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/7.3/sys/netinet/tcp_subr.c 201657 2010-01-06 16:56:31Z bz $");
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 = 1;
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
239 /*
240 * Target size of TCP PCB hash tables. Must be a power of two.
241 *
242 * Note that this can be overridden by the kernel environment
243 * variable net.inet.tcp.tcbhashsize
244 */
245 #ifndef TCBHASHSIZE
246 #define TCBHASHSIZE 512
247 #endif
248
249 /*
250 * XXX
251 * Callouts should be moved into struct tcp directly. They are currently
252 * separate because the tcpcb structure is exported to userland for sysctl
253 * parsing purposes, which do not know about callouts.
254 */
255 struct tcpcb_mem {
256 struct tcpcb tcb;
257 struct tcp_timer tt;
258 };
259
260 static uma_zone_t tcpcb_zone;
261 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
262 struct callout isn_callout;
263 static struct mtx isn_mtx;
264
265 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
266 #define ISN_LOCK() mtx_lock(&isn_mtx)
267 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
268
269 /*
270 * TCP initialization.
271 */
272 static void
273 tcp_zone_change(void *tag)
274 {
275
276 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
277 uma_zone_set_max(tcpcb_zone, maxsockets);
278 tcp_tw_zone_change();
279 }
280
281 static int
282 tcp_inpcb_init(void *mem, int size, int flags)
283 {
284 struct inpcb *inp = mem;
285
286 INP_LOCK_INIT(inp, "inp", "tcpinp");
287 return (0);
288 }
289
290 void
291 tcp_init(void)
292 {
293
294 int hashsize = TCBHASHSIZE;
295 tcp_delacktime = TCPTV_DELACK;
296 tcp_keepinit = TCPTV_KEEP_INIT;
297 tcp_keepidle = TCPTV_KEEP_IDLE;
298 tcp_keepintvl = TCPTV_KEEPINTVL;
299 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
300 tcp_msl = TCPTV_MSL;
301 tcp_rexmit_min = TCPTV_MIN;
302 if (tcp_rexmit_min < 1)
303 tcp_rexmit_min = 1;
304 tcp_rexmit_slop = TCPTV_CPU_VAR;
305 tcp_inflight_rttthresh = TCPTV_INFLIGHT_RTTTHRESH;
306 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
307
308 INP_INFO_LOCK_INIT(&tcbinfo, "tcp");
309 LIST_INIT(&tcb);
310 tcbinfo.ipi_listhead = &tcb;
311 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
312 if (!powerof2(hashsize)) {
313 printf("WARNING: TCB hash size not a power of 2\n");
314 hashsize = 512; /* safe default */
315 }
316 tcp_tcbhashsize = hashsize;
317 tcbinfo.ipi_hashbase = hashinit(hashsize, M_PCB,
318 &tcbinfo.ipi_hashmask);
319 tcbinfo.ipi_porthashbase = hashinit(hashsize, M_PCB,
320 &tcbinfo.ipi_porthashmask);
321 tcbinfo.ipi_zone = uma_zcreate("inpcb", sizeof(struct inpcb),
322 NULL, NULL, tcp_inpcb_init, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
323 uma_zone_set_max(tcbinfo.ipi_zone, maxsockets);
324 #ifdef INET6
325 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
326 #else /* INET6 */
327 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
328 #endif /* INET6 */
329 if (max_protohdr < TCP_MINPROTOHDR)
330 max_protohdr = TCP_MINPROTOHDR;
331 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
332 panic("tcp_init");
333 #undef TCP_MINPROTOHDR
334 /*
335 * These have to be type stable for the benefit of the timers.
336 */
337 tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
338 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
339 uma_zone_set_max(tcpcb_zone, maxsockets);
340 tcp_tw_init();
341 syncache_init();
342 tcp_hc_init();
343 tcp_reass_init();
344 ISN_LOCK_INIT();
345 callout_init(&isn_callout, CALLOUT_MPSAFE);
346 tcp_isn_tick(NULL);
347 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
348 SHUTDOWN_PRI_DEFAULT);
349 sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
350 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
351 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
352 EVENTHANDLER_PRI_ANY);
353 }
354
355 void
356 tcp_fini(void *xtp)
357 {
358
359 callout_stop(&isn_callout);
360 }
361
362 /*
363 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
364 * tcp_template used to store this data in mbufs, but we now recopy it out
365 * of the tcpcb each time to conserve mbufs.
366 */
367 void
368 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
369 {
370 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
371
372 INP_WLOCK_ASSERT(inp);
373
374 #ifdef INET6
375 if ((inp->inp_vflag & INP_IPV6) != 0) {
376 struct ip6_hdr *ip6;
377
378 ip6 = (struct ip6_hdr *)ip_ptr;
379 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
380 (inp->inp_flow & IPV6_FLOWINFO_MASK);
381 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
382 (IPV6_VERSION & IPV6_VERSION_MASK);
383 ip6->ip6_nxt = IPPROTO_TCP;
384 ip6->ip6_plen = htons(sizeof(struct tcphdr));
385 ip6->ip6_src = inp->in6p_laddr;
386 ip6->ip6_dst = inp->in6p_faddr;
387 } else
388 #endif
389 {
390 struct ip *ip;
391
392 ip = (struct ip *)ip_ptr;
393 ip->ip_v = IPVERSION;
394 ip->ip_hl = 5;
395 ip->ip_tos = inp->inp_ip_tos;
396 ip->ip_len = 0;
397 ip->ip_id = 0;
398 ip->ip_off = 0;
399 ip->ip_ttl = inp->inp_ip_ttl;
400 ip->ip_sum = 0;
401 ip->ip_p = IPPROTO_TCP;
402 ip->ip_src = inp->inp_laddr;
403 ip->ip_dst = inp->inp_faddr;
404 }
405 th->th_sport = inp->inp_lport;
406 th->th_dport = inp->inp_fport;
407 th->th_seq = 0;
408 th->th_ack = 0;
409 th->th_x2 = 0;
410 th->th_off = 5;
411 th->th_flags = 0;
412 th->th_win = 0;
413 th->th_urp = 0;
414 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
415 }
416
417 /*
418 * Create template to be used to send tcp packets on a connection.
419 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
420 * use for this function is in keepalives, which use tcp_respond.
421 */
422 struct tcptemp *
423 tcpip_maketemplate(struct inpcb *inp)
424 {
425 struct tcptemp *t;
426
427 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
428 if (t == NULL)
429 return (NULL);
430 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
431 return (t);
432 }
433
434 /*
435 * Send a single message to the TCP at address specified by
436 * the given TCP/IP header. If m == NULL, then we make a copy
437 * of the tcpiphdr at ti and send directly to the addressed host.
438 * This is used to force keep alive messages out using the TCP
439 * template for a connection. If flags are given then we send
440 * a message back to the TCP which originated the * segment ti,
441 * and discard the mbuf containing it and any other attached mbufs.
442 *
443 * In any case the ack and sequence number of the transmitted
444 * segment are as specified by the parameters.
445 *
446 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
447 */
448 void
449 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
450 tcp_seq ack, tcp_seq seq, int flags)
451 {
452 int tlen;
453 int win = 0;
454 struct ip *ip;
455 struct tcphdr *nth;
456 #ifdef INET6
457 struct ip6_hdr *ip6;
458 int isipv6;
459 #endif /* INET6 */
460 int ipflags = 0;
461 struct inpcb *inp;
462
463 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
464
465 #ifdef INET6
466 isipv6 = ((struct ip *)ipgen)->ip_v == 6;
467 ip6 = ipgen;
468 #endif /* INET6 */
469 ip = ipgen;
470
471 if (tp != NULL) {
472 inp = tp->t_inpcb;
473 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
474 INP_WLOCK_ASSERT(inp);
475 } else
476 inp = NULL;
477
478 if (tp != NULL) {
479 if (!(flags & TH_RST)) {
480 win = sbspace(&inp->inp_socket->so_rcv);
481 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
482 win = (long)TCP_MAXWIN << tp->rcv_scale;
483 }
484 }
485 if (m == NULL) {
486 m = m_gethdr(M_DONTWAIT, MT_DATA);
487 if (m == NULL)
488 return;
489 tlen = 0;
490 m->m_data += max_linkhdr;
491 #ifdef INET6
492 if (isipv6) {
493 bcopy((caddr_t)ip6, mtod(m, caddr_t),
494 sizeof(struct ip6_hdr));
495 ip6 = mtod(m, struct ip6_hdr *);
496 nth = (struct tcphdr *)(ip6 + 1);
497 } else
498 #endif /* INET6 */
499 {
500 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
501 ip = mtod(m, struct ip *);
502 nth = (struct tcphdr *)(ip + 1);
503 }
504 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
505 flags = TH_ACK;
506 } else {
507 /*
508 * reuse the mbuf.
509 * XXX MRT We inherrit the FIB, which is lucky.
510 */
511 m_freem(m->m_next);
512 m->m_next = NULL;
513 m->m_data = (caddr_t)ipgen;
514 /* m_len is set later */
515 tlen = 0;
516 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
517 #ifdef INET6
518 if (isipv6) {
519 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
520 nth = (struct tcphdr *)(ip6 + 1);
521 } else
522 #endif /* INET6 */
523 {
524 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
525 nth = (struct tcphdr *)(ip + 1);
526 }
527 if (th != nth) {
528 /*
529 * this is usually a case when an extension header
530 * exists between the IPv6 header and the
531 * TCP header.
532 */
533 nth->th_sport = th->th_sport;
534 nth->th_dport = th->th_dport;
535 }
536 xchg(nth->th_dport, nth->th_sport, uint16_t);
537 #undef xchg
538 }
539 #ifdef INET6
540 if (isipv6) {
541 ip6->ip6_flow = 0;
542 ip6->ip6_vfc = IPV6_VERSION;
543 ip6->ip6_nxt = IPPROTO_TCP;
544 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
545 tlen));
546 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
547 } else
548 #endif
549 {
550 tlen += sizeof (struct tcpiphdr);
551 ip->ip_len = tlen;
552 ip->ip_ttl = ip_defttl;
553 if (path_mtu_discovery)
554 ip->ip_off |= IP_DF;
555 }
556 m->m_len = tlen;
557 m->m_pkthdr.len = tlen;
558 m->m_pkthdr.rcvif = NULL;
559 #ifdef MAC
560 if (inp != NULL) {
561 /*
562 * Packet is associated with a socket, so allow the
563 * label of the response to reflect the socket label.
564 */
565 INP_WLOCK_ASSERT(inp);
566 mac_create_mbuf_from_inpcb(inp, m);
567 } else {
568 /*
569 * Packet is not associated with a socket, so possibly
570 * update the label in place.
571 */
572 mac_reflect_mbuf_tcp(m);
573 }
574 #endif
575 nth->th_seq = htonl(seq);
576 nth->th_ack = htonl(ack);
577 nth->th_x2 = 0;
578 nth->th_off = sizeof (struct tcphdr) >> 2;
579 nth->th_flags = flags;
580 if (tp != NULL)
581 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
582 else
583 nth->th_win = htons((u_short)win);
584 nth->th_urp = 0;
585 #ifdef INET6
586 if (isipv6) {
587 nth->th_sum = 0;
588 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
589 sizeof(struct ip6_hdr),
590 tlen - sizeof(struct ip6_hdr));
591 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
592 NULL, NULL);
593 } else
594 #endif /* INET6 */
595 {
596 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
597 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
598 m->m_pkthdr.csum_flags = CSUM_TCP;
599 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
600 }
601 #ifdef TCPDEBUG
602 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
603 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
604 #endif
605 #ifdef INET6
606 if (isipv6)
607 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
608 else
609 #endif /* INET6 */
610 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
611 }
612
613 /*
614 * Create a new TCP control block, making an
615 * empty reassembly queue and hooking it to the argument
616 * protocol control block. The `inp' parameter must have
617 * come from the zone allocator set up in tcp_init().
618 */
619 struct tcpcb *
620 tcp_newtcpcb(struct inpcb *inp)
621 {
622 struct tcpcb_mem *tm;
623 struct tcpcb *tp;
624 #ifdef INET6
625 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
626 #endif /* INET6 */
627
628 tm = uma_zalloc(tcpcb_zone, M_NOWAIT | M_ZERO);
629 if (tm == NULL)
630 return (NULL);
631 tp = &tm->tcb;
632 tp->t_timers = &tm->tt;
633 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
634 tp->t_maxseg = tp->t_maxopd =
635 #ifdef INET6
636 isipv6 ? tcp_v6mssdflt :
637 #endif /* INET6 */
638 tcp_mssdflt;
639
640 /* Set up our timeouts. */
641 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
642 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
643 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
644 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
645 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
646
647 if (tcp_do_rfc1323)
648 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
649 if (tcp_do_sack)
650 tp->t_flags |= TF_SACK_PERMIT;
651 TAILQ_INIT(&tp->snd_holes);
652 tp->t_inpcb = inp; /* XXX */
653 /*
654 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
655 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
656 * reasonable initial retransmit time.
657 */
658 tp->t_srtt = TCPTV_SRTTBASE;
659 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
660 tp->t_rttmin = tcp_rexmit_min;
661 tp->t_rxtcur = TCPTV_RTOBASE;
662 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
663 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
664 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
665 tp->t_rcvtime = ticks;
666 tp->t_bw_rtttime = ticks;
667 /*
668 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
669 * because the socket may be bound to an IPv6 wildcard address,
670 * which may match an IPv4-mapped IPv6 address.
671 */
672 inp->inp_ip_ttl = ip_defttl;
673 inp->inp_ppcb = tp;
674 return (tp); /* XXX */
675 }
676
677 /*
678 * Drop a TCP connection, reporting
679 * the specified error. If connection is synchronized,
680 * then send a RST to peer.
681 */
682 struct tcpcb *
683 tcp_drop(struct tcpcb *tp, int errno)
684 {
685 struct socket *so = tp->t_inpcb->inp_socket;
686
687 INP_INFO_WLOCK_ASSERT(&tcbinfo);
688 INP_WLOCK_ASSERT(tp->t_inpcb);
689
690 if (TCPS_HAVERCVDSYN(tp->t_state)) {
691 tp->t_state = TCPS_CLOSED;
692 (void) tcp_output_reset(tp);
693 tcpstat.tcps_drops++;
694 } else
695 tcpstat.tcps_conndrops++;
696 if (errno == ETIMEDOUT && tp->t_softerror)
697 errno = tp->t_softerror;
698 so->so_error = errno;
699 return (tcp_close(tp));
700 }
701
702 void
703 tcp_discardcb(struct tcpcb *tp)
704 {
705 struct tseg_qent *q;
706 struct inpcb *inp = tp->t_inpcb;
707 struct socket *so = inp->inp_socket;
708 #ifdef INET6
709 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
710 #endif /* INET6 */
711
712 INP_WLOCK_ASSERT(inp);
713
714 /*
715 * Make sure that all of our timers are stopped before we
716 * delete the PCB.
717 */
718 callout_stop(&tp->t_timers->tt_rexmt);
719 callout_stop(&tp->t_timers->tt_persist);
720 callout_stop(&tp->t_timers->tt_keep);
721 callout_stop(&tp->t_timers->tt_2msl);
722 callout_stop(&tp->t_timers->tt_delack);
723
724 /*
725 * If we got enough samples through the srtt filter,
726 * save the rtt and rttvar in the routing entry.
727 * 'Enough' is arbitrarily defined as 4 rtt samples.
728 * 4 samples is enough for the srtt filter to converge
729 * to within enough % of the correct value; fewer samples
730 * and we could save a bogus rtt. The danger is not high
731 * as tcp quickly recovers from everything.
732 * XXX: Works very well but needs some more statistics!
733 */
734 if (tp->t_rttupdated >= 4) {
735 struct hc_metrics_lite metrics;
736 u_long ssthresh;
737
738 bzero(&metrics, sizeof(metrics));
739 /*
740 * Update the ssthresh always when the conditions below
741 * are satisfied. This gives us better new start value
742 * for the congestion avoidance for new connections.
743 * ssthresh is only set if packet loss occured on a session.
744 *
745 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
746 * being torn down. Ideally this code would not use 'so'.
747 */
748 ssthresh = tp->snd_ssthresh;
749 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
750 /*
751 * convert the limit from user data bytes to
752 * packets then to packet data bytes.
753 */
754 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
755 if (ssthresh < 2)
756 ssthresh = 2;
757 ssthresh *= (u_long)(tp->t_maxseg +
758 #ifdef INET6
759 (isipv6 ? sizeof (struct ip6_hdr) +
760 sizeof (struct tcphdr) :
761 #endif
762 sizeof (struct tcpiphdr)
763 #ifdef INET6
764 )
765 #endif
766 );
767 } else
768 ssthresh = 0;
769 metrics.rmx_ssthresh = ssthresh;
770
771 metrics.rmx_rtt = tp->t_srtt;
772 metrics.rmx_rttvar = tp->t_rttvar;
773 /* XXX: This wraps if the pipe is more than 4 Gbit per second */
774 metrics.rmx_bandwidth = tp->snd_bandwidth;
775 metrics.rmx_cwnd = tp->snd_cwnd;
776 metrics.rmx_sendpipe = 0;
777 metrics.rmx_recvpipe = 0;
778
779 tcp_hc_update(&inp->inp_inc, &metrics);
780 }
781
782 /* free the reassembly queue, if any */
783 while ((q = LIST_FIRST(&tp->t_segq)) != NULL) {
784 LIST_REMOVE(q, tqe_q);
785 m_freem(q->tqe_m);
786 uma_zfree(tcp_reass_zone, q);
787 tp->t_segqlen--;
788 tcp_reass_qsize--;
789 }
790 /* Disconnect offload device, if any. */
791 tcp_offload_detach(tp);
792
793 tcp_free_sackholes(tp);
794 inp->inp_ppcb = NULL;
795 tp->t_inpcb = NULL;
796 uma_zfree(tcpcb_zone, tp);
797 }
798
799 /*
800 * Attempt to close a TCP control block, marking it as dropped, and freeing
801 * the socket if we hold the only reference.
802 */
803 struct tcpcb *
804 tcp_close(struct tcpcb *tp)
805 {
806 struct inpcb *inp = tp->t_inpcb;
807 struct socket *so;
808
809 INP_INFO_WLOCK_ASSERT(&tcbinfo);
810 INP_WLOCK_ASSERT(inp);
811
812 /* Notify any offload devices of listener close */
813 if (tp->t_state == TCPS_LISTEN)
814 tcp_offload_listen_close(tp);
815 in_pcbdrop(inp);
816 tcpstat.tcps_closed++;
817 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
818 so = inp->inp_socket;
819 soisdisconnected(so);
820 if (inp->inp_flags & INP_SOCKREF) {
821 KASSERT(so->so_state & SS_PROTOREF,
822 ("tcp_close: !SS_PROTOREF"));
823 inp->inp_flags &= ~INP_SOCKREF;
824 INP_WUNLOCK(inp);
825 ACCEPT_LOCK();
826 SOCK_LOCK(so);
827 so->so_state &= ~SS_PROTOREF;
828 sofree(so);
829 return (NULL);
830 }
831 return (tp);
832 }
833
834 void
835 tcp_drain(void)
836 {
837
838 if (do_tcpdrain) {
839 struct inpcb *inpb;
840 struct tcpcb *tcpb;
841 struct tseg_qent *te;
842
843 /*
844 * Walk the tcpbs, if existing, and flush the reassembly queue,
845 * if there is one...
846 * XXX: The "Net/3" implementation doesn't imply that the TCP
847 * reassembly queue should be flushed, but in a situation
848 * where we're really low on mbufs, this is potentially
849 * usefull.
850 */
851 INP_INFO_RLOCK(&tcbinfo);
852 LIST_FOREACH(inpb, tcbinfo.ipi_listhead, inp_list) {
853 if (inpb->inp_flags & INP_TIMEWAIT)
854 continue;
855 INP_WLOCK(inpb);
856 if ((tcpb = intotcpcb(inpb)) != NULL) {
857 while ((te = LIST_FIRST(&tcpb->t_segq))
858 != NULL) {
859 LIST_REMOVE(te, tqe_q);
860 m_freem(te->tqe_m);
861 uma_zfree(tcp_reass_zone, te);
862 tcpb->t_segqlen--;
863 tcp_reass_qsize--;
864 }
865 tcp_clean_sackreport(tcpb);
866 }
867 INP_WUNLOCK(inpb);
868 }
869 INP_INFO_RUNLOCK(&tcbinfo);
870 }
871 }
872
873 /*
874 * Notify a tcp user of an asynchronous error;
875 * store error as soft error, but wake up user
876 * (for now, won't do anything until can select for soft error).
877 *
878 * Do not wake up user since there currently is no mechanism for
879 * reporting soft errors (yet - a kqueue filter may be added).
880 */
881 static struct inpcb *
882 tcp_notify(struct inpcb *inp, int error)
883 {
884 struct tcpcb *tp;
885
886 INP_INFO_WLOCK_ASSERT(&tcbinfo);
887 INP_WLOCK_ASSERT(inp);
888
889 if ((inp->inp_flags & INP_TIMEWAIT) ||
890 (inp->inp_flags & INP_DROPPED))
891 return (inp);
892
893 tp = intotcpcb(inp);
894 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
895
896 /*
897 * Ignore some errors if we are hooked up.
898 * If connection hasn't completed, has retransmitted several times,
899 * and receives a second error, give up now. This is better
900 * than waiting a long time to establish a connection that
901 * can never complete.
902 */
903 if (tp->t_state == TCPS_ESTABLISHED &&
904 (error == EHOSTUNREACH || error == ENETUNREACH ||
905 error == EHOSTDOWN)) {
906 return (inp);
907 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
908 tp->t_softerror) {
909 tp = tcp_drop(tp, error);
910 if (tp != NULL)
911 return (inp);
912 else
913 return (NULL);
914 } else {
915 tp->t_softerror = error;
916 return (inp);
917 }
918 #if 0
919 wakeup( &so->so_timeo);
920 sorwakeup(so);
921 sowwakeup(so);
922 #endif
923 }
924
925 static int
926 tcp_pcblist(SYSCTL_HANDLER_ARGS)
927 {
928 int error, i, m, n, pcb_count;
929 struct inpcb *inp, **inp_list;
930 inp_gen_t gencnt;
931 struct xinpgen xig;
932
933 /*
934 * The process of preparing the TCB list is too time-consuming and
935 * resource-intensive to repeat twice on every request.
936 */
937 if (req->oldptr == NULL) {
938 m = syncache_pcbcount();
939 n = tcbinfo.ipi_count;
940 req->oldidx = 2 * (sizeof xig)
941 + ((m + n) + n/8) * sizeof(struct xtcpcb);
942 return (0);
943 }
944
945 if (req->newptr != NULL)
946 return (EPERM);
947
948 /*
949 * OK, now we're committed to doing something.
950 */
951 INP_INFO_RLOCK(&tcbinfo);
952 gencnt = tcbinfo.ipi_gencnt;
953 n = tcbinfo.ipi_count;
954 INP_INFO_RUNLOCK(&tcbinfo);
955
956 m = syncache_pcbcount();
957
958 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
959 + (n + m) * sizeof(struct xtcpcb));
960 if (error != 0)
961 return (error);
962
963 xig.xig_len = sizeof xig;
964 xig.xig_count = n + m;
965 xig.xig_gen = gencnt;
966 xig.xig_sogen = so_gencnt;
967 error = SYSCTL_OUT(req, &xig, sizeof xig);
968 if (error)
969 return (error);
970
971 error = syncache_pcblist(req, m, &pcb_count);
972 if (error)
973 return (error);
974
975 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
976 if (inp_list == NULL)
977 return (ENOMEM);
978
979 INP_INFO_RLOCK(&tcbinfo);
980 for (inp = LIST_FIRST(tcbinfo.ipi_listhead), i = 0; inp != NULL && i
981 < n; inp = LIST_NEXT(inp, inp_list)) {
982 INP_RLOCK(inp);
983 if (inp->inp_gencnt <= gencnt) {
984 /*
985 * XXX: This use of cr_cansee(), introduced with
986 * TCP state changes, is not quite right, but for
987 * now, better than nothing.
988 */
989 if (inp->inp_flags & INP_TIMEWAIT) {
990 if (intotw(inp) != NULL)
991 error = cr_cansee(req->td->td_ucred,
992 intotw(inp)->tw_cred);
993 else
994 error = EINVAL; /* Skip this inp. */
995 } else
996 error = cr_canseeinpcb(req->td->td_ucred, inp);
997 if (error == 0)
998 inp_list[i++] = inp;
999 }
1000 INP_RUNLOCK(inp);
1001 }
1002 INP_INFO_RUNLOCK(&tcbinfo);
1003 n = i;
1004
1005 error = 0;
1006 for (i = 0; i < n; i++) {
1007 inp = inp_list[i];
1008 INP_RLOCK(inp);
1009 if (inp->inp_gencnt <= gencnt) {
1010 struct xtcpcb xt;
1011 void *inp_ppcb;
1012
1013 bzero(&xt, sizeof(xt));
1014 xt.xt_len = sizeof xt;
1015 /* XXX should avoid extra copy */
1016 bcopy(inp, &xt.xt_inp, sizeof *inp);
1017 inp_ppcb = inp->inp_ppcb;
1018 if (inp_ppcb == NULL)
1019 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1020 else if (inp->inp_flags & INP_TIMEWAIT) {
1021 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1022 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1023 } else
1024 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1025 if (inp->inp_socket != NULL)
1026 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1027 else {
1028 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1029 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1030 }
1031 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1032 INP_RUNLOCK(inp);
1033 error = SYSCTL_OUT(req, &xt, sizeof xt);
1034 } else
1035 INP_RUNLOCK(inp);
1036
1037 }
1038 if (!error) {
1039 /*
1040 * Give the user an updated idea of our state.
1041 * If the generation differs from what we told
1042 * her before, she knows that something happened
1043 * while we were processing this request, and it
1044 * might be necessary to retry.
1045 */
1046 INP_INFO_RLOCK(&tcbinfo);
1047 xig.xig_gen = tcbinfo.ipi_gencnt;
1048 xig.xig_sogen = so_gencnt;
1049 xig.xig_count = tcbinfo.ipi_count + pcb_count;
1050 INP_INFO_RUNLOCK(&tcbinfo);
1051 error = SYSCTL_OUT(req, &xig, sizeof xig);
1052 }
1053 free(inp_list, M_TEMP);
1054 return (error);
1055 }
1056
1057 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist, CTLFLAG_RD, 0, 0,
1058 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1059
1060 static int
1061 tcp_getcred(SYSCTL_HANDLER_ARGS)
1062 {
1063 struct xucred xuc;
1064 struct sockaddr_in addrs[2];
1065 struct inpcb *inp;
1066 int error;
1067
1068 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1069 if (error)
1070 return (error);
1071 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1072 if (error)
1073 return (error);
1074 INP_INFO_RLOCK(&tcbinfo);
1075 inp = in_pcblookup_hash(&tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1076 addrs[0].sin_addr, addrs[0].sin_port, 0, NULL);
1077 if (inp != NULL) {
1078 INP_RLOCK(inp);
1079 INP_INFO_RUNLOCK(&tcbinfo);
1080 if (inp->inp_socket == NULL)
1081 error = ENOENT;
1082 if (error == 0)
1083 error = cr_canseeinpcb(req->td->td_ucred, inp);
1084 if (error == 0)
1085 cru2x(inp->inp_cred, &xuc);
1086 INP_RUNLOCK(inp);
1087 } else {
1088 INP_INFO_RUNLOCK(&tcbinfo);
1089 error = ENOENT;
1090 }
1091 if (error == 0)
1092 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1093 return (error);
1094 }
1095
1096 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1097 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1098 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1099
1100 #ifdef INET6
1101 static int
1102 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1103 {
1104 struct xucred xuc;
1105 struct sockaddr_in6 addrs[2];
1106 struct inpcb *inp;
1107 int error, mapped = 0;
1108
1109 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1110 if (error)
1111 return (error);
1112 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1113 if (error)
1114 return (error);
1115 if ((error = sa6_embedscope(&addrs[0], ip6_use_defzone)) != 0 ||
1116 (error = sa6_embedscope(&addrs[1], ip6_use_defzone)) != 0) {
1117 return (error);
1118 }
1119 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1120 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1121 mapped = 1;
1122 else
1123 return (EINVAL);
1124 }
1125
1126 INP_INFO_RLOCK(&tcbinfo);
1127 if (mapped == 1)
1128 inp = in_pcblookup_hash(&tcbinfo,
1129 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1130 addrs[1].sin6_port,
1131 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1132 addrs[0].sin6_port,
1133 0, NULL);
1134 else
1135 inp = in6_pcblookup_hash(&tcbinfo,
1136 &addrs[1].sin6_addr, addrs[1].sin6_port,
1137 &addrs[0].sin6_addr, addrs[0].sin6_port, 0, NULL);
1138 if (inp != NULL) {
1139 INP_RLOCK(inp);
1140 INP_INFO_RUNLOCK(&tcbinfo);
1141 if (inp->inp_socket == NULL)
1142 error = ENOENT;
1143 if (error == 0)
1144 error = cr_canseeinpcb(req->td->td_ucred, inp);
1145 if (error == 0)
1146 cru2x(inp->inp_cred, &xuc);
1147 INP_RUNLOCK(inp);
1148 } else {
1149 INP_INFO_RUNLOCK(&tcbinfo);
1150 error = ENOENT;
1151 }
1152 if (error == 0)
1153 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1154 return (error);
1155 }
1156
1157 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1158 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1159 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1160 #endif
1161
1162
1163 void
1164 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1165 {
1166 struct ip *ip = vip;
1167 struct tcphdr *th;
1168 struct in_addr faddr;
1169 struct inpcb *inp;
1170 struct tcpcb *tp;
1171 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1172 struct icmp *icp;
1173 struct in_conninfo inc;
1174 tcp_seq icmp_tcp_seq;
1175 int mtu;
1176
1177 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1178 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1179 return;
1180
1181 if (cmd == PRC_MSGSIZE)
1182 notify = tcp_mtudisc;
1183 else if (icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1184 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1185 notify = tcp_drop_syn_sent;
1186 /*
1187 * Redirects don't need to be handled up here.
1188 */
1189 else if (PRC_IS_REDIRECT(cmd))
1190 return;
1191 /*
1192 * Source quench is depreciated.
1193 */
1194 else if (cmd == PRC_QUENCH)
1195 return;
1196 /*
1197 * Hostdead is ugly because it goes linearly through all PCBs.
1198 * XXX: We never get this from ICMP, otherwise it makes an
1199 * excellent DoS attack on machines with many connections.
1200 */
1201 else if (cmd == PRC_HOSTDEAD)
1202 ip = NULL;
1203 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1204 return;
1205 if (ip != NULL) {
1206 icp = (struct icmp *)((caddr_t)ip
1207 - offsetof(struct icmp, icmp_ip));
1208 th = (struct tcphdr *)((caddr_t)ip
1209 + (ip->ip_hl << 2));
1210 INP_INFO_WLOCK(&tcbinfo);
1211 inp = in_pcblookup_hash(&tcbinfo, faddr, th->th_dport,
1212 ip->ip_src, th->th_sport, 0, NULL);
1213 if (inp != NULL) {
1214 INP_WLOCK(inp);
1215 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1216 !(inp->inp_flags & INP_DROPPED) &&
1217 !(inp->inp_socket == NULL)) {
1218 icmp_tcp_seq = htonl(th->th_seq);
1219 tp = intotcpcb(inp);
1220 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1221 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1222 if (cmd == PRC_MSGSIZE) {
1223 /*
1224 * MTU discovery:
1225 * If we got a needfrag set the MTU
1226 * in the route to the suggested new
1227 * value (if given) and then notify.
1228 */
1229 bzero(&inc, sizeof(inc));
1230 inc.inc_faddr = faddr;
1231 inc.inc_fibnum =
1232 inp->inp_inc.inc_fibnum;
1233
1234 mtu = ntohs(icp->icmp_nextmtu);
1235 /*
1236 * If no alternative MTU was
1237 * proposed, try the next smaller
1238 * one. ip->ip_len has already
1239 * been swapped in icmp_input().
1240 */
1241 if (!mtu)
1242 mtu = ip_next_mtu(ip->ip_len,
1243 1);
1244 if (mtu < max(296, (tcp_minmss)
1245 + sizeof(struct tcpiphdr)))
1246 mtu = 0;
1247 if (!mtu)
1248 mtu = tcp_mssdflt
1249 + sizeof(struct tcpiphdr);
1250 /*
1251 * Only cache the the MTU if it
1252 * is smaller than the interface
1253 * or route MTU. tcp_mtudisc()
1254 * will do right thing by itself.
1255 */
1256 if (mtu <= tcp_maxmtu(&inc, NULL))
1257 tcp_hc_updatemtu(&inc, mtu);
1258 }
1259
1260 inp = (*notify)(inp, inetctlerrmap[cmd]);
1261 }
1262 }
1263 if (inp != NULL)
1264 INP_WUNLOCK(inp);
1265 } else {
1266 bzero(&inc, sizeof(inc));
1267 inc.inc_fport = th->th_dport;
1268 inc.inc_lport = th->th_sport;
1269 inc.inc_faddr = faddr;
1270 inc.inc_laddr = ip->ip_src;
1271 syncache_unreach(&inc, th);
1272 }
1273 INP_INFO_WUNLOCK(&tcbinfo);
1274 } else
1275 in_pcbnotifyall(&tcbinfo, faddr, inetctlerrmap[cmd], notify);
1276 }
1277
1278 #ifdef INET6
1279 void
1280 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1281 {
1282 struct tcphdr th;
1283 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1284 struct ip6_hdr *ip6;
1285 struct mbuf *m;
1286 struct ip6ctlparam *ip6cp = NULL;
1287 const struct sockaddr_in6 *sa6_src = NULL;
1288 int off;
1289 struct tcp_portonly {
1290 u_int16_t th_sport;
1291 u_int16_t th_dport;
1292 } *thp;
1293
1294 if (sa->sa_family != AF_INET6 ||
1295 sa->sa_len != sizeof(struct sockaddr_in6))
1296 return;
1297
1298 if (cmd == PRC_MSGSIZE)
1299 notify = tcp_mtudisc;
1300 else if (!PRC_IS_REDIRECT(cmd) &&
1301 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1302 return;
1303 /* Source quench is depreciated. */
1304 else if (cmd == PRC_QUENCH)
1305 return;
1306
1307 /* if the parameter is from icmp6, decode it. */
1308 if (d != NULL) {
1309 ip6cp = (struct ip6ctlparam *)d;
1310 m = ip6cp->ip6c_m;
1311 ip6 = ip6cp->ip6c_ip6;
1312 off = ip6cp->ip6c_off;
1313 sa6_src = ip6cp->ip6c_src;
1314 } else {
1315 m = NULL;
1316 ip6 = NULL;
1317 off = 0; /* fool gcc */
1318 sa6_src = &sa6_any;
1319 }
1320
1321 if (ip6 != NULL) {
1322 struct in_conninfo inc;
1323 /*
1324 * XXX: We assume that when IPV6 is non NULL,
1325 * M and OFF are valid.
1326 */
1327
1328 /* check if we can safely examine src and dst ports */
1329 if (m->m_pkthdr.len < off + sizeof(*thp))
1330 return;
1331
1332 bzero(&th, sizeof(th));
1333 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1334
1335 in6_pcbnotify(&tcbinfo, sa, th.th_dport,
1336 (struct sockaddr *)ip6cp->ip6c_src,
1337 th.th_sport, cmd, NULL, notify);
1338
1339 bzero(&inc, sizeof(inc));
1340 inc.inc_fport = th.th_dport;
1341 inc.inc_lport = th.th_sport;
1342 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1343 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1344 inc.inc_flags |= INC_ISIPV6;
1345 INP_INFO_WLOCK(&tcbinfo);
1346 syncache_unreach(&inc, &th);
1347 INP_INFO_WUNLOCK(&tcbinfo);
1348 } else
1349 in6_pcbnotify(&tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1350 0, cmd, NULL, notify);
1351 }
1352 #endif /* INET6 */
1353
1354
1355 /*
1356 * Following is where TCP initial sequence number generation occurs.
1357 *
1358 * There are two places where we must use initial sequence numbers:
1359 * 1. In SYN-ACK packets.
1360 * 2. In SYN packets.
1361 *
1362 * All ISNs for SYN-ACK packets are generated by the syncache. See
1363 * tcp_syncache.c for details.
1364 *
1365 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1366 * depends on this property. In addition, these ISNs should be
1367 * unguessable so as to prevent connection hijacking. To satisfy
1368 * the requirements of this situation, the algorithm outlined in
1369 * RFC 1948 is used, with only small modifications.
1370 *
1371 * Implementation details:
1372 *
1373 * Time is based off the system timer, and is corrected so that it
1374 * increases by one megabyte per second. This allows for proper
1375 * recycling on high speed LANs while still leaving over an hour
1376 * before rollover.
1377 *
1378 * As reading the *exact* system time is too expensive to be done
1379 * whenever setting up a TCP connection, we increment the time
1380 * offset in two ways. First, a small random positive increment
1381 * is added to isn_offset for each connection that is set up.
1382 * Second, the function tcp_isn_tick fires once per clock tick
1383 * and increments isn_offset as necessary so that sequence numbers
1384 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1385 * random positive increments serve only to ensure that the same
1386 * exact sequence number is never sent out twice (as could otherwise
1387 * happen when a port is recycled in less than the system tick
1388 * interval.)
1389 *
1390 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1391 * between seeding of isn_secret. This is normally set to zero,
1392 * as reseeding should not be necessary.
1393 *
1394 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1395 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1396 * general, this means holding an exclusive (write) lock.
1397 */
1398
1399 #define ISN_BYTES_PER_SECOND 1048576
1400 #define ISN_STATIC_INCREMENT 4096
1401 #define ISN_RANDOM_INCREMENT (4096 - 1)
1402
1403 static u_char isn_secret[32];
1404 static int isn_last_reseed;
1405 static u_int32_t isn_offset, isn_offset_old;
1406 static MD5_CTX isn_ctx;
1407
1408 tcp_seq
1409 tcp_new_isn(struct tcpcb *tp)
1410 {
1411 u_int32_t md5_buffer[4];
1412 tcp_seq new_isn;
1413
1414 INP_WLOCK_ASSERT(tp->t_inpcb);
1415
1416 ISN_LOCK();
1417 /* Seed if this is the first use, reseed if requested. */
1418 if ((isn_last_reseed == 0) || ((tcp_isn_reseed_interval > 0) &&
1419 (((u_int)isn_last_reseed + (u_int)tcp_isn_reseed_interval*hz)
1420 < (u_int)ticks))) {
1421 read_random(&isn_secret, sizeof(isn_secret));
1422 isn_last_reseed = ticks;
1423 }
1424
1425 /* Compute the md5 hash and return the ISN. */
1426 MD5Init(&isn_ctx);
1427 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1428 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1429 #ifdef INET6
1430 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1431 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1432 sizeof(struct in6_addr));
1433 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1434 sizeof(struct in6_addr));
1435 } else
1436 #endif
1437 {
1438 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1439 sizeof(struct in_addr));
1440 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1441 sizeof(struct in_addr));
1442 }
1443 MD5Update(&isn_ctx, (u_char *) &isn_secret, sizeof(isn_secret));
1444 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1445 new_isn = (tcp_seq) md5_buffer[0];
1446 isn_offset += ISN_STATIC_INCREMENT +
1447 (arc4random() & ISN_RANDOM_INCREMENT);
1448 new_isn += isn_offset;
1449 ISN_UNLOCK();
1450 return (new_isn);
1451 }
1452
1453 /*
1454 * Increment the offset to the next ISN_BYTES_PER_SECOND / 100 boundary
1455 * to keep time flowing at a relatively constant rate. If the random
1456 * increments have already pushed us past the projected offset, do nothing.
1457 */
1458 static void
1459 tcp_isn_tick(void *xtp)
1460 {
1461 u_int32_t projected_offset;
1462
1463 ISN_LOCK();
1464 projected_offset = isn_offset_old + ISN_BYTES_PER_SECOND / 100;
1465
1466 if (SEQ_GT(projected_offset, isn_offset))
1467 isn_offset = projected_offset;
1468
1469 isn_offset_old = isn_offset;
1470 callout_reset(&isn_callout, hz/100, tcp_isn_tick, NULL);
1471 ISN_UNLOCK();
1472 }
1473
1474 /*
1475 * When a specific ICMP unreachable message is received and the
1476 * connection state is SYN-SENT, drop the connection. This behavior
1477 * is controlled by the icmp_may_rst sysctl.
1478 */
1479 struct inpcb *
1480 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1481 {
1482 struct tcpcb *tp;
1483
1484 INP_INFO_WLOCK_ASSERT(&tcbinfo);
1485 INP_WLOCK_ASSERT(inp);
1486
1487 if ((inp->inp_flags & INP_TIMEWAIT) ||
1488 (inp->inp_flags & INP_DROPPED))
1489 return (inp);
1490
1491 tp = intotcpcb(inp);
1492 if (tp->t_state != TCPS_SYN_SENT)
1493 return (inp);
1494
1495 tp = tcp_drop(tp, errno);
1496 if (tp != NULL)
1497 return (inp);
1498 else
1499 return (NULL);
1500 }
1501
1502 /*
1503 * When `need fragmentation' ICMP is received, update our idea of the MSS
1504 * based on the new value in the route. Also nudge TCP to send something,
1505 * since we know the packet we just sent was dropped.
1506 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1507 */
1508 struct inpcb *
1509 tcp_mtudisc(struct inpcb *inp, int errno)
1510 {
1511 struct tcpcb *tp;
1512 struct socket *so;
1513
1514 INP_WLOCK_ASSERT(inp);
1515 if ((inp->inp_flags & INP_TIMEWAIT) ||
1516 (inp->inp_flags & INP_DROPPED))
1517 return (inp);
1518
1519 tp = intotcpcb(inp);
1520 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1521
1522 tcp_mss_update(tp, -1, NULL, NULL);
1523
1524 so = inp->inp_socket;
1525 SOCKBUF_LOCK(&so->so_snd);
1526 /* If the mss is larger than the socket buffer, decrease the mss. */
1527 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1528 tp->t_maxseg = so->so_snd.sb_hiwat;
1529 SOCKBUF_UNLOCK(&so->so_snd);
1530
1531 tcpstat.tcps_mturesent++;
1532 tp->t_rtttime = 0;
1533 tp->snd_nxt = tp->snd_una;
1534 tcp_free_sackholes(tp);
1535 tp->snd_recover = tp->snd_max;
1536 if (tp->t_flags & TF_SACK_PERMIT)
1537 EXIT_FASTRECOVERY(tp);
1538 tcp_output_send(tp);
1539 return (inp);
1540 }
1541
1542 /*
1543 * Look-up the routing entry to the peer of this inpcb. If no route
1544 * is found and it cannot be allocated, then return 0. This routine
1545 * is called by TCP routines that access the rmx structure and by
1546 * tcp_mss_update to get the peer/interface MTU.
1547 */
1548 u_long
1549 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1550 {
1551 struct route sro;
1552 struct sockaddr_in *dst;
1553 struct ifnet *ifp;
1554 u_long maxmtu = 0;
1555
1556 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1557
1558 bzero(&sro, sizeof(sro));
1559 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1560 dst = (struct sockaddr_in *)&sro.ro_dst;
1561 dst->sin_family = AF_INET;
1562 dst->sin_len = sizeof(*dst);
1563 dst->sin_addr = inc->inc_faddr;
1564 in_rtalloc_ign(&sro, RTF_CLONING, inc->inc_fibnum);
1565 }
1566 if (sro.ro_rt != NULL) {
1567 ifp = sro.ro_rt->rt_ifp;
1568 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1569 maxmtu = ifp->if_mtu;
1570 else
1571 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1572
1573 /* Report additional interface capabilities. */
1574 if (flags != NULL) {
1575 if (ifp->if_capenable & IFCAP_TSO4 &&
1576 ifp->if_hwassist & CSUM_TSO)
1577 *flags |= CSUM_TSO;
1578 }
1579 RTFREE(sro.ro_rt);
1580 }
1581 return (maxmtu);
1582 }
1583
1584 #ifdef INET6
1585 u_long
1586 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1587 {
1588 struct route_in6 sro6;
1589 struct ifnet *ifp;
1590 u_long maxmtu = 0;
1591
1592 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1593
1594 bzero(&sro6, sizeof(sro6));
1595 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1596 sro6.ro_dst.sin6_family = AF_INET6;
1597 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1598 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1599 rtalloc_ign((struct route *)&sro6, RTF_CLONING);
1600 }
1601 if (sro6.ro_rt != NULL) {
1602 ifp = sro6.ro_rt->rt_ifp;
1603 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1604 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1605 else
1606 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1607 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1608
1609 /* Report additional interface capabilities. */
1610 if (flags != NULL) {
1611 if (ifp->if_capenable & IFCAP_TSO6 &&
1612 ifp->if_hwassist & CSUM_TSO)
1613 *flags |= CSUM_TSO;
1614 }
1615 RTFREE(sro6.ro_rt);
1616 }
1617
1618 return (maxmtu);
1619 }
1620 #endif /* INET6 */
1621
1622 #ifdef IPSEC
1623 /* compute ESP/AH header size for TCP, including outer IP header. */
1624 size_t
1625 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1626 {
1627 struct inpcb *inp;
1628 struct mbuf *m;
1629 size_t hdrsiz;
1630 struct ip *ip;
1631 #ifdef INET6
1632 struct ip6_hdr *ip6;
1633 #endif
1634 struct tcphdr *th;
1635
1636 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1637 return (0);
1638 MGETHDR(m, M_DONTWAIT, MT_DATA);
1639 if (!m)
1640 return (0);
1641
1642 #ifdef INET6
1643 if ((inp->inp_vflag & INP_IPV6) != 0) {
1644 ip6 = mtod(m, struct ip6_hdr *);
1645 th = (struct tcphdr *)(ip6 + 1);
1646 m->m_pkthdr.len = m->m_len =
1647 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1648 tcpip_fillheaders(inp, ip6, th);
1649 hdrsiz = ipsec6_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1650 } else
1651 #endif /* INET6 */
1652 {
1653 ip = mtod(m, struct ip *);
1654 th = (struct tcphdr *)(ip + 1);
1655 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1656 tcpip_fillheaders(inp, ip, th);
1657 hdrsiz = ipsec4_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1658 }
1659
1660 m_free(m);
1661 return (hdrsiz);
1662 }
1663 #endif /* IPSEC */
1664
1665 /*
1666 * TCP BANDWIDTH DELAY PRODUCT WINDOW LIMITING
1667 *
1668 * This code attempts to calculate the bandwidth-delay product as a
1669 * means of determining the optimal window size to maximize bandwidth,
1670 * minimize RTT, and avoid the over-allocation of buffers on interfaces and
1671 * routers. This code also does a fairly good job keeping RTTs in check
1672 * across slow links like modems. We implement an algorithm which is very
1673 * similar (but not meant to be) TCP/Vegas. The code operates on the
1674 * transmitter side of a TCP connection and so only effects the transmit
1675 * side of the connection.
1676 *
1677 * BACKGROUND: TCP makes no provision for the management of buffer space
1678 * at the end points or at the intermediate routers and switches. A TCP
1679 * stream, whether using NewReno or not, will eventually buffer as
1680 * many packets as it is able and the only reason this typically works is
1681 * due to the fairly small default buffers made available for a connection
1682 * (typicaly 16K or 32K). As machines use larger windows and/or window
1683 * scaling it is now fairly easy for even a single TCP connection to blow-out
1684 * all available buffer space not only on the local interface, but on
1685 * intermediate routers and switches as well. NewReno makes a misguided
1686 * attempt to 'solve' this problem by waiting for an actual failure to occur,
1687 * then backing off, then steadily increasing the window again until another
1688 * failure occurs, ad-infinitum. This results in terrible oscillation that
1689 * is only made worse as network loads increase and the idea of intentionally
1690 * blowing out network buffers is, frankly, a terrible way to manage network
1691 * resources.
1692 *
1693 * It is far better to limit the transmit window prior to the failure
1694 * condition being achieved. There are two general ways to do this: First
1695 * you can 'scan' through different transmit window sizes and locate the
1696 * point where the RTT stops increasing, indicating that you have filled the
1697 * pipe, then scan backwards until you note that RTT stops decreasing, then
1698 * repeat ad-infinitum. This method works in principle but has severe
1699 * implementation issues due to RTT variances, timer granularity, and
1700 * instability in the algorithm which can lead to many false positives and
1701 * create oscillations as well as interact badly with other TCP streams
1702 * implementing the same algorithm.
1703 *
1704 * The second method is to limit the window to the bandwidth delay product
1705 * of the link. This is the method we implement. RTT variances and our
1706 * own manipulation of the congestion window, bwnd, can potentially
1707 * destabilize the algorithm. For this reason we have to stabilize the
1708 * elements used to calculate the window. We do this by using the minimum
1709 * observed RTT, the long term average of the observed bandwidth, and
1710 * by adding two segments worth of slop. It isn't perfect but it is able
1711 * to react to changing conditions and gives us a very stable basis on
1712 * which to extend the algorithm.
1713 */
1714 void
1715 tcp_xmit_bandwidth_limit(struct tcpcb *tp, tcp_seq ack_seq)
1716 {
1717 u_long bw;
1718 u_long bwnd;
1719 int save_ticks;
1720
1721 INP_WLOCK_ASSERT(tp->t_inpcb);
1722
1723 /*
1724 * If inflight_enable is disabled in the middle of a tcp connection,
1725 * make sure snd_bwnd is effectively disabled.
1726 */
1727 if (tcp_inflight_enable == 0 || tp->t_rttlow < tcp_inflight_rttthresh) {
1728 tp->snd_bwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
1729 tp->snd_bandwidth = 0;
1730 return;
1731 }
1732
1733 /*
1734 * Figure out the bandwidth. Due to the tick granularity this
1735 * is a very rough number and it MUST be averaged over a fairly
1736 * long period of time. XXX we need to take into account a link
1737 * that is not using all available bandwidth, but for now our
1738 * slop will ramp us up if this case occurs and the bandwidth later
1739 * increases.
1740 *
1741 * Note: if ticks rollover 'bw' may wind up negative. We must
1742 * effectively reset t_bw_rtttime for this case.
1743 */
1744 save_ticks = ticks;
1745 if ((u_int)(save_ticks - tp->t_bw_rtttime) < 1)
1746 return;
1747
1748 bw = (int64_t)(ack_seq - tp->t_bw_rtseq) * hz /
1749 (save_ticks - tp->t_bw_rtttime);
1750 tp->t_bw_rtttime = save_ticks;
1751 tp->t_bw_rtseq = ack_seq;
1752 if (tp->t_bw_rtttime == 0 || (int)bw < 0)
1753 return;
1754 bw = ((int64_t)tp->snd_bandwidth * 15 + bw) >> 4;
1755
1756 tp->snd_bandwidth = bw;
1757
1758 /*
1759 * Calculate the semi-static bandwidth delay product, plus two maximal
1760 * segments. The additional slop puts us squarely in the sweet
1761 * spot and also handles the bandwidth run-up case and stabilization.
1762 * Without the slop we could be locking ourselves into a lower
1763 * bandwidth.
1764 *
1765 * Situations Handled:
1766 * (1) Prevents over-queueing of packets on LANs, especially on
1767 * high speed LANs, allowing larger TCP buffers to be
1768 * specified, and also does a good job preventing
1769 * over-queueing of packets over choke points like modems
1770 * (at least for the transmit side).
1771 *
1772 * (2) Is able to handle changing network loads (bandwidth
1773 * drops so bwnd drops, bandwidth increases so bwnd
1774 * increases).
1775 *
1776 * (3) Theoretically should stabilize in the face of multiple
1777 * connections implementing the same algorithm (this may need
1778 * a little work).
1779 *
1780 * (4) Stability value (defaults to 20 = 2 maximal packets) can
1781 * be adjusted with a sysctl but typically only needs to be
1782 * on very slow connections. A value no smaller then 5
1783 * should be used, but only reduce this default if you have
1784 * no other choice.
1785 */
1786 #define USERTT ((tp->t_srtt + tp->t_rttbest) / 2)
1787 bwnd = (int64_t)bw * USERTT / (hz << TCP_RTT_SHIFT) + tcp_inflight_stab * tp->t_maxseg / 10;
1788 #undef USERTT
1789
1790 if (tcp_inflight_debug > 0) {
1791 static int ltime;
1792 if ((u_int)(ticks - ltime) >= hz / tcp_inflight_debug) {
1793 ltime = ticks;
1794 printf("%p bw %ld rttbest %d srtt %d bwnd %ld\n",
1795 tp,
1796 bw,
1797 tp->t_rttbest,
1798 tp->t_srtt,
1799 bwnd
1800 );
1801 }
1802 }
1803 if ((long)bwnd < tcp_inflight_min)
1804 bwnd = tcp_inflight_min;
1805 if (bwnd > tcp_inflight_max)
1806 bwnd = tcp_inflight_max;
1807 if ((long)bwnd < tp->t_maxseg * 2)
1808 bwnd = tp->t_maxseg * 2;
1809 tp->snd_bwnd = bwnd;
1810 }
1811
1812 #ifdef TCP_SIGNATURE
1813 /*
1814 * Callback function invoked by m_apply() to digest TCP segment data
1815 * contained within an mbuf chain.
1816 */
1817 static int
1818 tcp_signature_apply(void *fstate, void *data, u_int len)
1819 {
1820
1821 MD5Update(fstate, (u_char *)data, len);
1822 return (0);
1823 }
1824
1825 /*
1826 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1827 *
1828 * Parameters:
1829 * m pointer to head of mbuf chain
1830 * _unused
1831 * len length of TCP segment data, excluding options
1832 * optlen length of TCP segment options
1833 * buf pointer to storage for computed MD5 digest
1834 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1835 *
1836 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1837 * When called from tcp_input(), we can be sure that th_sum has been
1838 * zeroed out and verified already.
1839 *
1840 * Return 0 if successful, otherwise return -1.
1841 *
1842 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1843 * search with the destination IP address, and a 'magic SPI' to be
1844 * determined by the application. This is hardcoded elsewhere to 1179
1845 * right now. Another branch of this code exists which uses the SPD to
1846 * specify per-application flows but it is unstable.
1847 */
1848 int
1849 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1850 u_char *buf, u_int direction)
1851 {
1852 union sockaddr_union dst;
1853 struct ippseudo ippseudo;
1854 MD5_CTX ctx;
1855 int doff;
1856 struct ip *ip;
1857 struct ipovly *ipovly;
1858 struct secasvar *sav;
1859 struct tcphdr *th;
1860 #ifdef INET6
1861 struct ip6_hdr *ip6;
1862 struct in6_addr in6;
1863 char ip6buf[INET6_ADDRSTRLEN];
1864 uint32_t plen;
1865 uint16_t nhdr;
1866 #endif
1867 u_short savecsum;
1868
1869 KASSERT(m != NULL, ("NULL mbuf chain"));
1870 KASSERT(buf != NULL, ("NULL signature pointer"));
1871
1872 /* Extract the destination from the IP header in the mbuf. */
1873 bzero(&dst, sizeof(union sockaddr_union));
1874 ip = mtod(m, struct ip *);
1875 #ifdef INET6
1876 ip6 = NULL; /* Make the compiler happy. */
1877 #endif
1878 switch (ip->ip_v) {
1879 case IPVERSION:
1880 dst.sa.sa_len = sizeof(struct sockaddr_in);
1881 dst.sa.sa_family = AF_INET;
1882 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1883 ip->ip_src : ip->ip_dst;
1884 break;
1885 #ifdef INET6
1886 case (IPV6_VERSION >> 4):
1887 ip6 = mtod(m, struct ip6_hdr *);
1888 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1889 dst.sa.sa_family = AF_INET6;
1890 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1891 ip6->ip6_src : ip6->ip6_dst;
1892 break;
1893 #endif
1894 default:
1895 return (EINVAL);
1896 /* NOTREACHED */
1897 break;
1898 }
1899
1900 /* Look up an SADB entry which matches the address of the peer. */
1901 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1902 if (sav == NULL) {
1903 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1904 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1905 #ifdef INET6
1906 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1907 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1908 #endif
1909 "(unsupported)"));
1910 return (EINVAL);
1911 }
1912
1913 MD5Init(&ctx);
1914 /*
1915 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1916 *
1917 * XXX The ippseudo header MUST be digested in network byte order,
1918 * or else we'll fail the regression test. Assume all fields we've
1919 * been doing arithmetic on have been in host byte order.
1920 * XXX One cannot depend on ipovly->ih_len here. When called from
1921 * tcp_output(), the underlying ip_len member has not yet been set.
1922 */
1923 switch (ip->ip_v) {
1924 case IPVERSION:
1925 ipovly = (struct ipovly *)ip;
1926 ippseudo.ippseudo_src = ipovly->ih_src;
1927 ippseudo.ippseudo_dst = ipovly->ih_dst;
1928 ippseudo.ippseudo_pad = 0;
1929 ippseudo.ippseudo_p = IPPROTO_TCP;
1930 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1931 optlen);
1932 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1933
1934 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1935 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1936 break;
1937 #ifdef INET6
1938 /*
1939 * RFC 2385, 2.0 Proposal
1940 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1941 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1942 * extended next header value (to form 32 bits), and 32-bit segment
1943 * length.
1944 * Note: Upper-Layer Packet Length comes before Next Header.
1945 */
1946 case (IPV6_VERSION >> 4):
1947 in6 = ip6->ip6_src;
1948 in6_clearscope(&in6);
1949 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1950 in6 = ip6->ip6_dst;
1951 in6_clearscope(&in6);
1952 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1953 plen = htonl(len + sizeof(struct tcphdr) + optlen);
1954 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1955 nhdr = 0;
1956 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1957 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1958 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1959 nhdr = IPPROTO_TCP;
1960 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1961
1962 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1963 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1964 break;
1965 #endif
1966 default:
1967 return (EINVAL);
1968 /* NOTREACHED */
1969 break;
1970 }
1971
1972
1973 /*
1974 * Step 2: Update MD5 hash with TCP header, excluding options.
1975 * The TCP checksum must be set to zero.
1976 */
1977 savecsum = th->th_sum;
1978 th->th_sum = 0;
1979 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1980 th->th_sum = savecsum;
1981
1982 /*
1983 * Step 3: Update MD5 hash with TCP segment data.
1984 * Use m_apply() to avoid an early m_pullup().
1985 */
1986 if (len > 0)
1987 m_apply(m, doff, len, tcp_signature_apply, &ctx);
1988
1989 /*
1990 * Step 4: Update MD5 hash with shared secret.
1991 */
1992 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
1993 MD5Final(buf, &ctx);
1994
1995 key_sa_recordxfer(sav, m);
1996 KEY_FREESAV(&sav);
1997 return (0);
1998 }
1999 #endif /* TCP_SIGNATURE */
2000
2001 static int
2002 sysctl_drop(SYSCTL_HANDLER_ARGS)
2003 {
2004 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2005 struct sockaddr_storage addrs[2];
2006 struct inpcb *inp;
2007 struct tcpcb *tp;
2008 struct tcptw *tw;
2009 struct sockaddr_in *fin, *lin;
2010 #ifdef INET6
2011 struct sockaddr_in6 *fin6, *lin6;
2012 struct in6_addr f6, l6;
2013 #endif
2014 int error;
2015
2016 inp = NULL;
2017 fin = lin = NULL;
2018 #ifdef INET6
2019 fin6 = lin6 = NULL;
2020 #endif
2021 error = 0;
2022
2023 if (req->oldptr != NULL || req->oldlen != 0)
2024 return (EINVAL);
2025 if (req->newptr == NULL)
2026 return (EPERM);
2027 if (req->newlen < sizeof(addrs))
2028 return (ENOMEM);
2029 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2030 if (error)
2031 return (error);
2032
2033 switch (addrs[0].ss_family) {
2034 #ifdef INET6
2035 case AF_INET6:
2036 fin6 = (struct sockaddr_in6 *)&addrs[0];
2037 lin6 = (struct sockaddr_in6 *)&addrs[1];
2038 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2039 lin6->sin6_len != sizeof(struct sockaddr_in6))
2040 return (EINVAL);
2041 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2042 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2043 return (EINVAL);
2044 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2045 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2046 fin = (struct sockaddr_in *)&addrs[0];
2047 lin = (struct sockaddr_in *)&addrs[1];
2048 break;
2049 }
2050 error = sa6_embedscope(fin6, ip6_use_defzone);
2051 if (error)
2052 return (error);
2053 error = sa6_embedscope(lin6, ip6_use_defzone);
2054 if (error)
2055 return (error);
2056 break;
2057 #endif
2058 case AF_INET:
2059 fin = (struct sockaddr_in *)&addrs[0];
2060 lin = (struct sockaddr_in *)&addrs[1];
2061 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2062 lin->sin_len != sizeof(struct sockaddr_in))
2063 return (EINVAL);
2064 break;
2065 default:
2066 return (EINVAL);
2067 }
2068 INP_INFO_WLOCK(&tcbinfo);
2069 switch (addrs[0].ss_family) {
2070 #ifdef INET6
2071 case AF_INET6:
2072 inp = in6_pcblookup_hash(&tcbinfo, &f6, fin6->sin6_port,
2073 &l6, lin6->sin6_port, 0, NULL);
2074 break;
2075 #endif
2076 case AF_INET:
2077 inp = in_pcblookup_hash(&tcbinfo, fin->sin_addr, fin->sin_port,
2078 lin->sin_addr, lin->sin_port, 0, NULL);
2079 break;
2080 }
2081 if (inp != NULL) {
2082 INP_WLOCK(inp);
2083 if (inp->inp_flags & INP_TIMEWAIT) {
2084 /*
2085 * XXXRW: There currently exists a state where an
2086 * inpcb is present, but its timewait state has been
2087 * discarded. For now, don't allow dropping of this
2088 * type of inpcb.
2089 */
2090 tw = intotw(inp);
2091 if (tw != NULL)
2092 tcp_twclose(tw, 0);
2093 else
2094 INP_WUNLOCK(inp);
2095 } else if (!(inp->inp_flags & INP_DROPPED) &&
2096 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2097 tp = intotcpcb(inp);
2098 tp = tcp_drop(tp, ECONNABORTED);
2099 if (tp != NULL)
2100 INP_WUNLOCK(inp);
2101 } else
2102 INP_WUNLOCK(inp);
2103 } else
2104 error = ESRCH;
2105 INP_INFO_WUNLOCK(&tcbinfo);
2106 return (error);
2107 }
2108
2109 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2110 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2111 0, sysctl_drop, "", "Drop TCP connection");
2112
2113 /*
2114 * Generate a standardized TCP log line for use throughout the
2115 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2116 * allow use in the interrupt context.
2117 *
2118 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2119 * NB: The function may return NULL if memory allocation failed.
2120 *
2121 * Due to header inclusion and ordering limitations the struct ip
2122 * and ip6_hdr pointers have to be passed as void pointers.
2123 */
2124 char *
2125 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2126 const void *ip6hdr)
2127 {
2128 char *s, *sp;
2129 size_t size;
2130 struct ip *ip;
2131 #ifdef INET6
2132 const struct ip6_hdr *ip6;
2133
2134 ip6 = (const struct ip6_hdr *)ip6hdr;
2135 #endif /* INET6 */
2136 ip = (struct ip *)ip4hdr;
2137
2138 /*
2139 * The log line looks like this:
2140 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2141 */
2142 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2143 sizeof(PRINT_TH_FLAGS) + 1 +
2144 #ifdef INET6
2145 2 * INET6_ADDRSTRLEN;
2146 #else
2147 2 * INET_ADDRSTRLEN;
2148 #endif /* INET6 */
2149
2150 /* Is logging enabled? */
2151 if (tcp_log_debug == 0 && tcp_log_in_vain == 0)
2152 return (NULL);
2153
2154 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2155 if (s == NULL)
2156 return (NULL);
2157
2158 strcat(s, "TCP: [");
2159 sp = s + strlen(s);
2160
2161 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2162 inet_ntoa_r(inc->inc_faddr, sp);
2163 sp = s + strlen(s);
2164 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2165 sp = s + strlen(s);
2166 inet_ntoa_r(inc->inc_laddr, sp);
2167 sp = s + strlen(s);
2168 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2169 #ifdef INET6
2170 } else if (inc) {
2171 ip6_sprintf(sp, &inc->inc6_faddr);
2172 sp = s + strlen(s);
2173 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2174 sp = s + strlen(s);
2175 ip6_sprintf(sp, &inc->inc6_laddr);
2176 sp = s + strlen(s);
2177 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2178 } else if (ip6 && th) {
2179 ip6_sprintf(sp, &ip6->ip6_src);
2180 sp = s + strlen(s);
2181 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2182 sp = s + strlen(s);
2183 ip6_sprintf(sp, &ip6->ip6_dst);
2184 sp = s + strlen(s);
2185 sprintf(sp, "]:%i", ntohs(th->th_dport));
2186 #endif /* INET6 */
2187 } else if (ip && th) {
2188 inet_ntoa_r(ip->ip_src, sp);
2189 sp = s + strlen(s);
2190 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2191 sp = s + strlen(s);
2192 inet_ntoa_r(ip->ip_dst, sp);
2193 sp = s + strlen(s);
2194 sprintf(sp, "]:%i", ntohs(th->th_dport));
2195 } else {
2196 free(s, M_TCPLOG);
2197 return (NULL);
2198 }
2199 sp = s + strlen(s);
2200 if (th)
2201 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2202 if (*(s + size - 1) != '\0')
2203 panic("%s: string too long", __func__);
2204 return (s);
2205 }
Cache object: 65ef3de7d0405ba8ea42113a6fc36150
|