to converge
698 * to within enough % of the correct value; fewer samples
699 * and we could save a bogus rtt. The danger is not high
700 * as tcp quickly recovers from everything.
701 * XXX: Works very well but needs some more statistics!
702 */
703 if (tp->t_rttupdated >= 4) {
704 struct hc_metrics_lite metrics;
705 u_long ssthresh;
706
707 bzero(&metrics, sizeof(metrics));
708 /*
709 * Update the ssthresh always when the conditions below
710 * are satisfied. This gives us better new start value
711 * for the congestion avoidance for new connections.
712 * ssthresh is only set if packet loss occured on a session.
713 */
714 ssthresh = tp->snd_ssthresh;
715 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
716 /*
717 * convert the limit from user data bytes to
718 * packets then to packet data bytes.
719 */
720 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
721 if (ssthresh < 2)
722 ssthresh = 2;
723 ssthresh *= (u_long)(tp->t_maxseg +
724 #ifdef INET6
725 (isipv6 ? sizeof (struct ip6_hdr) +
726 sizeof (struct tcphdr) :
727 #endif
728 sizeof (struct tcpiphdr)
729 #ifdef INET6
730 )
731 #endif
732 );
733 } else
734 ssthresh = 0;
735 metrics.rmx_ssthresh = ssthresh;
736
737 metrics.rmx_rtt = tp->t_srtt;
738 metrics.rmx_rttvar = tp->t_rttvar;
739 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
740 metrics.rmx_bandwidth = tp->snd_bandwidth;
741 metrics.rmx_cwnd = tp->snd_cwnd;
742 metrics.rmx_sendpipe = 0;
743 metrics.rmx_recvpipe = 0;
744
745 tcp_hc_update(&inp->inp_inc, &metrics);
746 }
747
748 /* free the reassembly queue, if any */
749 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
750 LIST_REMOVE(q, tqe_q);
751 m_freem(q->tqe_m);
752 uma_zfree(tcp_reass_zone, q);
753 tp->t_segqlen--;
754 tcp_reass_qsize--;
755 }
756 tcp_free_sackholes(tp);
757 inp->inp_ppcb = NULL;
758 tp->t_inpcb = NULL;
759 uma_zfree(tcpcb_zone, tp);
760 soisdisconnected(so);
761 }
762
763 /*
764 * Close a TCP control block:
765 * discard all space held by the tcp
766 * discard internet protocol block
767 * wake up any sleepers
768 */
769 struct tcpcb *
770 tcp_close(tp)
771 struct tcpcb *tp;
772 {
773 struct inpcb *inp = tp->t_inpcb;
774 #ifdef INET6
775 struct socket *so = inp->inp_socket;
776 #endif
777
778 INP_INFO_WLOCK_ASSERT(&tcbinfo);
779 INP_LOCK_ASSERT(inp);
780
781 tcp_discardcb(tp);
782 #ifdef INET6
783 if (INP_CHECK_SOCKAF(so, AF_INET6))
784 in6_pcbdetach(inp);
785 else
786 #endif
787 in_pcbdetach(inp);
788 tcpstat.tcps_closed++;
789 return (NULL);
790 }
791
792 void
793 tcp_drain()
794 {
795 if (do_tcpdrain)
796 {
797 struct inpcb *inpb;
798 struct tcpcb *tcpb;
799 struct tseg_qent *te;
800
801 /*
802 * Walk the tcpbs, if existing, and flush the reassembly queue,
803 * if there is one...
804 * XXX: The "Net/3" implementation doesn't imply that the TCP
805 * reassembly queue should be flushed, but in a situation
806 * where we're really low on mbufs, this is potentially
807 * usefull.
808 */
809 INP_INFO_RLOCK(&tcbinfo);
810 LIST_FOREACH(inpb, tcbinfo.listhead, inp_list) {
811 if (inpb->inp_vflag & INP_TIMEWAIT)
812 continue;
813 INP_LOCK(inpb);
814 if ((tcpb = intotcpcb(inpb)) != NULL) {
815 while ((te = LIST_FIRST(&tcpb->t_segq))
816 != NULL) {
817 LIST_REMOVE(te, tqe_q);
818 m_freem(te->tqe_m);
819 uma_zfree(tcp_reass_zone, te);
820 tcpb->t_segqlen--;
821 tcp_reass_qsize--;
822 }
823 tcp_clean_sackreport(tcpb);
824 }
825 INP_UNLOCK(inpb);
826 }
827 INP_INFO_RUNLOCK(&tcbinfo);
828 }
829 }
830
831 /*
832 * Notify a tcp user of an asynchronous error;
833 * store error as soft error, but wake up user
834 * (for now, won't do anything until can select for soft error).
835 *
836 * Do not wake up user since there currently is no mechanism for
837 * reporting soft errors (yet - a kqueue filter may be added).
838 */
839 static struct inpcb *
840 tcp_notify(inp, error)
841 struct inpcb *inp;
842 int error;
843 {
844 struct tcpcb *tp = (struct tcpcb *)inp->inp_ppcb;
845
846 INP_INFO_WLOCK_ASSERT(&tcbinfo);
847 INP_LOCK_ASSERT(inp);
848
849 /*
850 * Ignore some errors if we are hooked up.
851 * If connection hasn't completed, has retransmitted several times,
852 * and receives a second error, give up now. This is better
853 * than waiting a long time to establish a connection that
854 * can never complete.
855 */
856 if (tp->t_state == TCPS_ESTABLISHED &&
857 (error == EHOSTUNREACH || error == ENETUNREACH ||
858 error == EHOSTDOWN)) {
859 return (inp);
860 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
861 tp->t_softerror) {
862 tcp_drop(tp, error);
863 return (struct inpcb *)0;
864 } else {
865 tp->t_softerror = error;
866 return (inp);
867 }
868 #if 0
869 wakeup( &so->so_timeo);
870 sorwakeup(so);
871 sowwakeup(so);
872 #endif
873 }
874
875 static int
876 tcp_pcblist(SYSCTL_HANDLER_ARGS)
877 {
878 int error, i, n, s;
879 struct inpcb *inp, **inp_list;
880 inp_gen_t gencnt;
881 struct xinpgen xig;
882
883 /*
884 * The process of preparing the TCB list is too time-consuming and
885 * resource-intensive to repeat twice on every request.
886 */
887 if (req->oldptr == NULL) {
888 n = tcbinfo.ipi_count;
889 req->oldidx = 2 * (sizeof xig)
890 + (n + n/8) * sizeof(struct xtcpcb);
891 return (0);
892 }
893
894 if (req->newptr != NULL)
895 return (EPERM);
896
897 /*
898 * OK, now we're committed to doing something.
899 */
900 s = splnet();
901 INP_INFO_RLOCK(&tcbinfo);
902 gencnt = tcbinfo.ipi_gencnt;
903 n = tcbinfo.ipi_count;
904 INP_INFO_RUNLOCK(&tcbinfo);
905 splx(s);
906
907 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
908 + n * sizeof(struct xtcpcb));
909 if (error != 0)
910 return (error);
911
912 xig.xig_len = sizeof xig;
913 xig.xig_count = n;
914 xig.xig_gen = gencnt;
915 xig.xig_sogen = so_gencnt;
916 error = SYSCTL_OUT(req, &xig, sizeof xig);
917 if (error)
918 return (error);
919
920 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
921 if (inp_list == NULL)
922 return (ENOMEM);
923
924 s = splnet();
925 INP_INFO_RLOCK(&tcbinfo);
926 for (inp = LIST_FIRST(tcbinfo.listhead), i = 0; inp != NULL && i < n;
927 inp = LIST_NEXT(inp, inp_list)) {
928 INP_LOCK(inp);
929 if (inp->inp_gencnt <= gencnt) {
930 /*
931 * XXX: This use of cr_cansee(), introduced with
932 * TCP state changes, is not quite right, but for
933 * now, better than nothing.
934 */
935 if (inp->inp_vflag & INP_TIMEWAIT)
936 error = cr_cansee(req->td->td_ucred,
937 intotw(inp)->tw_cred);
938 else
939 error = cr_canseesocket(req->td->td_ucred,
940 inp->inp_socket);
941 if (error == 0)
942 inp_list[i++] = inp;
943 }
944 INP_UNLOCK(inp);
945 }
946 INP_INFO_RUNLOCK(&tcbinfo);
947 splx(s);
948 n = i;
949
950 error = 0;
951 for (i = 0; i < n; i++) {
952 inp = inp_list[i];
953 if (inp->inp_gencnt <= gencnt) {
954 struct xtcpcb xt;
955 caddr_t inp_ppcb;
956
957 bzero(&xt, sizeof(xt));
958 xt.xt_len = sizeof xt;
959 /* XXX should avoid extra copy */
960 bcopy(inp, &xt.xt_inp, sizeof *inp);
961 inp_ppcb = inp->inp_ppcb;
962 if (inp_ppcb == NULL)
963 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
964 else if (inp->inp_vflag & INP_TIMEWAIT) {
965 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
966 xt.xt_tp.t_state = TCPS_TIME_WAIT;
967 } else
968 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
969 if (inp->inp_socket != NULL)
970 sotoxsocket(inp->inp_socket, &xt.xt_socket);
971 else {
972 bzero(&xt.xt_socket, sizeof xt.xt_socket);
973 xt.xt_socket.xso_protocol = IPPROTO_TCP;
974 }
975 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
976 error = SYSCTL_OUT(req, &xt, sizeof xt);
977 }
978 }
979 if (!error) {
980 /*
981 * Give the user an updated idea of our state.
982 * If the generation differs from what we told
983 * her before, she knows that something happened
984 * while we were processing this request, and it
985 * might be necessary to retry.
986 */
987 s = splnet();
988 INP_INFO_RLOCK(&tcbinfo);
989 xig.xig_gen = tcbinfo.ipi_gencnt;
990 xig.xig_sogen = so_gencnt;
991 xig.xig_count = tcbinfo.ipi_count;
992 INP_INFO_RUNLOCK(&tcbinfo);
993 splx(s);
994 error = SYSCTL_OUT(req, &xig, sizeof xig);
995 }
996 free(inp_list, M_TEMP);
997 return (error);
998 }
999
1000 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1001 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1002
1003 static int
1004 tcp_getcred(SYSCTL_HANDLER_ARGS)
1005 {
1006 struct xucred xuc;
1007 struct sockaddr_in addrs[2];
1008 struct inpcb *inp;
1009 int error, s;
1010
1011 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1012 if (error)
1013 return (error);
1014 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1015 if (error)
1016 return (error);
1017 s = splnet();
1018 INP_INFO_RLOCK(&tcbinfo);
1019 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1020 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1021 if (inp == NULL) {
1022 error = ENOENT;
1023 goto outunlocked;
1024 }
1025 INP_LOCK(inp);
1026 if (inp->inp_socket == NULL) {
1027 error = ENOENT;
1028 goto out;
1029 }
1030 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1031 if (error)
1032 goto out;
1033 cru2x(inp->inp_socket->so_cred, &xuc);
1034 out:
1035 INP_UNLOCK(inp);
1036 outunlocked:
1037 INP_INFO_RUNLOCK(&tcbinfo);
1038 splx(s);
1039 if (error == 0)
1040 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1041 return (error);
1042 }
1043
1044 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1045 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1046 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1047
1048 #ifdef INET6
1049 static int
1050 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1051 {
1052 struct xucred xuc;
1053 struct sockaddr_in6 addrs[2];
1054 struct in6_addr a6[2];
1055 struct inpcb *inp;
1056 int error, s, mapped = 0;
1057
1058 error = suser_cred(req->td->td_ucred, SUSER_ALLOWJAIL);
1059 if (error)
1060 return (error);
1061 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1062 if (error)
1063 return (error);
1064 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1065 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1066 mapped = 1;
1067 else
1068 return (EINVAL);
1069 } else {
1070 error = in6_embedscope(&a6[0], &addrs[0], NULL, NULL);
1071 if (error)
1072 return (EINVAL);
1073 error = in6_embedscope(&a6[1], &addrs[1], NULL, NULL);
1074 if (error)
1075 return (EINVAL);
1076 }
1077 s = splnet();
1078 INP_INFO_RLOCK(&tcbinfo);
1079 if (mapped == 1)
1080 inp = in_pcblookup_hash(&tcbinfo,
1081 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1082 addrs[1].sin6_port,
1083 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1084 addrs[0].sin6_port,
1085 0, NULL);
1086 else
1087 inp = in6_pcblookup_hash(&tcbinfo, &a6[1], addrs[1].sin6_port,
1088 &a6[0], addrs[0].sin6_port, 0, NULL);
1089 if (inp == NULL) {
1090 error = ENOENT;
1091 goto outunlocked;
1092 }
1093 INP_LOCK(inp);
1094 if (inp->inp_socket == NULL) {
1095 error = ENOENT;
1096 goto out;
1097 }
1098 error = cr_canseesocket(req->td->td_ucred, inp->inp_socket);
1099 if (error)
1100 goto out;
1101 cru2x(inp->inp_socket->so_cred, &xuc);
1102 out:
1103 INP_UNLOCK(inp);
1104 outunlocked:
1105 INP_INFO_RUNLOCK(&tcbinfo);
1106 splx(s);
1107 if (error == 0)
1108 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1109 return (error);
1110 }
1111
1112 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1113 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1114 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1115 #endif
1116
1117
1118 void
1119 tcp_ctlinput(cmd, sa, vip)
1120 int cmd;
1121 struct sockaddr *sa;
1122 void *vip;
1123 {
1124 struct ip *ip = vip;
1125 struct tcphdr *th;
1126 struct in_addr faddr;
1127 struct inpcb *inp;
1128 struct tcpcb *tp;
1129 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1130 struct icmp *icp;
1131 struct in_conninfo inc;
1132 tcp_seq icmp_tcp_seq;
1133 int mtu, s;
1134
1135 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1136 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1137 return;
1138
1139 if (cmd == PRC_MSGSIZE)
1140 notify = tcp_mtudisc;
1141 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1142 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1143 notify = tcp_drop_syn_sent;
1144 /*
1145 * Redirects don't need to be handled up here.
1146 */
1147 else if (PRC_IS_REDIRECT(cmd))
1148 return;
1149 /*
1150 * Source quench is depreciated.
1151 */
1152 else if (cmd == PRC_QUENCH)
1153 return;
1154 /*
1155 * Hostdead is ugly because it goes linearly through all PCBs.
1156 * XXX: We never get this from ICMP, otherwise it makes an
1157 * excellent DoS attack on machines with many connections.
1158 */
1159 else if (cmd == PRC_HOSTDEAD)
1160 ip = NULL;
1161 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1162 return;
1163 if (ip != NULL) {
1164 s = splnet();
1165 icp = (struct icmp *)((caddr_t)ip
1166 - offsetof(struct icmp, icmp_ip));
1167 th = (struct tcphdr *)((caddr_t)ip
1168 + (ip->ip_hl << 2));
1169 INP_INFO_WLOCK(&tcbinfo);
1170 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1171 ip->ip_src, th->th_sport, 0, NULL);
1172 if (inp != NULL) {
1173 INP_LOCK(inp);
1174 if (inp->inp_socket != NULL) {
1175 icmp_tcp_seq = htonl(th->th_seq);
1176 tp = intotcpcb(inp);
1177 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1178 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1179 if (cmd == PRC_MSGSIZE) {
1180 /*
1181 * MTU discovery:
1182 * If we got a needfrag set the MTU
1183 * in the route to the suggested new
1184 * value (if given) and then notify.
1185 */
1186 bzero(&inc, sizeof(inc));
1187 inc.inc_flags = 0; /* IPv4 */
1188 inc.inc_faddr = faddr;
1189
1190 mtu = ntohs(icp->icmp_nextmtu);
1191 /*
1192 * If no alternative MTU was
1193 * proposed, try the next smaller
1194 * one. ip->ip_len has already
1195 * been swapped in icmp_input().
1196 */
1197 if (!mtu)
1198 mtu = ip_next_mtu(ip->ip_len,
1199 1);
1200 if (mtu < max(296, (tcp_minmss)
1201 + sizeof(struct tcpiphdr)))
1202 mtu = 0;
1203 if (!mtu)
1204 mtu = tcp_mssdflt
1205 + sizeof(struct tcpiphdr);
1206 /*
1207 * Only cache the the MTU if it
1208 * is smaller than the interface
1209 * or route MTU. tcp_mtudisc()
1210 * will do right thing by itself.
1211 */
1212 if (mtu <= tcp_maxmtu(&inc))
1213 tcp_hc_updatemtu(&inc, mtu);
1214 }
1215
1216 inp = (*notify)(inp, inetctlerrmap[cmd]);
1217 }
1218 }
1219 if (inp != NULL)
1220 INP_UNLOCK(inp);
1221 } else {
1222 inc.inc_fport = th->th_dport;
1223 inc.inc_lport = th->th_sport;
1224 inc.inc_faddr = faddr;
1225 inc.inc_laddr = ip->ip_src;
1226 #ifdef INET6
1227 inc.inc_isipv6 = 0;
1228 #endif
1229 syncache_unreach(&inc, th);
1230 }
1231 INP_INFO_WUNLOCK(&tcbinfo);
1232 splx(s);
1233 } else
1234 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1235 }
1236
1237 #ifdef INET6
1238 void
1239 tcp6_ctlinput(cmd, sa, d)
1240 int cmd;
1241 struct sockaddr *sa;
1242 void *d;
1243 {
1244 struct tcphdr th;
1245 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1246 struct ip6_hdr *ip6;
1247 struct mbuf *m;
1248 struct ip6ctlparam *ip6cp = NULL;
1249 const struct sockaddr_in6 *sa6_src = NULL;
1250 int off;
1251 struct tcp_portonly {
1252 u_int16_t th_sport;
1253 u_int16_t th_dport;
1254 } *thp;
1255
1256 if (sa->sa_family != AF_INET6 ||
1257 sa->sa_len != sizeof(struct sockaddr_in6))
1258 return;
1259
1260 if (cmd == PRC_MSGSIZE)
1261 notify = tcp_mtudisc;
1262 else if (!PRC_IS_REDIRECT(cmd) &&
1263 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1264 return;
1265 /* Source quench is depreciated. */
1266 else if (cmd == PRC_QUENCH)
1267 return;
1268
1269 /* if the parameter is from icmp6, decode it. */
1270 if (d != NULL) {
1271 ip6cp = (struct ip6ctlparam *)d;
1272 m = ip6cp->ip6c_m;
1273 ip6 = ip6cp->ip6c_ip6;
1274 off = ip6cp->ip6c_off;
1275 sa6_src = ip6cp->ip6c_src;
1276 } else {
1277 m = NULL;
1278 ip6 = NULL;
1279 off = 0; /* fool gcc */
1280 sa6_src = &sa6_any;
1281 }
1282
1283 if (ip6 != NULL) {
1284 struct in_conninfo inc;
1285 /*
1286 * XXX: We assume that when IPV6 is non NULL,
1287 * M and OFF are valid.
1288 */
1289
1290 /* check if we can safely examine src and dst ports */
1291 if (m->m_pkthdr.len < off + sizeof(*thp))
1292 return;
1293
1294 bzero(&th, sizeof(th));
1295 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1296
1297 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1298 (struct sockaddr *)ip6cp->ip6c_src,
1299 th.th_sport, cmd, NULL, notify);
1300
1301 inc.inc_fport = th.th_dport;
1302 inc.inc_lport = th.th_sport;
1303 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1304 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1305 inc.inc_isipv6 = 1;
1306 INP_INFO_WLOCK(&tcbinfo);
1307 syncache_unreach(&inc, &th);
1308 INP_INFO_WUNLOCK(&tcbinfo);
1309 } else
1310 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1311 0, cmd, NULL, notify);
1312 }
1313 #endif /* INET6 */
1314
1315
1316 /*
1317 * Following is where TCP initial sequence number generation occurs.
1318 *
1319 * There are two places where we must use initial sequence numbers:
1320 * 1. In SYN-ACK packets.
1321 * 2. In SYN packets.
1322 *
1323 * All ISNs for SYN-ACK packets are generated by the syncache. See
1324 * tcp_syncache.c for details.
1325 *
1326 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1327 * depends on this property. In addition, these ISNs should be
1328 * unguessable so as to prevent connection hijacking. To satisfy
1329 * the requirements of this situation, the algorithm outlined in
1330 * RFC 1948 is used, with only small modifications.
1331 *
1332 * Implementation details:
1333 *
1334 * Time is based off the system timer, and is corrected so that it
1335 * increases by one megabyte per second. This allows for proper
1336 * recycling on high speed LANs while still leaving over an hour
1337 * before rollover.
1338 *
1339 * As reading the *exact* system time is too expensive to be done
1340 * whenever setting up a TCP connection, we increment the time
1341 * offset in two ways. First, a small random positive increment
1342 * is added to isn_offset for each connection that is set up.
1343 * Second, the function tcp_isn_tick fires once per clock tick
1344 * and increments isn_offset as necessary so that sequence numbers
1345 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1346 * random positive increments serve only to ensure that the same
1347 * exact sequence number is never sent out twice (as could otherwise
1348 * happen when a port is recycled in less than the system tick
1349 * interval.)
1350 *
1351 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1352 * between seeding of isn_secret. This is normally set to zero,
1353 * as reseeding should not be necessary.
1354 *
1355 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1356 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1357 * general, this means holding an exclusive (write) lock.
1358 */
1359
1360 #define ISN_BYTES_PER_SECOND 1048576
1361 #define ISN_STATIC_INCREMENT 4096
1362 #define ISN_RANDOM_INCREMENT (4096 - 1)
1363
1364 static u_char isn_secret[32];
1365 static int isn_last_reseed;
1366 static u_int32_t isn_offset, isn_offset_old;
1367 static MD5_CTX isn_ctx;
1368
1369 tcp_seq
1370 tcp_new_isn(tp)
1371 struct tcpcb *tp;
1372 {
1373 u_int32_t md5_buffer[4];
1374 tcp_seq new_isn;
1375
1376 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1377 INP_LOCK_ASSERT(tp->t_inpcb);
1378
1379 /* Seed if this is the first use, reseed if requested. */
1380 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1381 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1382 < (u_int)ticks))) {
1383 read_random(&isn_secret, sizeof(isn_secret));
1384 isn_last_reseed = ticks;
1385 }
1386
1387 /* Compute the md5 hash and return the ISN. */
1388 MD5Init(&isn_ctx);
1389 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1390 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1391 #ifdef INET6
1392 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1393 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1394 sizeof(struct in6_addr));
1395 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1396 sizeof(struct in6_addr));
1397 } else
1398 #endif
1399 {
1400 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1401 sizeof(struct in_addr));
1402 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1403 sizeof(struct in_addr));
1404 }
1405 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1406 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1407 new_isn = (tcp_seq) md5_buffer[0];
1408 isn_offset += ISN_STATIC_INCREMENT +
1409 (arc4random() & ISN_RANDOM_INCREMENT);
1410 new_isn += isn_offset;
1411 return (new_isn);
1412 }
1413
1414 /*
1415 * Increment the offset to the next ISN_BYTES_PER_SECOND / hz boundary
1416 * to keep time flowing at a relatively constant rate. If the random
1417 * increments have already pushed us past the projected offset, do nothing.
1418 */
1419 static void
1420 tcp_isn_tick(xtp)
1421 void *xtp;
1422 {
1423 u_int32_t projected_offset;
1424
1425 INP_INFO_WLOCK(&tcbinfo);
1426 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1427
1428 if (projected_offset > isn_offset)
1429 isn_offset = projected_offset;
1430
1431 isn_offset_old = isn_offset;
1432 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1433 INP_INFO_WUNLOCK(&tcbinfo);
1434 }
1435
1436 /*
1437 * When a specific ICMP unreachable message is received and the
1438 * connection state is SYN-SENT, drop the connection. This behavior
1439 * is controlled by the icmp_may_rst sysctl.
1440 */
1441 struct inpcb *
1442 tcp_drop_syn_sent(inp, errno)
1443 struct inpcb *inp;
1444 int errno;
1445 {
1446 struct tcpcb *tp = intotcpcb(inp);
1447
1448 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1449 INP_LOCK_ASSERT(inp);
1450
1451 if (tp != NULL && tp->t_state == TCPS_SYN_SENT) {
1452 tcp_drop(tp, errno);
1453 return (NULL);
1454 }
1455 return (inp);
1456 }
1457
1458 /*
1459 * When `need fragmentation' ICMP is received, update our idea of the MSS
1460 * based on the new value in the route. Also nudge TCP to send something,
1461 * since we know the packet we just sent was dropped.
1462 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1463 */
1464 struct inpcb *
1465 tcp_mtudisc(inp, errno)
1466 struct inpcb *inp;
1467 int errno;
1468 {
1469 struct tcpcb *tp = intotcpcb(inp);
1470 struct rmxp_tao tao;
1471 struct socket *so = inp->inp_socket;
1472 u_int maxmtu;
1473 u_int romtu;
1474 int mss;
1475 #ifdef INET6
1476 int isipv6;
1477 #endif /* INET6 */
1478 bzero(&tao, sizeof(tao));
1479
1480 INP_LOCK_ASSERT(inp);
1481 if (tp != NULL) {
1482 #ifdef INET6
1483 isipv6 = (tp->t_inpcb->inp_vflag & INP_IPV6) != 0;
1484 #endif
1485 maxmtu = tcp_hc_getmtu(&inp->inp_inc); /* IPv4 and IPv6 */
1486 romtu =
1487 #ifdef INET6
1488 isipv6 ? tcp_maxmtu6(&inp->inp_inc) :
1489 #endif /* INET6 */
1490 tcp_maxmtu(&inp->inp_inc);
1491 if (!maxmtu)
1492 maxmtu = romtu;
1493 else
1494 maxmtu = min(maxmtu, romtu);
1495 if (!maxmtu) {
1496 tp->t_maxopd = tp->t_maxseg =
1497 #ifdef INET6
1498 isipv6 ? tcp_v6mssdflt :
1499 #endif /* INET6 */
1500 tcp_mssdflt;
1501 return (inp);
1502 }
1503 mss = maxmtu -
1504 #ifdef INET6
1505 (isipv6 ?
1506 sizeof(struct ip6_hdr) + sizeof(struct tcphdr) :
1507 #endif /* INET6 */
1508 sizeof(struct tcpiphdr)
1509 #ifdef INET6
1510 )
1511 #endif /* INET6 */
1512 ;
1513
1514 if (tcp_do_rfc1644) {
1515 tcp_hc_gettao(&inp->inp_inc, &tao);
1516 if (tao.tao_mssopt)
1517 mss = min(mss, tao.tao_mssopt);
1518 }
1519 /*
1520 * XXX - The above conditional probably violates the TCP
1521 * spec. The problem is that, since we don't know the
1522 * other end's MSS, we are supposed to use a conservative
1523 * default. But, if we do that, then MTU discovery will
1524 * never actually take place, because the conservative
1525 * default is much less than the MTUs typically seen
1526 * on the Internet today. For the moment, we'll sweep
1527 * this under the carpet.
1528 *
1529 * The conservative default might not actually be a problem
1530 * if the only case this occurs is when sending an initial
1531 * SYN with options and data to a host we've never talked
1532 * to before. Then, they will reply with an MSS value which
1533 * will get recorded and the new parameters should get
1534 * recomputed. For Further Study.
1535 */
1536 if (tp->t_maxopd <= mss)
1537 return (inp);
1538 tp->t_maxopd = mss;
1539
1540 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
1541 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP)
1542 mss -= TCPOLEN_TSTAMP_APPA;
1543 if ((tp->t_flags & (TF_REQ_CC|TF_NOOPT)) == TF_REQ_CC &&
1544 (tp->t_flags & TF_RCVD_CC) == TF_RCVD_CC)
1545 mss -= TCPOLEN_CC_APPA;
1546 #if (MCLBYTES & (MCLBYTES - 1)) == 0
1547 if (mss > MCLBYTES)
1548 mss &= ~(MCLBYTES-1);
1549 #else
1550 if (mss > MCLBYTES)
1551 mss = mss / MCLBYTES * MCLBYTES;
1552 #endif
1553 if (so->so_snd.sb_hiwat < mss)
1554 mss = so->so_snd.sb_hiwat;
1555
1556 tp->t_maxseg = mss;
1557
1558 tcpstat.tcps_mturesent++;
1559 tp->t_rtttime = 0;
1560 tp->snd_nxt = tp->snd_una;
1561 tcp_output(tp);
1562 }
1563 return (inp);
1564 }
1565
1566 /*
1567 * Look-up the routing entry to the peer of this inpcb. If no route
1568 * is found and it cannot be allocated, then return NULL. This routine
1569 * is called by TCP routines that access the rmx structure and by tcp_mss
1570 * to get the interface MTU.
1571 */
1572 u_long
1573 tcp_maxmtu(inc)
1574 struct in_conninfo *inc;
1575 {
1576 struct route sro;
1577 struct sockaddr_in *dst;
1578 struct ifnet *ifp;
1579 u_long maxmtu = 0;
1580
1581 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1582
1583 bzero(&sro, sizeof(sro));
1584 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1585 dst = (struct sockaddr_in *)&sro.ro_dst;
1586 dst->sin_family = AF_INET;
1587 dst->sin_len = sizeof(*dst);
1588 dst->sin_addr = inc->inc_faddr;
1589 rtalloc_ign(&sro, RTF_CLONING);
1590 }
1591 if (sro.ro_rt != NULL) {
1592 ifp = sro.ro_rt->rt_ifp;
1593 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1594 maxmtu = ifp->if_mtu;
1595 else
1596 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1597 RTFREE(sro.ro_rt);
1598 }
1599 return (maxmtu);
1600 }
1601
1602 #ifdef INET6
1603 u_long
1604 tcp_maxmtu6(inc)
1605 struct in_conninfo *inc;
1606 {
1607 struct route_in6 sro6;
1608 struct ifnet *ifp;
1609 u_long maxmtu = 0;
1610
1611 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1612
1613 bzero(&sro6, sizeof(sro6));
1614 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1615 sro6.ro_dst.sin6_family = AF_INET6;
1616 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1617 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1618 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1619 }
1620 if (sro6.ro_rt != NULL) {
1621 ifp = sro6.ro_rt->rt_ifp;
1622 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1623 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1624 else
1625 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1626 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1627 RTFREE(sro6.ro_rt);
1628 }
1629
1630 return (maxmtu);
1631 }
1632 #endif /* INET6 */
1633
1634 #ifdef IPSEC
1635 /* compute ESP/AH header size for TCP, including outer IP header. */
1636 size_t
1637 ipsec_hdrsiz_tcp(tp)
1638 struct tcpcb *tp;
1639 {
1640 struct inpcb *inp;
1641 struct mbuf *m;
1642 size_t hdrsiz;
1643 struct ip *ip;
1644 #ifdef INET6
1645 struct ip6_hdr *ip6;
1646 #endif
1647 struct tcphdr *th;
1648
1649 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1650 return (0);
1651 MGETHDR(m, M_DONTWAIT, MT_DATA);
1652 if (!m)
1653 return (0);
1654
1655 #ifdef INET6
1656 if ((inp->inp_vflag & INP_IPV6) != 0) {
1657 ip6 = mtod(m, struct ip6_hdr *);
1658 th = (struct tcphdr *)(ip6 + 1);
1659 m->m_pkthdr.len = m->m_len =
1660 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1661 tcpip_fillheaders(inp, ip6, th);
1662 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1663 } else
1664 #endif /* INET6 */
1665 {
1666 ip = mtod(m, struct ip *);
1667 th = (struct tcphdr *)(ip + 1);
1668 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1669 tcpip_fillheaders(inp, ip, th);
1670 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1671 }
1672
1673 m_free(m);
1674 return (hdrsiz);
1675 }
1676 #endif /*IPSEC*/
1677
1678 /*
1679 * Move a TCP connection into TIME_WAIT state.
1680 * tcbinfo is locked.
1681 * inp is locked, and is unlocked before returning.
1682 */
1683 void
1684 tcp_twstart(tp)
1685 struct tcpcb *tp;
1686 {
1687 struct tcptw *tw;
1688 struct inpcb *inp;
1689 int tw_time, acknow;
1690 struct socket *so;
1691
1692 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_reset(). */
1693 INP_LOCK_ASSERT(tp->t_inpcb);
1694
1695 tw = uma_zalloc(tcptw_zone, M_NOWAIT);
1696 if (tw == NULL) {
1697 tw = tcp_timer_2msl_tw(1);
1698 if (tw == NULL) {
1699 tcp_close(tp);
1700 return;
1701 }
1702 }
1703 inp = tp->t_inpcb;
1704 tw->tw_inpcb = inp;
1705
1706 /*
1707 * Recover last window size sent.
1708 */
1709 tw->last_win = (tp->rcv_adv - tp->rcv_nxt) >> tp->rcv_scale;
1710
1711 /*
1712 * Set t_recent if timestamps are used on the connection.
1713 */
1714 if ((tp->t_flags & (TF_REQ_TSTMP|TF_RCVD_TSTMP|TF_NOOPT)) ==
1715 (TF_REQ_TSTMP|TF_RCVD_TSTMP))
1716 tw->t_recent = tp->ts_recent;
1717 else
1718 tw->t_recent = 0;
1719
1720 tw->snd_nxt = tp->snd_nxt;
1721 tw->rcv_nxt = tp->rcv_nxt;
1722 tw->iss = tp->iss;
1723 tw->irs = tp->irs;
1724 tw->cc_recv = tp->cc_recv;
1725 tw->cc_send = tp->cc_send;
1726 tw->t_starttime = tp->t_starttime;
1727 tw->tw_time = 0;
1728
1729 /* XXX
1730 * If this code will
1731 * be used for fin-wait-2 state also, then we may need
1732 * a ts_recent from the last segment.
1733 */
1734 /* Shorten TIME_WAIT [RFC-1644, p.28] */
1735 if (tp->cc_recv != 0 && (ticks - tp->t_starttime) < tcp_msl) {
1736 tw_time = tp->t_rxtcur * TCPTV_TWTRUNC;
1737 /* For T/TCP client, force ACK now. */
1738 acknow = 1;
1739 } else {
1740 tw_time = 2 * tcp_msl;
1741 acknow = tp->t_flags & TF_ACKNOW;
1742 }
1743 tcp_discardcb(tp);
1744 so = inp->inp_socket;
1745 ACCEPT_LOCK();
1746 SOCK_LOCK(so);
1747 so->so_pcb = NULL;
1748 tw->tw_cred = crhold(so->so_cred);
1749 tw->tw_so_options = so->so_options;
1750 sotryfree(so);
1751 inp->inp_socket = NULL;
1752 if (acknow)
1753 tcp_twrespond(tw, TH_ACK);
1754 inp->inp_ppcb = (caddr_t)tw;
1755 inp->inp_vflag |= INP_TIMEWAIT;
1756 tcp_timer_2msl_reset(tw, tw_time);
1757 INP_UNLOCK(inp);
1758 }
1759
1760 /*
1761 * The appromixate rate of ISN increase of Microsoft TCP stacks;
1762 * the actual rate is slightly higher due to the addition of
1763 * random positive increments.
1764 *
1765 * Most other new OSes use semi-randomized ISN values, so we
1766 * do not need to worry about them.
1767 */
1768 #define MS_ISN_BYTES_PER_SECOND 250000
1769
1770 /*
1771 * Determine if the ISN we will generate has advanced beyond the last
1772 * sequence number used by the previous connection. If so, indicate
1773 * that it is safe to recycle this tw socket by returning 1.
1774 *
1775 * XXXRW: This function should assert the inpcb lock as it does multiple
1776 * non-atomic reads from the tcptw, but is currently called without it from
1777 * in_pcb.c:in_pcblookup_local().
1778 */
1779 int
1780 tcp_twrecycleable(struct tcptw *tw)
1781 {
1782 tcp_seq new_iss = tw->iss;
1783 tcp_seq new_irs = tw->irs;
1784
1785 new_iss += (ticks - tw->t_starttime) * (ISN_BYTES_PER_SECOND / hz);
1786 new_irs += (ticks - tw->t_starttime) * (MS_ISN_BYTES_PER_SECOND / hz);
1787
1788 if (SEQ_GT(new_iss, tw->snd_nxt) && SEQ_GT(new_irs, tw->rcv_nxt))
1789 return (1);
1790 else
1791 return (0);
1792 }
1793
1794 struct tcptw *
1795 tcp_twclose(struct tcptw *tw, int reuse)
1796 {
1797 struct inpcb *inp;
1798
1799 inp = tw->tw_inpcb;
1800 INP_INFO_WLOCK_ASSERT(&tcbinfo); /* tcp_timer_2msl_stop(). */
1801 INP_LOCK_ASSERT(inp);
1802
1803 tw->tw_inpcb = NULL;
1804 tcp_timer_2msl_stop(tw);
1805 inp->inp_ppcb = NULL;
1806 #ifdef INET6
1807 if (inp->inp_vflag & INP_IPV6PROTO)
1808 in6_pcbdetach(inp);
1809 else
1810 #endif
1811 in_pcbdetach(inp);
1812 tcpstat.tcps_closed++;
1813 crfree(tw->tw_cred);
1814 tw->tw_cred = NULL;
1815 if (reuse)
1816 return (tw);
1817 uma_zfree(tcptw_zone, tw);
1818 return (NULL);
1819 }
1820
1821 int
1822 tcp_twrespond(struct tcptw *tw, int flags)
1823 {
1824 struct inpcb *inp = tw->tw_inpcb;
1825 struct tcphdr *th;
1826 struct mbuf *m;
1827 struct ip *ip = NULL;
1828 u_int8_t *optp;
1829 u_int hdrlen, optlen;
1830 int error;
1831 #ifdef INET6
1832 struct ip6_hdr *ip6 = NULL;
1833 int isipv6 = inp->inp_inc.inc_isipv6;
1834 #endif
1835
1836 INP_LOCK_ASSERT(inp);
1837
1838 m = m_gethdr(M_DONTWAIT, MT_HEADER);
1839 if (m == NULL)
1840 return (ENOBUFS);
1841 m->m_data += max_linkhdr;
1842
1843 #ifdef MAC
1844 mac_create_mbuf_from_inpcb(inp, m);
1845 #endif
1846
1847 #ifdef INET6
1848 if (isipv6) {
1849 hdrlen = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1850 ip6 = mtod(m, struct ip6_hdr *);
1851 th = (struct tcphdr *)(ip6 + 1);
1852 tcpip_fillheaders(inp, ip6, th);
1853 } else
1854 #endif
1855 {
1856 hdrlen = sizeof(struct tcpiphdr);
1857 ip = mtod(m, struct ip *);
1858 th = (struct tcphdr *)(ip + 1);
1859 tcpip_fillheaders(inp, ip, th);
1860 }
1861 optp = (u_int8_t *)(th + 1);
1862
1863 /*
1864 * Send a timestamp and echo-reply if both our side and our peer
1865 * have sent timestamps in our SYN's and this is not a RST.
1866 */
1867 if (tw->t_recent && flags == TH_ACK) {
1868 u_int32_t *lp = (u_int32_t *)optp;
1869
1870 /* Form timestamp option as shown in appendix A of RFC 1323. */
1871 *lp++ = htonl(TCPOPT_TSTAMP_HDR);
1872 *lp++ = htonl(ticks);
1873 *lp = htonl(tw->t_recent);
1874 optp += TCPOLEN_TSTAMP_APPA;
1875 }
1876
1877 /*
1878 * Send `CC-family' options if needed, and it's not a RST.
1879 */
1880 if (tw->cc_recv != 0 && flags == TH_ACK) {
1881 u_int32_t *lp = (u_int32_t *)optp;
1882
1883 *lp++ = htonl(TCPOPT_CC_HDR(TCPOPT_CC));
1884 *lp = htonl(tw->cc_send);
1885 optp += TCPOLEN_CC_APPA;
1886 }
1887 optlen = optp - (u_int8_t *)(th + 1);
1888
1889 m->m_len = hdrlen + optlen;
1890 m->m_pkthdr.len = m->m_len;
1891
1892 KASSERT(max_linkhdr + m->m_len <= MHLEN, ("tcptw: mbuf too small"));
1893
1894 th->th_seq = htonl(tw->snd_nxt);
1895 th->th_ack = htonl(tw->rcv_nxt);
1896 th->th_off = (sizeof(struct tcphdr) + optlen) >> 2;
1897 th->th_flags = flags;
1898 th->th_win = htons(tw->last_win);
1899
1900 #ifdef INET6
1901 if (isipv6) {
1902 th->th_sum = in6_cksum(m, IPPROTO_TCP, sizeof(struct ip6_hdr),
1903 sizeof(struct tcphdr) + optlen);
1904 ip6->ip6_hlim = in6_selecthlim(inp, NULL);
1905 error = ip6_output(m, inp->in6p_outputopts, NULL,
1906 (tw->tw_so_options & SO_DONTROUTE), NULL, NULL, inp);
1907 } else
1908 #endif
1909 {
1910 th->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
1911 htons(sizeof(struct tcphdr) + optlen + IPPROTO_TCP));
1912 m->m_pkthdr.csum_flags = CSUM_TCP;
1913 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
1914 ip->ip_len = m->m_pkthdr.len;
1915 if (path_mtu_discovery)
1916 ip->ip_off |= IP_DF;
1917 error = ip_output(m, inp->inp_options, NULL,
1918 (tw->tw_so_options & SO_DONTROUTE), NULL, inp);
1919 }
1920 if (flags & TH_ACK)
1921 tcpstat.tcps_sndacks++;
1922 else
1923 tcpstat.tcps_sndctrl++;
1924 tcpstat.tcps_sndtotal++;
1925 return (error);
1926 }
1927
1928 /*
1929 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1930 *
1931 * This code attempts to calculate the bandwidth-delay product as a
1932 * means of determining the optimal window size to maximize bandwidth,
1933 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1934 * routers. This code also does a fairly good job keeping RTTs in check
1935 * across slow links like modems. We implement an algorithm which is very
1936 * similar (but not meant to be) TCP/Vegas. The code operates on the
1937 * transmitter side of a TCP connection and so only effects the transmit
1938 * side of the connection.
1939 *
1940 * BACKGROUND: TCP makes no provision for the management of buffer space
1941 * at the end points or at the intermediate routers and switches. A TCP
1942 * stream, whether using NewReno or not, will eventually buffer as
1943 * many packets as it is able and the only reason this typically works is
1944 * due to the fairly small default buffers made available for a connection
1945 * (typicaly 16K or 32K). As machines use larger windows and/or window
1946 * scaling it is now fairly easy for even a single TCP connection to blow-out
1947 * all available buffer space not only on the local interface, but on
1948 * intermediate routers and switches as well. NewReno makes a misguided
1949 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1950 * then backing off, then steadily increasing the window again until another
1951 * failure occurs, ad-infinitum. This results in terrible oscillation that
1952 * is only made worse as network loads increase and the idea of intentionally
1953 * blowing out network buffers is, frankly, a terrible way to manage network
1954 * resources.
1955 *
1956 * It is far better to limit the transmit window prior to the failure
1957 * condition being achieved. There are two general ways to do this: First
1958 * you can 'scan' through different transmit window sizes and locate the
1959 * point where the RTT stops increasing, indicating that you have filled the
1960 * pipe, then scan backwards until you note that RTT stops decreasing, then
1961 * repeat ad-infinitum. This method works in principle but has severe
1962 * implementation issues due to RTT variances, timer granularity, and
1963 * instability in the algorithm which can lead to many false positives and
1964 * create oscillations as well as interact badly with other TCP streams
1965 * implementing the same algorithm.
1966 *
1967 * The second method is to limit the window to the bandwidth delay product
1968 * of the link. This is the method we implement. RTT variances and our
1969 * own manipulation of the congestion window, bwnd, can potentially
1970 * destabilize the algorithm. For this reason we have to stabilize the
1971 * elements used to calculate the window. We do this by using the minimum
1972 * observed RTT, the long term average of the observed bandwidth, and
1973 * by adding two segments worth of slop. It isn't perfect but it is able
1974 * to react to changing conditions and gives us a very stable basis on
1975 * which to extend the algorithm.
1976 */
1977 void
1978 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1979 {
1980 u_long bw;
1981 u_long bwnd;
1982 int save_ticks;
1983
1984 INP_LOCK_ASSERT(tp->t_inpcb);
1985
1986 /*
1987 * If inflight_enable is disabled in the middle of a tcp connection,
1988 * make sure snd_bwnd is effectively disabled.
1989 */
1990 if (tcp_inflight_enable == 0) {
1991 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1992 tp->snd_bandwidth = 0;
1993 return;
1994 }
1995
1996 /*
1997 * Figure out the bandwidth. Due to the tick granularity this
1998 * is a very rough number and it MUST be averaged over a fairly
1999 * long period of time. XXX we need to take into account a link
2000 * that is not using all available bandwidth, but for now our
2001 * slop will ramp us up if this case occurs and the bandwidth later
2002 * increases.
2003 *
2004 * Note: if ticks rollover 'bw' may wind up negative. We must
2005 * effectively reset t_bw_rtttime for this case.
2006 */
2007 save_ticks = ticks;
2008 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
2009 return;
2010
2011 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
2012 (save_ticks - tp->t_bw_rtttime);
2013 tp->t_bw_rtttime = save_ticks;
2014 tp->t_bw_rtseq = ack_seq;
2015 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
2016 return;
2017 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
2018
2019 tp->snd_bandwidth = bw;
2020
2021 /*
2022 * Calculate the semi-static bandwidth delay product, plus two maximal
2023 * segments. The additional slop puts us squarely in the sweet
2024 * spot and also handles the bandwidth run-up case and stabilization.
2025 * Without the slop we could be locking ourselves into a lower
2026 * bandwidth.
2027 *
2028 * Situations Handled:
2029 * (1) Prevents over-queueing of packets on LANs, especially on
2030 * high speed LANs, allowing larger TCP buffers to be
2031 * specified, and also does a good job preventing
2032 * over-queueing of packets over choke points like modems
2033 * (at least for the transmit side).
2034 *
2035 * (2) Is able to handle changing network loads (bandwidth
2036 * drops so bwnd drops, bandwidth increases so bwnd
2037 * increases).
2038 *
2039 * (3) Theoretically should stabilize in the face of multiple
2040 * connections implementing the same algorithm (this may need
2041 * a little work).
2042 *
2043 * (4) Stability value (defaults to 20 = 2 maximal packets) can
2044 * be adjusted with a sysctl but typically only needs to be
2045 * on very slow connections. A value no smaller then 5
2046 * should be used, but only reduce this default if you have
2047 * no other choice.
2048 */
2049 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
2050 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
2051 #undef USERTT
2052
2053 if (tcp_inflight_debug > 0) {
2054 static int ltime;
2055 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
2056 ltime = ticks;
2057 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
2058 tp,
2059 bw,
2060 tp->t_rttbest,
2061 tp->t_srtt,
2062 bwnd
2063 );
2064 }
2065 }
2066 if ((long)bwnd < tcp_inflight_min)
2067 bwnd = tcp_inflight_min;
2068 if (bwnd > tcp_inflight_max)
2069 bwnd = tcp_inflight_max;
2070 if ((long)bwnd < tp->t_maxseg * 2)
2071 bwnd = tp->t_maxseg * 2;
2072 tp->snd_bwnd = bwnd;
2073 }
2074
2075 #ifdef TCP_SIGNATURE
2076 /*
2077 * Callback function invoked by m_apply() to digest TCP segment data
2078 * contained within an mbuf chain.
2079 */
2080 static int
2081 tcp_signature_apply(void *fstate, void *data, u_int len)
2082 {
2083
2084 MD5Update(fstate, (u_char *)data, len);
2085 return (0);
2086 }
2087
2088 /*
2089 * Compute TCP-MD5 hash of a TCPv4 segment. (RFC2385)
2090 *
2091 * Parameters:
2092 * m pointer to head of mbuf chain
2093 * off0 offset to TCP header within the mbuf chain
2094 * len length of TCP segment data, excluding options
2095 * optlen length of TCP segment options
2096 * buf pointer to storage for computed MD5 digest
2097 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2098 *
2099 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2100 * When called from tcp_input(), we can be sure that th_sum has been
2101 * zeroed out and verified already.
2102 *
2103 * This function is for IPv4 use only. Calling this function with an
2104 * IPv6 packet in the mbuf chain will yield undefined results.
2105 *
2106 * Return 0 if successful, otherwise return -1.
2107 *
2108 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2109 * search with the destination IP address, and a 'magic SPI' to be
2110 * determined by the application. This is hardcoded elsewhere to 1179
2111 * right now. Another branch of this code exists which uses the SPD to
2112 * specify per-application flows but it is unstable.
2113 */
2114 int
2115 tcp_signature_compute(struct mbuf *m, int off0, int len, int optlen,
2116 u_char *buf, u_int direction)
2117 {
2118 union sockaddr_union dst;
2119 struct ippseudo ippseudo;
2120 MD5_CTX ctx;
2121 int doff;
2122 struct ip *ip;
2123 struct ipovly *ipovly;
2124 struct secasvar *sav;
2125 struct tcphdr *th;
2126 u_short savecsum;
2127
2128 KASSERT(m != NULL, ("NULL mbuf chain"));
2129 KASSERT(buf != NULL, ("NULL signature pointer"));
2130
2131 /* Extract the destination from the IP header in the mbuf. */
2132 ip = mtod(m, struct ip *);
2133 bzero(&dst, sizeof(union sockaddr_union));
2134 dst.sa.sa_len = sizeof(struct sockaddr_in);
2135 dst.sa.sa_family = AF_INET;
2136 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2137 ip->ip_src : ip->ip_dst;
2138
2139 /* Look up an SADB entry which matches the address of the peer. */
2140 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2141 if (sav == NULL) {
2142 printf("%s: SADB lookup failed for %s\n", __func__,
2143 inet_ntoa(dst.sin.sin_addr));
2144 return (EINVAL);
2145 }
2146
2147 MD5Init(&ctx);
2148 ipovly = (struct ipovly *)ip;
2149 th = (struct tcphdr *)((u_char *)ip + off0);
2150 doff = off0 + sizeof(struct tcphdr) + optlen;
2151
2152 /*
2153 * Step 1: Update MD5 hash with IP pseudo-header.
2154 *
2155 * XXX The ippseudo header MUST be digested in network byte order,
2156 * or else we'll fail the regression test. Assume all fields we've
2157 * been doing arithmetic on have been in host byte order.
2158 * XXX One cannot depend on ipovly->ih_len here. When called from
2159 * tcp_output(), the underlying ip_len member has not yet been set.
2160 */
2161 ippseudo.ippseudo_src = ipovly->ih_src;
2162 ippseudo.ippseudo_dst = ipovly->ih_dst;
2163 ippseudo.ippseudo_pad = 0;
2164 ippseudo.ippseudo_p = IPPROTO_TCP;
2165 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) + optlen);
2166 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2167
2168 /*
2169 * Step 2: Update MD5 hash with TCP header, excluding options.
2170 * The TCP checksum must be set to zero.
2171 */
2172 savecsum = th->th_sum;
2173 th->th_sum = 0;
2174 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2175 th->th_sum = savecsum;
2176
2177 /*
2178 * Step 3: Update MD5 hash with TCP segment data.
2179 * Use m_apply() to avoid an early m_pullup().
2180 */
2181 if (len > 0)
2182 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2183
2184 /*
2185 * Step 4: Update MD5 hash with shared secret.
2186 */
2187 MD5Update(&ctx, _KEYBUF(sav->key_auth), _KEYLEN(sav->key_auth));
2188 MD5Final(buf, &ctx);
2189
2190 key_sa_recordxfer(sav, m);
2191 KEY_FREESAV(&sav);
2192 return (0);
2193 }
2194 #endif /* TCP_SIGNATURE */
2195
2196 static int
2197 sysctl_drop(SYSCTL_HANDLER_ARGS)
2198 {
2199 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2200 struct sockaddr_storage addrs[2];
2201 struct inpcb *inp;
2202 struct tcpcb *tp;
2203 struct tcptw *tw;
2204 struct sockaddr_in *fin, *lin;
2205 #ifdef INET6
2206 struct sockaddr_in6 *fin6, *lin6;
2207 struct in6_addr f6, l6;
2208 #endif
2209 int error;
2210
2211 inp = NULL;
2212 fin = lin = NULL;
2213 #ifdef INET6
2214 fin6 = lin6 = NULL;
2215 #endif
2216 error = 0;
2217
2218 if (req->oldptr != NULL || req->oldlen != 0)
2219 return (EINVAL);
2220 if (req->newptr == NULL)
2221 return (EPERM);
2222 if (req->newlen < sizeof(addrs))
2223 return (ENOMEM);
2224 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2225 if (error)
2226 return (error);
2227
2228 switch (addrs[0].ss_family) {
2229 #ifdef INET6
2230 case AF_INET6:
2231 fin6 = (struct sockaddr_in6 *)&addrs[0];
2232 lin6 = (struct sockaddr_in6 *)&addrs[1];
2233 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2234 lin6->sin6_len != sizeof(struct sockaddr_in6))
2235 return (EINVAL);
2236 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2237 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2238 return (EINVAL);
2239 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2240 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2241 fin = (struct sockaddr_in *)&addrs[0];
2242 lin = (struct sockaddr_in *)&addrs[1];
2243 break;
2244 }
2245 error = in6_embedscope(&f6, fin6, NULL, NULL);
2246 if (error)
2247 return (EINVAL);
2248 error = in6_embedscope(&l6, lin6, NULL, NULL);
2249 if (error)
2250 return (EINVAL);
2251 break;
2252 #endif
2253 case AF_INET:
2254 fin = (struct sockaddr_in *)&addrs[0];
2255 lin = (struct sockaddr_in *)&addrs[1];
2256 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2257 lin->sin_len != sizeof(struct sockaddr_in))
2258 return (EINVAL);
2259 break;
2260 default:
2261 return (EINVAL);
2262 }
2263 INP_INFO_WLOCK(&tcbinfo);
2264 switch (addrs[0].ss_family) {
2265 #ifdef INET6
2266 case AF_INET6:
2267 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2268 &l6, lin6->sin6_port, 0, NULL);
2269 break;
2270 #endif
2271 case AF_INET:
2272 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2273 lin->sin_addr, lin->sin_port, 0, NULL);
2274 break;
2275 }
2276 if (inp != NULL) {
2277 INP_LOCK(inp);
2278 if ((tw = intotw(inp)) &&
2279 (inp->inp_vflag & INP_TIMEWAIT) != 0) {
2280 (void) tcp_twclose(tw, 0);
2281 } else if ((tp = intotcpcb(inp)) &&
2282 ((inp->inp_socket->so_options & SO_ACCEPTCONN) == 0)) {
2283 tp = tcp_drop(tp, ECONNABORTED);
2284 if (tp != NULL)
2285 INP_UNLOCK(inp);
2286 } else
2287 INP_UNLOCK(inp);
2288 } else
2289 error = ESRCH;
2290 INP_INFO_WUNLOCK(&tcbinfo);
2291 return (error);
2292 }
2293
2294 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2295 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2296 0, sysctl_drop, "", "Drop TCP connection");
Cache object: e2a783cee173ed4aa7b3db30a6275c0a
|