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: head/sys/netinet/tcp_subr.c 240233 2012-09-08 06:41:54Z glebius $");
34
35#include "opt_compat.h"
36#include "opt_inet.h"
37#include "opt_inet6.h"
38#include "opt_ipsec.h"
39#include "opt_tcpdebug.h"
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/callout.h>
44#include <sys/hhook.h>
45#include <sys/kernel.h>
46#include <sys/khelp.h>
47#include <sys/sysctl.h>
48#include <sys/jail.h>
49#include <sys/malloc.h>
50#include <sys/mbuf.h>
51#ifdef INET6
52#include <sys/domain.h>
53#endif
54#include <sys/priv.h>
55#include <sys/proc.h>
56#include <sys/socket.h>
57#include <sys/socketvar.h>
58#include <sys/protosw.h>
59#include <sys/random.h>
60
61#include <vm/uma.h>
62
63#include <net/route.h>
64#include <net/if.h>
65#include <net/vnet.h>
66
67#include <netinet/cc.h>
68#include <netinet/in.h>
69#include <netinet/in_pcb.h>
70#include <netinet/in_systm.h>
71#include <netinet/in_var.h>
72#include <netinet/ip.h>
73#include <netinet/ip_icmp.h>
74#include <netinet/ip_var.h>
75#ifdef INET6
76#include <netinet/ip6.h>
77#include <netinet6/in6_pcb.h>
78#include <netinet6/ip6_var.h>
79#include <netinet6/scope6_var.h>
80#include <netinet6/nd6.h>
81#endif
82
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#ifdef INET6
89#include <netinet6/tcp6_var.h>
90#endif
91#include <netinet/tcpip.h>
92#ifdef TCPDEBUG
93#include <netinet/tcp_debug.h>
94#endif
95#ifdef INET6
96#include <netinet6/ip6protosw.h>
97#endif
98#ifdef TCP_OFFLOAD
99#include <netinet/tcp_offload.h>
100#endif
101
102#ifdef IPSEC
103#include <netipsec/ipsec.h>
104#include <netipsec/xform.h>
105#ifdef INET6
106#include <netipsec/ipsec6.h>
107#endif
108#include <netipsec/key.h>
109#include <sys/syslog.h>
110#endif /*IPSEC*/
111
112#include <machine/in_cksum.h>
113#include <sys/md5.h>
114
115#include <security/mac/mac_framework.h>
116
117VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
118#ifdef INET6
119VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
120#endif
121
122static int
123sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
124{
125 int error, new;
126
127 new = V_tcp_mssdflt;
128 error = sysctl_handle_int(oidp, &new, 0, req);
129 if (error == 0 && req->newptr) {
130 if (new < TCP_MINMSS)
131 error = EINVAL;
132 else
133 V_tcp_mssdflt = new;
134 }
135 return (error);
136}
137
138SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
139 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
140 &sysctl_net_inet_tcp_mss_check, "I",
141 "Default TCP Maximum Segment Size");
142
143#ifdef INET6
144static int
145sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
146{
147 int error, new;
148
149 new = V_tcp_v6mssdflt;
150 error = sysctl_handle_int(oidp, &new, 0, req);
151 if (error == 0 && req->newptr) {
152 if (new < TCP_MINMSS)
153 error = EINVAL;
154 else
155 V_tcp_v6mssdflt = new;
156 }
157 return (error);
158}
159
160SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
161 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
162 &sysctl_net_inet_tcp_mss_v6_check, "I",
163 "Default TCP Maximum Segment Size for IPv6");
164#endif /* INET6 */
165
166/*
167 * Minimum MSS we accept and use. This prevents DoS attacks where
168 * we are forced to a ridiculous low MSS like 20 and send hundreds
169 * of packets instead of one. The effect scales with the available
170 * bandwidth and quickly saturates the CPU and network interface
171 * with packet generation and sending. Set to zero to disable MINMSS
172 * checking. This setting prevents us from sending too small packets.
173 */
174VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
175SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
176 &VNET_NAME(tcp_minmss), 0,
177 "Minmum TCP Maximum Segment Size");
178
179VNET_DEFINE(int, tcp_do_rfc1323) = 1;
180SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_rfc1323), 0,
182 "Enable rfc1323 (high performance TCP) extensions");
183
184static int tcp_log_debug = 0;
185SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
186 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
187
188static int tcp_tcbhashsize = 0;
189SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
190 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
191
192static int do_tcpdrain = 1;
193SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
194 "Enable tcp_drain routine for extra help when low on mbufs");
195
196SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
197 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
198
199static VNET_DEFINE(int, icmp_may_rst) = 1;
200#define V_icmp_may_rst VNET(icmp_may_rst)
201SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
202 &VNET_NAME(icmp_may_rst), 0,
203 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
204
205static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
206#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
207SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
208 &VNET_NAME(tcp_isn_reseed_interval), 0,
209 "Seconds between reseeding of ISN secret");
210
211static int tcp_soreceive_stream = 0;
212SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
213 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
214
215#ifdef TCP_SIGNATURE
216static int tcp_sig_checksigs = 1;
217SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
218 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
219#endif
220
221VNET_DEFINE(uma_zone_t, sack_hole_zone);
222#define V_sack_hole_zone VNET(sack_hole_zone)
223
224VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
225
226static struct inpcb *tcp_notify(struct inpcb *, int);
227static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
228static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
229 void *ip4hdr, const void *ip6hdr);
230
231/*
232 * Target size of TCP PCB hash tables. Must be a power of two.
233 *
234 * Note that this can be overridden by the kernel environment
235 * variable net.inet.tcp.tcbhashsize
236 */
237#ifndef TCBHASHSIZE
238#define TCBHASHSIZE 512
239#endif
240
241/*
242 * XXX
243 * Callouts should be moved into struct tcp directly. They are currently
244 * separate because the tcpcb structure is exported to userland for sysctl
245 * parsing purposes, which do not know about callouts.
246 */
247struct tcpcb_mem {
248 struct tcpcb tcb;
249 struct tcp_timer tt;
250 struct cc_var ccv;
251 struct osd osd;
252};
253
254static VNET_DEFINE(uma_zone_t, tcpcb_zone);
255#define V_tcpcb_zone VNET(tcpcb_zone)
256
257MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
258static struct mtx isn_mtx;
259
260#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
261#define ISN_LOCK() mtx_lock(&isn_mtx)
262#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
263
264/*
265 * TCP initialization.
266 */
267static void
268tcp_zone_change(void *tag)
269{
270
271 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
272 uma_zone_set_max(V_tcpcb_zone, maxsockets);
273 tcp_tw_zone_change();
274}
275
276static int
277tcp_inpcb_init(void *mem, int size, int flags)
278{
279 struct inpcb *inp = mem;
280
281 INP_LOCK_INIT(inp, "inp", "tcpinp");
282 return (0);
283}
284
285void
286tcp_init(void)
287{
288 int hashsize;
289
290 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
291 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
292 printf("%s: WARNING: unable to register helper hook\n", __func__);
293 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
294 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
295 printf("%s: WARNING: unable to register helper hook\n", __func__);
296
297 hashsize = TCBHASHSIZE;
298 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
299 if (!powerof2(hashsize)) {
300 printf("WARNING: TCB hash size not a power of 2\n");
301 hashsize = 512; /* safe default */
302 }
303 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
304 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
305 IPI_HASHFIELDS_4TUPLE);
306
307 /*
308 * These have to be type stable for the benefit of the timers.
309 */
310 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
311 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
312 uma_zone_set_max(V_tcpcb_zone, maxsockets);
313
314 tcp_tw_init();
315 syncache_init();
316 tcp_hc_init();
317 tcp_reass_init();
318
319 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
320 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
321 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
322
323 /* Skip initialization of globals for non-default instances. */
324 if (!IS_DEFAULT_VNET(curvnet))
325 return;
326
327 /* XXX virtualize those bellow? */
328 tcp_delacktime = TCPTV_DELACK;
329 tcp_keepinit = TCPTV_KEEP_INIT;
330 tcp_keepidle = TCPTV_KEEP_IDLE;
331 tcp_keepintvl = TCPTV_KEEPINTVL;
332 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
333 tcp_msl = TCPTV_MSL;
334 tcp_rexmit_min = TCPTV_MIN;
335 if (tcp_rexmit_min < 1)
336 tcp_rexmit_min = 1;
337 tcp_rexmit_slop = TCPTV_CPU_VAR;
338 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
339 tcp_tcbhashsize = hashsize;
340
341 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
342 if (tcp_soreceive_stream) {
343#ifdef INET
344 tcp_usrreqs.pru_soreceive = soreceive_stream;
345#endif
346#ifdef INET6
347 tcp6_usrreqs.pru_soreceive = soreceive_stream;
348#endif /* INET6 */
349 }
350
351#ifdef INET6
352#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
353#else /* INET6 */
354#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
355#endif /* INET6 */
356 if (max_protohdr < TCP_MINPROTOHDR)
357 max_protohdr = TCP_MINPROTOHDR;
358 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
359 panic("tcp_init");
360#undef TCP_MINPROTOHDR
361
362 ISN_LOCK_INIT();
363 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
364 SHUTDOWN_PRI_DEFAULT);
365 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
366 EVENTHANDLER_PRI_ANY);
367}
368
369#ifdef VIMAGE
370void
371tcp_destroy(void)
372{
373
374 tcp_reass_destroy();
375 tcp_hc_destroy();
376 syncache_destroy();
377 tcp_tw_destroy();
378 in_pcbinfo_destroy(&V_tcbinfo);
379 uma_zdestroy(V_sack_hole_zone);
380 uma_zdestroy(V_tcpcb_zone);
381}
382#endif
383
384void
385tcp_fini(void *xtp)
386{
387
388}
389
390/*
391 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
392 * tcp_template used to store this data in mbufs, but we now recopy it out
393 * of the tcpcb each time to conserve mbufs.
394 */
395void
396tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
397{
398 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
399
400 INP_WLOCK_ASSERT(inp);
401
402#ifdef INET6
403 if ((inp->inp_vflag & INP_IPV6) != 0) {
404 struct ip6_hdr *ip6;
405
406 ip6 = (struct ip6_hdr *)ip_ptr;
407 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
408 (inp->inp_flow & IPV6_FLOWINFO_MASK);
409 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
410 (IPV6_VERSION & IPV6_VERSION_MASK);
411 ip6->ip6_nxt = IPPROTO_TCP;
412 ip6->ip6_plen = htons(sizeof(struct tcphdr));
413 ip6->ip6_src = inp->in6p_laddr;
414 ip6->ip6_dst = inp->in6p_faddr;
415 }
416#endif /* INET6 */
417#if defined(INET6) && defined(INET)
418 else
419#endif
420#ifdef INET
421 {
422 struct ip *ip;
423
424 ip = (struct ip *)ip_ptr;
425 ip->ip_v = IPVERSION;
426 ip->ip_hl = 5;
427 ip->ip_tos = inp->inp_ip_tos;
428 ip->ip_len = 0;
429 ip->ip_id = 0;
430 ip->ip_off = 0;
431 ip->ip_ttl = inp->inp_ip_ttl;
432 ip->ip_sum = 0;
433 ip->ip_p = IPPROTO_TCP;
434 ip->ip_src = inp->inp_laddr;
435 ip->ip_dst = inp->inp_faddr;
436 }
437#endif /* INET */
438 th->th_sport = inp->inp_lport;
439 th->th_dport = inp->inp_fport;
440 th->th_seq = 0;
441 th->th_ack = 0;
442 th->th_x2 = 0;
443 th->th_off = 5;
444 th->th_flags = 0;
445 th->th_win = 0;
446 th->th_urp = 0;
447 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
448}
449
450/*
451 * Create template to be used to send tcp packets on a connection.
452 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
453 * use for this function is in keepalives, which use tcp_respond.
454 */
455struct tcptemp *
456tcpip_maketemplate(struct inpcb *inp)
457{
458 struct tcptemp *t;
459
460 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
461 if (t == NULL)
462 return (NULL);
463 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
464 return (t);
465}
466
467/*
468 * Send a single message to the TCP at address specified by
469 * the given TCP/IP header. If m == NULL, then we make a copy
470 * of the tcpiphdr at ti and send directly to the addressed host.
471 * This is used to force keep alive messages out using the TCP
472 * template for a connection. If flags are given then we send
473 * a message back to the TCP which originated the * segment ti,
474 * and discard the mbuf containing it and any other attached mbufs.
475 *
476 * In any case the ack and sequence number of the transmitted
477 * segment are as specified by the parameters.
478 *
479 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
480 */
481void
482tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
483 tcp_seq ack, tcp_seq seq, int flags)
484{
485 int tlen;
486 int win = 0;
487 struct ip *ip;
488 struct tcphdr *nth;
489#ifdef INET6
490 struct ip6_hdr *ip6;
491 int isipv6;
492#endif /* INET6 */
493 int ipflags = 0;
494 struct inpcb *inp;
495
496 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
497
498#ifdef INET6
499 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
500 ip6 = ipgen;
501#endif /* INET6 */
502 ip = ipgen;
503
504 if (tp != NULL) {
505 inp = tp->t_inpcb;
506 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
507 INP_WLOCK_ASSERT(inp);
508 } else
509 inp = NULL;
510
511 if (tp != NULL) {
512 if (!(flags & TH_RST)) {
513 win = sbspace(&inp->inp_socket->so_rcv);
514 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
515 win = (long)TCP_MAXWIN << tp->rcv_scale;
516 }
517 }
518 if (m == NULL) {
519 m = m_gethdr(M_DONTWAIT, MT_DATA);
520 if (m == NULL)
521 return;
522 tlen = 0;
523 m->m_data += max_linkhdr;
524#ifdef INET6
525 if (isipv6) {
526 bcopy((caddr_t)ip6, mtod(m, caddr_t),
527 sizeof(struct ip6_hdr));
528 ip6 = mtod(m, struct ip6_hdr *);
529 nth = (struct tcphdr *)(ip6 + 1);
530 } else
531#endif /* INET6 */
532 {
533 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
534 ip = mtod(m, struct ip *);
535 nth = (struct tcphdr *)(ip + 1);
536 }
537 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
538 flags = TH_ACK;
539 } else {
540 /*
541 * reuse the mbuf.
542 * XXX MRT We inherrit the FIB, which is lucky.
543 */
544 m_freem(m->m_next);
545 m->m_next = NULL;
546 m->m_data = (caddr_t)ipgen;
547 /* m_len is set later */
548 tlen = 0;
549#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
550#ifdef INET6
551 if (isipv6) {
552 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
553 nth = (struct tcphdr *)(ip6 + 1);
554 } else
555#endif /* INET6 */
556 {
557 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
558 nth = (struct tcphdr *)(ip + 1);
559 }
560 if (th != nth) {
561 /*
562 * this is usually a case when an extension header
563 * exists between the IPv6 header and the
564 * TCP header.
565 */
566 nth->th_sport = th->th_sport;
567 nth->th_dport = th->th_dport;
568 }
569 xchg(nth->th_dport, nth->th_sport, uint16_t);
570#undef xchg
571 }
572#ifdef INET6
573 if (isipv6) {
574 ip6->ip6_flow = 0;
575 ip6->ip6_vfc = IPV6_VERSION;
576 ip6->ip6_nxt = IPPROTO_TCP;
577 ip6->ip6_plen = 0; /* Set in ip6_output(). */
578 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
579 }
580#endif
581#if defined(INET) && defined(INET6)
582 else
583#endif
584#ifdef INET
585 {
586 tlen += sizeof (struct tcpiphdr);
587 ip->ip_len = tlen;
588 ip->ip_ttl = V_ip_defttl;
589 if (V_path_mtu_discovery)
590 ip->ip_off |= IP_DF;
591 }
592#endif
593 m->m_len = tlen;
594 m->m_pkthdr.len = tlen;
595 m->m_pkthdr.rcvif = NULL;
596#ifdef MAC
597 if (inp != NULL) {
598 /*
599 * Packet is associated with a socket, so allow the
600 * label of the response to reflect the socket label.
601 */
602 INP_WLOCK_ASSERT(inp);
603 mac_inpcb_create_mbuf(inp, m);
604 } else {
605 /*
606 * Packet is not associated with a socket, so possibly
607 * update the label in place.
608 */
609 mac_netinet_tcp_reply(m);
610 }
611#endif
612 nth->th_seq = htonl(seq);
613 nth->th_ack = htonl(ack);
614 nth->th_x2 = 0;
615 nth->th_off = sizeof (struct tcphdr) >> 2;
616 nth->th_flags = flags;
617 if (tp != NULL)
618 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
619 else
620 nth->th_win = htons((u_short)win);
621 nth->th_urp = 0;
622
623 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
624#ifdef INET6
625 if (isipv6) {
626 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
627 nth->th_sum = in6_cksum_pseudo(ip6,
628 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
629 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
630 NULL, NULL);
631 }
632#endif /* INET6 */
633#if defined(INET6) && defined(INET)
634 else
635#endif
636#ifdef INET
637 {
638 m->m_pkthdr.csum_flags = CSUM_TCP;
639 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
640 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
641 }
642#endif /* INET */
643#ifdef TCPDEBUG
644 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
645 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
646#endif
647#ifdef INET6
648 if (isipv6)
649 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
650#endif /* INET6 */
651#if defined(INET) && defined(INET6)
652 else
653#endif
654#ifdef INET
655 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
656#endif
657}
658
659/*
660 * Create a new TCP control block, making an
661 * empty reassembly queue and hooking it to the argument
662 * protocol control block. The `inp' parameter must have
663 * come from the zone allocator set up in tcp_init().
664 */
665struct tcpcb *
666tcp_newtcpcb(struct inpcb *inp)
667{
668 struct tcpcb_mem *tm;
669 struct tcpcb *tp;
670#ifdef INET6
671 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
672#endif /* INET6 */
673
674 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
675 if (tm == NULL)
676 return (NULL);
677 tp = &tm->tcb;
678
679 /* Initialise cc_var struct for this tcpcb. */
680 tp->ccv = &tm->ccv;
681 tp->ccv->type = IPPROTO_TCP;
682 tp->ccv->ccvc.tcp = tp;
683
684 /*
685 * Use the current system default CC algorithm.
686 */
687 CC_LIST_RLOCK();
688 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
689 CC_ALGO(tp) = CC_DEFAULT();
690 CC_LIST_RUNLOCK();
691
692 if (CC_ALGO(tp)->cb_init != NULL)
693 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
694 uma_zfree(V_tcpcb_zone, tm);
695 return (NULL);
696 }
697
698 tp->osd = &tm->osd;
699 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
700 uma_zfree(V_tcpcb_zone, tm);
701 return (NULL);
702 }
703
704#ifdef VIMAGE
705 tp->t_vnet = inp->inp_vnet;
706#endif
707 tp->t_timers = &tm->tt;
708 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
709 tp->t_maxseg = tp->t_maxopd =
710#ifdef INET6
711 isipv6 ? V_tcp_v6mssdflt :
712#endif /* INET6 */
713 V_tcp_mssdflt;
714
715 /* Set up our timeouts. */
716 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
717 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
718 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
719 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
720 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
721
722 if (V_tcp_do_rfc1323)
723 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
724 if (V_tcp_do_sack)
725 tp->t_flags |= TF_SACK_PERMIT;
726 TAILQ_INIT(&tp->snd_holes);
727 tp->t_inpcb = inp; /* XXX */
728 /*
729 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
730 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
731 * reasonable initial retransmit time.
732 */
733 tp->t_srtt = TCPTV_SRTTBASE;
734 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
735 tp->t_rttmin = tcp_rexmit_min;
736 tp->t_rxtcur = TCPTV_RTOBASE;
737 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
738 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
739 tp->t_rcvtime = ticks;
740 /*
741 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
742 * because the socket may be bound to an IPv6 wildcard address,
743 * which may match an IPv4-mapped IPv6 address.
744 */
745 inp->inp_ip_ttl = V_ip_defttl;
746 inp->inp_ppcb = tp;
747 return (tp); /* XXX */
748}
749
750/*
751 * Switch the congestion control algorithm back to NewReno for any active
752 * control blocks using an algorithm which is about to go away.
753 * This ensures the CC framework can allow the unload to proceed without leaving
754 * any dangling pointers which would trigger a panic.
755 * Returning non-zero would inform the CC framework that something went wrong
756 * and it would be unsafe to allow the unload to proceed. However, there is no
757 * way for this to occur with this implementation so we always return zero.
758 */
759int
760tcp_ccalgounload(struct cc_algo *unload_algo)
761{
762 struct cc_algo *tmpalgo;
763 struct inpcb *inp;
764 struct tcpcb *tp;
765 VNET_ITERATOR_DECL(vnet_iter);
766
767 /*
768 * Check all active control blocks across all network stacks and change
769 * any that are using "unload_algo" back to NewReno. If "unload_algo"
770 * requires cleanup code to be run, call it.
771 */
772 VNET_LIST_RLOCK();
773 VNET_FOREACH(vnet_iter) {
774 CURVNET_SET(vnet_iter);
775 INP_INFO_RLOCK(&V_tcbinfo);
776 /*
777 * New connections already part way through being initialised
778 * with the CC algo we're removing will not race with this code
779 * because the INP_INFO_WLOCK is held during initialisation. We
780 * therefore don't enter the loop below until the connection
781 * list has stabilised.
782 */
783 LIST_FOREACH(inp, &V_tcb, inp_list) {
784 INP_WLOCK(inp);
785 /* Important to skip tcptw structs. */
786 if (!(inp->inp_flags & INP_TIMEWAIT) &&
787 (tp = intotcpcb(inp)) != NULL) {
788 /*
789 * By holding INP_WLOCK here, we are assured
790 * that the connection is not currently
791 * executing inside the CC module's functions
792 * i.e. it is safe to make the switch back to
793 * NewReno.
794 */
795 if (CC_ALGO(tp) == unload_algo) {
796 tmpalgo = CC_ALGO(tp);
797 /* NewReno does not require any init. */
798 CC_ALGO(tp) = &newreno_cc_algo;
799 if (tmpalgo->cb_destroy != NULL)
800 tmpalgo->cb_destroy(tp->ccv);
801 }
802 }
803 INP_WUNLOCK(inp);
804 }
805 INP_INFO_RUNLOCK(&V_tcbinfo);
806 CURVNET_RESTORE();
807 }
808 VNET_LIST_RUNLOCK();
809
810 return (0);
811}
812
813/*
814 * Drop a TCP connection, reporting
815 * the specified error. If connection is synchronized,
816 * then send a RST to peer.
817 */
818struct tcpcb *
819tcp_drop(struct tcpcb *tp, int errno)
820{
821 struct socket *so = tp->t_inpcb->inp_socket;
822
823 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
824 INP_WLOCK_ASSERT(tp->t_inpcb);
825
826 if (TCPS_HAVERCVDSYN(tp->t_state)) {
827 tp->t_state = TCPS_CLOSED;
828 (void) tcp_output(tp);
829 TCPSTAT_INC(tcps_drops);
830 } else
831 TCPSTAT_INC(tcps_conndrops);
832 if (errno == ETIMEDOUT && tp->t_softerror)
833 errno = tp->t_softerror;
834 so->so_error = errno;
835 return (tcp_close(tp));
836}
837
838void
839tcp_discardcb(struct tcpcb *tp)
840{
841 struct inpcb *inp = tp->t_inpcb;
842 struct socket *so = inp->inp_socket;
843#ifdef INET6
844 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
845#endif /* INET6 */
846
847 INP_WLOCK_ASSERT(inp);
848
849 /*
850 * Make sure that all of our timers are stopped before we delete the
851 * PCB.
852 *
853 * XXXRW: Really, we would like to use callout_drain() here in order
854 * to avoid races experienced in tcp_timer.c where a timer is already
855 * executing at this point. However, we can't, both because we're
856 * running in a context where we can't sleep, and also because we
857 * hold locks required by the timers. What we instead need to do is
858 * test to see if callout_drain() is required, and if so, defer some
859 * portion of the remainder of tcp_discardcb() to an asynchronous
860 * context that can callout_drain() and then continue. Some care
861 * will be required to ensure that no further processing takes place
862 * on the tcpcb, even though it hasn't been freed (a flag?).
863 */
864 callout_stop(&tp->t_timers->tt_rexmt);
865 callout_stop(&tp->t_timers->tt_persist);
866 callout_stop(&tp->t_timers->tt_keep);
867 callout_stop(&tp->t_timers->tt_2msl);
868 callout_stop(&tp->t_timers->tt_delack);
869
870 /*
871 * If we got enough samples through the srtt filter,
872 * save the rtt and rttvar in the routing entry.
873 * 'Enough' is arbitrarily defined as 4 rtt samples.
874 * 4 samples is enough for the srtt filter to converge
875 * to within enough % of the correct value; fewer samples
876 * and we could save a bogus rtt. The danger is not high
877 * as tcp quickly recovers from everything.
878 * XXX: Works very well but needs some more statistics!
879 */
880 if (tp->t_rttupdated >= 4) {
881 struct hc_metrics_lite metrics;
882 u_long ssthresh;
883
884 bzero(&metrics, sizeof(metrics));
885 /*
886 * Update the ssthresh always when the conditions below
887 * are satisfied. This gives us better new start value
888 * for the congestion avoidance for new connections.
889 * ssthresh is only set if packet loss occured on a session.
890 *
891 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
892 * being torn down. Ideally this code would not use 'so'.
893 */
894 ssthresh = tp->snd_ssthresh;
895 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
896 /*
897 * convert the limit from user data bytes to
898 * packets then to packet data bytes.
899 */
900 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
901 if (ssthresh < 2)
902 ssthresh = 2;
903 ssthresh *= (u_long)(tp->t_maxseg +
904#ifdef INET6
905 (isipv6 ? sizeof (struct ip6_hdr) +
906 sizeof (struct tcphdr) :
907#endif
908 sizeof (struct tcpiphdr)
909#ifdef INET6
910 )
911#endif
912 );
913 } else
914 ssthresh = 0;
915 metrics.rmx_ssthresh = ssthresh;
916
917 metrics.rmx_rtt = tp->t_srtt;
918 metrics.rmx_rttvar = tp->t_rttvar;
919 metrics.rmx_cwnd = tp->snd_cwnd;
920 metrics.rmx_sendpipe = 0;
921 metrics.rmx_recvpipe = 0;
922
923 tcp_hc_update(&inp->inp_inc, &metrics);
924 }
925
926 /* free the reassembly queue, if any */
927 tcp_reass_flush(tp);
928
929#ifdef TCP_OFFLOAD
930 /* Disconnect offload device, if any. */
931 if (tp->t_flags & TF_TOE)
932 tcp_offload_detach(tp);
933#endif
934
935 tcp_free_sackholes(tp);
936
937 /* Allow the CC algorithm to clean up after itself. */
938 if (CC_ALGO(tp)->cb_destroy != NULL)
939 CC_ALGO(tp)->cb_destroy(tp->ccv);
940
941 khelp_destroy_osd(tp->osd);
942
943 CC_ALGO(tp) = NULL;
944 inp->inp_ppcb = NULL;
945 tp->t_inpcb = NULL;
946 uma_zfree(V_tcpcb_zone, tp);
947}
948
949/*
950 * Attempt to close a TCP control block, marking it as dropped, and freeing
951 * the socket if we hold the only reference.
952 */
953struct tcpcb *
954tcp_close(struct tcpcb *tp)
955{
956 struct inpcb *inp = tp->t_inpcb;
957 struct socket *so;
958
959 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
960 INP_WLOCK_ASSERT(inp);
961
962#ifdef TCP_OFFLOAD
963 if (tp->t_state == TCPS_LISTEN)
964 tcp_offload_listen_stop(tp);
965#endif
966 in_pcbdrop(inp);
967 TCPSTAT_INC(tcps_closed);
968 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
969 so = inp->inp_socket;
970 soisdisconnected(so);
971 if (inp->inp_flags & INP_SOCKREF) {
972 KASSERT(so->so_state & SS_PROTOREF,
973 ("tcp_close: !SS_PROTOREF"));
974 inp->inp_flags &= ~INP_SOCKREF;
975 INP_WUNLOCK(inp);
976 ACCEPT_LOCK();
977 SOCK_LOCK(so);
978 so->so_state &= ~SS_PROTOREF;
979 sofree(so);
980 return (NULL);
981 }
982 return (tp);
983}
984
985void
986tcp_drain(void)
987{
988 VNET_ITERATOR_DECL(vnet_iter);
989
990 if (!do_tcpdrain)
991 return;
992
993 VNET_LIST_RLOCK_NOSLEEP();
994 VNET_FOREACH(vnet_iter) {
995 CURVNET_SET(vnet_iter);
996 struct inpcb *inpb;
997 struct tcpcb *tcpb;
998
999 /*
1000 * Walk the tcpbs, if existing, and flush the reassembly queue,
1001 * if there is one...
1002 * XXX: The "Net/3" implementation doesn't imply that the TCP
1003 * reassembly queue should be flushed, but in a situation
1004 * where we're really low on mbufs, this is potentially
1005 * usefull.
1006 */
1007 INP_INFO_RLOCK(&V_tcbinfo);
1008 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1009 if (inpb->inp_flags & INP_TIMEWAIT)
1010 continue;
1011 INP_WLOCK(inpb);
1012 if ((tcpb = intotcpcb(inpb)) != NULL) {
1013 tcp_reass_flush(tcpb);
1014 tcp_clean_sackreport(tcpb);
1015 }
1016 INP_WUNLOCK(inpb);
1017 }
1018 INP_INFO_RUNLOCK(&V_tcbinfo);
1019 CURVNET_RESTORE();
1020 }
1021 VNET_LIST_RUNLOCK_NOSLEEP();
1022}
1023
1024/*
1025 * Notify a tcp user of an asynchronous error;
1026 * store error as soft error, but wake up user
1027 * (for now, won't do anything until can select for soft error).
1028 *
1029 * Do not wake up user since there currently is no mechanism for
1030 * reporting soft errors (yet - a kqueue filter may be added).
1031 */
1032static struct inpcb *
1033tcp_notify(struct inpcb *inp, int error)
1034{
1035 struct tcpcb *tp;
1036
1037 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1038 INP_WLOCK_ASSERT(inp);
1039
1040 if ((inp->inp_flags & INP_TIMEWAIT) ||
1041 (inp->inp_flags & INP_DROPPED))
1042 return (inp);
1043
1044 tp = intotcpcb(inp);
1045 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1046
1047 /*
1048 * Ignore some errors if we are hooked up.
1049 * If connection hasn't completed, has retransmitted several times,
1050 * and receives a second error, give up now. This is better
1051 * than waiting a long time to establish a connection that
1052 * can never complete.
1053 */
1054 if (tp->t_state == TCPS_ESTABLISHED &&
1055 (error == EHOSTUNREACH || error == ENETUNREACH ||
1056 error == EHOSTDOWN)) {
1057 return (inp);
1058 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1059 tp->t_softerror) {
1060 tp = tcp_drop(tp, error);
1061 if (tp != NULL)
1062 return (inp);
1063 else
1064 return (NULL);
1065 } else {
1066 tp->t_softerror = error;
1067 return (inp);
1068 }
1069#if 0
1070 wakeup( &so->so_timeo);
1071 sorwakeup(so);
1072 sowwakeup(so);
1073#endif
1074}
1075
1076static int
1077tcp_pcblist(SYSCTL_HANDLER_ARGS)
1078{
1079 int error, i, m, n, pcb_count;
1080 struct inpcb *inp, **inp_list;
1081 inp_gen_t gencnt;
1082 struct xinpgen xig;
1083
1084 /*
1085 * The process of preparing the TCB list is too time-consuming and
1086 * resource-intensive to repeat twice on every request.
1087 */
1088 if (req->oldptr == NULL) {
1089 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1090 n += imax(n / 8, 10);
1091 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1092 return (0);
1093 }
1094
1095 if (req->newptr != NULL)
1096 return (EPERM);
1097
1098 /*
1099 * OK, now we're committed to doing something.
1100 */
1101 INP_INFO_RLOCK(&V_tcbinfo);
1102 gencnt = V_tcbinfo.ipi_gencnt;
1103 n = V_tcbinfo.ipi_count;
1104 INP_INFO_RUNLOCK(&V_tcbinfo);
1105
1106 m = syncache_pcbcount();
1107
1108 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1109 + (n + m) * sizeof(struct xtcpcb));
1110 if (error != 0)
1111 return (error);
1112
1113 xig.xig_len = sizeof xig;
1114 xig.xig_count = n + m;
1115 xig.xig_gen = gencnt;
1116 xig.xig_sogen = so_gencnt;
1117 error = SYSCTL_OUT(req, &xig, sizeof xig);
1118 if (error)
1119 return (error);
1120
1121 error = syncache_pcblist(req, m, &pcb_count);
1122 if (error)
1123 return (error);
1124
1125 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1126 if (inp_list == NULL)
1127 return (ENOMEM);
1128
1129 INP_INFO_RLOCK(&V_tcbinfo);
1130 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1131 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1132 INP_WLOCK(inp);
1133 if (inp->inp_gencnt <= gencnt) {
1134 /*
1135 * XXX: This use of cr_cansee(), introduced with
1136 * TCP state changes, is not quite right, but for
1137 * now, better than nothing.
1138 */
1139 if (inp->inp_flags & INP_TIMEWAIT) {
1140 if (intotw(inp) != NULL)
1141 error = cr_cansee(req->td->td_ucred,
1142 intotw(inp)->tw_cred);
1143 else
1144 error = EINVAL; /* Skip this inp. */
1145 } else
1146 error = cr_canseeinpcb(req->td->td_ucred, inp);
1147 if (error == 0) {
1148 in_pcbref(inp);
1149 inp_list[i++] = inp;
1150 }
1151 }
1152 INP_WUNLOCK(inp);
1153 }
1154 INP_INFO_RUNLOCK(&V_tcbinfo);
1155 n = i;
1156
1157 error = 0;
1158 for (i = 0; i < n; i++) {
1159 inp = inp_list[i];
1160 INP_RLOCK(inp);
1161 if (inp->inp_gencnt <= gencnt) {
1162 struct xtcpcb xt;
1163 void *inp_ppcb;
1164
1165 bzero(&xt, sizeof(xt));
1166 xt.xt_len = sizeof xt;
1167 /* XXX should avoid extra copy */
1168 bcopy(inp, &xt.xt_inp, sizeof *inp);
1169 inp_ppcb = inp->inp_ppcb;
1170 if (inp_ppcb == NULL)
1171 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1172 else if (inp->inp_flags & INP_TIMEWAIT) {
1173 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1174 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1175 } else {
1176 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1177 if (xt.xt_tp.t_timers)
1178 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1179 }
1180 if (inp->inp_socket != NULL)
1181 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1182 else {
1183 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1184 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1185 }
1186 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1187 INP_RUNLOCK(inp);
1188 error = SYSCTL_OUT(req, &xt, sizeof xt);
1189 } else
1190 INP_RUNLOCK(inp);
1191 }
1192 INP_INFO_WLOCK(&V_tcbinfo);
1193 for (i = 0; i < n; i++) {
1194 inp = inp_list[i];
1195 INP_RLOCK(inp);
1196 if (!in_pcbrele_rlocked(inp))
1197 INP_RUNLOCK(inp);
1198 }
1199 INP_INFO_WUNLOCK(&V_tcbinfo);
1200
1201 if (!error) {
1202 /*
1203 * Give the user an updated idea of our state.
1204 * If the generation differs from what we told
1205 * her before, she knows that something happened
1206 * while we were processing this request, and it
1207 * might be necessary to retry.
1208 */
1209 INP_INFO_RLOCK(&V_tcbinfo);
1210 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1211 xig.xig_sogen = so_gencnt;
1212 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1213 INP_INFO_RUNLOCK(&V_tcbinfo);
1214 error = SYSCTL_OUT(req, &xig, sizeof xig);
1215 }
1216 free(inp_list, M_TEMP);
1217 return (error);
1218}
1219
1220SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1221 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1222 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1223
1224#ifdef INET
1225static int
1226tcp_getcred(SYSCTL_HANDLER_ARGS)
1227{
1228 struct xucred xuc;
1229 struct sockaddr_in addrs[2];
1230 struct inpcb *inp;
1231 int error;
1232
1233 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1234 if (error)
1235 return (error);
1236 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1237 if (error)
1238 return (error);
1239 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1240 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1241 if (inp != NULL) {
1242 if (inp->inp_socket == NULL)
1243 error = ENOENT;
1244 if (error == 0)
1245 error = cr_canseeinpcb(req->td->td_ucred, inp);
1246 if (error == 0)
1247 cru2x(inp->inp_cred, &xuc);
1248 INP_RUNLOCK(inp);
1249 } else
1250 error = ENOENT;
1251 if (error == 0)
1252 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1253 return (error);
1254}
1255
1256SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1257 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1258 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1259#endif /* INET */
1260
1261#ifdef INET6
1262static int
1263tcp6_getcred(SYSCTL_HANDLER_ARGS)
1264{
1265 struct xucred xuc;
1266 struct sockaddr_in6 addrs[2];
1267 struct inpcb *inp;
1268 int error;
1269#ifdef INET
1270 int mapped = 0;
1271#endif
1272
1273 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1274 if (error)
1275 return (error);
1276 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1277 if (error)
1278 return (error);
1279 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1280 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1281 return (error);
1282 }
1283 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1284#ifdef INET
1285 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1286 mapped = 1;
1287 else
1288#endif
1289 return (EINVAL);
1290 }
1291
1292#ifdef INET
1293 if (mapped == 1)
1294 inp = in_pcblookup(&V_tcbinfo,
1295 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1296 addrs[1].sin6_port,
1297 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1298 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1299 else
1300#endif
1301 inp = in6_pcblookup(&V_tcbinfo,
1302 &addrs[1].sin6_addr, addrs[1].sin6_port,
1303 &addrs[0].sin6_addr, addrs[0].sin6_port,
1304 INPLOOKUP_RLOCKPCB, NULL);
1305 if (inp != NULL) {
1306 if (inp->inp_socket == NULL)
1307 error = ENOENT;
1308 if (error == 0)
1309 error = cr_canseeinpcb(req->td->td_ucred, inp);
1310 if (error == 0)
1311 cru2x(inp->inp_cred, &xuc);
1312 INP_RUNLOCK(inp);
1313 } else
1314 error = ENOENT;
1315 if (error == 0)
1316 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1317 return (error);
1318}
1319
1320SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1321 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1322 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1323#endif /* INET6 */
1324
1325
1326#ifdef INET
1327void
1328tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1329{
1330 struct ip *ip = vip;
1331 struct tcphdr *th;
1332 struct in_addr faddr;
1333 struct inpcb *inp;
1334 struct tcpcb *tp;
1335 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1336 struct icmp *icp;
1337 struct in_conninfo inc;
1338 tcp_seq icmp_tcp_seq;
1339 int mtu;
1340
1341 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1342 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1343 return;
1344
1345 if (cmd == PRC_MSGSIZE)
1346 notify = tcp_mtudisc_notify;
1347 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1348 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1349 notify = tcp_drop_syn_sent;
1350 /*
1351 * Redirects don't need to be handled up here.
1352 */
1353 else if (PRC_IS_REDIRECT(cmd))
1354 return;
1355 /*
1356 * Source quench is depreciated.
1357 */
1358 else if (cmd == PRC_QUENCH)
1359 return;
1360 /*
1361 * Hostdead is ugly because it goes linearly through all PCBs.
1362 * XXX: We never get this from ICMP, otherwise it makes an
1363 * excellent DoS attack on machines with many connections.
1364 */
1365 else if (cmd == PRC_HOSTDEAD)
1366 ip = NULL;
1367 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1368 return;
1369 if (ip != NULL) {
1370 icp = (struct icmp *)((caddr_t)ip
1371 - offsetof(struct icmp, icmp_ip));
1372 th = (struct tcphdr *)((caddr_t)ip
1373 + (ip->ip_hl << 2));
1374 INP_INFO_WLOCK(&V_tcbinfo);
1375 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1376 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1377 if (inp != NULL) {
1378 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1379 !(inp->inp_flags & INP_DROPPED) &&
1380 !(inp->inp_socket == NULL)) {
1381 icmp_tcp_seq = htonl(th->th_seq);
1382 tp = intotcpcb(inp);
1383 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1384 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1385 if (cmd == PRC_MSGSIZE) {
1386 /*
1387 * MTU discovery:
1388 * If we got a needfrag set the MTU
1389 * in the route to the suggested new
1390 * value (if given) and then notify.
1391 */
1392 bzero(&inc, sizeof(inc));
1393 inc.inc_faddr = faddr;
1394 inc.inc_fibnum =
1395 inp->inp_inc.inc_fibnum;
1396
1397 mtu = ntohs(icp->icmp_nextmtu);
1398 /*
1399 * If no alternative MTU was
1400 * proposed, try the next smaller
1401 * one. ip->ip_len has already
1402 * been swapped in icmp_input().
1403 */
1404 if (!mtu)
1405 mtu = ip_next_mtu(ip->ip_len,
1406 1);
1407 if (mtu < V_tcp_minmss
1408 + sizeof(struct tcpiphdr))
1409 mtu = V_tcp_minmss
1410 + sizeof(struct tcpiphdr);
1411 /*
1412 * Only cache the MTU if it
1413 * is smaller than the interface
1414 * or route MTU. tcp_mtudisc()
1415 * will do right thing by itself.
1416 */
1417 if (mtu <= tcp_maxmtu(&inc, NULL))
1418 tcp_hc_updatemtu(&inc, mtu);
1419 tcp_mtudisc(inp, mtu);
1420 } else
1421 inp = (*notify)(inp,
1422 inetctlerrmap[cmd]);
1423 }
1424 }
1425 if (inp != NULL)
1426 INP_WUNLOCK(inp);
1427 } else {
1428 bzero(&inc, sizeof(inc));
1429 inc.inc_fport = th->th_dport;
1430 inc.inc_lport = th->th_sport;
1431 inc.inc_faddr = faddr;
1432 inc.inc_laddr = ip->ip_src;
1433 syncache_unreach(&inc, th);
1434 }
1435 INP_INFO_WUNLOCK(&V_tcbinfo);
1436 } else
1437 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1438}
1439#endif /* INET */
1440
1441#ifdef INET6
1442void
1443tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1444{
1445 struct tcphdr th;
1446 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1447 struct ip6_hdr *ip6;
1448 struct mbuf *m;
1449 struct ip6ctlparam *ip6cp = NULL;
1450 const struct sockaddr_in6 *sa6_src = NULL;
1451 int off;
1452 struct tcp_portonly {
1453 u_int16_t th_sport;
1454 u_int16_t th_dport;
1455 } *thp;
1456
1457 if (sa->sa_family != AF_INET6 ||
1458 sa->sa_len != sizeof(struct sockaddr_in6))
1459 return;
1460
1461 if (cmd == PRC_MSGSIZE)
1462 notify = tcp_mtudisc_notify;
1463 else if (!PRC_IS_REDIRECT(cmd) &&
1464 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1465 return;
1466 /* Source quench is depreciated. */
1467 else if (cmd == PRC_QUENCH)
1468 return;
1469
1470 /* if the parameter is from icmp6, decode it. */
1471 if (d != NULL) {
1472 ip6cp = (struct ip6ctlparam *)d;
1473 m = ip6cp->ip6c_m;
1474 ip6 = ip6cp->ip6c_ip6;
1475 off = ip6cp->ip6c_off;
1476 sa6_src = ip6cp->ip6c_src;
1477 } else {
1478 m = NULL;
1479 ip6 = NULL;
1480 off = 0; /* fool gcc */
1481 sa6_src = &sa6_any;
1482 }
1483
1484 if (ip6 != NULL) {
1485 struct in_conninfo inc;
1486 /*
1487 * XXX: We assume that when IPV6 is non NULL,
1488 * M and OFF are valid.
1489 */
1490
1491 /* check if we can safely examine src and dst ports */
1492 if (m->m_pkthdr.len < off + sizeof(*thp))
1493 return;
1494
1495 bzero(&th, sizeof(th));
1496 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1497
1498 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1499 (struct sockaddr *)ip6cp->ip6c_src,
1500 th.th_sport, cmd, NULL, notify);
1501
1502 bzero(&inc, sizeof(inc));
1503 inc.inc_fport = th.th_dport;
1504 inc.inc_lport = th.th_sport;
1505 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1506 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1507 inc.inc_flags |= INC_ISIPV6;
1508 INP_INFO_WLOCK(&V_tcbinfo);
1509 syncache_unreach(&inc, &th);
1510 INP_INFO_WUNLOCK(&V_tcbinfo);
1511 } else
1512 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1513 0, cmd, NULL, notify);
1514}
1515#endif /* INET6 */
1516
1517
1518/*
1519 * Following is where TCP initial sequence number generation occurs.
1520 *
1521 * There are two places where we must use initial sequence numbers:
1522 * 1. In SYN-ACK packets.
1523 * 2. In SYN packets.
1524 *
1525 * All ISNs for SYN-ACK packets are generated by the syncache. See
1526 * tcp_syncache.c for details.
1527 *
1528 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1529 * depends on this property. In addition, these ISNs should be
1530 * unguessable so as to prevent connection hijacking. To satisfy
1531 * the requirements of this situation, the algorithm outlined in
1532 * RFC 1948 is used, with only small modifications.
1533 *
1534 * Implementation details:
1535 *
1536 * Time is based off the system timer, and is corrected so that it
1537 * increases by one megabyte per second. This allows for proper
1538 * recycling on high speed LANs while still leaving over an hour
1539 * before rollover.
1540 *
1541 * As reading the *exact* system time is too expensive to be done
1542 * whenever setting up a TCP connection, we increment the time
1543 * offset in two ways. First, a small random positive increment
1544 * is added to isn_offset for each connection that is set up.
1545 * Second, the function tcp_isn_tick fires once per clock tick
1546 * and increments isn_offset as necessary so that sequence numbers
1547 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1548 * random positive increments serve only to ensure that the same
1549 * exact sequence number is never sent out twice (as could otherwise
1550 * happen when a port is recycled in less than the system tick
1551 * interval.)
1552 *
1553 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1554 * between seeding of isn_secret. This is normally set to zero,
1555 * as reseeding should not be necessary.
1556 *
1557 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1558 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1559 * general, this means holding an exclusive (write) lock.
1560 */
1561
1562#define ISN_BYTES_PER_SECOND 1048576
1563#define ISN_STATIC_INCREMENT 4096
1564#define ISN_RANDOM_INCREMENT (4096 - 1)
1565
1566static VNET_DEFINE(u_char, isn_secret[32]);
1567static VNET_DEFINE(int, isn_last);
1568static VNET_DEFINE(int, isn_last_reseed);
1569static VNET_DEFINE(u_int32_t, isn_offset);
1570static VNET_DEFINE(u_int32_t, isn_offset_old);
1571
1572#define V_isn_secret VNET(isn_secret)
1573#define V_isn_last VNET(isn_last)
1574#define V_isn_last_reseed VNET(isn_last_reseed)
1575#define V_isn_offset VNET(isn_offset)
1576#define V_isn_offset_old VNET(isn_offset_old)
1577
1578tcp_seq
1579tcp_new_isn(struct tcpcb *tp)
1580{
1581 MD5_CTX isn_ctx;
1582 u_int32_t md5_buffer[4];
1583 tcp_seq new_isn;
1584 u_int32_t projected_offset;
1585
1586 INP_WLOCK_ASSERT(tp->t_inpcb);
1587
1588 ISN_LOCK();
1589 /* Seed if this is the first use, reseed if requested. */
1590 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1591 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1592 < (u_int)ticks))) {
1593 read_random(&V_isn_secret, sizeof(V_isn_secret));
1594 V_isn_last_reseed = ticks;
1595 }
1596
1597 /* Compute the md5 hash and return the ISN. */
1598 MD5Init(&isn_ctx);
1599 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1600 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1601#ifdef INET6
1602 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1603 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1604 sizeof(struct in6_addr));
1605 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1606 sizeof(struct in6_addr));
1607 } else
1608#endif
1609 {
1610 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1611 sizeof(struct in_addr));
1612 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1613 sizeof(struct in_addr));
1614 }
1615 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1616 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1617 new_isn = (tcp_seq) md5_buffer[0];
1618 V_isn_offset += ISN_STATIC_INCREMENT +
1619 (arc4random() & ISN_RANDOM_INCREMENT);
1620 if (ticks != V_isn_last) {
1621 projected_offset = V_isn_offset_old +
1622 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1623 if (SEQ_GT(projected_offset, V_isn_offset))
1624 V_isn_offset = projected_offset;
1625 V_isn_offset_old = V_isn_offset;
1626 V_isn_last = ticks;
1627 }
1628 new_isn += V_isn_offset;
1629 ISN_UNLOCK();
1630 return (new_isn);
1631}
1632
1633/*
1634 * When a specific ICMP unreachable message is received and the
1635 * connection state is SYN-SENT, drop the connection. This behavior
1636 * is controlled by the icmp_may_rst sysctl.
1637 */
1638struct inpcb *
1639tcp_drop_syn_sent(struct inpcb *inp, int errno)
1640{
1641 struct tcpcb *tp;
1642
1643 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1644 INP_WLOCK_ASSERT(inp);
1645
1646 if ((inp->inp_flags & INP_TIMEWAIT) ||
1647 (inp->inp_flags & INP_DROPPED))
1648 return (inp);
1649
1650 tp = intotcpcb(inp);
1651 if (tp->t_state != TCPS_SYN_SENT)
1652 return (inp);
1653
1654 tp = tcp_drop(tp, errno);
1655 if (tp != NULL)
1656 return (inp);
1657 else
1658 return (NULL);
1659}
1660
1661/*
1662 * When `need fragmentation' ICMP is received, update our idea of the MSS
1663 * based on the new value. Also nudge TCP to send something, since we
1664 * know the packet we just sent was dropped.
1665 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1666 */
1667static struct inpcb *
1668tcp_mtudisc_notify(struct inpcb *inp, int error)
1669{
1670
1671 return (tcp_mtudisc(inp, -1));
1672}
1673
1674struct inpcb *
1675tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1676{
1677 struct tcpcb *tp;
1678 struct socket *so;
1679
1680 INP_WLOCK_ASSERT(inp);
1681 if ((inp->inp_flags & INP_TIMEWAIT) ||
1682 (inp->inp_flags & INP_DROPPED))
1683 return (inp);
1684
1685 tp = intotcpcb(inp);
1686 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1687
1688 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1689
1690 so = inp->inp_socket;
1691 SOCKBUF_LOCK(&so->so_snd);
1692 /* If the mss is larger than the socket buffer, decrease the mss. */
1693 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1694 tp->t_maxseg = so->so_snd.sb_hiwat;
1695 SOCKBUF_UNLOCK(&so->so_snd);
1696
1697 TCPSTAT_INC(tcps_mturesent);
1698 tp->t_rtttime = 0;
1699 tp->snd_nxt = tp->snd_una;
1700 tcp_free_sackholes(tp);
1701 tp->snd_recover = tp->snd_max;
1702 if (tp->t_flags & TF_SACK_PERMIT)
1703 EXIT_FASTRECOVERY(tp->t_flags);
1704 tcp_output(tp);
1705 return (inp);
1706}
1707
1708#ifdef INET
1709/*
1710 * Look-up the routing entry to the peer of this inpcb. If no route
1711 * is found and it cannot be allocated, then return 0. This routine
1712 * is called by TCP routines that access the rmx structure and by
1713 * tcp_mss_update to get the peer/interface MTU.
1714 */
1715u_long
1716tcp_maxmtu(struct in_conninfo *inc, int *flags)
1717{
1718 struct route sro;
1719 struct sockaddr_in *dst;
1720 struct ifnet *ifp;
1721 u_long maxmtu = 0;
1722
1723 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1724
1725 bzero(&sro, sizeof(sro));
1726 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1727 dst = (struct sockaddr_in *)&sro.ro_dst;
1728 dst->sin_family = AF_INET;
1729 dst->sin_len = sizeof(*dst);
1730 dst->sin_addr = inc->inc_faddr;
1731 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1732 }
1733 if (sro.ro_rt != NULL) {
1734 ifp = sro.ro_rt->rt_ifp;
1735 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1736 maxmtu = ifp->if_mtu;
1737 else
1738 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1739
1740 /* Report additional interface capabilities. */
1741 if (flags != NULL) {
1742 if (ifp->if_capenable & IFCAP_TSO4 &&
1743 ifp->if_hwassist & CSUM_TSO)
1744 *flags |= CSUM_TSO;
1745 }
1746 RTFREE(sro.ro_rt);
1747 }
1748 return (maxmtu);
1749}
1750#endif /* INET */
1751
1752#ifdef INET6
1753u_long
1754tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1755{
1756 struct route_in6 sro6;
1757 struct ifnet *ifp;
1758 u_long maxmtu = 0;
1759
1760 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1761
1762 bzero(&sro6, sizeof(sro6));
1763 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1764 sro6.ro_dst.sin6_family = AF_INET6;
1765 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1766 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1767 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1768 }
1769 if (sro6.ro_rt != NULL) {
1770 ifp = sro6.ro_rt->rt_ifp;
1771 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1772 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1773 else
1774 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1775 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1776
1777 /* Report additional interface capabilities. */
1778 if (flags != NULL) {
1779 if (ifp->if_capenable & IFCAP_TSO6 &&
1780 ifp->if_hwassist & CSUM_TSO)
1781 *flags |= CSUM_TSO;
1782 }
1783 RTFREE(sro6.ro_rt);
1784 }
1785
1786 return (maxmtu);
1787}
1788#endif /* INET6 */
1789
1790#ifdef IPSEC
1791/* compute ESP/AH header size for TCP, including outer IP header. */
1792size_t
1793ipsec_hdrsiz_tcp(struct tcpcb *tp)
1794{
1795 struct inpcb *inp;
1796 struct mbuf *m;
1797 size_t hdrsiz;
1798 struct ip *ip;
1799#ifdef INET6
1800 struct ip6_hdr *ip6;
1801#endif
1802 struct tcphdr *th;
1803
1804 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1805 return (0);
1806 MGETHDR(m, M_DONTWAIT, MT_DATA);
1807 if (!m)
1808 return (0);
1809
1810#ifdef INET6
1811 if ((inp->inp_vflag & INP_IPV6) != 0) {
1812 ip6 = mtod(m, struct ip6_hdr *);
1813 th = (struct tcphdr *)(ip6 + 1);
1814 m->m_pkthdr.len = m->m_len =
1815 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1816 tcpip_fillheaders(inp, ip6, th);
1817 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1818 } else
1819#endif /* INET6 */
1820 {
1821 ip = mtod(m, struct ip *);
1822 th = (struct tcphdr *)(ip + 1);
1823 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1824 tcpip_fillheaders(inp, ip, th);
1825 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1826 }
1827
1828 m_free(m);
1829 return (hdrsiz);
1830}
1831#endif /* IPSEC */
1832
1833#ifdef TCP_SIGNATURE
1834/*
1835 * Callback function invoked by m_apply() to digest TCP segment data
1836 * contained within an mbuf chain.
1837 */
1838static int
1839tcp_signature_apply(void *fstate, void *data, u_int len)
1840{
1841
1842 MD5Update(fstate, (u_char *)data, len);
1843 return (0);
1844}
1845
1846/*
1847 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1848 *
1849 * Parameters:
1850 * m pointer to head of mbuf chain
1851 * _unused
1852 * len length of TCP segment data, excluding options
1853 * optlen length of TCP segment options
1854 * buf pointer to storage for computed MD5 digest
1855 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1856 *
1857 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1858 * When called from tcp_input(), we can be sure that th_sum has been
1859 * zeroed out and verified already.
1860 *
1861 * Return 0 if successful, otherwise return -1.
1862 *
1863 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1864 * search with the destination IP address, and a 'magic SPI' to be
1865 * determined by the application. This is hardcoded elsewhere to 1179
1866 * right now. Another branch of this code exists which uses the SPD to
1867 * specify per-application flows but it is unstable.
1868 */
1869int
1870tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1871 u_char *buf, u_int direction)
1872{
1873 union sockaddr_union dst;
1874#ifdef INET
1875 struct ippseudo ippseudo;
1876#endif
1877 MD5_CTX ctx;
1878 int doff;
1879 struct ip *ip;
1880#ifdef INET
1881 struct ipovly *ipovly;
1882#endif
1883 struct secasvar *sav;
1884 struct tcphdr *th;
1885#ifdef INET6
1886 struct ip6_hdr *ip6;
1887 struct in6_addr in6;
1888 char ip6buf[INET6_ADDRSTRLEN];
1889 uint32_t plen;
1890 uint16_t nhdr;
1891#endif
1892 u_short savecsum;
1893
1894 KASSERT(m != NULL, ("NULL mbuf chain"));
1895 KASSERT(buf != NULL, ("NULL signature pointer"));
1896
1897 /* Extract the destination from the IP header in the mbuf. */
1898 bzero(&dst, sizeof(union sockaddr_union));
1899 ip = mtod(m, struct ip *);
1900#ifdef INET6
1901 ip6 = NULL; /* Make the compiler happy. */
1902#endif
1903 switch (ip->ip_v) {
1904#ifdef INET
1905 case IPVERSION:
1906 dst.sa.sa_len = sizeof(struct sockaddr_in);
1907 dst.sa.sa_family = AF_INET;
1908 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1909 ip->ip_src : ip->ip_dst;
1910 break;
1911#endif
1912#ifdef INET6
1913 case (IPV6_VERSION >> 4):
1914 ip6 = mtod(m, struct ip6_hdr *);
1915 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1916 dst.sa.sa_family = AF_INET6;
1917 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1918 ip6->ip6_src : ip6->ip6_dst;
1919 break;
1920#endif
1921 default:
1922 return (EINVAL);
1923 /* NOTREACHED */
1924 break;
1925 }
1926
1927 /* Look up an SADB entry which matches the address of the peer. */
1928 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1929 if (sav == NULL) {
1930 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1931 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1932#ifdef INET6
1933 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1934 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1935#endif
1936 "(unsupported)"));
1937 return (EINVAL);
1938 }
1939
1940 MD5Init(&ctx);
1941 /*
1942 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1943 *
1944 * XXX The ippseudo header MUST be digested in network byte order,
1945 * or else we'll fail the regression test. Assume all fields we've
1946 * been doing arithmetic on have been in host byte order.
1947 * XXX One cannot depend on ipovly->ih_len here. When called from
1948 * tcp_output(), the underlying ip_len member has not yet been set.
1949 */
1950 switch (ip->ip_v) {
1951#ifdef INET
1952 case IPVERSION:
1953 ipovly = (struct ipovly *)ip;
1954 ippseudo.ippseudo_src = ipovly->ih_src;
1955 ippseudo.ippseudo_dst = ipovly->ih_dst;
1956 ippseudo.ippseudo_pad = 0;
1957 ippseudo.ippseudo_p = IPPROTO_TCP;
1958 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1959 optlen);
1960 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1961
1962 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1963 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1964 break;
1965#endif
1966#ifdef INET6
1967 /*
1968 * RFC 2385, 2.0 Proposal
1969 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1970 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1971 * extended next header value (to form 32 bits), and 32-bit segment
1972 * length.
1973 * Note: Upper-Layer Packet Length comes before Next Header.
1974 */
1975 case (IPV6_VERSION >> 4):
1976 in6 = ip6->ip6_src;
1977 in6_clearscope(&in6);
1978 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1979 in6 = ip6->ip6_dst;
1980 in6_clearscope(&in6);
1981 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1982 plen = htonl(len + sizeof(struct tcphdr) + optlen);
1983 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1984 nhdr = 0;
1985 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1986 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1987 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1988 nhdr = IPPROTO_TCP;
1989 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1990
1991 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1992 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1993 break;
1994#endif
1995 default:
1996 return (EINVAL);
1997 /* NOTREACHED */
1998 break;
1999 }
2000
2001
2002 /*
2003 * Step 2: Update MD5 hash with TCP header, excluding options.
2004 * The TCP checksum must be set to zero.
2005 */
2006 savecsum = th->th_sum;
2007 th->th_sum = 0;
2008 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2009 th->th_sum = savecsum;
2010
2011 /*
2012 * Step 3: Update MD5 hash with TCP segment data.
2013 * Use m_apply() to avoid an early m_pullup().
2014 */
2015 if (len > 0)
2016 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2017
2018 /*
2019 * Step 4: Update MD5 hash with shared secret.
2020 */
2021 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2022 MD5Final(buf, &ctx);
2023
2024 key_sa_recordxfer(sav, m);
2025 KEY_FREESAV(&sav);
2026 return (0);
2027}
2028
2029/*
2030 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2031 *
2032 * Parameters:
2033 * m pointer to head of mbuf chain
2034 * len length of TCP segment data, excluding options
2035 * optlen length of TCP segment options
2036 * buf pointer to storage for computed MD5 digest
2037 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2038 *
2039 * Return 1 if successful, otherwise return 0.
2040 */
2041int
2042tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2043 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2044{
2045 char tmpdigest[TCP_SIGLEN];
2046
2047 if (tcp_sig_checksigs == 0)
2048 return (1);
2049 if ((tcpbflag & TF_SIGNATURE) == 0) {
2050 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2051
2052 /*
2053 * If this socket is not expecting signature but
2054 * the segment contains signature just fail.
2055 */
2056 TCPSTAT_INC(tcps_sig_err_sigopt);
2057 TCPSTAT_INC(tcps_sig_rcvbadsig);
2058 return (0);
2059 }
2060
2061 /* Signature is not expected, and not present in segment. */
2062 return (1);
2063 }
2064
2065 /*
2066 * If this socket is expecting signature but the segment does not
2067 * contain any just fail.
2068 */
2069 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2070 TCPSTAT_INC(tcps_sig_err_nosigopt);
2071 TCPSTAT_INC(tcps_sig_rcvbadsig);
2072 return (0);
2073 }
2074 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2075 IPSEC_DIR_INBOUND) == -1) {
2076 TCPSTAT_INC(tcps_sig_err_buildsig);
2077 TCPSTAT_INC(tcps_sig_rcvbadsig);
2078 return (0);
2079 }
2080
2081 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2082 TCPSTAT_INC(tcps_sig_rcvbadsig);
2083 return (0);
2084 }
2085 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2086 return (1);
2087}
2088#endif /* TCP_SIGNATURE */
2089
2090static int
2091sysctl_drop(SYSCTL_HANDLER_ARGS)
2092{
2093 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2094 struct sockaddr_storage addrs[2];
2095 struct inpcb *inp;
2096 struct tcpcb *tp;
2097 struct tcptw *tw;
2098 struct sockaddr_in *fin, *lin;
2099#ifdef INET6
2100 struct sockaddr_in6 *fin6, *lin6;
2101#endif
2102 int error;
2103
2104 inp = NULL;
2105 fin = lin = NULL;
2106#ifdef INET6
2107 fin6 = lin6 = NULL;
2108#endif
2109 error = 0;
2110
2111 if (req->oldptr != NULL || req->oldlen != 0)
2112 return (EINVAL);
2113 if (req->newptr == NULL)
2114 return (EPERM);
2115 if (req->newlen < sizeof(addrs))
2116 return (ENOMEM);
2117 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2118 if (error)
2119 return (error);
2120
2121 switch (addrs[0].ss_family) {
2122#ifdef INET6
2123 case AF_INET6:
2124 fin6 = (struct sockaddr_in6 *)&addrs[0];
2125 lin6 = (struct sockaddr_in6 *)&addrs[1];
2126 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2127 lin6->sin6_len != sizeof(struct sockaddr_in6))
2128 return (EINVAL);
2129 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2130 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2131 return (EINVAL);
2132 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2133 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2134 fin = (struct sockaddr_in *)&addrs[0];
2135 lin = (struct sockaddr_in *)&addrs[1];
2136 break;
2137 }
2138 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2139 if (error)
2140 return (error);
2141 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2142 if (error)
2143 return (error);
2144 break;
2145#endif
2146#ifdef INET
2147 case AF_INET:
2148 fin = (struct sockaddr_in *)&addrs[0];
2149 lin = (struct sockaddr_in *)&addrs[1];
2150 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2151 lin->sin_len != sizeof(struct sockaddr_in))
2152 return (EINVAL);
2153 break;
2154#endif
2155 default:
2156 return (EINVAL);
2157 }
2158 INP_INFO_WLOCK(&V_tcbinfo);
2159 switch (addrs[0].ss_family) {
2160#ifdef INET6
2161 case AF_INET6:
2162 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2163 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2164 INPLOOKUP_WLOCKPCB, NULL);
2165 break;
2166#endif
2167#ifdef INET
2168 case AF_INET:
2169 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2170 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2171 break;
2172#endif
2173 }
2174 if (inp != NULL) {
2175 if (inp->inp_flags & INP_TIMEWAIT) {
2176 /*
2177 * XXXRW: There currently exists a state where an
2178 * inpcb is present, but its timewait state has been
2179 * discarded. For now, don't allow dropping of this
2180 * type of inpcb.
2181 */
2182 tw = intotw(inp);
2183 if (tw != NULL)
2184 tcp_twclose(tw, 0);
2185 else
2186 INP_WUNLOCK(inp);
2187 } else if (!(inp->inp_flags & INP_DROPPED) &&
2188 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2189 tp = intotcpcb(inp);
2190 tp = tcp_drop(tp, ECONNABORTED);
2191 if (tp != NULL)
2192 INP_WUNLOCK(inp);
2193 } else
2194 INP_WUNLOCK(inp);
2195 } else
2196 error = ESRCH;
2197 INP_INFO_WUNLOCK(&V_tcbinfo);
2198 return (error);
2199}
2200
2201SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2202 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2203 0, sysctl_drop, "", "Drop TCP connection");
2204
2205/*
2206 * Generate a standardized TCP log line for use throughout the
2207 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2208 * allow use in the interrupt context.
2209 *
2210 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2211 * NB: The function may return NULL if memory allocation failed.
2212 *
2213 * Due to header inclusion and ordering limitations the struct ip
2214 * and ip6_hdr pointers have to be passed as void pointers.
2215 */
2216char *
2217tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2218 const void *ip6hdr)
2219{
2220
2221 /* Is logging enabled? */
2222 if (tcp_log_in_vain == 0)
2223 return (NULL);
2224
2225 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2226}
2227
2228char *
2229tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2230 const void *ip6hdr)
2231{
2232
2233 /* Is logging enabled? */
2234 if (tcp_log_debug == 0)
2235 return (NULL);
2236
2237 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2238}
2239
2240static char *
2241tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2242 const void *ip6hdr)
2243{
2244 char *s, *sp;
2245 size_t size;
2246 struct ip *ip;
2247#ifdef INET6
2248 const struct ip6_hdr *ip6;
2249
2250 ip6 = (const struct ip6_hdr *)ip6hdr;
2251#endif /* INET6 */
2252 ip = (struct ip *)ip4hdr;
2253
2254 /*
2255 * The log line looks like this:
2256 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2257 */
2258 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2259 sizeof(PRINT_TH_FLAGS) + 1 +
2260#ifdef INET6
2261 2 * INET6_ADDRSTRLEN;
2262#else
2263 2 * INET_ADDRSTRLEN;
2264#endif /* INET6 */
2265
2266 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2267 if (s == NULL)
2268 return (NULL);
2269
2270 strcat(s, "TCP: [");
2271 sp = s + strlen(s);
2272
2273 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2274 inet_ntoa_r(inc->inc_faddr, sp);
2275 sp = s + strlen(s);
2276 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2277 sp = s + strlen(s);
2278 inet_ntoa_r(inc->inc_laddr, sp);
2279 sp = s + strlen(s);
2280 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2281#ifdef INET6
2282 } else if (inc) {
2283 ip6_sprintf(sp, &inc->inc6_faddr);
2284 sp = s + strlen(s);
2285 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2286 sp = s + strlen(s);
2287 ip6_sprintf(sp, &inc->inc6_laddr);
2288 sp = s + strlen(s);
2289 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2290 } else if (ip6 && th) {
2291 ip6_sprintf(sp, &ip6->ip6_src);
2292 sp = s + strlen(s);
2293 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2294 sp = s + strlen(s);
2295 ip6_sprintf(sp, &ip6->ip6_dst);
2296 sp = s + strlen(s);
2297 sprintf(sp, "]:%i", ntohs(th->th_dport));
2298#endif /* INET6 */
2299#ifdef INET
2300 } else if (ip && th) {
2301 inet_ntoa_r(ip->ip_src, sp);
2302 sp = s + strlen(s);
2303 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2304 sp = s + strlen(s);
2305 inet_ntoa_r(ip->ip_dst, sp);
2306 sp = s + strlen(s);
2307 sprintf(sp, "]:%i", ntohs(th->th_dport));
2308#endif /* INET */
2309 } else {
2310 free(s, M_TCPLOG);
2311 return (NULL);
2312 }
2313 sp = s + strlen(s);
2314 if (th)
2315 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2316 if (*(s + size - 1) != '\0')
2317 panic("%s: string too long", __func__);
2318 return (s);
2319}