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 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/7.4/sys/netinet/tcp_subr.c 214878 2010-11-06 13:46:58Z lstewart $");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_mac.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/kernel.h>
46 #include <sys/sysctl.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #ifdef INET6
50 #include <sys/domain.h>
51 #endif
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/protosw.h>
57 #include <sys/random.h>
58
59 #include <vm/uma.h>
60
61 #include <net/route.h>
62 #include <net/if.h>
63
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef INET6
68 #include <netinet/ip6.h>
69 #endif
70 #include <netinet/in_pcb.h>
71 #ifdef INET6
72 #include <netinet6/in6_pcb.h>
73 #endif
74 #include <netinet/in_var.h>
75 #include <netinet/ip_var.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #include <netinet6/nd6.h>
80 #endif
81 #include <netinet/ip_icmp.h>
82 #include <netinet/tcp.h>
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #include <netinet/tcp_offload.h>
89 #ifdef INET6
90 #include <netinet6/tcp6_var.h>
91 #endif
92 #include <netinet/tcpip.h>
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #include <netinet6/ip6protosw.h>
97
98 #ifdef IPSEC
99 #include <netipsec/ipsec.h>
100 #include <netipsec/xform.h>
101 #ifdef INET6
102 #include <netipsec/ipsec6.h>
103 #endif
104 #include <netipsec/key.h>
105 #include <sys/syslog.h>
106 #endif /*IPSEC*/
107
108 #include <machine/in_cksum.h>
109 #include <sys/md5.h>
110
111 #include <security/mac/mac_framework.h>
112
113 int tcp_mssdflt = TCP_MSS;
114 #ifdef INET6
115 int tcp_v6mssdflt = TCP6_MSS;
116 #endif
117
118 static int
119 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
120 {
121 int error, new;
122
123 new = tcp_mssdflt;
124 error = sysctl_handle_int(oidp, &new, 0, req);
125 if (error == 0 && req->newptr) {
126 if (new < TCP_MINMSS)
127 error = EINVAL;
128 else
129 tcp_mssdflt = new;
130 }
131 return (error);
132 }
133
134 SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt, CTLTYPE_INT|CTLFLAG_RW,
135 &tcp_mssdflt, 0, &sysctl_net_inet_tcp_mss_check, "I",
136 "Default TCP Maximum Segment Size");
137
138 #ifdef INET6
139 static int
140 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
141 {
142 int error, new;
143
144 new = tcp_v6mssdflt;
145 error = sysctl_handle_int(oidp, &new, 0, req);
146 if (error == 0 && req->newptr) {
147 if (new < TCP_MINMSS)
148 error = EINVAL;
149 else
150 tcp_v6mssdflt = new;
151 }
152 return (error);
153 }
154
155 SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt, CTLTYPE_INT|CTLFLAG_RW,
156 &tcp_v6mssdflt, 0, &sysctl_net_inet_tcp_mss_v6_check, "I",
157 "Default TCP Maximum Segment Size for IPv6");
158 #endif
159
160 /*
161 * Minimum MSS we accept and use. This prevents DoS attacks where
162 * we are forced to a ridiculous low MSS like 20 and send hundreds
163 * of packets instead of one. The effect scales with the available
164 * bandwidth and quickly saturates the CPU and network interface
165 * with packet generation and sending. Set to zero to disable MINMSS
166 * checking. This setting prevents us from sending too small packets.
167 */
168 int tcp_minmss = TCP_MINMSS;
169 SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
170 &tcp_minmss , 0, "Minmum TCP Maximum Segment Size");
171
172 int tcp_do_rfc1323 = 1;
173 SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
174 &tcp_do_rfc1323, 0, "Enable rfc1323 (high performance TCP) extensions");
175
176 static int tcp_log_debug = 0;
177 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
178 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
179
180 static int tcp_tcbhashsize = 0;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
182 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
183
184 static int do_tcpdrain = 1;
185 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW,
186 &do_tcpdrain, 0,
187 "Enable tcp_drain routine for extra help when low on mbufs");
188
189 SYSCTL_INT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
190 &tcbinfo.ipi_count, 0, "Number of active PCBs");
191
192 static int icmp_may_rst = 1;
193 SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
194 &icmp_may_rst, 0,
195 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
196
197 static int tcp_isn_reseed_interval = 0;
198 SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
199 &tcp_isn_reseed_interval, 0, "Seconds between reseeding of ISN secret");
200
201 /*
202 * TCP bandwidth limiting sysctls. Note that the default lower bound of
203 * 1024 exists only for debugging. A good production default would be
204 * something like 6100.
205 */
206 SYSCTL_NODE(_net_inet_tcp, OID_AUTO, inflight, CTLFLAG_RW, 0,
207 "TCP inflight data limiting");
208
209 static int tcp_inflight_enable = 0;
210 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, enable, CTLFLAG_RW,
211 &tcp_inflight_enable, 0, "Enable automatic TCP inflight data limiting");
212
213 static int tcp_inflight_debug = 0;
214 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, debug, CTLFLAG_RW,
215 &tcp_inflight_debug, 0, "Debug TCP inflight calculations");
216
217 static int tcp_inflight_rttthresh;
218 SYSCTL_PROC(_net_inet_tcp_inflight, OID_AUTO, rttthresh, CTLTYPE_INT|CTLFLAG_RW,
219 &tcp_inflight_rttthresh, 0, sysctl_msec_to_ticks, "I",
220 "RTT threshold below which inflight will deactivate itself");
221
222 static int tcp_inflight_min = 6144;
223 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, min, CTLFLAG_RW,
224 &tcp_inflight_min, 0, "Lower-bound for TCP inflight window");
225
226 static int tcp_inflight_max = TCP_MAXWIN << TCP_MAX_WINSHIFT;
227 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, max, CTLFLAG_RW,
228 &tcp_inflight_max, 0, "Upper-bound for TCP inflight window");
229
230 static int tcp_inflight_stab = 20;
231 SYSCTL_INT(_net_inet_tcp_inflight, OID_AUTO, stab, CTLFLAG_RW,
232 &tcp_inflight_stab, 0, "Inflight Algorithm Stabilization 20 = 2 packets");
233
234 uma_zone_t sack_hole_zone;
235
236 static struct inpcb *tcp_notify(struct inpcb *, int);
237 static void tcp_isn_tick(void *);
238 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
239 void *ip4hdr, const void *ip6hdr);
240
241 /*
242 * Target size of TCP PCB hash tables. Must be a power of two.
243 *
244 * Note that this can be overridden by the kernel environment
245 * variable net.inet.tcp.tcbhashsize
246 */
247 #ifndef TCBHASHSIZE
248 #define TCBHASHSIZE 512
249 #endif
250
251 /*
252 * XXX
253 * Callouts should be moved into struct tcp directly. They are currently
254 * separate because the tcpcb structure is exported to userland for sysctl
255 * parsing purposes, which do not know about callouts.
256 */
257 struct tcpcb_mem {
258 struct tcpcb tcb;
259 struct tcp_timer tt;
260 };
261
262 static uma_zone_t tcpcb_zone;
263 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
264 struct callout isn_callout;
265 static struct mtx isn_mtx;
266
267 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
268 #define ISN_LOCK() mtx_lock(&isn_mtx)
269 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
270
271 /*
272 * TCP initialization.
273 */
274 static void
275 tcp_zone_change(void *tag)
276 {
277
278 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
279 uma_zone_set_max(tcpcb_zone, maxsockets);
280 tcp_tw_zone_change();
281 }
282
283 static int
284 tcp_inpcb_init(void *mem, int size, int flags)
285 {
286 struct inpcb *inp = mem;
287
288 INP_LOCK_INIT(inp, "inp", "tcpinp");
289 return (0);
290 }
291
292 void
293 tcp_init(void)
294 {
295
296 int hashsize = TCBHASHSIZE;
297 tcp_delacktime = TCPTV_DELACK;
298 tcp_keepinit = TCPTV_KEEP_INIT;
299 tcp_keepidle = TCPTV_KEEP_IDLE;
300 tcp_keepintvl = TCPTV_KEEPINTVL;
301 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
302 tcp_msl = TCPTV_MSL;
303 tcp_rexmit_min = TCPTV_MIN;
304 if (tcp_rexmit_min < 1)
305 tcp_rexmit_min = 1;
306 tcp_rexmit_slop = TCPTV_CPU_VAR;
307 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
308 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
309
310 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
311 LIST_INIT(&tcb);
312 tcbinfo.ipi_listhead = &tcb;
313 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
314 if (!powerof2(hashsize)) {
315 printf("WARNING: TCB hash size not a power of 2\n");
316 hashsize = 512; /* safe default */
317 }
318 tcp_tcbhashsize = hashsize;
319 tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
320 &tcbinfo.ipi_hashmask);
321 tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
322 &tcbinfo.ipi_porthashmask);
323 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
324 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
325 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
326 #ifdef INET6
327 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
328 #else /* INET6 */
329 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
330 #endif /* INET6 */
331 if (max_protohdr < TCP_MINPROTOHDR)
332 max_protohdr = TCP_MINPROTOHDR;
333 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
334 panic("tcp_init");
335 #undef TCP_MINPROTOHDR
336 /*
337 * These have to be type stable for the benefit of the timers.
338 */
339 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
340 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
341 uma_zone_set_max(tcpcb_zone, maxsockets);
342 tcp_tw_init();
343 syncache_init();
344 tcp_hc_init();
345 tcp_reass_init();
346 ISN_LOCK_INIT();
347 callout_init(&isn_callout, CALLOUT_MPSAFE);
348 tcp_isn_tick(NULL);
349 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
350 SHUTDOWN_PRI_DEFAULT);
351 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
352 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
353 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
354 EVENTHANDLER_PRI_ANY);
355 }
356
357 void
358 tcp_fini(void *xtp)
359 {
360
361 callout_stop(&isn_callout);
362 }
363
364 /*
365 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
366 * tcp_template used to store this data in mbufs, but we now recopy it out
367 * of the tcpcb each time to conserve mbufs.
368 */
369 void
370 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
371 {
372 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
373
374 INP_WLOCK_ASSERT(inp);
375
376 #ifdef INET6
377 if ((inp->inp_vflag & INP_IPV6) != 0) {
378 struct ip6_hdr *ip6;
379
380 ip6 = (struct ip6_hdr *)ip_ptr;
381 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
382 (inp->inp_flow & IPV6_FLOWINFO_MASK);
383 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
384 (IPV6_VERSION & IPV6_VERSION_MASK);
385 ip6->ip6_nxt = IPPROTO_TCP;
386 ip6->ip6_plen = htons(sizeof(struct tcphdr));
387 ip6->ip6_src = inp->in6p_laddr;
388 ip6->ip6_dst = inp->in6p_faddr;
389 } else
390 #endif
391 {
392 struct ip *ip;
393
394 ip = (struct ip *)ip_ptr;
395 ip->ip_v = IPVERSION;
396 ip->ip_hl = 5;
397 ip->ip_tos = inp->inp_ip_tos;
398 ip->ip_len = 0;
399 ip->ip_id = 0;
400 ip->ip_off = 0;
401 ip->ip_ttl = inp->inp_ip_ttl;
402 ip->ip_sum = 0;
403 ip->ip_p = IPPROTO_TCP;
404 ip->ip_src = inp->inp_laddr;
405 ip->ip_dst = inp->inp_faddr;
406 }
407 th->th_sport = inp->inp_lport;
408 th->th_dport = inp->inp_fport;
409 th->th_seq = 0;
410 th->th_ack = 0;
411 th->th_x2 = 0;
412 th->th_off = 5;
413 th->th_flags = 0;
414 th->th_win = 0;
415 th->th_urp = 0;
416 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
417 }
418
419 /*
420 * Create template to be used to send tcp packets on a connection.
421 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
422 * use for this function is in keepalives, which use tcp_respond.
423 */
424 struct tcptemp *
425 tcpip_maketemplate(struct inpcb *inp)
426 {
427 struct tcptemp *t;
428
429 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
430 if (t == NULL)
431 return (NULL);
432 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
433 return (t);
434 }
435
436 /*
437 * Send a single message to the TCP at address specified by
438 * the given TCP/IP header. If m == NULL, then we make a copy
439 * of the tcpiphdr at ti and send directly to the addressed host.
440 * This is used to force keep alive messages out using the TCP
441 * template for a connection. If flags are given then we send
442 * a message back to the TCP which originated the * segment ti,
443 * and discard the mbuf containing it and any other attached mbufs.
444 *
445 * In any case the ack and sequence number of the transmitted
446 * segment are as specified by the parameters.
447 *
448 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
449 */
450 void
451 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
452 tcp_seq ack, tcp_seq seq, int flags)
453 {
454 int tlen;
455 int win = 0;
456 struct ip *ip;
457 struct tcphdr *nth;
458 #ifdef INET6
459 struct ip6_hdr *ip6;
460 int isipv6;
461 #endif /* INET6 */
462 int ipflags = 0;
463 struct inpcb *inp;
464
465 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
466
467 #ifdef INET6
468 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
469 ip6 = ipgen;
470 #endif /* INET6 */
471 ip = ipgen;
472
473 if (tp != NULL) {
474 inp = tp->t_inpcb;
475 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
476 INP_WLOCK_ASSERT(inp);
477 } else
478 inp = NULL;
479
480 if (tp != NULL) {
481 if (!(flags & TH_RST)) {
482 win = sbspace(&inp->inp_socket->so_rcv);
483 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
484 win = (long)TCP_MAXWIN << tp->rcv_scale;
485 }
486 }
487 if (m == NULL) {
488 m = m_gethdr(M_DONTWAIT, MT_DATA);
489 if (m == NULL)
490 return;
491 tlen = 0;
492 m->m_data += max_linkhdr;
493 #ifdef INET6
494 if (isipv6) {
495 bcopy((caddr_t)ip6, mtod(m, caddr_t),
496 sizeof(struct ip6_hdr));
497 ip6 = mtod(m, struct ip6_hdr *);
498 nth = (struct tcphdr *)(ip6 + 1);
499 } else
500 #endif /* INET6 */
501 {
502 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
503 ip = mtod(m, struct ip *);
504 nth = (struct tcphdr *)(ip + 1);
505 }
506 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
507 flags = TH_ACK;
508 } else {
509 /*
510 * reuse the mbuf.
511 * XXX MRT We inherrit the FIB, which is lucky.
512 */
513 m_freem(m->m_next);
514 m->m_next = NULL;
515 m->m_data = (caddr_t)ipgen;
516 /* m_len is set later */
517 tlen = 0;
518 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
519 #ifdef INET6
520 if (isipv6) {
521 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
522 nth = (struct tcphdr *)(ip6 + 1);
523 } else
524 #endif /* INET6 */
525 {
526 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
527 nth = (struct tcphdr *)(ip + 1);
528 }
529 if (th != nth) {
530 /*
531 * this is usually a case when an extension header
532 * exists between the IPv6 header and the
533 * TCP header.
534 */
535 nth->th_sport = th->th_sport;
536 nth->th_dport = th->th_dport;
537 }
538 xchg(nth->th_dport, nth->th_sport, uint16_t);
539 #undef xchg
540 }
541 #ifdef INET6
542 if (isipv6) {
543 ip6->ip6_flow = 0;
544 ip6->ip6_vfc = IPV6_VERSION;
545 ip6->ip6_nxt = IPPROTO_TCP;
546 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
547 tlen));
548 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
549 } else
550 #endif
551 {
552 tlen += sizeof (struct tcpiphdr);
553 ip->ip_len = tlen;
554 ip->ip_ttl = ip_defttl;
555 if (path_mtu_discovery)
556 ip->ip_off |= IP_DF;
557 }
558 m->m_len = tlen;
559 m->m_pkthdr.len = tlen;
560 m->m_pkthdr.rcvif = NULL;
561 #ifdef MAC
562 if (inp != NULL) {
563 /*
564 * Packet is associated with a socket, so allow the
565 * label of the response to reflect the socket label.
566 */
567 INP_WLOCK_ASSERT(inp);
568 mac_create_mbuf_from_inpcb(inp, m);
569 } else {
570 /*
571 * Packet is not associated with a socket, so possibly
572 * update the label in place.
573 */
574 mac_reflect_mbuf_tcp(m);
575 }
576 #endif
577 nth->th_seq = htonl(seq);
578 nth->th_ack = htonl(ack);
579 nth->th_x2 = 0;
580 nth->th_off = sizeof (struct tcphdr) >> 2;
581 nth->th_flags = flags;
582 if (tp != NULL)
583 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
584 else
585 nth->th_win = htons((u_short)win);
586 nth->th_urp = 0;
587 #ifdef INET6
588 if (isipv6) {
589 nth->th_sum = 0;
590 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
591 sizeof(struct ip6_hdr),
592 tlen - sizeof(struct ip6_hdr));
593 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
594 NULL, NULL);
595 } else
596 #endif /* INET6 */
597 {
598 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
599 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
600 m->m_pkthdr.csum_flags = CSUM_TCP;
601 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
602 }
603 #ifdef TCPDEBUG
604 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
605 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
606 #endif
607 #ifdef INET6
608 if (isipv6)
609 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
610 else
611 #endif /* INET6 */
612 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
613 }
614
615 /*
616 * Create a new TCP control block, making an
617 * empty reassembly queue and hooking it to the argument
618 * protocol control block. The `inp' parameter must have
619 * come from the zone allocator set up in tcp_init().
620 */
621 struct tcpcb *
622 tcp_newtcpcb(struct inpcb *inp)
623 {
624 struct tcpcb_mem *tm;
625 struct tcpcb *tp;
626 #ifdef INET6
627 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
628 #endif /* INET6 */
629
630 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
631 if (tm == NULL)
632 return (NULL);
633 tp = &tm->tcb;
634 tp->t_timers = &tm->tt;
635 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
636 tp->t_maxseg = tp->t_maxopd =
637 #ifdef INET6
638 isipv6 ? tcp_v6mssdflt :
639 #endif /* INET6 */
640 tcp_mssdflt;
641
642 /* Set up our timeouts. */
643 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
644 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
645 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
646 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
647 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
648
649 if (tcp_do_rfc1323)
650 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
651 if (tcp_do_sack)
652 tp->t_flags |= TF_SACK_PERMIT;
653 TAILQ_INIT(&tp->snd_holes);
654 tp->t_inpcb = inp; /* XXX */
655 /*
656 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
657 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
658 * reasonable initial retransmit time.
659 */
660 tp->t_srtt = TCPTV_SRTTBASE;
661 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
662 tp->t_rttmin = tcp_rexmit_min;
663 tp->t_rxtcur = TCPTV_RTOBASE;
664 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
665 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
666 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
667 tp->t_rcvtime = ticks;
668 tp->t_bw_rtttime = ticks;
669 /*
670 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
671 * because the socket may be bound to an IPv6 wildcard address,
672 * which may match an IPv4-mapped IPv6 address.
673 */
674 inp->inp_ip_ttl = ip_defttl;
675 inp->inp_ppcb = tp;
676 return (tp); /* XXX */
677 }
678
679 /*
680 * Drop a TCP connection, reporting
681 * the specified error. If connection is synchronized,
682 * then send a RST to peer.
683 */
684 struct tcpcb *
685 tcp_drop(struct tcpcb *tp, int errno)
686 {
687 struct socket *so = tp->t_inpcb->inp_socket;
688
689 INP_INFO_WLOCK_ASSERT(&tcbinfo);
690 INP_WLOCK_ASSERT(tp->t_inpcb);
691
692 if (TCPS_HAVERCVDSYN(tp->t_state)) {
693 tp->t_state = TCPS_CLOSED;
694 (void) tcp_output_reset(tp);
695 tcpstat.tcps_drops++;
696 } else
697 tcpstat.tcps_conndrops++;
698 if (errno == ETIMEDOUT && tp->t_softerror)
699 errno = tp->t_softerror;
700 so->so_error = errno;
701 return (tcp_close(tp));
702 }
703
704 void
705 tcp_discardcb(struct tcpcb *tp)
706 {
707 struct inpcb *inp = tp->t_inpcb;
708 struct socket *so = inp->inp_socket;
709 #ifdef INET6
710 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
711 #endif /* INET6 */
712
713 INP_WLOCK_ASSERT(inp);
714
715 /*
716 * Make sure that all of our timers are stopped before we
717 * delete the PCB.
718 */
719 callout_stop(&tp->t_timers->tt_rexmt);
720 callout_stop(&tp->t_timers->tt_persist);
721 callout_stop(&tp->t_timers->tt_keep);
722 callout_stop(&tp->t_timers->tt_2msl);
723 callout_stop(&tp->t_timers->tt_delack);
724
725 /*
726 * If we got enough samples through the srtt filter,
727 * save the rtt and rttvar in the routing entry.
728 * 'Enough' is arbitrarily defined as 4 rtt samples.
729 * 4 samples is enough for the srtt filter to converge
730 * to within enough % of the correct value; fewer samples
731 * and we could save a bogus rtt. The danger is not high
732 * as tcp quickly recovers from everything.
733 * XXX: Works very well but needs some more statistics!
734 */
735 if (tp->t_rttupdated >= 4) {
736 struct hc_metrics_lite metrics;
737 u_long ssthresh;
738
739 bzero(&metrics, sizeof(metrics));
740 /*
741 * Update the ssthresh always when the conditions below
742 * are satisfied. This gives us better new start value
743 * for the congestion avoidance for new connections.
744 * ssthresh is only set if packet loss occured on a session.
745 *
746 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
747 * being torn down. Ideally this code would not use 'so'.
748 */
749 ssthresh = tp->snd_ssthresh;
750 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
751 /*
752 * convert the limit from user data bytes to
753 * packets then to packet data bytes.
754 */
755 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
756 if (ssthresh < 2)
757 ssthresh = 2;
758 ssthresh *= (u_long)(tp->t_maxseg +
759 #ifdef INET6
760 (isipv6 ? sizeof (struct ip6_hdr) +
761 sizeof (struct tcphdr) :
762 #endif
763 sizeof (struct tcpiphdr)
764 #ifdef INET6
765 )
766 #endif
767 );
768 } else
769 ssthresh = 0;
770 metrics.rmx_ssthresh = ssthresh;
771
772 metrics.rmx_rtt = tp->t_srtt;
773 metrics.rmx_rttvar = tp->t_rttvar;
774 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
775 metrics.rmx_bandwidth = tp->snd_bandwidth;
776 metrics.rmx_cwnd = tp->snd_cwnd;
777 metrics.rmx_sendpipe = 0;
778 metrics.rmx_recvpipe = 0;
779
780 tcp_hc_update(&inp->inp_inc, &metrics);
781 }
782
783 /* free the reassembly queue, if any */
784 tcp_reass_flush(tp);
785 /* Disconnect offload device, if any. */
786 tcp_offload_detach(tp);
787
788 tcp_free_sackholes(tp);
789 inp->inp_ppcb = NULL;
790 tp->t_inpcb = NULL;
791 uma_zfree(tcpcb_zone, tp);
792 }
793
794 /*
795 * Attempt to close a TCP control block, marking it as dropped, and freeing
796 * the socket if we hold the only reference.
797 */
798 struct tcpcb *
799 tcp_close(struct tcpcb *tp)
800 {
801 struct inpcb *inp = tp->t_inpcb;
802 struct socket *so;
803
804 INP_INFO_WLOCK_ASSERT(&tcbinfo);
805 INP_WLOCK_ASSERT(inp);
806
807 /* Notify any offload devices of listener close */
808 if (tp->t_state == TCPS_LISTEN)
809 tcp_offload_listen_close(tp);
810 in_pcbdrop(inp);
811 tcpstat.tcps_closed++;
812 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
813 so = inp->inp_socket;
814 soisdisconnected(so);
815 if (inp->inp_flags & INP_SOCKREF) {
816 KASSERT(so->so_state & SS_PROTOREF,
817 ("tcp_close: !SS_PROTOREF"));
818 inp->inp_flags &= ~INP_SOCKREF;
819 INP_WUNLOCK(inp);
820 ACCEPT_LOCK();
821 SOCK_LOCK(so);
822 so->so_state &= ~SS_PROTOREF;
823 sofree(so);
824 return (NULL);
825 }
826 return (tp);
827 }
828
829 void
830 tcp_drain(void)
831 {
832
833 if (do_tcpdrain) {
834 struct inpcb *inpb;
835 struct tcpcb *tcpb;
836
837 /*
838 * Walk the tcpbs, if existing, and flush the reassembly queue,
839 * if there is one...
840 * XXX: The "Net/3" implementation doesn't imply that the TCP
841 * reassembly queue should be flushed, but in a situation
842 * where we're really low on mbufs, this is potentially
843 * usefull.
844 */
845 INP_INFO_RLOCK(&tcbinfo);
846 LIST_FOREACH(inpb, tcbinfo.ipi_listhead, inp_list) {
847 if (inpb->inp_flags & INP_TIMEWAIT)
848 continue;
849 INP_WLOCK(inpb);
850 if ((tcpb = intotcpcb(inpb)) != NULL) {
851 tcp_reass_flush(tcpb);
852 tcp_clean_sackreport(tcpb);
853 }
854 INP_WUNLOCK(inpb);
855 }
856 INP_INFO_RUNLOCK(&tcbinfo);
857 }
858 }
859
860 /*
861 * Notify a tcp user of an asynchronous error;
862 * store error as soft error, but wake up user
863 * (for now, won't do anything until can select for soft error).
864 *
865 * Do not wake up user since there currently is no mechanism for
866 * reporting soft errors (yet - a kqueue filter may be added).
867 */
868 static struct inpcb *
869 tcp_notify(struct inpcb *inp, int error)
870 {
871 struct tcpcb *tp;
872
873 INP_INFO_WLOCK_ASSERT(&tcbinfo);
874 INP_WLOCK_ASSERT(inp);
875
876 if ((inp->inp_flags & INP_TIMEWAIT) ||
877 (inp->inp_flags & INP_DROPPED))
878 return (inp);
879
880 tp = intotcpcb(inp);
881 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
882
883 /*
884 * Ignore some errors if we are hooked up.
885 * If connection hasn't completed, has retransmitted several times,
886 * and receives a second error, give up now. This is better
887 * than waiting a long time to establish a connection that
888 * can never complete.
889 */
890 if (tp->t_state == TCPS_ESTABLISHED &&
891 (error == EHOSTUNREACH || error == ENETUNREACH ||
892 error == EHOSTDOWN)) {
893 return (inp);
894 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
895 tp->t_softerror) {
896 tp = tcp_drop(tp, error);
897 if (tp != NULL)
898 return (inp);
899 else
900 return (NULL);
901 } else {
902 tp->t_softerror = error;
903 return (inp);
904 }
905 #if 0
906 wakeup( &so->so_timeo);
907 sorwakeup(so);
908 sowwakeup(so);
909 #endif
910 }
911
912 static int
913 tcp_pcblist(SYSCTL_HANDLER_ARGS)
914 {
915 int error, i, m, n, pcb_count;
916 struct inpcb *inp, **inp_list;
917 inp_gen_t gencnt;
918 struct xinpgen xig;
919
920 /*
921 * The process of preparing the TCB list is too time-consuming and
922 * resource-intensive to repeat twice on every request.
923 */
924 if (req->oldptr == NULL) {
925 n = tcbinfo.ipi_count + syncache_pcbcount();
926 n += imax(n / 8, 10);
927 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
928 return (0);
929 }
930
931 if (req->newptr != NULL)
932 return (EPERM);
933
934 /*
935 * OK, now we're committed to doing something.
936 */
937 INP_INFO_RLOCK(&tcbinfo);
938 gencnt = tcbinfo.ipi_gencnt;
939 n = tcbinfo.ipi_count;
940 INP_INFO_RUNLOCK(&tcbinfo);
941
942 m = syncache_pcbcount();
943
944 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
945 + (n + m) * sizeof(struct xtcpcb));
946 if (error != 0)
947 return (error);
948
949 xig.xig_len = sizeof xig;
950 xig.xig_count = n + m;
951 xig.xig_gen = gencnt;
952 xig.xig_sogen = so_gencnt;
953 error = SYSCTL_OUT(req, &xig, sizeof xig);
954 if (error)
955 return (error);
956
957 error = syncache_pcblist(req, m, &pcb_count);
958 if (error)
959 return (error);
960
961 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
962 if (inp_list == NULL)
963 return (ENOMEM);
964
965 INP_INFO_RLOCK(&tcbinfo);
966 for (inp = LIST_FIRST(tcbinfo.ipi_listhead), i = 0; inp != NULL && i
967 < n; inp = LIST_NEXT(inp, inp_list)) {
968 INP_RLOCK(inp);
969 if (inp->inp_gencnt <= gencnt) {
970 /*
971 * XXX: This use of cr_cansee(), introduced with
972 * TCP state changes, is not quite right, but for
973 * now, better than nothing.
974 */
975 if (inp->inp_flags & INP_TIMEWAIT) {
976 if (intotw(inp) != NULL)
977 error = cr_cansee(req->td->td_ucred,
978 intotw(inp)->tw_cred);
979 else
980 error = EINVAL; /* Skip this inp. */
981 } else
982 error = cr_canseeinpcb(req->td->td_ucred, inp);
983 if (error == 0)
984 inp_list[i++] = inp;
985 }
986 INP_RUNLOCK(inp);
987 }
988 INP_INFO_RUNLOCK(&tcbinfo);
989 n = i;
990
991 error = 0;
992 for (i = 0; i < n; i++) {
993 inp = inp_list[i];
994 INP_RLOCK(inp);
995 if (inp->inp_gencnt <= gencnt) {
996 struct xtcpcb xt;
997 void *inp_ppcb;
998
999 bzero(&xt, sizeof(xt));
1000 xt.xt_len = sizeof xt;
1001 /* XXX should avoid extra copy */
1002 bcopy(inp, &xt.xt_inp, sizeof *inp);
1003 inp_ppcb = inp->inp_ppcb;
1004 if (inp_ppcb == NULL)
1005 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1006 else if (inp->inp_flags & INP_TIMEWAIT) {
1007 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1008 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1009 } else
1010 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1011 if (inp->inp_socket != NULL)
1012 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1013 else {
1014 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1015 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1016 }
1017 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1018 INP_RUNLOCK(inp);
1019 error = SYSCTL_OUT(req, &xt, sizeof xt);
1020 } else
1021 INP_RUNLOCK(inp);
1022
1023 }
1024 if (!error) {
1025 /*
1026 * Give the user an updated idea of our state.
1027 * If the generation differs from what we told
1028 * her before, she knows that something happened
1029 * while we were processing this request, and it
1030 * might be necessary to retry.
1031 */
1032 INP_INFO_RLOCK(&tcbinfo);
1033 xig.xig_gen = tcbinfo.ipi_gencnt;
1034 xig.xig_sogen = so_gencnt;
1035 xig.xig_count = tcbinfo.ipi_count + pcb_count;
1036 INP_INFO_RUNLOCK(&tcbinfo);
1037 error = SYSCTL_OUT(req, &xig, sizeof xig);
1038 }
1039 free(inp_list, M_TEMP);
1040 return (error);
1041 }
1042
1043 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1044 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1045
1046 static int
1047 tcp_getcred(SYSCTL_HANDLER_ARGS)
1048 {
1049 struct xucred xuc;
1050 struct sockaddr_in addrs[2];
1051 struct inpcb *inp;
1052 int error;
1053
1054 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1055 if (error)
1056 return (error);
1057 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1058 if (error)
1059 return (error);
1060 INP_INFO_RLOCK(&tcbinfo);
1061 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1062 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1063 if (inp != NULL) {
1064 INP_RLOCK(inp);
1065 INP_INFO_RUNLOCK(&tcbinfo);
1066 if (inp->inp_socket == NULL)
1067 error = ENOENT;
1068 if (error == 0)
1069 error = cr_canseeinpcb(req->td->td_ucred, inp);
1070 if (error == 0)
1071 cru2x(inp->inp_cred, &xuc);
1072 INP_RUNLOCK(inp);
1073 } else {
1074 INP_INFO_RUNLOCK(&tcbinfo);
1075 error = ENOENT;
1076 }
1077 if (error == 0)
1078 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1079 return (error);
1080 }
1081
1082 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1083 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1084 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1085
1086 #ifdef INET6
1087 static int
1088 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1089 {
1090 struct xucred xuc;
1091 struct sockaddr_in6 addrs[2];
1092 struct inpcb *inp;
1093 int error, mapped = 0;
1094
1095 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1096 if (error)
1097 return (error);
1098 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1099 if (error)
1100 return (error);
1101 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1102 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1103 return (error);
1104 }
1105 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1106 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1107 mapped = 1;
1108 else
1109 return (EINVAL);
1110 }
1111
1112 INP_INFO_RLOCK(&tcbinfo);
1113 if (mapped == 1)
1114 inp = in_pcblookup_hash(&tcbinfo,
1115 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1116 addrs[1].sin6_port,
1117 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1118 addrs[0].sin6_port,
1119 0, NULL);
1120 else
1121 inp = in6_pcblookup_hash(&tcbinfo,
1122 &addrs[1].sin6_addr, addrs[1].sin6_port,
1123 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1124 if (inp != NULL) {
1125 INP_RLOCK(inp);
1126 INP_INFO_RUNLOCK(&tcbinfo);
1127 if (inp->inp_socket == NULL)
1128 error = ENOENT;
1129 if (error == 0)
1130 error = cr_canseeinpcb(req->td->td_ucred, inp);
1131 if (error == 0)
1132 cru2x(inp->inp_cred, &xuc);
1133 INP_RUNLOCK(inp);
1134 } else {
1135 INP_INFO_RUNLOCK(&tcbinfo);
1136 error = ENOENT;
1137 }
1138 if (error == 0)
1139 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1140 return (error);
1141 }
1142
1143 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1144 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1145 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1146 #endif
1147
1148
1149 void
1150 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1151 {
1152 struct ip *ip = vip;
1153 struct tcphdr *th;
1154 struct in_addr faddr;
1155 struct inpcb *inp;
1156 struct tcpcb *tp;
1157 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1158 struct icmp *icp;
1159 struct in_conninfo inc;
1160 tcp_seq icmp_tcp_seq;
1161 int mtu;
1162
1163 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1164 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1165 return;
1166
1167 if (cmd == PRC_MSGSIZE)
1168 notify = tcp_mtudisc;
1169 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1170 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1171 notify = tcp_drop_syn_sent;
1172 /*
1173 * Redirects don't need to be handled up here.
1174 */
1175 else if (PRC_IS_REDIRECT(cmd))
1176 return;
1177 /*
1178 * Source quench is depreciated.
1179 */
1180 else if (cmd == PRC_QUENCH)
1181 return;
1182 /*
1183 * Hostdead is ugly because it goes linearly through all PCBs.
1184 * XXX: We never get this from ICMP, otherwise it makes an
1185 * excellent DoS attack on machines with many connections.
1186 */
1187 else if (cmd == PRC_HOSTDEAD)
1188 ip = NULL;
1189 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1190 return;
1191 if (ip != NULL) {
1192 icp = (struct icmp *)((caddr_t)ip
1193 - offsetof(struct icmp, icmp_ip));
1194 th = (struct tcphdr *)((caddr_t)ip
1195 + (ip->ip_hl << 2));
1196 INP_INFO_WLOCK(&tcbinfo);
1197 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1198 ip->ip_src, th->th_sport, 0, NULL);
1199 if (inp != NULL) {
1200 INP_WLOCK(inp);
1201 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1202 !(inp->inp_flags & INP_DROPPED) &&
1203 !(inp->inp_socket == NULL)) {
1204 icmp_tcp_seq = htonl(th->th_seq);
1205 tp = intotcpcb(inp);
1206 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1207 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1208 if (cmd == PRC_MSGSIZE) {
1209 /*
1210 * MTU discovery:
1211 * If we got a needfrag set the MTU
1212 * in the route to the suggested new
1213 * value (if given) and then notify.
1214 */
1215 bzero(&inc, sizeof(inc));
1216 inc.inc_faddr = faddr;
1217 inc.inc_fibnum =
1218 inp->inp_inc.inc_fibnum;
1219
1220 mtu = ntohs(icp->icmp_nextmtu);
1221 /*
1222 * If no alternative MTU was
1223 * proposed, try the next smaller
1224 * one. ip->ip_len has already
1225 * been swapped in icmp_input().
1226 */
1227 if (!mtu)
1228 mtu = ip_next_mtu(ip->ip_len,
1229 1);
1230 if (mtu < tcp_minmss
1231 + sizeof(struct tcpiphdr))
1232 mtu = tcp_minmss
1233 + sizeof(struct tcpiphdr);
1234 /*
1235 * Only cache the the MTU if it
1236 * is smaller than the interface
1237 * or route MTU. tcp_mtudisc()
1238 * will do right thing by itself.
1239 */
1240 if (mtu <= tcp_maxmtu(&inc, NULL))
1241 tcp_hc_updatemtu(&inc, mtu);
1242 }
1243
1244 inp = (*notify)(inp, inetctlerrmap[cmd]);
1245 }
1246 }
1247 if (inp != NULL)
1248 INP_WUNLOCK(inp);
1249 } else {
1250 bzero(&inc, sizeof(inc));
1251 inc.inc_fport = th->th_dport;
1252 inc.inc_lport = th->th_sport;
1253 inc.inc_faddr = faddr;
1254 inc.inc_laddr = ip->ip_src;
1255 syncache_unreach(&inc, th);
1256 }
1257 INP_INFO_WUNLOCK(&tcbinfo);
1258 } else
1259 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1260 }
1261
1262 #ifdef INET6
1263 void
1264 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1265 {
1266 struct tcphdr th;
1267 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1268 struct ip6_hdr *ip6;
1269 struct mbuf *m;
1270 struct ip6ctlparam *ip6cp = NULL;
1271 const struct sockaddr_in6 *sa6_src = NULL;
1272 int off;
1273 struct tcp_portonly {
1274 u_int16_t th_sport;
1275 u_int16_t th_dport;
1276 } *thp;
1277
1278 if (sa->sa_family != AF_INET6 ||
1279 sa->sa_len != sizeof(struct sockaddr_in6))
1280 return;
1281
1282 if (cmd == PRC_MSGSIZE)
1283 notify = tcp_mtudisc;
1284 else if (!PRC_IS_REDIRECT(cmd) &&
1285 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1286 return;
1287 /* Source quench is depreciated. */
1288 else if (cmd == PRC_QUENCH)
1289 return;
1290
1291 /* if the parameter is from icmp6, decode it. */
1292 if (d != NULL) {
1293 ip6cp = (struct ip6ctlparam *)d;
1294 m = ip6cp->ip6c_m;
1295 ip6 = ip6cp->ip6c_ip6;
1296 off = ip6cp->ip6c_off;
1297 sa6_src = ip6cp->ip6c_src;
1298 } else {
1299 m = NULL;
1300 ip6 = NULL;
1301 off = 0; /* fool gcc */
1302 sa6_src = &sa6_any;
1303 }
1304
1305 if (ip6 != NULL) {
1306 struct in_conninfo inc;
1307 /*
1308 * XXX: We assume that when IPV6 is non NULL,
1309 * M and OFF are valid.
1310 */
1311
1312 /* check if we can safely examine src and dst ports */
1313 if (m->m_pkthdr.len < off + sizeof(*thp))
1314 return;
1315
1316 bzero(&th, sizeof(th));
1317 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1318
1319 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1320 (struct sockaddr *)ip6cp->ip6c_src,
1321 th.th_sport, cmd, NULL, notify);
1322
1323 bzero(&inc, sizeof(inc));
1324 inc.inc_fport = th.th_dport;
1325 inc.inc_lport = th.th_sport;
1326 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1327 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1328 inc.inc_flags |= INC_ISIPV6;
1329 INP_INFO_WLOCK(&tcbinfo);
1330 syncache_unreach(&inc, &th);
1331 INP_INFO_WUNLOCK(&tcbinfo);
1332 } else
1333 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1334 0, cmd, NULL, notify);
1335 }
1336 #endif /* INET6 */
1337
1338
1339 /*
1340 * Following is where TCP initial sequence number generation occurs.
1341 *
1342 * There are two places where we must use initial sequence numbers:
1343 * 1. In SYN-ACK packets.
1344 * 2. In SYN packets.
1345 *
1346 * All ISNs for SYN-ACK packets are generated by the syncache. See
1347 * tcp_syncache.c for details.
1348 *
1349 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1350 * depends on this property. In addition, these ISNs should be
1351 * unguessable so as to prevent connection hijacking. To satisfy
1352 * the requirements of this situation, the algorithm outlined in
1353 * RFC 1948 is used, with only small modifications.
1354 *
1355 * Implementation details:
1356 *
1357 * Time is based off the system timer, and is corrected so that it
1358 * increases by one megabyte per second. This allows for proper
1359 * recycling on high speed LANs while still leaving over an hour
1360 * before rollover.
1361 *
1362 * As reading the *exact* system time is too expensive to be done
1363 * whenever setting up a TCP connection, we increment the time
1364 * offset in two ways. First, a small random positive increment
1365 * is added to isn_offset for each connection that is set up.
1366 * Second, the function tcp_isn_tick fires once per clock tick
1367 * and increments isn_offset as necessary so that sequence numbers
1368 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1369 * random positive increments serve only to ensure that the same
1370 * exact sequence number is never sent out twice (as could otherwise
1371 * happen when a port is recycled in less than the system tick
1372 * interval.)
1373 *
1374 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1375 * between seeding of isn_secret. This is normally set to zero,
1376 * as reseeding should not be necessary.
1377 *
1378 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1379 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1380 * general, this means holding an exclusive (write) lock.
1381 */
1382
1383 #define ISN_BYTES_PER_SECOND 1048576
1384 #define ISN_STATIC_INCREMENT 4096
1385 #define ISN_RANDOM_INCREMENT (4096 - 1)
1386
1387 static u_char isn_secret[32];
1388 static int isn_last_reseed;
1389 static u_int32_t isn_offset, isn_offset_old;
1390 static MD5_CTX isn_ctx;
1391
1392 tcp_seq
1393 tcp_new_isn(struct tcpcb *tp)
1394 {
1395 u_int32_t md5_buffer[4];
1396 tcp_seq new_isn;
1397
1398 INP_WLOCK_ASSERT(tp->t_inpcb);
1399
1400 ISN_LOCK();
1401 /* Seed if this is the first use, reseed if requested. */
1402 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1403 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1404 < (u_int)ticks))) {
1405 read_random(&isn_secret, sizeof(isn_secret));
1406 isn_last_reseed = ticks;
1407 }
1408
1409 /* Compute the md5 hash and return the ISN. */
1410 MD5Init(&isn_ctx);
1411 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1412 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1413 #ifdef INET6
1414 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1415 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1416 sizeof(struct in6_addr));
1417 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1418 sizeof(struct in6_addr));
1419 } else
1420 #endif
1421 {
1422 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1423 sizeof(struct in_addr));
1424 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1425 sizeof(struct in_addr));
1426 }
1427 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1428 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1429 new_isn = (tcp_seq) md5_buffer[0];
1430 isn_offset += ISN_STATIC_INCREMENT +
1431 (arc4random() & ISN_RANDOM_INCREMENT);
1432 new_isn += isn_offset;
1433 ISN_UNLOCK();
1434 return (new_isn);
1435 }
1436
1437 /*
1438 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1439 * to keep time flowing at a relatively constant rate. If the random
1440 * increments have already pushed us past the projected offset, do nothing.
1441 */
1442 static void
1443 tcp_isn_tick(void *xtp)
1444 {
1445 u_int32_t projected_offset;
1446
1447 ISN_LOCK();
1448 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1449
1450 if (SEQ_GT(projected_offset, isn_offset))
1451 isn_offset = projected_offset;
1452
1453 isn_offset_old = isn_offset;
1454 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1455 ISN_UNLOCK();
1456 }
1457
1458 /*
1459 * When a specific ICMP unreachable message is received and the
1460 * connection state is SYN-SENT, drop the connection. This behavior
1461 * is controlled by the icmp_may_rst sysctl.
1462 */
1463 struct inpcb *
1464 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1465 {
1466 struct tcpcb *tp;
1467
1468 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1469 INP_WLOCK_ASSERT(inp);
1470
1471 if ((inp->inp_flags & INP_TIMEWAIT) ||
1472 (inp->inp_flags & INP_DROPPED))
1473 return (inp);
1474
1475 tp = intotcpcb(inp);
1476 if (tp->t_state != TCPS_SYN_SENT)
1477 return (inp);
1478
1479 tp = tcp_drop(tp, errno);
1480 if (tp != NULL)
1481 return (inp);
1482 else
1483 return (NULL);
1484 }
1485
1486 /*
1487 * When `need fragmentation' ICMP is received, update our idea of the MSS
1488 * based on the new value in the route. Also nudge TCP to send something,
1489 * since we know the packet we just sent was dropped.
1490 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1491 */
1492 struct inpcb *
1493 tcp_mtudisc(struct inpcb *inp, int errno)
1494 {
1495 struct tcpcb *tp;
1496 struct socket *so;
1497
1498 INP_WLOCK_ASSERT(inp);
1499 if ((inp->inp_flags & INP_TIMEWAIT) ||
1500 (inp->inp_flags & INP_DROPPED))
1501 return (inp);
1502
1503 tp = intotcpcb(inp);
1504 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1505
1506 tcp_mss_update(tp, -1, NULL, NULL);
1507
1508 so = inp->inp_socket;
1509 SOCKBUF_LOCK(&so->so_snd);
1510 /* If the mss is larger than the socket buffer, decrease the mss. */
1511 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1512 tp->t_maxseg = so->so_snd.sb_hiwat;
1513 SOCKBUF_UNLOCK(&so->so_snd);
1514
1515 tcpstat.tcps_mturesent++;
1516 tp->t_rtttime = 0;
1517 tp->snd_nxt = tp->snd_una;
1518 tcp_free_sackholes(tp);
1519 tp->snd_recover = tp->snd_max;
1520 if (tp->t_flags & TF_SACK_PERMIT)
1521 EXIT_FASTRECOVERY(tp);
1522 tcp_output_send(tp);
1523 return (inp);
1524 }
1525
1526 /*
1527 * Look-up the routing entry to the peer of this inpcb. If no route
1528 * is found and it cannot be allocated, then return 0. This routine
1529 * is called by TCP routines that access the rmx structure and by
1530 * tcp_mss_update to get the peer/interface MTU.
1531 */
1532 u_long
1533 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1534 {
1535 struct route sro;
1536 struct sockaddr_in *dst;
1537 struct ifnet *ifp;
1538 u_long maxmtu = 0;
1539
1540 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1541
1542 bzero(&sro, sizeof(sro));
1543 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1544 dst = (struct sockaddr_in *)&sro.ro_dst;
1545 dst->sin_family = AF_INET;
1546 dst->sin_len = sizeof(*dst);
1547 dst->sin_addr = inc->inc_faddr;
1548 in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1549 }
1550 if (sro.ro_rt != NULL) {
1551 ifp = sro.ro_rt->rt_ifp;
1552 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1553 maxmtu = ifp->if_mtu;
1554 else
1555 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1556
1557 /* Report additional interface capabilities. */
1558 if (flags != NULL) {
1559 if (ifp->if_capenable & IFCAP_TSO4 &&
1560 ifp->if_hwassist & CSUM_TSO)
1561 *flags |= CSUM_TSO;
1562 }
1563 RTFREE(sro.ro_rt);
1564 }
1565 return (maxmtu);
1566 }
1567
1568 #ifdef INET6
1569 u_long
1570 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
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
1593 /* Report additional interface capabilities. */
1594 if (flags != NULL) {
1595 if (ifp->if_capenable & IFCAP_TSO6 &&
1596 ifp->if_hwassist & CSUM_TSO)
1597 *flags |= CSUM_TSO;
1598 }
1599 RTFREE(sro6.ro_rt);
1600 }
1601
1602 return (maxmtu);
1603 }
1604 #endif /* INET6 */
1605
1606 #ifdef IPSEC
1607 /* compute ESP/AH header size for TCP, including outer IP header. */
1608 size_t
1609 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1610 {
1611 struct inpcb *inp;
1612 struct mbuf *m;
1613 size_t hdrsiz;
1614 struct ip *ip;
1615 #ifdef INET6
1616 struct ip6_hdr *ip6;
1617 #endif
1618 struct tcphdr *th;
1619
1620 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1621 return (0);
1622 MGETHDR(m, M_DONTWAIT, MT_DATA);
1623 if (!m)
1624 return (0);
1625
1626 #ifdef INET6
1627 if ((inp->inp_vflag & INP_IPV6) != 0) {
1628 ip6 = mtod(m, struct ip6_hdr *);
1629 th = (struct tcphdr *)(ip6 + 1);
1630 m->m_pkthdr.len = m->m_len =
1631 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1632 tcpip_fillheaders(inp, ip6, th);
1633 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1634 } else
1635 #endif /* INET6 */
1636 {
1637 ip = mtod(m, struct ip *);
1638 th = (struct tcphdr *)(ip + 1);
1639 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1640 tcpip_fillheaders(inp, ip, th);
1641 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1642 }
1643
1644 m_free(m);
1645 return (hdrsiz);
1646 }
1647 #endif /* IPSEC */
1648
1649 /*
1650 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1651 *
1652 * This code attempts to calculate the bandwidth-delay product as a
1653 * means of determining the optimal window size to maximize bandwidth,
1654 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1655 * routers. This code also does a fairly good job keeping RTTs in check
1656 * across slow links like modems. We implement an algorithm which is very
1657 * similar (but not meant to be) TCP/Vegas. The code operates on the
1658 * transmitter side of a TCP connection and so only effects the transmit
1659 * side of the connection.
1660 *
1661 * BACKGROUND: TCP makes no provision for the management of buffer space
1662 * at the end points or at the intermediate routers and switches. A TCP
1663 * stream, whether using NewReno or not, will eventually buffer as
1664 * many packets as it is able and the only reason this typically works is
1665 * due to the fairly small default buffers made available for a connection
1666 * (typicaly 16K or 32K). As machines use larger windows and/or window
1667 * scaling it is now fairly easy for even a single TCP connection to blow-out
1668 * all available buffer space not only on the local interface, but on
1669 * intermediate routers and switches as well. NewReno makes a misguided
1670 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1671 * then backing off, then steadily increasing the window again until another
1672 * failure occurs, ad-infinitum. This results in terrible oscillation that
1673 * is only made worse as network loads increase and the idea of intentionally
1674 * blowing out network buffers is, frankly, a terrible way to manage network
1675 * resources.
1676 *
1677 * It is far better to limit the transmit window prior to the failure
1678 * condition being achieved. There are two general ways to do this: First
1679 * you can 'scan' through different transmit window sizes and locate the
1680 * point where the RTT stops increasing, indicating that you have filled the
1681 * pipe, then scan backwards until you note that RTT stops decreasing, then
1682 * repeat ad-infinitum. This method works in principle but has severe
1683 * implementation issues due to RTT variances, timer granularity, and
1684 * instability in the algorithm which can lead to many false positives and
1685 * create oscillations as well as interact badly with other TCP streams
1686 * implementing the same algorithm.
1687 *
1688 * The second method is to limit the window to the bandwidth delay product
1689 * of the link. This is the method we implement. RTT variances and our
1690 * own manipulation of the congestion window, bwnd, can potentially
1691 * destabilize the algorithm. For this reason we have to stabilize the
1692 * elements used to calculate the window. We do this by using the minimum
1693 * observed RTT, the long term average of the observed bandwidth, and
1694 * by adding two segments worth of slop. It isn't perfect but it is able
1695 * to react to changing conditions and gives us a very stable basis on
1696 * which to extend the algorithm.
1697 */
1698 void
1699 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1700 {
1701 u_long bw;
1702 u_long bwnd;
1703 int save_ticks;
1704
1705 INP_WLOCK_ASSERT(tp->t_inpcb);
1706
1707 /*
1708 * If inflight_enable is disabled in the middle of a tcp connection,
1709 * make sure snd_bwnd is effectively disabled.
1710 */
1711 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
1712 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1713 tp->snd_bandwidth = 0;
1714 return;
1715 }
1716
1717 /*
1718 * Figure out the bandwidth. Due to the tick granularity this
1719 * is a very rough number and it MUST be averaged over a fairly
1720 * long period of time. XXX we need to take into account a link
1721 * that is not using all available bandwidth, but for now our
1722 * slop will ramp us up if this case occurs and the bandwidth later
1723 * increases.
1724 *
1725 * Note: if ticks rollover 'bw' may wind up negative. We must
1726 * effectively reset t_bw_rtttime for this case.
1727 */
1728 save_ticks = ticks;
1729 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1730 return;
1731
1732 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1733 (save_ticks - tp->t_bw_rtttime);
1734 tp->t_bw_rtttime = save_ticks;
1735 tp->t_bw_rtseq = ack_seq;
1736 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1737 return;
1738 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1739
1740 tp->snd_bandwidth = bw;
1741
1742 /*
1743 * Calculate the semi-static bandwidth delay product, plus two maximal
1744 * segments. The additional slop puts us squarely in the sweet
1745 * spot and also handles the bandwidth run-up case and stabilization.
1746 * Without the slop we could be locking ourselves into a lower
1747 * bandwidth.
1748 *
1749 * Situations Handled:
1750 * (1) Prevents over-queueing of packets on LANs, especially on
1751 * high speed LANs, allowing larger TCP buffers to be
1752 * specified, and also does a good job preventing
1753 * over-queueing of packets over choke points like modems
1754 * (at least for the transmit side).
1755 *
1756 * (2) Is able to handle changing network loads (bandwidth
1757 * drops so bwnd drops, bandwidth increases so bwnd
1758 * increases).
1759 *
1760 * (3) Theoretically should stabilize in the face of multiple
1761 * connections implementing the same algorithm (this may need
1762 * a little work).
1763 *
1764 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1765 * be adjusted with a sysctl but typically only needs to be
1766 * on very slow connections. A value no smaller then 5
1767 * should be used, but only reduce this default if you have
1768 * no other choice.
1769 */
1770 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1771 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
1772 #undef USERTT
1773
1774 if (tcp_inflight_debug > 0) {
1775 static int ltime;
1776 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1777 ltime = ticks;
1778 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1779 tp,
1780 bw,
1781 tp->t_rttbest,
1782 tp->t_srtt,
1783 bwnd
1784 );
1785 }
1786 }
1787 if ((long)bwnd < tcp_inflight_min)
1788 bwnd = tcp_inflight_min;
1789 if (bwnd > tcp_inflight_max)
1790 bwnd = tcp_inflight_max;
1791 if ((long)bwnd < tp->t_maxseg * 2)
1792 bwnd = tp->t_maxseg * 2;
1793 tp->snd_bwnd = bwnd;
1794 }
1795
1796 #ifdef TCP_SIGNATURE
1797 /*
1798 * Callback function invoked by m_apply() to digest TCP segment data
1799 * contained within an mbuf chain.
1800 */
1801 static int
1802 tcp_signature_apply(void *fstate, void *data, u_int len)
1803 {
1804
1805 MD5Update(fstate, (u_char *)data, len);
1806 return (0);
1807 }
1808
1809 /*
1810 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1811 *
1812 * Parameters:
1813 * m pointer to head of mbuf chain
1814 * _unused
1815 * len length of TCP segment data, excluding options
1816 * optlen length of TCP segment options
1817 * buf pointer to storage for computed MD5 digest
1818 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1819 *
1820 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1821 * When called from tcp_input(), we can be sure that th_sum has been
1822 * zeroed out and verified already.
1823 *
1824 * Return 0 if successful, otherwise return -1.
1825 *
1826 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1827 * search with the destination IP address, and a 'magic SPI' to be
1828 * determined by the application. This is hardcoded elsewhere to 1179
1829 * right now. Another branch of this code exists which uses the SPD to
1830 * specify per-application flows but it is unstable.
1831 */
1832 int
1833 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1834 u_char *buf, u_int direction)
1835 {
1836 union sockaddr_union dst;
1837 struct ippseudo ippseudo;
1838 MD5_CTX ctx;
1839 int doff;
1840 struct ip *ip;
1841 struct ipovly *ipovly;
1842 struct secasvar *sav;
1843 struct tcphdr *th;
1844 #ifdef INET6
1845 struct ip6_hdr *ip6;
1846 struct in6_addr in6;
1847 char ip6buf[INET6_ADDRSTRLEN];
1848 uint32_t plen;
1849 uint16_t nhdr;
1850 #endif
1851 u_short savecsum;
1852
1853 KASSERT(m != NULL, ("NULL mbuf chain"));
1854 KASSERT(buf != NULL, ("NULL signature pointer"));
1855
1856 /* Extract the destination from the IP header in the mbuf. */
1857 bzero(&dst, sizeof(union sockaddr_union));
1858 ip = mtod(m, struct ip *);
1859 #ifdef INET6
1860 ip6 = NULL; /* Make the compiler happy. */
1861 #endif
1862 switch (ip->ip_v) {
1863 case IPVERSION:
1864 dst.sa.sa_len = sizeof(struct sockaddr_in);
1865 dst.sa.sa_family = AF_INET;
1866 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1867 ip->ip_src : ip->ip_dst;
1868 break;
1869 #ifdef INET6
1870 case (IPV6_VERSION >> 4):
1871 ip6 = mtod(m, struct ip6_hdr *);
1872 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1873 dst.sa.sa_family = AF_INET6;
1874 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1875 ip6->ip6_src : ip6->ip6_dst;
1876 break;
1877 #endif
1878 default:
1879 return (EINVAL);
1880 /* NOTREACHED */
1881 break;
1882 }
1883
1884 /* Look up an SADB entry which matches the address of the peer. */
1885 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1886 if (sav == NULL) {
1887 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1888 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1889 #ifdef INET6
1890 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1891 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1892 #endif
1893 "(unsupported)"));
1894 return (EINVAL);
1895 }
1896
1897 MD5Init(&ctx);
1898 /*
1899 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1900 *
1901 * XXX The ippseudo header MUST be digested in network byte order,
1902 * or else we'll fail the regression test. Assume all fields we've
1903 * been doing arithmetic on have been in host byte order.
1904 * XXX One cannot depend on ipovly->ih_len here. When called from
1905 * tcp_output(), the underlying ip_len member has not yet been set.
1906 */
1907 switch (ip->ip_v) {
1908 case IPVERSION:
1909 ipovly = (struct ipovly *)ip;
1910 ippseudo.ippseudo_src = ipovly->ih_src;
1911 ippseudo.ippseudo_dst = ipovly->ih_dst;
1912 ippseudo.ippseudo_pad = 0;
1913 ippseudo.ippseudo_p = IPPROTO_TCP;
1914 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1915 optlen);
1916 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1917
1918 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1919 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1920 break;
1921 #ifdef INET6
1922 /*
1923 * RFC 2385, 2.0 Proposal
1924 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1925 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1926 * extended next header value (to form 32 bits), and 32-bit segment
1927 * length.
1928 * Note: Upper-Layer Packet Length comes before Next Header.
1929 */
1930 case (IPV6_VERSION >> 4):
1931 in6 = ip6->ip6_src;
1932 in6_clearscope(&in6);
1933 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1934 in6 = ip6->ip6_dst;
1935 in6_clearscope(&in6);
1936 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1937 plen = htonl(len + sizeof(struct tcphdr) + optlen);
1938 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1939 nhdr = 0;
1940 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1941 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1942 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1943 nhdr = IPPROTO_TCP;
1944 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1945
1946 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1947 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1948 break;
1949 #endif
1950 default:
1951 return (EINVAL);
1952 /* NOTREACHED */
1953 break;
1954 }
1955
1956
1957 /*
1958 * Step 2: Update MD5 hash with TCP header, excluding options.
1959 * The TCP checksum must be set to zero.
1960 */
1961 savecsum = th->th_sum;
1962 th->th_sum = 0;
1963 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1964 th->th_sum = savecsum;
1965
1966 /*
1967 * Step 3: Update MD5 hash with TCP segment data.
1968 * Use m_apply() to avoid an early m_pullup().
1969 */
1970 if (len > 0)
1971 m_apply(m, doff, len, tcp_signature_apply, &ctx);
1972
1973 /*
1974 * Step 4: Update MD5 hash with shared secret.
1975 */
1976 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1977 MD5Final(buf, &ctx);
1978
1979 key_sa_recordxfer(sav, m);
1980 KEY_FREESAV(&sav);
1981 return (0);
1982 }
1983 #endif /* TCP_SIGNATURE */
1984
1985 static int
1986 sysctl_drop(SYSCTL_HANDLER_ARGS)
1987 {
1988 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
1989 struct sockaddr_storage addrs[2];
1990 struct inpcb *inp;
1991 struct tcpcb *tp;
1992 struct tcptw *tw;
1993 struct sockaddr_in *fin, *lin;
1994 #ifdef INET6
1995 struct sockaddr_in6 *fin6, *lin6;
1996 struct in6_addr f6, l6;
1997 #endif
1998 int error;
1999
2000 inp = NULL;
2001 fin = lin = NULL;
2002 #ifdef INET6
2003 fin6 = lin6 = NULL;
2004 #endif
2005 error = 0;
2006
2007 if (req->oldptr != NULL || req->oldlen != 0)
2008 return (EINVAL);
2009 if (req->newptr == NULL)
2010 return (EPERM);
2011 if (req->newlen < sizeof(addrs))
2012 return (ENOMEM);
2013 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2014 if (error)
2015 return (error);
2016
2017 switch (addrs[0].ss_family) {
2018 #ifdef INET6
2019 case AF_INET6:
2020 fin6 = (struct sockaddr_in6 *)&addrs[0];
2021 lin6 = (struct sockaddr_in6 *)&addrs[1];
2022 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2023 lin6->sin6_len != sizeof(struct sockaddr_in6))
2024 return (EINVAL);
2025 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2026 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2027 return (EINVAL);
2028 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2029 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2030 fin = (struct sockaddr_in *)&addrs[0];
2031 lin = (struct sockaddr_in *)&addrs[1];
2032 break;
2033 }
2034 error = sa6_embedscope(fin6, ip6_use_defzone);
2035 if (error)
2036 return (error);
2037 error = sa6_embedscope(lin6, ip6_use_defzone);
2038 if (error)
2039 return (error);
2040 break;
2041 #endif
2042 case AF_INET:
2043 fin = (struct sockaddr_in *)&addrs[0];
2044 lin = (struct sockaddr_in *)&addrs[1];
2045 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2046 lin->sin_len != sizeof(struct sockaddr_in))
2047 return (EINVAL);
2048 break;
2049 default:
2050 return (EINVAL);
2051 }
2052 INP_INFO_WLOCK(&tcbinfo);
2053 switch (addrs[0].ss_family) {
2054 #ifdef INET6
2055 case AF_INET6:
2056 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2057 &l6, lin6->sin6_port, 0, NULL);
2058 break;
2059 #endif
2060 case AF_INET:
2061 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2062 lin->sin_addr, lin->sin_port, 0, NULL);
2063 break;
2064 }
2065 if (inp != NULL) {
2066 INP_WLOCK(inp);
2067 if (inp->inp_flags & INP_TIMEWAIT) {
2068 /*
2069 * XXXRW: There currently exists a state where an
2070 * inpcb is present, but its timewait state has been
2071 * discarded. For now, don't allow dropping of this
2072 * type of inpcb.
2073 */
2074 tw = intotw(inp);
2075 if (tw != NULL)
2076 tcp_twclose(tw, 0);
2077 else
2078 INP_WUNLOCK(inp);
2079 } else if (!(inp->inp_flags & INP_DROPPED) &&
2080 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2081 tp = intotcpcb(inp);
2082 tp = tcp_drop(tp, ECONNABORTED);
2083 if (tp != NULL)
2084 INP_WUNLOCK(inp);
2085 } else
2086 INP_WUNLOCK(inp);
2087 } else
2088 error = ESRCH;
2089 INP_INFO_WUNLOCK(&tcbinfo);
2090 return (error);
2091 }
2092
2093 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2094 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2095 0, sysctl_drop, "", "Drop TCP connection");
2096
2097 /*
2098 * Generate a standardized TCP log line for use throughout the
2099 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2100 * allow use in the interrupt context.
2101 *
2102 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2103 * NB: The function may return NULL if memory allocation failed.
2104 *
2105 * Due to header inclusion and ordering limitations the struct ip
2106 * and ip6_hdr pointers have to be passed as void pointers.
2107 */
2108 char *
2109 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2110 const void *ip6hdr)
2111 {
2112
2113 /* Is logging enabled? */
2114 if (tcp_log_in_vain == 0)
2115 return (NULL);
2116
2117 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2118 }
2119
2120 char *
2121 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2122 const void *ip6hdr)
2123 {
2124
2125 /* Is logging enabled? */
2126 if (tcp_log_debug == 0)
2127 return (NULL);
2128
2129 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2130 }
2131
2132 static char *
2133 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2134 const void *ip6hdr)
2135 {
2136 char *s, *sp;
2137 size_t size;
2138 struct ip *ip;
2139 #ifdef INET6
2140 const struct ip6_hdr *ip6;
2141
2142 ip6 = (const struct ip6_hdr *)ip6hdr;
2143 #endif /* INET6 */
2144 ip = (struct ip *)ip4hdr;
2145
2146 /*
2147 * The log line looks like this:
2148 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2149 */
2150 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2151 sizeof(PRINT_TH_FLAGS) + 1 +
2152 #ifdef INET6
2153 2 * INET6_ADDRSTRLEN;
2154 #else
2155 2 * INET_ADDRSTRLEN;
2156 #endif /* INET6 */
2157
2158 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2159 if (s == NULL)
2160 return (NULL);
2161
2162 strcat(s, "TCP: [");
2163 sp = s + strlen(s);
2164
2165 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2166 inet_ntoa_r(inc->inc_faddr, sp);
2167 sp = s + strlen(s);
2168 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2169 sp = s + strlen(s);
2170 inet_ntoa_r(inc->inc_laddr, sp);
2171 sp = s + strlen(s);
2172 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2173 #ifdef INET6
2174 } else if (inc) {
2175 ip6_sprintf(sp, &inc->inc6_faddr);
2176 sp = s + strlen(s);
2177 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2178 sp = s + strlen(s);
2179 ip6_sprintf(sp, &inc->inc6_laddr);
2180 sp = s + strlen(s);
2181 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2182 } else if (ip6 && th) {
2183 ip6_sprintf(sp, &ip6->ip6_src);
2184 sp = s + strlen(s);
2185 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2186 sp = s + strlen(s);
2187 ip6_sprintf(sp, &ip6->ip6_dst);
2188 sp = s + strlen(s);
2189 sprintf(sp, "]:%i", ntohs(th->th_dport));
2190 #endif /* INET6 */
2191 } else if (ip && th) {
2192 inet_ntoa_r(ip->ip_src, sp);
2193 sp = s + strlen(s);
2194 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2195 sp = s + strlen(s);
2196 inet_ntoa_r(ip->ip_dst, sp);
2197 sp = s + strlen(s);
2198 sprintf(sp, "]:%i", ntohs(th->th_dport));
2199 } else {
2200 free(s, M_TCPLOG);
2201 return (NULL);
2202 }
2203 sp = s + strlen(s);
2204 if (th)
2205 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2206 if (*(s + size - 1) != '\0')
2207 panic("%s: string too long", __func__);
2208 return (s);
2209 }
Cache object: 4a0b78745848fdc3ce24f4b8dc2ffd3f
|