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 289276 2015-10-14 00:35:37Z hiren $");
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/sdt.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/protosw.h>
60#include <sys/random.h>
61
62#include <vm/uma.h>
63
64#include <net/route.h>
65#include <net/if.h>
66#include <net/if_var.h>
67#include <net/vnet.h>
68
69#include <netinet/cc.h>
70#include <netinet/in.h>
71#include <netinet/in_kdtrace.h>
72#include <netinet/in_pcb.h>
73#include <netinet/in_systm.h>
74#include <netinet/in_var.h>
75#include <netinet/ip.h>
76#include <netinet/ip_icmp.h>
77#include <netinet/ip_var.h>
78#ifdef INET6
79#include <netinet/ip6.h>
80#include <netinet6/in6_pcb.h>
81#include <netinet6/ip6_var.h>
82#include <netinet6/scope6_var.h>
83#include <netinet6/nd6.h>
84#endif
85
86#include <netinet/tcp_fsm.h>
87#include <netinet/tcp_seq.h>
88#include <netinet/tcp_timer.h>
89#include <netinet/tcp_var.h>
90#include <netinet/tcp_syncache.h>
91#ifdef INET6
92#include <netinet6/tcp6_var.h>
93#endif
94#include <netinet/tcpip.h>
95#ifdef TCPPCAP
96#include <netinet/tcp_pcap.h>
97#endif
98#ifdef TCPDEBUG
99#include <netinet/tcp_debug.h>
100#endif
101#ifdef INET6
102#include <netinet6/ip6protosw.h>
103#endif
104#ifdef TCP_OFFLOAD
105#include <netinet/tcp_offload.h>
106#endif
107
108#ifdef IPSEC
109#include <netipsec/ipsec.h>
110#include <netipsec/xform.h>
111#ifdef INET6
112#include <netipsec/ipsec6.h>
113#endif
114#include <netipsec/key.h>
115#include <sys/syslog.h>
116#endif /*IPSEC*/
117
118#include <machine/in_cksum.h>
119#include <sys/md5.h>
120
121#include <security/mac/mac_framework.h>
122
123VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
124#ifdef INET6
125VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
126#endif
127
128static int
129sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
130{
131 int error, new;
132
133 new = V_tcp_mssdflt;
134 error = sysctl_handle_int(oidp, &new, 0, req);
135 if (error == 0 && req->newptr) {
136 if (new < TCP_MINMSS)
137 error = EINVAL;
138 else
139 V_tcp_mssdflt = new;
140 }
141 return (error);
142}
143
144SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
145 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
146 &sysctl_net_inet_tcp_mss_check, "I",
147 "Default TCP Maximum Segment Size");
148
149#ifdef INET6
150static int
151sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
152{
153 int error, new;
154
155 new = V_tcp_v6mssdflt;
156 error = sysctl_handle_int(oidp, &new, 0, req);
157 if (error == 0 && req->newptr) {
158 if (new < TCP_MINMSS)
159 error = EINVAL;
160 else
161 V_tcp_v6mssdflt = new;
162 }
163 return (error);
164}
165
166SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
167 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
168 &sysctl_net_inet_tcp_mss_v6_check, "I",
169 "Default TCP Maximum Segment Size for IPv6");
170#endif /* INET6 */
171
172/*
173 * Minimum MSS we accept and use. This prevents DoS attacks where
174 * we are forced to a ridiculous low MSS like 20 and send hundreds
175 * of packets instead of one. The effect scales with the available
176 * bandwidth and quickly saturates the CPU and network interface
177 * with packet generation and sending. Set to zero to disable MINMSS
178 * checking. This setting prevents us from sending too small packets.
179 */
180VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
181SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
182 &VNET_NAME(tcp_minmss), 0,
183 "Minimum TCP Maximum Segment Size");
184
185VNET_DEFINE(int, tcp_do_rfc1323) = 1;
186SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc1323), 0,
188 "Enable rfc1323 (high performance TCP) extensions");
189
190static int tcp_log_debug = 0;
191SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
192 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
193
194static int tcp_tcbhashsize;
195SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
196 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
197
198static int do_tcpdrain = 1;
199SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
200 "Enable tcp_drain routine for extra help when low on mbufs");
201
202SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
203 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
204
205static VNET_DEFINE(int, icmp_may_rst) = 1;
206#define V_icmp_may_rst VNET(icmp_may_rst)
207SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(icmp_may_rst), 0,
209 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
210
211static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
212#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
213SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_isn_reseed_interval), 0,
215 "Seconds between reseeding of ISN secret");
216
217static int tcp_soreceive_stream;
218SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
219 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
220
221#ifdef TCP_SIGNATURE
222static int tcp_sig_checksigs = 1;
223SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
224 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
225#endif
226
227VNET_DEFINE(uma_zone_t, sack_hole_zone);
228#define V_sack_hole_zone VNET(sack_hole_zone)
229
230VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
231
232static struct inpcb *tcp_notify(struct inpcb *, int);
233static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234static void tcp_mtudisc(struct inpcb *, int);
235static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
236 void *ip4hdr, const void *ip6hdr);
237static void tcp_timer_discard(struct tcpcb *, uint32_t);
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 0
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 */
255struct tcpcb_mem {
256 struct tcpcb tcb;
257 struct tcp_timer tt;
258 struct cc_var ccv;
259 struct osd osd;
260};
261
262static VNET_DEFINE(uma_zone_t, tcpcb_zone);
263#define V_tcpcb_zone VNET(tcpcb_zone)
264
265MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
266static struct mtx isn_mtx;
267
268#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
269#define ISN_LOCK() mtx_lock(&isn_mtx)
270#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
271
272/*
273 * TCP initialization.
274 */
275static void
276tcp_zone_change(void *tag)
277{
278
279 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
280 uma_zone_set_max(V_tcpcb_zone, maxsockets);
281 tcp_tw_zone_change();
282}
283
284static int
285tcp_inpcb_init(void *mem, int size, int flags)
286{
287 struct inpcb *inp = mem;
288
289 INP_LOCK_INIT(inp, "inp", "tcpinp");
290 return (0);
291}
292
293/*
294 * Take a value and get the next power of 2 that doesn't overflow.
295 * Used to size the tcp_inpcb hash buckets.
296 */
297static int
298maketcp_hashsize(int size)
299{
300 int hashsize;
301
302 /*
303 * auto tune.
304 * get the next power of 2 higher than maxsockets.
305 */
306 hashsize = 1 << fls(size);
307 /* catch overflow, and just go one power of 2 smaller */
308 if (hashsize < size) {
309 hashsize = 1 << (fls(size) - 1);
310 }
311 return (hashsize);
312}
313
314void
315tcp_init(void)
316{
317 const char *tcbhash_tuneable;
318 int hashsize;
319
320 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
321
322 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
323 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
324 printf("%s: WARNING: unable to register helper hook\n", __func__);
325 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
326 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
327 printf("%s: WARNING: unable to register helper hook\n", __func__);
328
329 hashsize = TCBHASHSIZE;
330 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
331 if (hashsize == 0) {
332 /*
333 * Auto tune the hash size based on maxsockets.
334 * A perfect hash would have a 1:1 mapping
335 * (hashsize = maxsockets) however it's been
336 * suggested that O(2) average is better.
337 */
338 hashsize = maketcp_hashsize(maxsockets / 4);
339 /*
340 * Our historical default is 512,
341 * do not autotune lower than this.
342 */
343 if (hashsize < 512)
344 hashsize = 512;
345 if (bootverbose)
346 printf("%s: %s auto tuned to %d\n", __func__,
347 tcbhash_tuneable, hashsize);
348 }
349 /*
350 * We require a hashsize to be a power of two.
351 * Previously if it was not a power of two we would just reset it
352 * back to 512, which could be a nasty surprise if you did not notice
353 * the error message.
354 * Instead what we do is clip it to the closest power of two lower
355 * than the specified hash value.
356 */
357 if (!powerof2(hashsize)) {
358 int oldhashsize = hashsize;
359
360 hashsize = maketcp_hashsize(hashsize);
361 /* prevent absurdly low value */
362 if (hashsize < 16)
363 hashsize = 16;
364 printf("%s: WARNING: TCB hash size not a power of 2, "
365 "clipped from %d to %d.\n", __func__, oldhashsize,
366 hashsize);
367 }
368 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
369 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
370 IPI_HASHFIELDS_4TUPLE);
371
372 /*
373 * These have to be type stable for the benefit of the timers.
374 */
375 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
376 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
377 uma_zone_set_max(V_tcpcb_zone, maxsockets);
378 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
379
380 tcp_tw_init();
381 syncache_init();
382 tcp_hc_init();
383
384 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
385 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
386 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
387
388 /* Skip initialization of globals for non-default instances. */
389 if (!IS_DEFAULT_VNET(curvnet))
390 return;
391
392 tcp_reass_global_init();
393
394 /* XXX virtualize those bellow? */
395 tcp_delacktime = TCPTV_DELACK;
396 tcp_keepinit = TCPTV_KEEP_INIT;
397 tcp_keepidle = TCPTV_KEEP_IDLE;
398 tcp_keepintvl = TCPTV_KEEPINTVL;
399 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
400 tcp_msl = TCPTV_MSL;
401 tcp_rexmit_min = TCPTV_MIN;
402 if (tcp_rexmit_min < 1)
403 tcp_rexmit_min = 1;
404 tcp_rexmit_slop = TCPTV_CPU_VAR;
405 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
406 tcp_tcbhashsize = hashsize;
407
408 if (tcp_soreceive_stream) {
409#ifdef INET
410 tcp_usrreqs.pru_soreceive = soreceive_stream;
411#endif
412#ifdef INET6
413 tcp6_usrreqs.pru_soreceive = soreceive_stream;
414#endif /* INET6 */
415 }
416
417#ifdef INET6
418#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
419#else /* INET6 */
420#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
421#endif /* INET6 */
422 if (max_protohdr < TCP_MINPROTOHDR)
423 max_protohdr = TCP_MINPROTOHDR;
424 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
425 panic("tcp_init");
426#undef TCP_MINPROTOHDR
427
428 ISN_LOCK_INIT();
429 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
430 SHUTDOWN_PRI_DEFAULT);
431 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
432 EVENTHANDLER_PRI_ANY);
433#ifdef TCPPCAP
434 tcp_pcap_init();
435#endif
436}
437
438#ifdef VIMAGE
439void
440tcp_destroy(void)
441{
442 int error;
443
444 tcp_hc_destroy();
445 syncache_destroy();
446 tcp_tw_destroy();
447 in_pcbinfo_destroy(&V_tcbinfo);
448 uma_zdestroy(V_sack_hole_zone);
449 uma_zdestroy(V_tcpcb_zone);
450
451 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
452 if (error != 0) {
453 printf("%s: WARNING: unable to deregister helper hook "
454 "type=%d, id=%d: error %d returned\n", __func__,
455 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
456 }
457 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
458 if (error != 0) {
459 printf("%s: WARNING: unable to deregister helper hook "
460 "type=%d, id=%d: error %d returned\n", __func__,
461 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
462 }
463}
464#endif
465
466void
467tcp_fini(void *xtp)
468{
469
470}
471
472/*
473 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
474 * tcp_template used to store this data in mbufs, but we now recopy it out
475 * of the tcpcb each time to conserve mbufs.
476 */
477void
478tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
479{
480 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
481
482 INP_WLOCK_ASSERT(inp);
483
484#ifdef INET6
485 if ((inp->inp_vflag & INP_IPV6) != 0) {
486 struct ip6_hdr *ip6;
487
488 ip6 = (struct ip6_hdr *)ip_ptr;
489 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
490 (inp->inp_flow & IPV6_FLOWINFO_MASK);
491 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
492 (IPV6_VERSION & IPV6_VERSION_MASK);
493 ip6->ip6_nxt = IPPROTO_TCP;
494 ip6->ip6_plen = htons(sizeof(struct tcphdr));
495 ip6->ip6_src = inp->in6p_laddr;
496 ip6->ip6_dst = inp->in6p_faddr;
497 }
498#endif /* INET6 */
499#if defined(INET6) && defined(INET)
500 else
501#endif
502#ifdef INET
503 {
504 struct ip *ip;
505
506 ip = (struct ip *)ip_ptr;
507 ip->ip_v = IPVERSION;
508 ip->ip_hl = 5;
509 ip->ip_tos = inp->inp_ip_tos;
510 ip->ip_len = 0;
511 ip->ip_id = 0;
512 ip->ip_off = 0;
513 ip->ip_ttl = inp->inp_ip_ttl;
514 ip->ip_sum = 0;
515 ip->ip_p = IPPROTO_TCP;
516 ip->ip_src = inp->inp_laddr;
517 ip->ip_dst = inp->inp_faddr;
518 }
519#endif /* INET */
520 th->th_sport = inp->inp_lport;
521 th->th_dport = inp->inp_fport;
522 th->th_seq = 0;
523 th->th_ack = 0;
524 th->th_x2 = 0;
525 th->th_off = 5;
526 th->th_flags = 0;
527 th->th_win = 0;
528 th->th_urp = 0;
529 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
530}
531
532/*
533 * Create template to be used to send tcp packets on a connection.
534 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
535 * use for this function is in keepalives, which use tcp_respond.
536 */
537struct tcptemp *
538tcpip_maketemplate(struct inpcb *inp)
539{
540 struct tcptemp *t;
541
542 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
543 if (t == NULL)
544 return (NULL);
545 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
546 return (t);
547}
548
549/*
550 * Send a single message to the TCP at address specified by
551 * the given TCP/IP header. If m == NULL, then we make a copy
552 * of the tcpiphdr at th and send directly to the addressed host.
553 * This is used to force keep alive messages out using the TCP
554 * template for a connection. If flags are given then we send
555 * a message back to the TCP which originated the segment th,
556 * and discard the mbuf containing it and any other attached mbufs.
557 *
558 * In any case the ack and sequence number of the transmitted
559 * segment are as specified by the parameters.
560 *
561 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
562 */
563void
564tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
565 tcp_seq ack, tcp_seq seq, int flags)
566{
567 int tlen;
568 int win = 0;
569 struct ip *ip;
570 struct tcphdr *nth;
571#ifdef INET6
572 struct ip6_hdr *ip6;
573 int isipv6;
574#endif /* INET6 */
575 int ipflags = 0;
576 struct inpcb *inp;
577
578 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
579
580#ifdef INET6
581 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
582 ip6 = ipgen;
583#endif /* INET6 */
584 ip = ipgen;
585
586 if (tp != NULL) {
587 inp = tp->t_inpcb;
588 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
589 INP_WLOCK_ASSERT(inp);
590 } else
591 inp = NULL;
592
593 if (tp != NULL) {
594 if (!(flags & TH_RST)) {
595 win = sbspace(&inp->inp_socket->so_rcv);
596 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
597 win = (long)TCP_MAXWIN << tp->rcv_scale;
598 }
599 }
600 if (m == NULL) {
601 m = m_gethdr(M_NOWAIT, MT_DATA);
602 if (m == NULL)
603 return;
604 tlen = 0;
605 m->m_data += max_linkhdr;
606#ifdef INET6
607 if (isipv6) {
608 bcopy((caddr_t)ip6, mtod(m, caddr_t),
609 sizeof(struct ip6_hdr));
610 ip6 = mtod(m, struct ip6_hdr *);
611 nth = (struct tcphdr *)(ip6 + 1);
612 } else
613#endif /* INET6 */
614 {
615 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
616 ip = mtod(m, struct ip *);
617 nth = (struct tcphdr *)(ip + 1);
618 }
619 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
620 flags = TH_ACK;
621 } else {
622 /*
623 * reuse the mbuf.
624 * XXX MRT We inherrit the FIB, which is lucky.
625 */
626 m_freem(m->m_next);
627 m->m_next = NULL;
628 m->m_data = (caddr_t)ipgen;
629 /* m_len is set later */
630 tlen = 0;
631#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
632#ifdef INET6
633 if (isipv6) {
634 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
635 nth = (struct tcphdr *)(ip6 + 1);
636 } else
637#endif /* INET6 */
638 {
639 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
640 nth = (struct tcphdr *)(ip + 1);
641 }
642 if (th != nth) {
643 /*
644 * this is usually a case when an extension header
645 * exists between the IPv6 header and the
646 * TCP header.
647 */
648 nth->th_sport = th->th_sport;
649 nth->th_dport = th->th_dport;
650 }
651 xchg(nth->th_dport, nth->th_sport, uint16_t);
652#undef xchg
653 }
654#ifdef INET6
655 if (isipv6) {
656 ip6->ip6_flow = 0;
657 ip6->ip6_vfc = IPV6_VERSION;
658 ip6->ip6_nxt = IPPROTO_TCP;
659 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
660 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
661 }
662#endif
663#if defined(INET) && defined(INET6)
664 else
665#endif
666#ifdef INET
667 {
668 tlen += sizeof (struct tcpiphdr);
669 ip->ip_len = htons(tlen);
670 ip->ip_ttl = V_ip_defttl;
671 if (V_path_mtu_discovery)
672 ip->ip_off |= htons(IP_DF);
673 }
674#endif
675 m->m_len = tlen;
676 m->m_pkthdr.len = tlen;
677 m->m_pkthdr.rcvif = NULL;
678#ifdef MAC
679 if (inp != NULL) {
680 /*
681 * Packet is associated with a socket, so allow the
682 * label of the response to reflect the socket label.
683 */
684 INP_WLOCK_ASSERT(inp);
685 mac_inpcb_create_mbuf(inp, m);
686 } else {
687 /*
688 * Packet is not associated with a socket, so possibly
689 * update the label in place.
690 */
691 mac_netinet_tcp_reply(m);
692 }
693#endif
694 nth->th_seq = htonl(seq);
695 nth->th_ack = htonl(ack);
696 nth->th_x2 = 0;
697 nth->th_off = sizeof (struct tcphdr) >> 2;
698 nth->th_flags = flags;
699 if (tp != NULL)
700 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
701 else
702 nth->th_win = htons((u_short)win);
703 nth->th_urp = 0;
704
705 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
706#ifdef INET6
707 if (isipv6) {
708 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
709 nth->th_sum = in6_cksum_pseudo(ip6,
710 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
711 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
712 NULL, NULL);
713 }
714#endif /* INET6 */
715#if defined(INET6) && defined(INET)
716 else
717#endif
718#ifdef INET
719 {
720 m->m_pkthdr.csum_flags = CSUM_TCP;
721 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
722 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
723 }
724#endif /* INET */
725#ifdef TCPDEBUG
726 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
727 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
728#endif
729 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
730 if (flags & TH_RST)
731 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
732 tp, nth);
733
734 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
735#ifdef INET6
736 if (isipv6)
737 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
738#endif /* INET6 */
739#if defined(INET) && defined(INET6)
740 else
741#endif
742#ifdef INET
743 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
744#endif
745}
746
747/*
748 * Create a new TCP control block, making an
749 * empty reassembly queue and hooking it to the argument
750 * protocol control block. The `inp' parameter must have
751 * come from the zone allocator set up in tcp_init().
752 */
753struct tcpcb *
754tcp_newtcpcb(struct inpcb *inp)
755{
756 struct tcpcb_mem *tm;
757 struct tcpcb *tp;
758#ifdef INET6
759 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
760#endif /* INET6 */
761
762 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
763 if (tm == NULL)
764 return (NULL);
765 tp = &tm->tcb;
766
767 /* Initialise cc_var struct for this tcpcb. */
768 tp->ccv = &tm->ccv;
769 tp->ccv->type = IPPROTO_TCP;
770 tp->ccv->ccvc.tcp = tp;
771
772 /*
773 * Use the current system default CC algorithm.
774 */
775 CC_LIST_RLOCK();
776 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
777 CC_ALGO(tp) = CC_DEFAULT();
778 CC_LIST_RUNLOCK();
779
780 if (CC_ALGO(tp)->cb_init != NULL)
781 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
782 uma_zfree(V_tcpcb_zone, tm);
783 return (NULL);
784 }
785
786 tp->osd = &tm->osd;
787 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
788 uma_zfree(V_tcpcb_zone, tm);
789 return (NULL);
790 }
791
792#ifdef VIMAGE
793 tp->t_vnet = inp->inp_vnet;
794#endif
795 tp->t_timers = &tm->tt;
796 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
797 tp->t_maxseg = tp->t_maxopd =
798#ifdef INET6
799 isipv6 ? V_tcp_v6mssdflt :
800#endif /* INET6 */
801 V_tcp_mssdflt;
802
803 /* Set up our timeouts. */
804 callout_init(&tp->t_timers->tt_rexmt, 1);
805 callout_init(&tp->t_timers->tt_persist, 1);
806 callout_init(&tp->t_timers->tt_keep, 1);
807 callout_init(&tp->t_timers->tt_2msl, 1);
808 callout_init(&tp->t_timers->tt_delack, 1);
809
810 if (V_tcp_do_rfc1323)
811 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
812 if (V_tcp_do_sack)
813 tp->t_flags |= TF_SACK_PERMIT;
814 TAILQ_INIT(&tp->snd_holes);
815 /*
816 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
817 * is called.
818 */
819 in_pcbref(inp); /* Reference for tcpcb */
820 tp->t_inpcb = inp;
821
822 /*
823 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
824 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
825 * reasonable initial retransmit time.
826 */
827 tp->t_srtt = TCPTV_SRTTBASE;
828 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
829 tp->t_rttmin = tcp_rexmit_min;
830 tp->t_rxtcur = TCPTV_RTOBASE;
831 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
832 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
833 tp->t_rcvtime = ticks;
834 /*
835 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
836 * because the socket may be bound to an IPv6 wildcard address,
837 * which may match an IPv4-mapped IPv6 address.
838 */
839 inp->inp_ip_ttl = V_ip_defttl;
840 inp->inp_ppcb = tp;
841#ifdef TCPPCAP
842 /*
843 * Init the TCP PCAP queues.
844 */
845 tcp_pcap_tcpcb_init(tp);
846#endif
847 return (tp); /* XXX */
848}
849
850/*
851 * Switch the congestion control algorithm back to NewReno for any active
852 * control blocks using an algorithm which is about to go away.
853 * This ensures the CC framework can allow the unload to proceed without leaving
854 * any dangling pointers which would trigger a panic.
855 * Returning non-zero would inform the CC framework that something went wrong
856 * and it would be unsafe to allow the unload to proceed. However, there is no
857 * way for this to occur with this implementation so we always return zero.
858 */
859int
860tcp_ccalgounload(struct cc_algo *unload_algo)
861{
862 struct cc_algo *tmpalgo;
863 struct inpcb *inp;
864 struct tcpcb *tp;
865 VNET_ITERATOR_DECL(vnet_iter);
866
867 /*
868 * Check all active control blocks across all network stacks and change
869 * any that are using "unload_algo" back to NewReno. If "unload_algo"
870 * requires cleanup code to be run, call it.
871 */
872 VNET_LIST_RLOCK();
873 VNET_FOREACH(vnet_iter) {
874 CURVNET_SET(vnet_iter);
875 INP_INFO_WLOCK(&V_tcbinfo);
876 /*
877 * New connections already part way through being initialised
878 * with the CC algo we're removing will not race with this code
879 * because the INP_INFO_WLOCK is held during initialisation. We
880 * therefore don't enter the loop below until the connection
881 * list has stabilised.
882 */
883 LIST_FOREACH(inp, &V_tcb, inp_list) {
884 INP_WLOCK(inp);
885 /* Important to skip tcptw structs. */
886 if (!(inp->inp_flags & INP_TIMEWAIT) &&
887 (tp = intotcpcb(inp)) != NULL) {
888 /*
889 * By holding INP_WLOCK here, we are assured
890 * that the connection is not currently
891 * executing inside the CC module's functions
892 * i.e. it is safe to make the switch back to
893 * NewReno.
894 */
895 if (CC_ALGO(tp) == unload_algo) {
896 tmpalgo = CC_ALGO(tp);
897 /* NewReno does not require any init. */
898 CC_ALGO(tp) = &newreno_cc_algo;
899 if (tmpalgo->cb_destroy != NULL)
900 tmpalgo->cb_destroy(tp->ccv);
901 }
902 }
903 INP_WUNLOCK(inp);
904 }
905 INP_INFO_WUNLOCK(&V_tcbinfo);
906 CURVNET_RESTORE();
907 }
908 VNET_LIST_RUNLOCK();
909
910 return (0);
911}
912
913/*
914 * Drop a TCP connection, reporting
915 * the specified error. If connection is synchronized,
916 * then send a RST to peer.
917 */
918struct tcpcb *
919tcp_drop(struct tcpcb *tp, int errno)
920{
921 struct socket *so = tp->t_inpcb->inp_socket;
922
923 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
924 INP_WLOCK_ASSERT(tp->t_inpcb);
925
926 if (TCPS_HAVERCVDSYN(tp->t_state)) {
927 tcp_state_change(tp, TCPS_CLOSED);
928 (void) tcp_output(tp);
929 TCPSTAT_INC(tcps_drops);
930 } else
931 TCPSTAT_INC(tcps_conndrops);
932 if (errno == ETIMEDOUT && tp->t_softerror)
933 errno = tp->t_softerror;
934 so->so_error = errno;
935 return (tcp_close(tp));
936}
937
938void
939tcp_discardcb(struct tcpcb *tp)
940{
941 struct inpcb *inp = tp->t_inpcb;
942 struct socket *so = inp->inp_socket;
943#ifdef INET6
944 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
945#endif /* INET6 */
946 int released;
947
948 INP_WLOCK_ASSERT(inp);
949
950 /*
951 * Make sure that all of our timers are stopped before we delete the
952 * PCB.
953 *
954 * If stopping a timer fails, we schedule a discard function in same
955 * callout, and the last discard function called will take care of
956 * deleting the tcpcb.
957 */
958 tcp_timer_stop(tp, TT_REXMT);
959 tcp_timer_stop(tp, TT_PERSIST);
960 tcp_timer_stop(tp, TT_KEEP);
961 tcp_timer_stop(tp, TT_2MSL);
962 tcp_timer_stop(tp, TT_DELACK);
963
964 /*
965 * If we got enough samples through the srtt filter,
966 * save the rtt and rttvar in the routing entry.
967 * 'Enough' is arbitrarily defined as 4 rtt samples.
968 * 4 samples is enough for the srtt filter to converge
969 * to within enough % of the correct value; fewer samples
970 * and we could save a bogus rtt. The danger is not high
971 * as tcp quickly recovers from everything.
972 * XXX: Works very well but needs some more statistics!
973 */
974 if (tp->t_rttupdated >= 4) {
975 struct hc_metrics_lite metrics;
976 u_long ssthresh;
977
978 bzero(&metrics, sizeof(metrics));
979 /*
980 * Update the ssthresh always when the conditions below
981 * are satisfied. This gives us better new start value
982 * for the congestion avoidance for new connections.
983 * ssthresh is only set if packet loss occured on a session.
984 *
985 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
986 * being torn down. Ideally this code would not use 'so'.
987 */
988 ssthresh = tp->snd_ssthresh;
989 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
990 /*
991 * convert the limit from user data bytes to
992 * packets then to packet data bytes.
993 */
994 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
995 if (ssthresh < 2)
996 ssthresh = 2;
997 ssthresh *= (u_long)(tp->t_maxseg +
998#ifdef INET6
999 (isipv6 ? sizeof (struct ip6_hdr) +
1000 sizeof (struct tcphdr) :
1001#endif
1002 sizeof (struct tcpiphdr)
1003#ifdef INET6
1004 )
1005#endif
1006 );
1007 } else
1008 ssthresh = 0;
1009 metrics.rmx_ssthresh = ssthresh;
1010
1011 metrics.rmx_rtt = tp->t_srtt;
1012 metrics.rmx_rttvar = tp->t_rttvar;
1013 metrics.rmx_cwnd = tp->snd_cwnd;
1014 metrics.rmx_sendpipe = 0;
1015 metrics.rmx_recvpipe = 0;
1016
1017 tcp_hc_update(&inp->inp_inc, &metrics);
1018 }
1019
1020 /* free the reassembly queue, if any */
1021 tcp_reass_flush(tp);
1022
1023#ifdef TCP_OFFLOAD
1024 /* Disconnect offload device, if any. */
1025 if (tp->t_flags & TF_TOE)
1026 tcp_offload_detach(tp);
1027#endif
1028
1029 tcp_free_sackholes(tp);
1030
1031#ifdef TCPPCAP
1032 /* Free the TCP PCAP queues. */
1033 tcp_pcap_drain(&(tp->t_inpkts));
1034 tcp_pcap_drain(&(tp->t_outpkts));
1035#endif
1036
1037 /* Allow the CC algorithm to clean up after itself. */
1038 if (CC_ALGO(tp)->cb_destroy != NULL)
1039 CC_ALGO(tp)->cb_destroy(tp->ccv);
1040
1041 khelp_destroy_osd(tp->osd);
1042
1043 CC_ALGO(tp) = NULL;
1044 inp->inp_ppcb = NULL;
1045 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1046 /* We own the last reference on tcpcb, let's free it. */
1047 tp->t_inpcb = NULL;
1048 uma_zfree(V_tcpcb_zone, tp);
1049 released = in_pcbrele_wlocked(inp);
1050 KASSERT(!released, ("%s: inp %p should not have been released "
1051 "here", __func__, inp));
1052 }
1053}
1054
1055void
1056tcp_timer_2msl_discard(void *xtp)
1057{
1058
1059 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1060}
1061
1062void
1063tcp_timer_keep_discard(void *xtp)
1064{
1065
1066 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1067}
1068
1069void
1070tcp_timer_persist_discard(void *xtp)
1071{
1072
1073 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1074}
1075
1076void
1077tcp_timer_rexmt_discard(void *xtp)
1078{
1079
1080 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1081}
1082
1083void
1084tcp_timer_delack_discard(void *xtp)
1085{
1086
1087 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1088}
1089
1090void
1091tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1092{
1093 struct inpcb *inp;
1094
1095 CURVNET_SET(tp->t_vnet);
1096 INP_INFO_RLOCK(&V_tcbinfo);
1097 inp = tp->t_inpcb;
1098 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1099 __func__, tp));
1100 INP_WLOCK(inp);
1101 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1102 ("%s: tcpcb has to be stopped here", __func__));
1103 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1104 ("%s: discard callout should be running", __func__));
1105 tp->t_timers->tt_flags &= ~timer_type;
1106 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1107 /* We own the last reference on this tcpcb, let's free it. */
1108 tp->t_inpcb = NULL;
1109 uma_zfree(V_tcpcb_zone, tp);
1110 if (in_pcbrele_wlocked(inp)) {
1111 INP_INFO_RUNLOCK(&V_tcbinfo);
1112 CURVNET_RESTORE();
1113 return;
1114 }
1115 }
1116 INP_WUNLOCK(inp);
1117 INP_INFO_RUNLOCK(&V_tcbinfo);
1118 CURVNET_RESTORE();
1119}
1120
1121/*
1122 * Attempt to close a TCP control block, marking it as dropped, and freeing
1123 * the socket if we hold the only reference.
1124 */
1125struct tcpcb *
1126tcp_close(struct tcpcb *tp)
1127{
1128 struct inpcb *inp = tp->t_inpcb;
1129 struct socket *so;
1130
1131 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1132 INP_WLOCK_ASSERT(inp);
1133
1134#ifdef TCP_OFFLOAD
1135 if (tp->t_state == TCPS_LISTEN)
1136 tcp_offload_listen_stop(tp);
1137#endif
1138 in_pcbdrop(inp);
1139 TCPSTAT_INC(tcps_closed);
1140 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1141 so = inp->inp_socket;
1142 soisdisconnected(so);
1143 if (inp->inp_flags & INP_SOCKREF) {
1144 KASSERT(so->so_state & SS_PROTOREF,
1145 ("tcp_close: !SS_PROTOREF"));
1146 inp->inp_flags &= ~INP_SOCKREF;
1147 INP_WUNLOCK(inp);
1148 ACCEPT_LOCK();
1149 SOCK_LOCK(so);
1150 so->so_state &= ~SS_PROTOREF;
1151 sofree(so);
1152 return (NULL);
1153 }
1154 return (tp);
1155}
1156
1157void
1158tcp_drain(void)
1159{
1160 VNET_ITERATOR_DECL(vnet_iter);
1161
1162 if (!do_tcpdrain)
1163 return;
1164
1165 VNET_LIST_RLOCK_NOSLEEP();
1166 VNET_FOREACH(vnet_iter) {
1167 CURVNET_SET(vnet_iter);
1168 struct inpcb *inpb;
1169 struct tcpcb *tcpb;
1170
1171 /*
1172 * Walk the tcpbs, if existing, and flush the reassembly queue,
1173 * if there is one...
1174 * XXX: The "Net/3" implementation doesn't imply that the TCP
1175 * reassembly queue should be flushed, but in a situation
1176 * where we're really low on mbufs, this is potentially
1177 * useful.
1178 */
1179 INP_INFO_WLOCK(&V_tcbinfo);
1180 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1181 if (inpb->inp_flags & INP_TIMEWAIT)
1182 continue;
1183 INP_WLOCK(inpb);
1184 if ((tcpb = intotcpcb(inpb)) != NULL) {
1185 tcp_reass_flush(tcpb);
1186 tcp_clean_sackreport(tcpb);
1187 }
1188 INP_WUNLOCK(inpb);
1189 }
1190 INP_INFO_WUNLOCK(&V_tcbinfo);
1191 CURVNET_RESTORE();
1192 }
1193 VNET_LIST_RUNLOCK_NOSLEEP();
1194}
1195
1196/*
1197 * Notify a tcp user of an asynchronous error;
1198 * store error as soft error, but wake up user
1199 * (for now, won't do anything until can select for soft error).
1200 *
1201 * Do not wake up user since there currently is no mechanism for
1202 * reporting soft errors (yet - a kqueue filter may be added).
1203 */
1204static struct inpcb *
1205tcp_notify(struct inpcb *inp, int error)
1206{
1207 struct tcpcb *tp;
1208
1209 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1210 INP_WLOCK_ASSERT(inp);
1211
1212 if ((inp->inp_flags & INP_TIMEWAIT) ||
1213 (inp->inp_flags & INP_DROPPED))
1214 return (inp);
1215
1216 tp = intotcpcb(inp);
1217 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1218
1219 /*
1220 * Ignore some errors if we are hooked up.
1221 * If connection hasn't completed, has retransmitted several times,
1222 * and receives a second error, give up now. This is better
1223 * than waiting a long time to establish a connection that
1224 * can never complete.
1225 */
1226 if (tp->t_state == TCPS_ESTABLISHED &&
1227 (error == EHOSTUNREACH || error == ENETUNREACH ||
1228 error == EHOSTDOWN)) {
1229 return (inp);
1230 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1231 tp->t_softerror) {
1232 tp = tcp_drop(tp, error);
1233 if (tp != NULL)
1234 return (inp);
1235 else
1236 return (NULL);
1237 } else {
1238 tp->t_softerror = error;
1239 return (inp);
1240 }
1241#if 0
1242 wakeup( &so->so_timeo);
1243 sorwakeup(so);
1244 sowwakeup(so);
1245#endif
1246}
1247
1248static int
1249tcp_pcblist(SYSCTL_HANDLER_ARGS)
1250{
1251 int error, i, m, n, pcb_count;
1252 struct inpcb *inp, **inp_list;
1253 inp_gen_t gencnt;
1254 struct xinpgen xig;
1255
1256 /*
1257 * The process of preparing the TCB list is too time-consuming and
1258 * resource-intensive to repeat twice on every request.
1259 */
1260 if (req->oldptr == NULL) {
1261 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1262 n += imax(n / 8, 10);
1263 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1264 return (0);
1265 }
1266
1267 if (req->newptr != NULL)
1268 return (EPERM);
1269
1270 /*
1271 * OK, now we're committed to doing something.
1272 */
1273 INP_LIST_RLOCK(&V_tcbinfo);
1274 gencnt = V_tcbinfo.ipi_gencnt;
1275 n = V_tcbinfo.ipi_count;
1276 INP_LIST_RUNLOCK(&V_tcbinfo);
1277
1278 m = syncache_pcbcount();
1279
1280 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1281 + (n + m) * sizeof(struct xtcpcb));
1282 if (error != 0)
1283 return (error);
1284
1285 xig.xig_len = sizeof xig;
1286 xig.xig_count = n + m;
1287 xig.xig_gen = gencnt;
1288 xig.xig_sogen = so_gencnt;
1289 error = SYSCTL_OUT(req, &xig, sizeof xig);
1290 if (error)
1291 return (error);
1292
1293 error = syncache_pcblist(req, m, &pcb_count);
1294 if (error)
1295 return (error);
1296
1297 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1298 if (inp_list == NULL)
1299 return (ENOMEM);
1300
1301 INP_INFO_WLOCK(&V_tcbinfo);
1302 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1303 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1304 INP_WLOCK(inp);
1305 if (inp->inp_gencnt <= gencnt) {
1306 /*
1307 * XXX: This use of cr_cansee(), introduced with
1308 * TCP state changes, is not quite right, but for
1309 * now, better than nothing.
1310 */
1311 if (inp->inp_flags & INP_TIMEWAIT) {
1312 if (intotw(inp) != NULL)
1313 error = cr_cansee(req->td->td_ucred,
1314 intotw(inp)->tw_cred);
1315 else
1316 error = EINVAL; /* Skip this inp. */
1317 } else
1318 error = cr_canseeinpcb(req->td->td_ucred, inp);
1319 if (error == 0) {
1320 in_pcbref(inp);
1321 inp_list[i++] = inp;
1322 }
1323 }
1324 INP_WUNLOCK(inp);
1325 }
1326 INP_INFO_WUNLOCK(&V_tcbinfo);
1327 n = i;
1328
1329 error = 0;
1330 for (i = 0; i < n; i++) {
1331 inp = inp_list[i];
1332 INP_RLOCK(inp);
1333 if (inp->inp_gencnt <= gencnt) {
1334 struct xtcpcb xt;
1335 void *inp_ppcb;
1336
1337 bzero(&xt, sizeof(xt));
1338 xt.xt_len = sizeof xt;
1339 /* XXX should avoid extra copy */
1340 bcopy(inp, &xt.xt_inp, sizeof *inp);
1341 inp_ppcb = inp->inp_ppcb;
1342 if (inp_ppcb == NULL)
1343 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1344 else if (inp->inp_flags & INP_TIMEWAIT) {
1345 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1346 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1347 } else {
1348 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1349 if (xt.xt_tp.t_timers)
1350 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1351 }
1352 if (inp->inp_socket != NULL)
1353 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1354 else {
1355 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1356 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1357 }
1358 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1359 INP_RUNLOCK(inp);
1360 error = SYSCTL_OUT(req, &xt, sizeof xt);
1361 } else
1362 INP_RUNLOCK(inp);
1363 }
1364 INP_INFO_RLOCK(&V_tcbinfo);
1365 for (i = 0; i < n; i++) {
1366 inp = inp_list[i];
1367 INP_RLOCK(inp);
1368 if (!in_pcbrele_rlocked(inp))
1369 INP_RUNLOCK(inp);
1370 }
1371 INP_INFO_RUNLOCK(&V_tcbinfo);
1372
1373 if (!error) {
1374 /*
1375 * Give the user an updated idea of our state.
1376 * If the generation differs from what we told
1377 * her before, she knows that something happened
1378 * while we were processing this request, and it
1379 * might be necessary to retry.
1380 */
1381 INP_LIST_RLOCK(&V_tcbinfo);
1382 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1383 xig.xig_sogen = so_gencnt;
1384 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1385 INP_LIST_RUNLOCK(&V_tcbinfo);
1386 error = SYSCTL_OUT(req, &xig, sizeof xig);
1387 }
1388 free(inp_list, M_TEMP);
1389 return (error);
1390}
1391
1392SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1393 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1394 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1395
1396#ifdef INET
1397static int
1398tcp_getcred(SYSCTL_HANDLER_ARGS)
1399{
1400 struct xucred xuc;
1401 struct sockaddr_in addrs[2];
1402 struct inpcb *inp;
1403 int error;
1404
1405 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1406 if (error)
1407 return (error);
1408 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1409 if (error)
1410 return (error);
1411 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1412 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1413 if (inp != NULL) {
1414 if (inp->inp_socket == NULL)
1415 error = ENOENT;
1416 if (error == 0)
1417 error = cr_canseeinpcb(req->td->td_ucred, inp);
1418 if (error == 0)
1419 cru2x(inp->inp_cred, &xuc);
1420 INP_RUNLOCK(inp);
1421 } else
1422 error = ENOENT;
1423 if (error == 0)
1424 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1425 return (error);
1426}
1427
1428SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1429 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1430 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1431#endif /* INET */
1432
1433#ifdef INET6
1434static int
1435tcp6_getcred(SYSCTL_HANDLER_ARGS)
1436{
1437 struct xucred xuc;
1438 struct sockaddr_in6 addrs[2];
1439 struct inpcb *inp;
1440 int error;
1441#ifdef INET
1442 int mapped = 0;
1443#endif
1444
1445 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1446 if (error)
1447 return (error);
1448 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1449 if (error)
1450 return (error);
1451 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1452 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1453 return (error);
1454 }
1455 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1456#ifdef INET
1457 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1458 mapped = 1;
1459 else
1460#endif
1461 return (EINVAL);
1462 }
1463
1464#ifdef INET
1465 if (mapped == 1)
1466 inp = in_pcblookup(&V_tcbinfo,
1467 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1468 addrs[1].sin6_port,
1469 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1470 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1471 else
1472#endif
1473 inp = in6_pcblookup(&V_tcbinfo,
1474 &addrs[1].sin6_addr, addrs[1].sin6_port,
1475 &addrs[0].sin6_addr, addrs[0].sin6_port,
1476 INPLOOKUP_RLOCKPCB, NULL);
1477 if (inp != NULL) {
1478 if (inp->inp_socket == NULL)
1479 error = ENOENT;
1480 if (error == 0)
1481 error = cr_canseeinpcb(req->td->td_ucred, inp);
1482 if (error == 0)
1483 cru2x(inp->inp_cred, &xuc);
1484 INP_RUNLOCK(inp);
1485 } else
1486 error = ENOENT;
1487 if (error == 0)
1488 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1489 return (error);
1490}
1491
1492SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1493 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1494 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1495#endif /* INET6 */
1496
1497
1498#ifdef INET
1499void
1500tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1501{
1502 struct ip *ip = vip;
1503 struct tcphdr *th;
1504 struct in_addr faddr;
1505 struct inpcb *inp;
1506 struct tcpcb *tp;
1507 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1508 struct icmp *icp;
1509 struct in_conninfo inc;
1510 tcp_seq icmp_tcp_seq;
1511 int mtu;
1512
1513 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1514 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1515 return;
1516
1517 if (cmd == PRC_MSGSIZE)
1518 notify = tcp_mtudisc_notify;
1519 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1520 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1521 notify = tcp_drop_syn_sent;
1522 /*
1523 * Redirects don't need to be handled up here.
1524 */
1525 else if (PRC_IS_REDIRECT(cmd))
1526 return;
1527 /*
1528 * Hostdead is ugly because it goes linearly through all PCBs.
1529 * XXX: We never get this from ICMP, otherwise it makes an
1530 * excellent DoS attack on machines with many connections.
1531 */
1532 else if (cmd == PRC_HOSTDEAD)
1533 ip = NULL;
1534 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1535 return;
1536
1537 if (ip == NULL) {
1538 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1539 return;
1540 }
1541
1542 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1543 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1544 INP_INFO_RLOCK(&V_tcbinfo);
1545 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1546 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1547 if (inp != NULL) {
1548 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1549 !(inp->inp_flags & INP_DROPPED) &&
1550 !(inp->inp_socket == NULL)) {
1551 icmp_tcp_seq = ntohl(th->th_seq);
1552 tp = intotcpcb(inp);
1553 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1554 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1555 if (cmd == PRC_MSGSIZE) {
1556 /*
1557 * MTU discovery:
1558 * If we got a needfrag set the MTU
1559 * in the route to the suggested new
1560 * value (if given) and then notify.
1561 */
1562 mtu = ntohs(icp->icmp_nextmtu);
1563 /*
1564 * If no alternative MTU was
1565 * proposed, try the next smaller
1566 * one.
1567 */
1568 if (!mtu)
1569 mtu = ip_next_mtu(
1570 ntohs(ip->ip_len), 1);
1571 if (mtu < V_tcp_minmss +
1572 sizeof(struct tcpiphdr))
1573 mtu = V_tcp_minmss +
1574 sizeof(struct tcpiphdr);
1575 /*
1576 * Only process the offered MTU if it
1577 * is smaller than the current one.
1578 */
1579 if (mtu < tp->t_maxopd +
1580 sizeof(struct tcpiphdr)) {
1581 bzero(&inc, sizeof(inc));
1582 inc.inc_faddr = faddr;
1583 inc.inc_fibnum =
1584 inp->inp_inc.inc_fibnum;
1585 tcp_hc_updatemtu(&inc, mtu);
1586 tcp_mtudisc(inp, mtu);
1587 }
1588 } else
1589 inp = (*notify)(inp,
1590 inetctlerrmap[cmd]);
1591 }
1592 }
1593 if (inp != NULL)
1594 INP_WUNLOCK(inp);
1595 } else {
1596 bzero(&inc, sizeof(inc));
1597 inc.inc_fport = th->th_dport;
1598 inc.inc_lport = th->th_sport;
1599 inc.inc_faddr = faddr;
1600 inc.inc_laddr = ip->ip_src;
1601 syncache_unreach(&inc, th);
1602 }
1603 INP_INFO_RUNLOCK(&V_tcbinfo);
1604}
1605#endif /* INET */
1606
1607#ifdef INET6
1608void
1609tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1610{
1611 struct tcphdr th;
1612 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1613 struct ip6_hdr *ip6;
1614 struct mbuf *m;
1615 struct ip6ctlparam *ip6cp = NULL;
1616 const struct sockaddr_in6 *sa6_src = NULL;
1617 int off;
1618 struct tcp_portonly {
1619 u_int16_t th_sport;
1620 u_int16_t th_dport;
1621 } *thp;
1622
1623 if (sa->sa_family != AF_INET6 ||
1624 sa->sa_len != sizeof(struct sockaddr_in6))
1625 return;
1626
1627 if (cmd == PRC_MSGSIZE)
1628 notify = tcp_mtudisc_notify;
1629 else if (!PRC_IS_REDIRECT(cmd) &&
1630 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1631 return;
1632
1633 /* if the parameter is from icmp6, decode it. */
1634 if (d != NULL) {
1635 ip6cp = (struct ip6ctlparam *)d;
1636 m = ip6cp->ip6c_m;
1637 ip6 = ip6cp->ip6c_ip6;
1638 off = ip6cp->ip6c_off;
1639 sa6_src = ip6cp->ip6c_src;
1640 } else {
1641 m = NULL;
1642 ip6 = NULL;
1643 off = 0; /* fool gcc */
1644 sa6_src = &sa6_any;
1645 }
1646
1647 if (ip6 != NULL) {
1648 struct in_conninfo inc;
1649 /*
1650 * XXX: We assume that when IPV6 is non NULL,
1651 * M and OFF are valid.
1652 */
1653
1654 /* check if we can safely examine src and dst ports */
1655 if (m->m_pkthdr.len < off + sizeof(*thp))
1656 return;
1657
1658 bzero(&th, sizeof(th));
1659 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1660
1661 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1662 (struct sockaddr *)ip6cp->ip6c_src,
1663 th.th_sport, cmd, NULL, notify);
1664
1665 bzero(&inc, sizeof(inc));
1666 inc.inc_fport = th.th_dport;
1667 inc.inc_lport = th.th_sport;
1668 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1669 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1670 inc.inc_flags |= INC_ISIPV6;
1671 INP_INFO_RLOCK(&V_tcbinfo);
1672 syncache_unreach(&inc, &th);
1673 INP_INFO_RUNLOCK(&V_tcbinfo);
1674 } else
1675 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1676 0, cmd, NULL, notify);
1677}
1678#endif /* INET6 */
1679
1680
1681/*
1682 * Following is where TCP initial sequence number generation occurs.
1683 *
1684 * There are two places where we must use initial sequence numbers:
1685 * 1. In SYN-ACK packets.
1686 * 2. In SYN packets.
1687 *
1688 * All ISNs for SYN-ACK packets are generated by the syncache. See
1689 * tcp_syncache.c for details.
1690 *
1691 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1692 * depends on this property. In addition, these ISNs should be
1693 * unguessable so as to prevent connection hijacking. To satisfy
1694 * the requirements of this situation, the algorithm outlined in
1695 * RFC 1948 is used, with only small modifications.
1696 *
1697 * Implementation details:
1698 *
1699 * Time is based off the system timer, and is corrected so that it
1700 * increases by one megabyte per second. This allows for proper
1701 * recycling on high speed LANs while still leaving over an hour
1702 * before rollover.
1703 *
1704 * As reading the *exact* system time is too expensive to be done
1705 * whenever setting up a TCP connection, we increment the time
1706 * offset in two ways. First, a small random positive increment
1707 * is added to isn_offset for each connection that is set up.
1708 * Second, the function tcp_isn_tick fires once per clock tick
1709 * and increments isn_offset as necessary so that sequence numbers
1710 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1711 * random positive increments serve only to ensure that the same
1712 * exact sequence number is never sent out twice (as could otherwise
1713 * happen when a port is recycled in less than the system tick
1714 * interval.)
1715 *
1716 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1717 * between seeding of isn_secret. This is normally set to zero,
1718 * as reseeding should not be necessary.
1719 *
1720 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1721 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1722 * general, this means holding an exclusive (write) lock.
1723 */
1724
1725#define ISN_BYTES_PER_SECOND 1048576
1726#define ISN_STATIC_INCREMENT 4096
1727#define ISN_RANDOM_INCREMENT (4096 - 1)
1728
1729static VNET_DEFINE(u_char, isn_secret[32]);
1730static VNET_DEFINE(int, isn_last);
1731static VNET_DEFINE(int, isn_last_reseed);
1732static VNET_DEFINE(u_int32_t, isn_offset);
1733static VNET_DEFINE(u_int32_t, isn_offset_old);
1734
1735#define V_isn_secret VNET(isn_secret)
1736#define V_isn_last VNET(isn_last)
1737#define V_isn_last_reseed VNET(isn_last_reseed)
1738#define V_isn_offset VNET(isn_offset)
1739#define V_isn_offset_old VNET(isn_offset_old)
1740
1741tcp_seq
1742tcp_new_isn(struct tcpcb *tp)
1743{
1744 MD5_CTX isn_ctx;
1745 u_int32_t md5_buffer[4];
1746 tcp_seq new_isn;
1747 u_int32_t projected_offset;
1748
1749 INP_WLOCK_ASSERT(tp->t_inpcb);
1750
1751 ISN_LOCK();
1752 /* Seed if this is the first use, reseed if requested. */
1753 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1754 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1755 < (u_int)ticks))) {
1756 read_random(&V_isn_secret, sizeof(V_isn_secret));
1757 V_isn_last_reseed = ticks;
1758 }
1759
1760 /* Compute the md5 hash and return the ISN. */
1761 MD5Init(&isn_ctx);
1762 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1763 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1764#ifdef INET6
1765 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1766 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1767 sizeof(struct in6_addr));
1768 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1769 sizeof(struct in6_addr));
1770 } else
1771#endif
1772 {
1773 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1774 sizeof(struct in_addr));
1775 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1776 sizeof(struct in_addr));
1777 }
1778 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1779 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1780 new_isn = (tcp_seq) md5_buffer[0];
1781 V_isn_offset += ISN_STATIC_INCREMENT +
1782 (arc4random() & ISN_RANDOM_INCREMENT);
1783 if (ticks != V_isn_last) {
1784 projected_offset = V_isn_offset_old +
1785 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1786 if (SEQ_GT(projected_offset, V_isn_offset))
1787 V_isn_offset = projected_offset;
1788 V_isn_offset_old = V_isn_offset;
1789 V_isn_last = ticks;
1790 }
1791 new_isn += V_isn_offset;
1792 ISN_UNLOCK();
1793 return (new_isn);
1794}
1795
1796/*
1797 * When a specific ICMP unreachable message is received and the
1798 * connection state is SYN-SENT, drop the connection. This behavior
1799 * is controlled by the icmp_may_rst sysctl.
1800 */
1801struct inpcb *
1802tcp_drop_syn_sent(struct inpcb *inp, int errno)
1803{
1804 struct tcpcb *tp;
1805
1806 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1807 INP_WLOCK_ASSERT(inp);
1808
1809 if ((inp->inp_flags & INP_TIMEWAIT) ||
1810 (inp->inp_flags & INP_DROPPED))
1811 return (inp);
1812
1813 tp = intotcpcb(inp);
1814 if (tp->t_state != TCPS_SYN_SENT)
1815 return (inp);
1816
1817 tp = tcp_drop(tp, errno);
1818 if (tp != NULL)
1819 return (inp);
1820 else
1821 return (NULL);
1822}
1823
1824/*
1825 * When `need fragmentation' ICMP is received, update our idea of the MSS
1826 * based on the new value. Also nudge TCP to send something, since we
1827 * know the packet we just sent was dropped.
1828 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1829 */
1830static struct inpcb *
1831tcp_mtudisc_notify(struct inpcb *inp, int error)
1832{
1833
1834 tcp_mtudisc(inp, -1);
1835 return (inp);
1836}
1837
1838static void
1839tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1840{
1841 struct tcpcb *tp;
1842 struct socket *so;
1843
1844 INP_WLOCK_ASSERT(inp);
1845 if ((inp->inp_flags & INP_TIMEWAIT) ||
1846 (inp->inp_flags & INP_DROPPED))
1847 return;
1848
1849 tp = intotcpcb(inp);
1850 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1851
1852 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1853
1854 so = inp->inp_socket;
1855 SOCKBUF_LOCK(&so->so_snd);
1856 /* If the mss is larger than the socket buffer, decrease the mss. */
1857 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1858 tp->t_maxseg = so->so_snd.sb_hiwat;
1859 SOCKBUF_UNLOCK(&so->so_snd);
1860
1861 TCPSTAT_INC(tcps_mturesent);
1862 tp->t_rtttime = 0;
1863 tp->snd_nxt = tp->snd_una;
1864 tcp_free_sackholes(tp);
1865 tp->snd_recover = tp->snd_max;
1866 if (tp->t_flags & TF_SACK_PERMIT)
1867 EXIT_FASTRECOVERY(tp->t_flags);
1868 tcp_output(tp);
1869}
1870
1871#ifdef INET
1872/*
1873 * Look-up the routing entry to the peer of this inpcb. If no route
1874 * is found and it cannot be allocated, then return 0. This routine
1875 * is called by TCP routines that access the rmx structure and by
1876 * tcp_mss_update to get the peer/interface MTU.
1877 */
1878u_long
1879tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1880{
1881 struct route sro;
1882 struct sockaddr_in *dst;
1883 struct ifnet *ifp;
1884 u_long maxmtu = 0;
1885
1886 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1887
1888 bzero(&sro, sizeof(sro));
1889 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1890 dst = (struct sockaddr_in *)&sro.ro_dst;
1891 dst->sin_family = AF_INET;
1892 dst->sin_len = sizeof(*dst);
1893 dst->sin_addr = inc->inc_faddr;
1894 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1895 }
1896 if (sro.ro_rt != NULL) {
1897 ifp = sro.ro_rt->rt_ifp;
1898 if (sro.ro_rt->rt_mtu == 0)
1899 maxmtu = ifp->if_mtu;
1900 else
1901 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1902
1903 /* Report additional interface capabilities. */
1904 if (cap != NULL) {
1905 if (ifp->if_capenable & IFCAP_TSO4 &&
1906 ifp->if_hwassist & CSUM_TSO) {
1907 cap->ifcap |= CSUM_TSO;
1908 cap->tsomax = ifp->if_hw_tsomax;
1909 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1910 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1911 }
1912 }
1913 RTFREE(sro.ro_rt);
1914 }
1915 return (maxmtu);
1916}
1917#endif /* INET */
1918
1919#ifdef INET6
1920u_long
1921tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1922{
1923 struct route_in6 sro6;
1924 struct ifnet *ifp;
1925 u_long maxmtu = 0;
1926
1927 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1928
1929 bzero(&sro6, sizeof(sro6));
1930 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1931 sro6.ro_dst.sin6_family = AF_INET6;
1932 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1933 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1934 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1935 }
1936 if (sro6.ro_rt != NULL) {
1937 ifp = sro6.ro_rt->rt_ifp;
1938 if (sro6.ro_rt->rt_mtu == 0)
1939 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1940 else
1941 maxmtu = min(sro6.ro_rt->rt_mtu,
1942 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1943
1944 /* Report additional interface capabilities. */
1945 if (cap != NULL) {
1946 if (ifp->if_capenable & IFCAP_TSO6 &&
1947 ifp->if_hwassist & CSUM_TSO) {
1948 cap->ifcap |= CSUM_TSO;
1949 cap->tsomax = ifp->if_hw_tsomax;
1950 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1951 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1952 }
1953 }
1954 RTFREE(sro6.ro_rt);
1955 }
1956
1957 return (maxmtu);
1958}
1959#endif /* INET6 */
1960
1961#ifdef IPSEC
1962/* compute ESP/AH header size for TCP, including outer IP header. */
1963size_t
1964ipsec_hdrsiz_tcp(struct tcpcb *tp)
1965{
1966 struct inpcb *inp;
1967 struct mbuf *m;
1968 size_t hdrsiz;
1969 struct ip *ip;
1970#ifdef INET6
1971 struct ip6_hdr *ip6;
1972#endif
1973 struct tcphdr *th;
1974
1975 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1976 return (0);
1977 m = m_gethdr(M_NOWAIT, MT_DATA);
1978 if (!m)
1979 return (0);
1980
1981#ifdef INET6
1982 if ((inp->inp_vflag & INP_IPV6) != 0) {
1983 ip6 = mtod(m, struct ip6_hdr *);
1984 th = (struct tcphdr *)(ip6 + 1);
1985 m->m_pkthdr.len = m->m_len =
1986 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1987 tcpip_fillheaders(inp, ip6, th);
1988 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1989 } else
1990#endif /* INET6 */
1991 {
1992 ip = mtod(m, struct ip *);
1993 th = (struct tcphdr *)(ip + 1);
1994 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1995 tcpip_fillheaders(inp, ip, th);
1996 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1997 }
1998
1999 m_free(m);
2000 return (hdrsiz);
2001}
2002#endif /* IPSEC */
2003
2004#ifdef TCP_SIGNATURE
2005/*
2006 * Callback function invoked by m_apply() to digest TCP segment data
2007 * contained within an mbuf chain.
2008 */
2009static int
2010tcp_signature_apply(void *fstate, void *data, u_int len)
2011{
2012
2013 MD5Update(fstate, (u_char *)data, len);
2014 return (0);
2015}
2016
2017/*
2018 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2019 * search with the destination IP address, and a 'magic SPI' to be
2020 * determined by the application. This is hardcoded elsewhere to 1179
2021*/
2022struct secasvar *
2023tcp_get_sav(struct mbuf *m, u_int direction)
2024{
2025 union sockaddr_union dst;
2026 struct secasvar *sav;
2027 struct ip *ip;
2028#ifdef INET6
2029 struct ip6_hdr *ip6;
2030 char ip6buf[INET6_ADDRSTRLEN];
2031#endif
2032
2033 /* Extract the destination from the IP header in the mbuf. */
2034 bzero(&dst, sizeof(union sockaddr_union));
2035 ip = mtod(m, struct ip *);
2036#ifdef INET6
2037 ip6 = NULL; /* Make the compiler happy. */
2038#endif
2039 switch (ip->ip_v) {
2040#ifdef INET
2041 case IPVERSION:
2042 dst.sa.sa_len = sizeof(struct sockaddr_in);
2043 dst.sa.sa_family = AF_INET;
2044 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2045 ip->ip_src : ip->ip_dst;
2046 break;
2047#endif
2048#ifdef INET6
2049 case (IPV6_VERSION >> 4):
2050 ip6 = mtod(m, struct ip6_hdr *);
2051 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2052 dst.sa.sa_family = AF_INET6;
2053 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2054 ip6->ip6_src : ip6->ip6_dst;
2055 break;
2056#endif
2057 default:
2058 return (NULL);
2059 /* NOTREACHED */
2060 break;
2061 }
2062
2063 /* Look up an SADB entry which matches the address of the peer. */
2064 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2065 if (sav == NULL) {
2066 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2067 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2068#ifdef INET6
2069 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2070 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2071#endif
2072 "(unsupported)"));
2073 }
2074
2075 return (sav);
2076}
2077
2078/*
2079 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2080 *
2081 * Parameters:
2082 * m pointer to head of mbuf chain
2083 * len length of TCP segment data, excluding options
2084 * optlen length of TCP segment options
2085 * buf pointer to storage for computed MD5 digest
2086 * sav pointer to security assosiation
2087 *
2088 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2089 * When called from tcp_input(), we can be sure that th_sum has been
2090 * zeroed out and verified already.
2091 *
2092 * Releases reference to SADB key before return.
2093 *
2094 * Return 0 if successful, otherwise return -1.
2095 *
2096 */
2097int
2098tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2099 u_char *buf, struct secasvar *sav)
2100{
2101#ifdef INET
2102 struct ippseudo ippseudo;
2103#endif
2104 MD5_CTX ctx;
2105 int doff;
2106 struct ip *ip;
2107#ifdef INET
2108 struct ipovly *ipovly;
2109#endif
2110 struct tcphdr *th;
2111#ifdef INET6
2112 struct ip6_hdr *ip6;
2113 struct in6_addr in6;
2114 uint32_t plen;
2115 uint16_t nhdr;
2116#endif
2117 u_short savecsum;
2118
2119 KASSERT(m != NULL, ("NULL mbuf chain"));
2120 KASSERT(buf != NULL, ("NULL signature pointer"));
2121
2122 /* Extract the destination from the IP header in the mbuf. */
2123 ip = mtod(m, struct ip *);
2124#ifdef INET6
2125 ip6 = NULL; /* Make the compiler happy. */
2126#endif
2127
2128 MD5Init(&ctx);
2129 /*
2130 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2131 *
2132 * XXX The ippseudo header MUST be digested in network byte order,
2133 * or else we'll fail the regression test. Assume all fields we've
2134 * been doing arithmetic on have been in host byte order.
2135 * XXX One cannot depend on ipovly->ih_len here. When called from
2136 * tcp_output(), the underlying ip_len member has not yet been set.
2137 */
2138 switch (ip->ip_v) {
2139#ifdef INET
2140 case IPVERSION:
2141 ipovly = (struct ipovly *)ip;
2142 ippseudo.ippseudo_src = ipovly->ih_src;
2143 ippseudo.ippseudo_dst = ipovly->ih_dst;
2144 ippseudo.ippseudo_pad = 0;
2145 ippseudo.ippseudo_p = IPPROTO_TCP;
2146 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2147 optlen);
2148 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2149
2150 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2151 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2152 break;
2153#endif
2154#ifdef INET6
2155 /*
2156 * RFC 2385, 2.0 Proposal
2157 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2158 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2159 * extended next header value (to form 32 bits), and 32-bit segment
2160 * length.
2161 * Note: Upper-Layer Packet Length comes before Next Header.
2162 */
2163 case (IPV6_VERSION >> 4):
2164 in6 = ip6->ip6_src;
2165 in6_clearscope(&in6);
2166 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2167 in6 = ip6->ip6_dst;
2168 in6_clearscope(&in6);
2169 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2170 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2171 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2172 nhdr = 0;
2173 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2174 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2175 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2176 nhdr = IPPROTO_TCP;
2177 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2178
2179 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2180 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2181 break;
2182#endif
2183 default:
2184 KEY_FREESAV(&sav);
2185 return (-1);
2186 /* NOTREACHED */
2187 break;
2188 }
2189
2190
2191 /*
2192 * Step 2: Update MD5 hash with TCP header, excluding options.
2193 * The TCP checksum must be set to zero.
2194 */
2195 savecsum = th->th_sum;
2196 th->th_sum = 0;
2197 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2198 th->th_sum = savecsum;
2199
2200 /*
2201 * Step 3: Update MD5 hash with TCP segment data.
2202 * Use m_apply() to avoid an early m_pullup().
2203 */
2204 if (len > 0)
2205 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2206
2207 /*
2208 * Step 4: Update MD5 hash with shared secret.
2209 */
2210 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2211 MD5Final(buf, &ctx);
2212
2213 key_sa_recordxfer(sav, m);
2214 KEY_FREESAV(&sav);
2215 return (0);
2216}
2217
2218/*
2219 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2220 *
2221 * Return 0 if successful, otherwise return -1.
2222 */
2223int
2224tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2225 u_char *buf, u_int direction)
2226{
2227 struct secasvar *sav;
2228
2229 if ((sav = tcp_get_sav(m, direction)) == NULL)
2230 return (-1);
2231
2232 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2233}
2234
2235/*
2236 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2237 *
2238 * Parameters:
2239 * m pointer to head of mbuf chain
2240 * len length of TCP segment data, excluding options
2241 * optlen length of TCP segment options
2242 * buf pointer to storage for computed MD5 digest
2243 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2244 *
2245 * Return 1 if successful, otherwise return 0.
2246 */
2247int
2248tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2249 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2250{
2251 char tmpdigest[TCP_SIGLEN];
2252
2253 if (tcp_sig_checksigs == 0)
2254 return (1);
2255 if ((tcpbflag & TF_SIGNATURE) == 0) {
2256 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2257
2258 /*
2259 * If this socket is not expecting signature but
2260 * the segment contains signature just fail.
2261 */
2262 TCPSTAT_INC(tcps_sig_err_sigopt);
2263 TCPSTAT_INC(tcps_sig_rcvbadsig);
2264 return (0);
2265 }
2266
2267 /* Signature is not expected, and not present in segment. */
2268 return (1);
2269 }
2270
2271 /*
2272 * If this socket is expecting signature but the segment does not
2273 * contain any just fail.
2274 */
2275 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2276 TCPSTAT_INC(tcps_sig_err_nosigopt);
2277 TCPSTAT_INC(tcps_sig_rcvbadsig);
2278 return (0);
2279 }
2280 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2281 IPSEC_DIR_INBOUND) == -1) {
2282 TCPSTAT_INC(tcps_sig_err_buildsig);
2283 TCPSTAT_INC(tcps_sig_rcvbadsig);
2284 return (0);
2285 }
2286
2287 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2288 TCPSTAT_INC(tcps_sig_rcvbadsig);
2289 return (0);
2290 }
2291 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2292 return (1);
2293}
2294#endif /* TCP_SIGNATURE */
2295
2296static int
2297sysctl_drop(SYSCTL_HANDLER_ARGS)
2298{
2299 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2300 struct sockaddr_storage addrs[2];
2301 struct inpcb *inp;
2302 struct tcpcb *tp;
2303 struct tcptw *tw;
2304 struct sockaddr_in *fin, *lin;
2305#ifdef INET6
2306 struct sockaddr_in6 *fin6, *lin6;
2307#endif
2308 int error;
2309
2310 inp = NULL;
2311 fin = lin = NULL;
2312#ifdef INET6
2313 fin6 = lin6 = NULL;
2314#endif
2315 error = 0;
2316
2317 if (req->oldptr != NULL || req->oldlen != 0)
2318 return (EINVAL);
2319 if (req->newptr == NULL)
2320 return (EPERM);
2321 if (req->newlen < sizeof(addrs))
2322 return (ENOMEM);
2323 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2324 if (error)
2325 return (error);
2326
2327 switch (addrs[0].ss_family) {
2328#ifdef INET6
2329 case AF_INET6:
2330 fin6 = (struct sockaddr_in6 *)&addrs[0];
2331 lin6 = (struct sockaddr_in6 *)&addrs[1];
2332 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2333 lin6->sin6_len != sizeof(struct sockaddr_in6))
2334 return (EINVAL);
2335 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2336 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2337 return (EINVAL);
2338 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2339 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2340 fin = (struct sockaddr_in *)&addrs[0];
2341 lin = (struct sockaddr_in *)&addrs[1];
2342 break;
2343 }
2344 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2345 if (error)
2346 return (error);
2347 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2348 if (error)
2349 return (error);
2350 break;
2351#endif
2352#ifdef INET
2353 case AF_INET:
2354 fin = (struct sockaddr_in *)&addrs[0];
2355 lin = (struct sockaddr_in *)&addrs[1];
2356 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2357 lin->sin_len != sizeof(struct sockaddr_in))
2358 return (EINVAL);
2359 break;
2360#endif
2361 default:
2362 return (EINVAL);
2363 }
2364 INP_INFO_RLOCK(&V_tcbinfo);
2365 switch (addrs[0].ss_family) {
2366#ifdef INET6
2367 case AF_INET6:
2368 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2369 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2370 INPLOOKUP_WLOCKPCB, NULL);
2371 break;
2372#endif
2373#ifdef INET
2374 case AF_INET:
2375 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2376 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2377 break;
2378#endif
2379 }
2380 if (inp != NULL) {
2381 if (inp->inp_flags & INP_TIMEWAIT) {
2382 /*
2383 * XXXRW: There currently exists a state where an
2384 * inpcb is present, but its timewait state has been
2385 * discarded. For now, don't allow dropping of this
2386 * type of inpcb.
2387 */
2388 tw = intotw(inp);
2389 if (tw != NULL)
2390 tcp_twclose(tw, 0);
2391 else
2392 INP_WUNLOCK(inp);
2393 } else if (!(inp->inp_flags & INP_DROPPED) &&
2394 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2395 tp = intotcpcb(inp);
2396 tp = tcp_drop(tp, ECONNABORTED);
2397 if (tp != NULL)
2398 INP_WUNLOCK(inp);
2399 } else
2400 INP_WUNLOCK(inp);
2401 } else
2402 error = ESRCH;
2403 INP_INFO_RUNLOCK(&V_tcbinfo);
2404 return (error);
2405}
2406
2407SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2408 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2409 0, sysctl_drop, "", "Drop TCP connection");
2410
2411/*
2412 * Generate a standardized TCP log line for use throughout the
2413 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2414 * allow use in the interrupt context.
2415 *
2416 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2417 * NB: The function may return NULL if memory allocation failed.
2418 *
2419 * Due to header inclusion and ordering limitations the struct ip
2420 * and ip6_hdr pointers have to be passed as void pointers.
2421 */
2422char *
2423tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2424 const void *ip6hdr)
2425{
2426
2427 /* Is logging enabled? */
2428 if (tcp_log_in_vain == 0)
2429 return (NULL);
2430
2431 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2432}
2433
2434char *
2435tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2436 const void *ip6hdr)
2437{
2438
2439 /* Is logging enabled? */
2440 if (tcp_log_debug == 0)
2441 return (NULL);
2442
2443 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2444}
2445
2446static char *
2447tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2448 const void *ip6hdr)
2449{
2450 char *s, *sp;
2451 size_t size;
2452 struct ip *ip;
2453#ifdef INET6
2454 const struct ip6_hdr *ip6;
2455
2456 ip6 = (const struct ip6_hdr *)ip6hdr;
2457#endif /* INET6 */
2458 ip = (struct ip *)ip4hdr;
2459
2460 /*
2461 * The log line looks like this:
2462 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2463 */
2464 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2465 sizeof(PRINT_TH_FLAGS) + 1 +
2466#ifdef INET6
2467 2 * INET6_ADDRSTRLEN;
2468#else
2469 2 * INET_ADDRSTRLEN;
2470#endif /* INET6 */
2471
2472 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2473 if (s == NULL)
2474 return (NULL);
2475
2476 strcat(s, "TCP: [");
2477 sp = s + strlen(s);
2478
2479 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2480 inet_ntoa_r(inc->inc_faddr, sp);
2481 sp = s + strlen(s);
2482 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2483 sp = s + strlen(s);
2484 inet_ntoa_r(inc->inc_laddr, sp);
2485 sp = s + strlen(s);
2486 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2487#ifdef INET6
2488 } else if (inc) {
2489 ip6_sprintf(sp, &inc->inc6_faddr);
2490 sp = s + strlen(s);
2491 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2492 sp = s + strlen(s);
2493 ip6_sprintf(sp, &inc->inc6_laddr);
2494 sp = s + strlen(s);
2495 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2496 } else if (ip6 && th) {
2497 ip6_sprintf(sp, &ip6->ip6_src);
2498 sp = s + strlen(s);
2499 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2500 sp = s + strlen(s);
2501 ip6_sprintf(sp, &ip6->ip6_dst);
2502 sp = s + strlen(s);
2503 sprintf(sp, "]:%i", ntohs(th->th_dport));
2504#endif /* INET6 */
2505#ifdef INET
2506 } else if (ip && th) {
2507 inet_ntoa_r(ip->ip_src, sp);
2508 sp = s + strlen(s);
2509 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2510 sp = s + strlen(s);
2511 inet_ntoa_r(ip->ip_dst, sp);
2512 sp = s + strlen(s);
2513 sprintf(sp, "]:%i", ntohs(th->th_dport));
2514#endif /* INET */
2515 } else {
2516 free(s, M_TCPLOG);
2517 return (NULL);
2518 }
2519 sp = s + strlen(s);
2520 if (th)
2521 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2522 if (*(s + size - 1) != '\0')
2523 panic("%s: string too long", __func__);
2524 return (s);
2525}
2526
2527/*
2528 * A subroutine which makes it easy to track TCP state changes with DTrace.
2529 * This function shouldn't be called for t_state initializations that don't
2530 * correspond to actual TCP state transitions.
2531 */
2532void
2533tcp_state_change(struct tcpcb *tp, int newstate)
2534{
2535#if defined(KDTRACE_HOOKS)
2536 int pstate = tp->t_state;
2537#endif
2538
2539 tp->t_state = newstate;
2540 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2541}
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 289276 2015-10-14 00:35:37Z hiren $");
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/sdt.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/protosw.h>
60#include <sys/random.h>
61
62#include <vm/uma.h>
63
64#include <net/route.h>
65#include <net/if.h>
66#include <net/if_var.h>
67#include <net/vnet.h>
68
69#include <netinet/cc.h>
70#include <netinet/in.h>
71#include <netinet/in_kdtrace.h>
72#include <netinet/in_pcb.h>
73#include <netinet/in_systm.h>
74#include <netinet/in_var.h>
75#include <netinet/ip.h>
76#include <netinet/ip_icmp.h>
77#include <netinet/ip_var.h>
78#ifdef INET6
79#include <netinet/ip6.h>
80#include <netinet6/in6_pcb.h>
81#include <netinet6/ip6_var.h>
82#include <netinet6/scope6_var.h>
83#include <netinet6/nd6.h>
84#endif
85
86#include <netinet/tcp_fsm.h>
87#include <netinet/tcp_seq.h>
88#include <netinet/tcp_timer.h>
89#include <netinet/tcp_var.h>
90#include <netinet/tcp_syncache.h>
91#ifdef INET6
92#include <netinet6/tcp6_var.h>
93#endif
94#include <netinet/tcpip.h>
95#ifdef TCPPCAP
96#include <netinet/tcp_pcap.h>
97#endif
98#ifdef TCPDEBUG
99#include <netinet/tcp_debug.h>
100#endif
101#ifdef INET6
102#include <netinet6/ip6protosw.h>
103#endif
104#ifdef TCP_OFFLOAD
105#include <netinet/tcp_offload.h>
106#endif
107
108#ifdef IPSEC
109#include <netipsec/ipsec.h>
110#include <netipsec/xform.h>
111#ifdef INET6
112#include <netipsec/ipsec6.h>
113#endif
114#include <netipsec/key.h>
115#include <sys/syslog.h>
116#endif /*IPSEC*/
117
118#include <machine/in_cksum.h>
119#include <sys/md5.h>
120
121#include <security/mac/mac_framework.h>
122
123VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
124#ifdef INET6
125VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
126#endif
127
128static int
129sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
130{
131 int error, new;
132
133 new = V_tcp_mssdflt;
134 error = sysctl_handle_int(oidp, &new, 0, req);
135 if (error == 0 && req->newptr) {
136 if (new < TCP_MINMSS)
137 error = EINVAL;
138 else
139 V_tcp_mssdflt = new;
140 }
141 return (error);
142}
143
144SYSCTL_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
145 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
146 &sysctl_net_inet_tcp_mss_check, "I",
147 "Default TCP Maximum Segment Size");
148
149#ifdef INET6
150static int
151sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
152{
153 int error, new;
154
155 new = V_tcp_v6mssdflt;
156 error = sysctl_handle_int(oidp, &new, 0, req);
157 if (error == 0 && req->newptr) {
158 if (new < TCP_MINMSS)
159 error = EINVAL;
160 else
161 V_tcp_v6mssdflt = new;
162 }
163 return (error);
164}
165
166SYSCTL_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
167 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
168 &sysctl_net_inet_tcp_mss_v6_check, "I",
169 "Default TCP Maximum Segment Size for IPv6");
170#endif /* INET6 */
171
172/*
173 * Minimum MSS we accept and use. This prevents DoS attacks where
174 * we are forced to a ridiculous low MSS like 20 and send hundreds
175 * of packets instead of one. The effect scales with the available
176 * bandwidth and quickly saturates the CPU and network interface
177 * with packet generation and sending. Set to zero to disable MINMSS
178 * checking. This setting prevents us from sending too small packets.
179 */
180VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
181SYSCTL_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_VNET | CTLFLAG_RW,
182 &VNET_NAME(tcp_minmss), 0,
183 "Minimum TCP Maximum Segment Size");
184
185VNET_DEFINE(int, tcp_do_rfc1323) = 1;
186SYSCTL_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc1323), 0,
188 "Enable rfc1323 (high performance TCP) extensions");
189
190static int tcp_log_debug = 0;
191SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
192 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
193
194static int tcp_tcbhashsize;
195SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,
196 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
197
198static int do_tcpdrain = 1;
199SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
200 "Enable tcp_drain routine for extra help when low on mbufs");
201
202SYSCTL_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_VNET | CTLFLAG_RD,
203 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
204
205static VNET_DEFINE(int, icmp_may_rst) = 1;
206#define V_icmp_may_rst VNET(icmp_may_rst)
207SYSCTL_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_VNET | CTLFLAG_RW,
208 &VNET_NAME(icmp_may_rst), 0,
209 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
210
211static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
212#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
213SYSCTL_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_VNET | CTLFLAG_RW,
214 &VNET_NAME(tcp_isn_reseed_interval), 0,
215 "Seconds between reseeding of ISN secret");
216
217static int tcp_soreceive_stream;
218SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
219 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
220
221#ifdef TCP_SIGNATURE
222static int tcp_sig_checksigs = 1;
223SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
224 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
225#endif
226
227VNET_DEFINE(uma_zone_t, sack_hole_zone);
228#define V_sack_hole_zone VNET(sack_hole_zone)
229
230VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
231
232static struct inpcb *tcp_notify(struct inpcb *, int);
233static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234static void tcp_mtudisc(struct inpcb *, int);
235static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
236 void *ip4hdr, const void *ip6hdr);
237static void tcp_timer_discard(struct tcpcb *, uint32_t);
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 0
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 */
255struct tcpcb_mem {
256 struct tcpcb tcb;
257 struct tcp_timer tt;
258 struct cc_var ccv;
259 struct osd osd;
260};
261
262static VNET_DEFINE(uma_zone_t, tcpcb_zone);
263#define V_tcpcb_zone VNET(tcpcb_zone)
264
265MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
266static struct mtx isn_mtx;
267
268#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
269#define ISN_LOCK() mtx_lock(&isn_mtx)
270#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
271
272/*
273 * TCP initialization.
274 */
275static void
276tcp_zone_change(void *tag)
277{
278
279 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
280 uma_zone_set_max(V_tcpcb_zone, maxsockets);
281 tcp_tw_zone_change();
282}
283
284static int
285tcp_inpcb_init(void *mem, int size, int flags)
286{
287 struct inpcb *inp = mem;
288
289 INP_LOCK_INIT(inp, "inp", "tcpinp");
290 return (0);
291}
292
293/*
294 * Take a value and get the next power of 2 that doesn't overflow.
295 * Used to size the tcp_inpcb hash buckets.
296 */
297static int
298maketcp_hashsize(int size)
299{
300 int hashsize;
301
302 /*
303 * auto tune.
304 * get the next power of 2 higher than maxsockets.
305 */
306 hashsize = 1 << fls(size);
307 /* catch overflow, and just go one power of 2 smaller */
308 if (hashsize < size) {
309 hashsize = 1 << (fls(size) - 1);
310 }
311 return (hashsize);
312}
313
314void
315tcp_init(void)
316{
317 const char *tcbhash_tuneable;
318 int hashsize;
319
320 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
321
322 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
323 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
324 printf("%s: WARNING: unable to register helper hook\n", __func__);
325 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
326 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
327 printf("%s: WARNING: unable to register helper hook\n", __func__);
328
329 hashsize = TCBHASHSIZE;
330 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
331 if (hashsize == 0) {
332 /*
333 * Auto tune the hash size based on maxsockets.
334 * A perfect hash would have a 1:1 mapping
335 * (hashsize = maxsockets) however it's been
336 * suggested that O(2) average is better.
337 */
338 hashsize = maketcp_hashsize(maxsockets / 4);
339 /*
340 * Our historical default is 512,
341 * do not autotune lower than this.
342 */
343 if (hashsize < 512)
344 hashsize = 512;
345 if (bootverbose)
346 printf("%s: %s auto tuned to %d\n", __func__,
347 tcbhash_tuneable, hashsize);
348 }
349 /*
350 * We require a hashsize to be a power of two.
351 * Previously if it was not a power of two we would just reset it
352 * back to 512, which could be a nasty surprise if you did not notice
353 * the error message.
354 * Instead what we do is clip it to the closest power of two lower
355 * than the specified hash value.
356 */
357 if (!powerof2(hashsize)) {
358 int oldhashsize = hashsize;
359
360 hashsize = maketcp_hashsize(hashsize);
361 /* prevent absurdly low value */
362 if (hashsize < 16)
363 hashsize = 16;
364 printf("%s: WARNING: TCB hash size not a power of 2, "
365 "clipped from %d to %d.\n", __func__, oldhashsize,
366 hashsize);
367 }
368 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
369 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
370 IPI_HASHFIELDS_4TUPLE);
371
372 /*
373 * These have to be type stable for the benefit of the timers.
374 */
375 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
376 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
377 uma_zone_set_max(V_tcpcb_zone, maxsockets);
378 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
379
380 tcp_tw_init();
381 syncache_init();
382 tcp_hc_init();
383
384 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
385 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
386 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
387
388 /* Skip initialization of globals for non-default instances. */
389 if (!IS_DEFAULT_VNET(curvnet))
390 return;
391
392 tcp_reass_global_init();
393
394 /* XXX virtualize those bellow? */
395 tcp_delacktime = TCPTV_DELACK;
396 tcp_keepinit = TCPTV_KEEP_INIT;
397 tcp_keepidle = TCPTV_KEEP_IDLE;
398 tcp_keepintvl = TCPTV_KEEPINTVL;
399 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
400 tcp_msl = TCPTV_MSL;
401 tcp_rexmit_min = TCPTV_MIN;
402 if (tcp_rexmit_min < 1)
403 tcp_rexmit_min = 1;
404 tcp_rexmit_slop = TCPTV_CPU_VAR;
405 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
406 tcp_tcbhashsize = hashsize;
407
408 if (tcp_soreceive_stream) {
409#ifdef INET
410 tcp_usrreqs.pru_soreceive = soreceive_stream;
411#endif
412#ifdef INET6
413 tcp6_usrreqs.pru_soreceive = soreceive_stream;
414#endif /* INET6 */
415 }
416
417#ifdef INET6
418#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
419#else /* INET6 */
420#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
421#endif /* INET6 */
422 if (max_protohdr < TCP_MINPROTOHDR)
423 max_protohdr = TCP_MINPROTOHDR;
424 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
425 panic("tcp_init");
426#undef TCP_MINPROTOHDR
427
428 ISN_LOCK_INIT();
429 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
430 SHUTDOWN_PRI_DEFAULT);
431 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
432 EVENTHANDLER_PRI_ANY);
433#ifdef TCPPCAP
434 tcp_pcap_init();
435#endif
436}
437
438#ifdef VIMAGE
439void
440tcp_destroy(void)
441{
442 int error;
443
444 tcp_hc_destroy();
445 syncache_destroy();
446 tcp_tw_destroy();
447 in_pcbinfo_destroy(&V_tcbinfo);
448 uma_zdestroy(V_sack_hole_zone);
449 uma_zdestroy(V_tcpcb_zone);
450
451 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
452 if (error != 0) {
453 printf("%s: WARNING: unable to deregister helper hook "
454 "type=%d, id=%d: error %d returned\n", __func__,
455 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
456 }
457 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
458 if (error != 0) {
459 printf("%s: WARNING: unable to deregister helper hook "
460 "type=%d, id=%d: error %d returned\n", __func__,
461 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
462 }
463}
464#endif
465
466void
467tcp_fini(void *xtp)
468{
469
470}
471
472/*
473 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
474 * tcp_template used to store this data in mbufs, but we now recopy it out
475 * of the tcpcb each time to conserve mbufs.
476 */
477void
478tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
479{
480 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
481
482 INP_WLOCK_ASSERT(inp);
483
484#ifdef INET6
485 if ((inp->inp_vflag & INP_IPV6) != 0) {
486 struct ip6_hdr *ip6;
487
488 ip6 = (struct ip6_hdr *)ip_ptr;
489 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
490 (inp->inp_flow & IPV6_FLOWINFO_MASK);
491 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
492 (IPV6_VERSION & IPV6_VERSION_MASK);
493 ip6->ip6_nxt = IPPROTO_TCP;
494 ip6->ip6_plen = htons(sizeof(struct tcphdr));
495 ip6->ip6_src = inp->in6p_laddr;
496 ip6->ip6_dst = inp->in6p_faddr;
497 }
498#endif /* INET6 */
499#if defined(INET6) && defined(INET)
500 else
501#endif
502#ifdef INET
503 {
504 struct ip *ip;
505
506 ip = (struct ip *)ip_ptr;
507 ip->ip_v = IPVERSION;
508 ip->ip_hl = 5;
509 ip->ip_tos = inp->inp_ip_tos;
510 ip->ip_len = 0;
511 ip->ip_id = 0;
512 ip->ip_off = 0;
513 ip->ip_ttl = inp->inp_ip_ttl;
514 ip->ip_sum = 0;
515 ip->ip_p = IPPROTO_TCP;
516 ip->ip_src = inp->inp_laddr;
517 ip->ip_dst = inp->inp_faddr;
518 }
519#endif /* INET */
520 th->th_sport = inp->inp_lport;
521 th->th_dport = inp->inp_fport;
522 th->th_seq = 0;
523 th->th_ack = 0;
524 th->th_x2 = 0;
525 th->th_off = 5;
526 th->th_flags = 0;
527 th->th_win = 0;
528 th->th_urp = 0;
529 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
530}
531
532/*
533 * Create template to be used to send tcp packets on a connection.
534 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
535 * use for this function is in keepalives, which use tcp_respond.
536 */
537struct tcptemp *
538tcpip_maketemplate(struct inpcb *inp)
539{
540 struct tcptemp *t;
541
542 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
543 if (t == NULL)
544 return (NULL);
545 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
546 return (t);
547}
548
549/*
550 * Send a single message to the TCP at address specified by
551 * the given TCP/IP header. If m == NULL, then we make a copy
552 * of the tcpiphdr at th and send directly to the addressed host.
553 * This is used to force keep alive messages out using the TCP
554 * template for a connection. If flags are given then we send
555 * a message back to the TCP which originated the segment th,
556 * and discard the mbuf containing it and any other attached mbufs.
557 *
558 * In any case the ack and sequence number of the transmitted
559 * segment are as specified by the parameters.
560 *
561 * NOTE: If m != NULL, then th must point to *inside* the mbuf.
562 */
563void
564tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
565 tcp_seq ack, tcp_seq seq, int flags)
566{
567 int tlen;
568 int win = 0;
569 struct ip *ip;
570 struct tcphdr *nth;
571#ifdef INET6
572 struct ip6_hdr *ip6;
573 int isipv6;
574#endif /* INET6 */
575 int ipflags = 0;
576 struct inpcb *inp;
577
578 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
579
580#ifdef INET6
581 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
582 ip6 = ipgen;
583#endif /* INET6 */
584 ip = ipgen;
585
586 if (tp != NULL) {
587 inp = tp->t_inpcb;
588 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
589 INP_WLOCK_ASSERT(inp);
590 } else
591 inp = NULL;
592
593 if (tp != NULL) {
594 if (!(flags & TH_RST)) {
595 win = sbspace(&inp->inp_socket->so_rcv);
596 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
597 win = (long)TCP_MAXWIN << tp->rcv_scale;
598 }
599 }
600 if (m == NULL) {
601 m = m_gethdr(M_NOWAIT, MT_DATA);
602 if (m == NULL)
603 return;
604 tlen = 0;
605 m->m_data += max_linkhdr;
606#ifdef INET6
607 if (isipv6) {
608 bcopy((caddr_t)ip6, mtod(m, caddr_t),
609 sizeof(struct ip6_hdr));
610 ip6 = mtod(m, struct ip6_hdr *);
611 nth = (struct tcphdr *)(ip6 + 1);
612 } else
613#endif /* INET6 */
614 {
615 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
616 ip = mtod(m, struct ip *);
617 nth = (struct tcphdr *)(ip + 1);
618 }
619 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
620 flags = TH_ACK;
621 } else {
622 /*
623 * reuse the mbuf.
624 * XXX MRT We inherrit the FIB, which is lucky.
625 */
626 m_freem(m->m_next);
627 m->m_next = NULL;
628 m->m_data = (caddr_t)ipgen;
629 /* m_len is set later */
630 tlen = 0;
631#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
632#ifdef INET6
633 if (isipv6) {
634 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
635 nth = (struct tcphdr *)(ip6 + 1);
636 } else
637#endif /* INET6 */
638 {
639 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
640 nth = (struct tcphdr *)(ip + 1);
641 }
642 if (th != nth) {
643 /*
644 * this is usually a case when an extension header
645 * exists between the IPv6 header and the
646 * TCP header.
647 */
648 nth->th_sport = th->th_sport;
649 nth->th_dport = th->th_dport;
650 }
651 xchg(nth->th_dport, nth->th_sport, uint16_t);
652#undef xchg
653 }
654#ifdef INET6
655 if (isipv6) {
656 ip6->ip6_flow = 0;
657 ip6->ip6_vfc = IPV6_VERSION;
658 ip6->ip6_nxt = IPPROTO_TCP;
659 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
660 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
661 }
662#endif
663#if defined(INET) && defined(INET6)
664 else
665#endif
666#ifdef INET
667 {
668 tlen += sizeof (struct tcpiphdr);
669 ip->ip_len = htons(tlen);
670 ip->ip_ttl = V_ip_defttl;
671 if (V_path_mtu_discovery)
672 ip->ip_off |= htons(IP_DF);
673 }
674#endif
675 m->m_len = tlen;
676 m->m_pkthdr.len = tlen;
677 m->m_pkthdr.rcvif = NULL;
678#ifdef MAC
679 if (inp != NULL) {
680 /*
681 * Packet is associated with a socket, so allow the
682 * label of the response to reflect the socket label.
683 */
684 INP_WLOCK_ASSERT(inp);
685 mac_inpcb_create_mbuf(inp, m);
686 } else {
687 /*
688 * Packet is not associated with a socket, so possibly
689 * update the label in place.
690 */
691 mac_netinet_tcp_reply(m);
692 }
693#endif
694 nth->th_seq = htonl(seq);
695 nth->th_ack = htonl(ack);
696 nth->th_x2 = 0;
697 nth->th_off = sizeof (struct tcphdr) >> 2;
698 nth->th_flags = flags;
699 if (tp != NULL)
700 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
701 else
702 nth->th_win = htons((u_short)win);
703 nth->th_urp = 0;
704
705 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
706#ifdef INET6
707 if (isipv6) {
708 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
709 nth->th_sum = in6_cksum_pseudo(ip6,
710 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
711 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
712 NULL, NULL);
713 }
714#endif /* INET6 */
715#if defined(INET6) && defined(INET)
716 else
717#endif
718#ifdef INET
719 {
720 m->m_pkthdr.csum_flags = CSUM_TCP;
721 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
722 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
723 }
724#endif /* INET */
725#ifdef TCPDEBUG
726 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
727 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
728#endif
729 TCP_PROBE3(debug__input, tp, th, mtod(m, const char *));
730 if (flags & TH_RST)
731 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
732 tp, nth);
733
734 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
735#ifdef INET6
736 if (isipv6)
737 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
738#endif /* INET6 */
739#if defined(INET) && defined(INET6)
740 else
741#endif
742#ifdef INET
743 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
744#endif
745}
746
747/*
748 * Create a new TCP control block, making an
749 * empty reassembly queue and hooking it to the argument
750 * protocol control block. The `inp' parameter must have
751 * come from the zone allocator set up in tcp_init().
752 */
753struct tcpcb *
754tcp_newtcpcb(struct inpcb *inp)
755{
756 struct tcpcb_mem *tm;
757 struct tcpcb *tp;
758#ifdef INET6
759 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
760#endif /* INET6 */
761
762 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
763 if (tm == NULL)
764 return (NULL);
765 tp = &tm->tcb;
766
767 /* Initialise cc_var struct for this tcpcb. */
768 tp->ccv = &tm->ccv;
769 tp->ccv->type = IPPROTO_TCP;
770 tp->ccv->ccvc.tcp = tp;
771
772 /*
773 * Use the current system default CC algorithm.
774 */
775 CC_LIST_RLOCK();
776 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
777 CC_ALGO(tp) = CC_DEFAULT();
778 CC_LIST_RUNLOCK();
779
780 if (CC_ALGO(tp)->cb_init != NULL)
781 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
782 uma_zfree(V_tcpcb_zone, tm);
783 return (NULL);
784 }
785
786 tp->osd = &tm->osd;
787 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
788 uma_zfree(V_tcpcb_zone, tm);
789 return (NULL);
790 }
791
792#ifdef VIMAGE
793 tp->t_vnet = inp->inp_vnet;
794#endif
795 tp->t_timers = &tm->tt;
796 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
797 tp->t_maxseg = tp->t_maxopd =
798#ifdef INET6
799 isipv6 ? V_tcp_v6mssdflt :
800#endif /* INET6 */
801 V_tcp_mssdflt;
802
803 /* Set up our timeouts. */
804 callout_init(&tp->t_timers->tt_rexmt, 1);
805 callout_init(&tp->t_timers->tt_persist, 1);
806 callout_init(&tp->t_timers->tt_keep, 1);
807 callout_init(&tp->t_timers->tt_2msl, 1);
808 callout_init(&tp->t_timers->tt_delack, 1);
809
810 if (V_tcp_do_rfc1323)
811 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
812 if (V_tcp_do_sack)
813 tp->t_flags |= TF_SACK_PERMIT;
814 TAILQ_INIT(&tp->snd_holes);
815 /*
816 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
817 * is called.
818 */
819 in_pcbref(inp); /* Reference for tcpcb */
820 tp->t_inpcb = inp;
821
822 /*
823 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
824 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
825 * reasonable initial retransmit time.
826 */
827 tp->t_srtt = TCPTV_SRTTBASE;
828 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
829 tp->t_rttmin = tcp_rexmit_min;
830 tp->t_rxtcur = TCPTV_RTOBASE;
831 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
832 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
833 tp->t_rcvtime = ticks;
834 /*
835 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
836 * because the socket may be bound to an IPv6 wildcard address,
837 * which may match an IPv4-mapped IPv6 address.
838 */
839 inp->inp_ip_ttl = V_ip_defttl;
840 inp->inp_ppcb = tp;
841#ifdef TCPPCAP
842 /*
843 * Init the TCP PCAP queues.
844 */
845 tcp_pcap_tcpcb_init(tp);
846#endif
847 return (tp); /* XXX */
848}
849
850/*
851 * Switch the congestion control algorithm back to NewReno for any active
852 * control blocks using an algorithm which is about to go away.
853 * This ensures the CC framework can allow the unload to proceed without leaving
854 * any dangling pointers which would trigger a panic.
855 * Returning non-zero would inform the CC framework that something went wrong
856 * and it would be unsafe to allow the unload to proceed. However, there is no
857 * way for this to occur with this implementation so we always return zero.
858 */
859int
860tcp_ccalgounload(struct cc_algo *unload_algo)
861{
862 struct cc_algo *tmpalgo;
863 struct inpcb *inp;
864 struct tcpcb *tp;
865 VNET_ITERATOR_DECL(vnet_iter);
866
867 /*
868 * Check all active control blocks across all network stacks and change
869 * any that are using "unload_algo" back to NewReno. If "unload_algo"
870 * requires cleanup code to be run, call it.
871 */
872 VNET_LIST_RLOCK();
873 VNET_FOREACH(vnet_iter) {
874 CURVNET_SET(vnet_iter);
875 INP_INFO_WLOCK(&V_tcbinfo);
876 /*
877 * New connections already part way through being initialised
878 * with the CC algo we're removing will not race with this code
879 * because the INP_INFO_WLOCK is held during initialisation. We
880 * therefore don't enter the loop below until the connection
881 * list has stabilised.
882 */
883 LIST_FOREACH(inp, &V_tcb, inp_list) {
884 INP_WLOCK(inp);
885 /* Important to skip tcptw structs. */
886 if (!(inp->inp_flags & INP_TIMEWAIT) &&
887 (tp = intotcpcb(inp)) != NULL) {
888 /*
889 * By holding INP_WLOCK here, we are assured
890 * that the connection is not currently
891 * executing inside the CC module's functions
892 * i.e. it is safe to make the switch back to
893 * NewReno.
894 */
895 if (CC_ALGO(tp) == unload_algo) {
896 tmpalgo = CC_ALGO(tp);
897 /* NewReno does not require any init. */
898 CC_ALGO(tp) = &newreno_cc_algo;
899 if (tmpalgo->cb_destroy != NULL)
900 tmpalgo->cb_destroy(tp->ccv);
901 }
902 }
903 INP_WUNLOCK(inp);
904 }
905 INP_INFO_WUNLOCK(&V_tcbinfo);
906 CURVNET_RESTORE();
907 }
908 VNET_LIST_RUNLOCK();
909
910 return (0);
911}
912
913/*
914 * Drop a TCP connection, reporting
915 * the specified error. If connection is synchronized,
916 * then send a RST to peer.
917 */
918struct tcpcb *
919tcp_drop(struct tcpcb *tp, int errno)
920{
921 struct socket *so = tp->t_inpcb->inp_socket;
922
923 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
924 INP_WLOCK_ASSERT(tp->t_inpcb);
925
926 if (TCPS_HAVERCVDSYN(tp->t_state)) {
927 tcp_state_change(tp, TCPS_CLOSED);
928 (void) tcp_output(tp);
929 TCPSTAT_INC(tcps_drops);
930 } else
931 TCPSTAT_INC(tcps_conndrops);
932 if (errno == ETIMEDOUT && tp->t_softerror)
933 errno = tp->t_softerror;
934 so->so_error = errno;
935 return (tcp_close(tp));
936}
937
938void
939tcp_discardcb(struct tcpcb *tp)
940{
941 struct inpcb *inp = tp->t_inpcb;
942 struct socket *so = inp->inp_socket;
943#ifdef INET6
944 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
945#endif /* INET6 */
946 int released;
947
948 INP_WLOCK_ASSERT(inp);
949
950 /*
951 * Make sure that all of our timers are stopped before we delete the
952 * PCB.
953 *
954 * If stopping a timer fails, we schedule a discard function in same
955 * callout, and the last discard function called will take care of
956 * deleting the tcpcb.
957 */
958 tcp_timer_stop(tp, TT_REXMT);
959 tcp_timer_stop(tp, TT_PERSIST);
960 tcp_timer_stop(tp, TT_KEEP);
961 tcp_timer_stop(tp, TT_2MSL);
962 tcp_timer_stop(tp, TT_DELACK);
963
964 /*
965 * If we got enough samples through the srtt filter,
966 * save the rtt and rttvar in the routing entry.
967 * 'Enough' is arbitrarily defined as 4 rtt samples.
968 * 4 samples is enough for the srtt filter to converge
969 * to within enough % of the correct value; fewer samples
970 * and we could save a bogus rtt. The danger is not high
971 * as tcp quickly recovers from everything.
972 * XXX: Works very well but needs some more statistics!
973 */
974 if (tp->t_rttupdated >= 4) {
975 struct hc_metrics_lite metrics;
976 u_long ssthresh;
977
978 bzero(&metrics, sizeof(metrics));
979 /*
980 * Update the ssthresh always when the conditions below
981 * are satisfied. This gives us better new start value
982 * for the congestion avoidance for new connections.
983 * ssthresh is only set if packet loss occured on a session.
984 *
985 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
986 * being torn down. Ideally this code would not use 'so'.
987 */
988 ssthresh = tp->snd_ssthresh;
989 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
990 /*
991 * convert the limit from user data bytes to
992 * packets then to packet data bytes.
993 */
994 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
995 if (ssthresh < 2)
996 ssthresh = 2;
997 ssthresh *= (u_long)(tp->t_maxseg +
998#ifdef INET6
999 (isipv6 ? sizeof (struct ip6_hdr) +
1000 sizeof (struct tcphdr) :
1001#endif
1002 sizeof (struct tcpiphdr)
1003#ifdef INET6
1004 )
1005#endif
1006 );
1007 } else
1008 ssthresh = 0;
1009 metrics.rmx_ssthresh = ssthresh;
1010
1011 metrics.rmx_rtt = tp->t_srtt;
1012 metrics.rmx_rttvar = tp->t_rttvar;
1013 metrics.rmx_cwnd = tp->snd_cwnd;
1014 metrics.rmx_sendpipe = 0;
1015 metrics.rmx_recvpipe = 0;
1016
1017 tcp_hc_update(&inp->inp_inc, &metrics);
1018 }
1019
1020 /* free the reassembly queue, if any */
1021 tcp_reass_flush(tp);
1022
1023#ifdef TCP_OFFLOAD
1024 /* Disconnect offload device, if any. */
1025 if (tp->t_flags & TF_TOE)
1026 tcp_offload_detach(tp);
1027#endif
1028
1029 tcp_free_sackholes(tp);
1030
1031#ifdef TCPPCAP
1032 /* Free the TCP PCAP queues. */
1033 tcp_pcap_drain(&(tp->t_inpkts));
1034 tcp_pcap_drain(&(tp->t_outpkts));
1035#endif
1036
1037 /* Allow the CC algorithm to clean up after itself. */
1038 if (CC_ALGO(tp)->cb_destroy != NULL)
1039 CC_ALGO(tp)->cb_destroy(tp->ccv);
1040
1041 khelp_destroy_osd(tp->osd);
1042
1043 CC_ALGO(tp) = NULL;
1044 inp->inp_ppcb = NULL;
1045 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1046 /* We own the last reference on tcpcb, let's free it. */
1047 tp->t_inpcb = NULL;
1048 uma_zfree(V_tcpcb_zone, tp);
1049 released = in_pcbrele_wlocked(inp);
1050 KASSERT(!released, ("%s: inp %p should not have been released "
1051 "here", __func__, inp));
1052 }
1053}
1054
1055void
1056tcp_timer_2msl_discard(void *xtp)
1057{
1058
1059 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1060}
1061
1062void
1063tcp_timer_keep_discard(void *xtp)
1064{
1065
1066 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1067}
1068
1069void
1070tcp_timer_persist_discard(void *xtp)
1071{
1072
1073 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1074}
1075
1076void
1077tcp_timer_rexmt_discard(void *xtp)
1078{
1079
1080 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1081}
1082
1083void
1084tcp_timer_delack_discard(void *xtp)
1085{
1086
1087 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1088}
1089
1090void
1091tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1092{
1093 struct inpcb *inp;
1094
1095 CURVNET_SET(tp->t_vnet);
1096 INP_INFO_RLOCK(&V_tcbinfo);
1097 inp = tp->t_inpcb;
1098 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1099 __func__, tp));
1100 INP_WLOCK(inp);
1101 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1102 ("%s: tcpcb has to be stopped here", __func__));
1103 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1104 ("%s: discard callout should be running", __func__));
1105 tp->t_timers->tt_flags &= ~timer_type;
1106 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1107 /* We own the last reference on this tcpcb, let's free it. */
1108 tp->t_inpcb = NULL;
1109 uma_zfree(V_tcpcb_zone, tp);
1110 if (in_pcbrele_wlocked(inp)) {
1111 INP_INFO_RUNLOCK(&V_tcbinfo);
1112 CURVNET_RESTORE();
1113 return;
1114 }
1115 }
1116 INP_WUNLOCK(inp);
1117 INP_INFO_RUNLOCK(&V_tcbinfo);
1118 CURVNET_RESTORE();
1119}
1120
1121/*
1122 * Attempt to close a TCP control block, marking it as dropped, and freeing
1123 * the socket if we hold the only reference.
1124 */
1125struct tcpcb *
1126tcp_close(struct tcpcb *tp)
1127{
1128 struct inpcb *inp = tp->t_inpcb;
1129 struct socket *so;
1130
1131 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1132 INP_WLOCK_ASSERT(inp);
1133
1134#ifdef TCP_OFFLOAD
1135 if (tp->t_state == TCPS_LISTEN)
1136 tcp_offload_listen_stop(tp);
1137#endif
1138 in_pcbdrop(inp);
1139 TCPSTAT_INC(tcps_closed);
1140 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1141 so = inp->inp_socket;
1142 soisdisconnected(so);
1143 if (inp->inp_flags & INP_SOCKREF) {
1144 KASSERT(so->so_state & SS_PROTOREF,
1145 ("tcp_close: !SS_PROTOREF"));
1146 inp->inp_flags &= ~INP_SOCKREF;
1147 INP_WUNLOCK(inp);
1148 ACCEPT_LOCK();
1149 SOCK_LOCK(so);
1150 so->so_state &= ~SS_PROTOREF;
1151 sofree(so);
1152 return (NULL);
1153 }
1154 return (tp);
1155}
1156
1157void
1158tcp_drain(void)
1159{
1160 VNET_ITERATOR_DECL(vnet_iter);
1161
1162 if (!do_tcpdrain)
1163 return;
1164
1165 VNET_LIST_RLOCK_NOSLEEP();
1166 VNET_FOREACH(vnet_iter) {
1167 CURVNET_SET(vnet_iter);
1168 struct inpcb *inpb;
1169 struct tcpcb *tcpb;
1170
1171 /*
1172 * Walk the tcpbs, if existing, and flush the reassembly queue,
1173 * if there is one...
1174 * XXX: The "Net/3" implementation doesn't imply that the TCP
1175 * reassembly queue should be flushed, but in a situation
1176 * where we're really low on mbufs, this is potentially
1177 * useful.
1178 */
1179 INP_INFO_WLOCK(&V_tcbinfo);
1180 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1181 if (inpb->inp_flags & INP_TIMEWAIT)
1182 continue;
1183 INP_WLOCK(inpb);
1184 if ((tcpb = intotcpcb(inpb)) != NULL) {
1185 tcp_reass_flush(tcpb);
1186 tcp_clean_sackreport(tcpb);
1187 }
1188 INP_WUNLOCK(inpb);
1189 }
1190 INP_INFO_WUNLOCK(&V_tcbinfo);
1191 CURVNET_RESTORE();
1192 }
1193 VNET_LIST_RUNLOCK_NOSLEEP();
1194}
1195
1196/*
1197 * Notify a tcp user of an asynchronous error;
1198 * store error as soft error, but wake up user
1199 * (for now, won't do anything until can select for soft error).
1200 *
1201 * Do not wake up user since there currently is no mechanism for
1202 * reporting soft errors (yet - a kqueue filter may be added).
1203 */
1204static struct inpcb *
1205tcp_notify(struct inpcb *inp, int error)
1206{
1207 struct tcpcb *tp;
1208
1209 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1210 INP_WLOCK_ASSERT(inp);
1211
1212 if ((inp->inp_flags & INP_TIMEWAIT) ||
1213 (inp->inp_flags & INP_DROPPED))
1214 return (inp);
1215
1216 tp = intotcpcb(inp);
1217 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1218
1219 /*
1220 * Ignore some errors if we are hooked up.
1221 * If connection hasn't completed, has retransmitted several times,
1222 * and receives a second error, give up now. This is better
1223 * than waiting a long time to establish a connection that
1224 * can never complete.
1225 */
1226 if (tp->t_state == TCPS_ESTABLISHED &&
1227 (error == EHOSTUNREACH || error == ENETUNREACH ||
1228 error == EHOSTDOWN)) {
1229 return (inp);
1230 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1231 tp->t_softerror) {
1232 tp = tcp_drop(tp, error);
1233 if (tp != NULL)
1234 return (inp);
1235 else
1236 return (NULL);
1237 } else {
1238 tp->t_softerror = error;
1239 return (inp);
1240 }
1241#if 0
1242 wakeup( &so->so_timeo);
1243 sorwakeup(so);
1244 sowwakeup(so);
1245#endif
1246}
1247
1248static int
1249tcp_pcblist(SYSCTL_HANDLER_ARGS)
1250{
1251 int error, i, m, n, pcb_count;
1252 struct inpcb *inp, **inp_list;
1253 inp_gen_t gencnt;
1254 struct xinpgen xig;
1255
1256 /*
1257 * The process of preparing the TCB list is too time-consuming and
1258 * resource-intensive to repeat twice on every request.
1259 */
1260 if (req->oldptr == NULL) {
1261 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1262 n += imax(n / 8, 10);
1263 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1264 return (0);
1265 }
1266
1267 if (req->newptr != NULL)
1268 return (EPERM);
1269
1270 /*
1271 * OK, now we're committed to doing something.
1272 */
1273 INP_LIST_RLOCK(&V_tcbinfo);
1274 gencnt = V_tcbinfo.ipi_gencnt;
1275 n = V_tcbinfo.ipi_count;
1276 INP_LIST_RUNLOCK(&V_tcbinfo);
1277
1278 m = syncache_pcbcount();
1279
1280 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1281 + (n + m) * sizeof(struct xtcpcb));
1282 if (error != 0)
1283 return (error);
1284
1285 xig.xig_len = sizeof xig;
1286 xig.xig_count = n + m;
1287 xig.xig_gen = gencnt;
1288 xig.xig_sogen = so_gencnt;
1289 error = SYSCTL_OUT(req, &xig, sizeof xig);
1290 if (error)
1291 return (error);
1292
1293 error = syncache_pcblist(req, m, &pcb_count);
1294 if (error)
1295 return (error);
1296
1297 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1298 if (inp_list == NULL)
1299 return (ENOMEM);
1300
1301 INP_INFO_WLOCK(&V_tcbinfo);
1302 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1303 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1304 INP_WLOCK(inp);
1305 if (inp->inp_gencnt <= gencnt) {
1306 /*
1307 * XXX: This use of cr_cansee(), introduced with
1308 * TCP state changes, is not quite right, but for
1309 * now, better than nothing.
1310 */
1311 if (inp->inp_flags & INP_TIMEWAIT) {
1312 if (intotw(inp) != NULL)
1313 error = cr_cansee(req->td->td_ucred,
1314 intotw(inp)->tw_cred);
1315 else
1316 error = EINVAL; /* Skip this inp. */
1317 } else
1318 error = cr_canseeinpcb(req->td->td_ucred, inp);
1319 if (error == 0) {
1320 in_pcbref(inp);
1321 inp_list[i++] = inp;
1322 }
1323 }
1324 INP_WUNLOCK(inp);
1325 }
1326 INP_INFO_WUNLOCK(&V_tcbinfo);
1327 n = i;
1328
1329 error = 0;
1330 for (i = 0; i < n; i++) {
1331 inp = inp_list[i];
1332 INP_RLOCK(inp);
1333 if (inp->inp_gencnt <= gencnt) {
1334 struct xtcpcb xt;
1335 void *inp_ppcb;
1336
1337 bzero(&xt, sizeof(xt));
1338 xt.xt_len = sizeof xt;
1339 /* XXX should avoid extra copy */
1340 bcopy(inp, &xt.xt_inp, sizeof *inp);
1341 inp_ppcb = inp->inp_ppcb;
1342 if (inp_ppcb == NULL)
1343 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1344 else if (inp->inp_flags & INP_TIMEWAIT) {
1345 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1346 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1347 } else {
1348 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1349 if (xt.xt_tp.t_timers)
1350 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1351 }
1352 if (inp->inp_socket != NULL)
1353 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1354 else {
1355 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1356 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1357 }
1358 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1359 INP_RUNLOCK(inp);
1360 error = SYSCTL_OUT(req, &xt, sizeof xt);
1361 } else
1362 INP_RUNLOCK(inp);
1363 }
1364 INP_INFO_RLOCK(&V_tcbinfo);
1365 for (i = 0; i < n; i++) {
1366 inp = inp_list[i];
1367 INP_RLOCK(inp);
1368 if (!in_pcbrele_rlocked(inp))
1369 INP_RUNLOCK(inp);
1370 }
1371 INP_INFO_RUNLOCK(&V_tcbinfo);
1372
1373 if (!error) {
1374 /*
1375 * Give the user an updated idea of our state.
1376 * If the generation differs from what we told
1377 * her before, she knows that something happened
1378 * while we were processing this request, and it
1379 * might be necessary to retry.
1380 */
1381 INP_LIST_RLOCK(&V_tcbinfo);
1382 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1383 xig.xig_sogen = so_gencnt;
1384 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1385 INP_LIST_RUNLOCK(&V_tcbinfo);
1386 error = SYSCTL_OUT(req, &xig, sizeof xig);
1387 }
1388 free(inp_list, M_TEMP);
1389 return (error);
1390}
1391
1392SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1393 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1394 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1395
1396#ifdef INET
1397static int
1398tcp_getcred(SYSCTL_HANDLER_ARGS)
1399{
1400 struct xucred xuc;
1401 struct sockaddr_in addrs[2];
1402 struct inpcb *inp;
1403 int error;
1404
1405 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1406 if (error)
1407 return (error);
1408 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1409 if (error)
1410 return (error);
1411 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1412 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1413 if (inp != NULL) {
1414 if (inp->inp_socket == NULL)
1415 error = ENOENT;
1416 if (error == 0)
1417 error = cr_canseeinpcb(req->td->td_ucred, inp);
1418 if (error == 0)
1419 cru2x(inp->inp_cred, &xuc);
1420 INP_RUNLOCK(inp);
1421 } else
1422 error = ENOENT;
1423 if (error == 0)
1424 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1425 return (error);
1426}
1427
1428SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1429 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1430 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1431#endif /* INET */
1432
1433#ifdef INET6
1434static int
1435tcp6_getcred(SYSCTL_HANDLER_ARGS)
1436{
1437 struct xucred xuc;
1438 struct sockaddr_in6 addrs[2];
1439 struct inpcb *inp;
1440 int error;
1441#ifdef INET
1442 int mapped = 0;
1443#endif
1444
1445 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1446 if (error)
1447 return (error);
1448 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1449 if (error)
1450 return (error);
1451 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1452 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1453 return (error);
1454 }
1455 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1456#ifdef INET
1457 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1458 mapped = 1;
1459 else
1460#endif
1461 return (EINVAL);
1462 }
1463
1464#ifdef INET
1465 if (mapped == 1)
1466 inp = in_pcblookup(&V_tcbinfo,
1467 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1468 addrs[1].sin6_port,
1469 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1470 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1471 else
1472#endif
1473 inp = in6_pcblookup(&V_tcbinfo,
1474 &addrs[1].sin6_addr, addrs[1].sin6_port,
1475 &addrs[0].sin6_addr, addrs[0].sin6_port,
1476 INPLOOKUP_RLOCKPCB, NULL);
1477 if (inp != NULL) {
1478 if (inp->inp_socket == NULL)
1479 error = ENOENT;
1480 if (error == 0)
1481 error = cr_canseeinpcb(req->td->td_ucred, inp);
1482 if (error == 0)
1483 cru2x(inp->inp_cred, &xuc);
1484 INP_RUNLOCK(inp);
1485 } else
1486 error = ENOENT;
1487 if (error == 0)
1488 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1489 return (error);
1490}
1491
1492SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1493 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1494 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1495#endif /* INET6 */
1496
1497
1498#ifdef INET
1499void
1500tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1501{
1502 struct ip *ip = vip;
1503 struct tcphdr *th;
1504 struct in_addr faddr;
1505 struct inpcb *inp;
1506 struct tcpcb *tp;
1507 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1508 struct icmp *icp;
1509 struct in_conninfo inc;
1510 tcp_seq icmp_tcp_seq;
1511 int mtu;
1512
1513 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1514 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1515 return;
1516
1517 if (cmd == PRC_MSGSIZE)
1518 notify = tcp_mtudisc_notify;
1519 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1520 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1521 notify = tcp_drop_syn_sent;
1522 /*
1523 * Redirects don't need to be handled up here.
1524 */
1525 else if (PRC_IS_REDIRECT(cmd))
1526 return;
1527 /*
1528 * Hostdead is ugly because it goes linearly through all PCBs.
1529 * XXX: We never get this from ICMP, otherwise it makes an
1530 * excellent DoS attack on machines with many connections.
1531 */
1532 else if (cmd == PRC_HOSTDEAD)
1533 ip = NULL;
1534 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1535 return;
1536
1537 if (ip == NULL) {
1538 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1539 return;
1540 }
1541
1542 icp = (struct icmp *)((caddr_t)ip - offsetof(struct icmp, icmp_ip));
1543 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1544 INP_INFO_RLOCK(&V_tcbinfo);
1545 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport, ip->ip_src,
1546 th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1547 if (inp != NULL) {
1548 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1549 !(inp->inp_flags & INP_DROPPED) &&
1550 !(inp->inp_socket == NULL)) {
1551 icmp_tcp_seq = ntohl(th->th_seq);
1552 tp = intotcpcb(inp);
1553 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1554 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1555 if (cmd == PRC_MSGSIZE) {
1556 /*
1557 * MTU discovery:
1558 * If we got a needfrag set the MTU
1559 * in the route to the suggested new
1560 * value (if given) and then notify.
1561 */
1562 mtu = ntohs(icp->icmp_nextmtu);
1563 /*
1564 * If no alternative MTU was
1565 * proposed, try the next smaller
1566 * one.
1567 */
1568 if (!mtu)
1569 mtu = ip_next_mtu(
1570 ntohs(ip->ip_len), 1);
1571 if (mtu < V_tcp_minmss +
1572 sizeof(struct tcpiphdr))
1573 mtu = V_tcp_minmss +
1574 sizeof(struct tcpiphdr);
1575 /*
1576 * Only process the offered MTU if it
1577 * is smaller than the current one.
1578 */
1579 if (mtu < tp->t_maxopd +
1580 sizeof(struct tcpiphdr)) {
1581 bzero(&inc, sizeof(inc));
1582 inc.inc_faddr = faddr;
1583 inc.inc_fibnum =
1584 inp->inp_inc.inc_fibnum;
1585 tcp_hc_updatemtu(&inc, mtu);
1586 tcp_mtudisc(inp, mtu);
1587 }
1588 } else
1589 inp = (*notify)(inp,
1590 inetctlerrmap[cmd]);
1591 }
1592 }
1593 if (inp != NULL)
1594 INP_WUNLOCK(inp);
1595 } else {
1596 bzero(&inc, sizeof(inc));
1597 inc.inc_fport = th->th_dport;
1598 inc.inc_lport = th->th_sport;
1599 inc.inc_faddr = faddr;
1600 inc.inc_laddr = ip->ip_src;
1601 syncache_unreach(&inc, th);
1602 }
1603 INP_INFO_RUNLOCK(&V_tcbinfo);
1604}
1605#endif /* INET */
1606
1607#ifdef INET6
1608void
1609tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1610{
1611 struct tcphdr th;
1612 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1613 struct ip6_hdr *ip6;
1614 struct mbuf *m;
1615 struct ip6ctlparam *ip6cp = NULL;
1616 const struct sockaddr_in6 *sa6_src = NULL;
1617 int off;
1618 struct tcp_portonly {
1619 u_int16_t th_sport;
1620 u_int16_t th_dport;
1621 } *thp;
1622
1623 if (sa->sa_family != AF_INET6 ||
1624 sa->sa_len != sizeof(struct sockaddr_in6))
1625 return;
1626
1627 if (cmd == PRC_MSGSIZE)
1628 notify = tcp_mtudisc_notify;
1629 else if (!PRC_IS_REDIRECT(cmd) &&
1630 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1631 return;
1632
1633 /* if the parameter is from icmp6, decode it. */
1634 if (d != NULL) {
1635 ip6cp = (struct ip6ctlparam *)d;
1636 m = ip6cp->ip6c_m;
1637 ip6 = ip6cp->ip6c_ip6;
1638 off = ip6cp->ip6c_off;
1639 sa6_src = ip6cp->ip6c_src;
1640 } else {
1641 m = NULL;
1642 ip6 = NULL;
1643 off = 0; /* fool gcc */
1644 sa6_src = &sa6_any;
1645 }
1646
1647 if (ip6 != NULL) {
1648 struct in_conninfo inc;
1649 /*
1650 * XXX: We assume that when IPV6 is non NULL,
1651 * M and OFF are valid.
1652 */
1653
1654 /* check if we can safely examine src and dst ports */
1655 if (m->m_pkthdr.len < off + sizeof(*thp))
1656 return;
1657
1658 bzero(&th, sizeof(th));
1659 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1660
1661 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1662 (struct sockaddr *)ip6cp->ip6c_src,
1663 th.th_sport, cmd, NULL, notify);
1664
1665 bzero(&inc, sizeof(inc));
1666 inc.inc_fport = th.th_dport;
1667 inc.inc_lport = th.th_sport;
1668 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1669 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1670 inc.inc_flags |= INC_ISIPV6;
1671 INP_INFO_RLOCK(&V_tcbinfo);
1672 syncache_unreach(&inc, &th);
1673 INP_INFO_RUNLOCK(&V_tcbinfo);
1674 } else
1675 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1676 0, cmd, NULL, notify);
1677}
1678#endif /* INET6 */
1679
1680
1681/*
1682 * Following is where TCP initial sequence number generation occurs.
1683 *
1684 * There are two places where we must use initial sequence numbers:
1685 * 1. In SYN-ACK packets.
1686 * 2. In SYN packets.
1687 *
1688 * All ISNs for SYN-ACK packets are generated by the syncache. See
1689 * tcp_syncache.c for details.
1690 *
1691 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1692 * depends on this property. In addition, these ISNs should be
1693 * unguessable so as to prevent connection hijacking. To satisfy
1694 * the requirements of this situation, the algorithm outlined in
1695 * RFC 1948 is used, with only small modifications.
1696 *
1697 * Implementation details:
1698 *
1699 * Time is based off the system timer, and is corrected so that it
1700 * increases by one megabyte per second. This allows for proper
1701 * recycling on high speed LANs while still leaving over an hour
1702 * before rollover.
1703 *
1704 * As reading the *exact* system time is too expensive to be done
1705 * whenever setting up a TCP connection, we increment the time
1706 * offset in two ways. First, a small random positive increment
1707 * is added to isn_offset for each connection that is set up.
1708 * Second, the function tcp_isn_tick fires once per clock tick
1709 * and increments isn_offset as necessary so that sequence numbers
1710 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1711 * random positive increments serve only to ensure that the same
1712 * exact sequence number is never sent out twice (as could otherwise
1713 * happen when a port is recycled in less than the system tick
1714 * interval.)
1715 *
1716 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1717 * between seeding of isn_secret. This is normally set to zero,
1718 * as reseeding should not be necessary.
1719 *
1720 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1721 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1722 * general, this means holding an exclusive (write) lock.
1723 */
1724
1725#define ISN_BYTES_PER_SECOND 1048576
1726#define ISN_STATIC_INCREMENT 4096
1727#define ISN_RANDOM_INCREMENT (4096 - 1)
1728
1729static VNET_DEFINE(u_char, isn_secret[32]);
1730static VNET_DEFINE(int, isn_last);
1731static VNET_DEFINE(int, isn_last_reseed);
1732static VNET_DEFINE(u_int32_t, isn_offset);
1733static VNET_DEFINE(u_int32_t, isn_offset_old);
1734
1735#define V_isn_secret VNET(isn_secret)
1736#define V_isn_last VNET(isn_last)
1737#define V_isn_last_reseed VNET(isn_last_reseed)
1738#define V_isn_offset VNET(isn_offset)
1739#define V_isn_offset_old VNET(isn_offset_old)
1740
1741tcp_seq
1742tcp_new_isn(struct tcpcb *tp)
1743{
1744 MD5_CTX isn_ctx;
1745 u_int32_t md5_buffer[4];
1746 tcp_seq new_isn;
1747 u_int32_t projected_offset;
1748
1749 INP_WLOCK_ASSERT(tp->t_inpcb);
1750
1751 ISN_LOCK();
1752 /* Seed if this is the first use, reseed if requested. */
1753 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1754 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1755 < (u_int)ticks))) {
1756 read_random(&V_isn_secret, sizeof(V_isn_secret));
1757 V_isn_last_reseed = ticks;
1758 }
1759
1760 /* Compute the md5 hash and return the ISN. */
1761 MD5Init(&isn_ctx);
1762 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1763 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1764#ifdef INET6
1765 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1766 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1767 sizeof(struct in6_addr));
1768 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1769 sizeof(struct in6_addr));
1770 } else
1771#endif
1772 {
1773 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1774 sizeof(struct in_addr));
1775 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1776 sizeof(struct in_addr));
1777 }
1778 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1779 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1780 new_isn = (tcp_seq) md5_buffer[0];
1781 V_isn_offset += ISN_STATIC_INCREMENT +
1782 (arc4random() & ISN_RANDOM_INCREMENT);
1783 if (ticks != V_isn_last) {
1784 projected_offset = V_isn_offset_old +
1785 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1786 if (SEQ_GT(projected_offset, V_isn_offset))
1787 V_isn_offset = projected_offset;
1788 V_isn_offset_old = V_isn_offset;
1789 V_isn_last = ticks;
1790 }
1791 new_isn += V_isn_offset;
1792 ISN_UNLOCK();
1793 return (new_isn);
1794}
1795
1796/*
1797 * When a specific ICMP unreachable message is received and the
1798 * connection state is SYN-SENT, drop the connection. This behavior
1799 * is controlled by the icmp_may_rst sysctl.
1800 */
1801struct inpcb *
1802tcp_drop_syn_sent(struct inpcb *inp, int errno)
1803{
1804 struct tcpcb *tp;
1805
1806 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1807 INP_WLOCK_ASSERT(inp);
1808
1809 if ((inp->inp_flags & INP_TIMEWAIT) ||
1810 (inp->inp_flags & INP_DROPPED))
1811 return (inp);
1812
1813 tp = intotcpcb(inp);
1814 if (tp->t_state != TCPS_SYN_SENT)
1815 return (inp);
1816
1817 tp = tcp_drop(tp, errno);
1818 if (tp != NULL)
1819 return (inp);
1820 else
1821 return (NULL);
1822}
1823
1824/*
1825 * When `need fragmentation' ICMP is received, update our idea of the MSS
1826 * based on the new value. Also nudge TCP to send something, since we
1827 * know the packet we just sent was dropped.
1828 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1829 */
1830static struct inpcb *
1831tcp_mtudisc_notify(struct inpcb *inp, int error)
1832{
1833
1834 tcp_mtudisc(inp, -1);
1835 return (inp);
1836}
1837
1838static void
1839tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1840{
1841 struct tcpcb *tp;
1842 struct socket *so;
1843
1844 INP_WLOCK_ASSERT(inp);
1845 if ((inp->inp_flags & INP_TIMEWAIT) ||
1846 (inp->inp_flags & INP_DROPPED))
1847 return;
1848
1849 tp = intotcpcb(inp);
1850 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1851
1852 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1853
1854 so = inp->inp_socket;
1855 SOCKBUF_LOCK(&so->so_snd);
1856 /* If the mss is larger than the socket buffer, decrease the mss. */
1857 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1858 tp->t_maxseg = so->so_snd.sb_hiwat;
1859 SOCKBUF_UNLOCK(&so->so_snd);
1860
1861 TCPSTAT_INC(tcps_mturesent);
1862 tp->t_rtttime = 0;
1863 tp->snd_nxt = tp->snd_una;
1864 tcp_free_sackholes(tp);
1865 tp->snd_recover = tp->snd_max;
1866 if (tp->t_flags & TF_SACK_PERMIT)
1867 EXIT_FASTRECOVERY(tp->t_flags);
1868 tcp_output(tp);
1869}
1870
1871#ifdef INET
1872/*
1873 * Look-up the routing entry to the peer of this inpcb. If no route
1874 * is found and it cannot be allocated, then return 0. This routine
1875 * is called by TCP routines that access the rmx structure and by
1876 * tcp_mss_update to get the peer/interface MTU.
1877 */
1878u_long
1879tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1880{
1881 struct route sro;
1882 struct sockaddr_in *dst;
1883 struct ifnet *ifp;
1884 u_long maxmtu = 0;
1885
1886 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1887
1888 bzero(&sro, sizeof(sro));
1889 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1890 dst = (struct sockaddr_in *)&sro.ro_dst;
1891 dst->sin_family = AF_INET;
1892 dst->sin_len = sizeof(*dst);
1893 dst->sin_addr = inc->inc_faddr;
1894 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1895 }
1896 if (sro.ro_rt != NULL) {
1897 ifp = sro.ro_rt->rt_ifp;
1898 if (sro.ro_rt->rt_mtu == 0)
1899 maxmtu = ifp->if_mtu;
1900 else
1901 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1902
1903 /* Report additional interface capabilities. */
1904 if (cap != NULL) {
1905 if (ifp->if_capenable & IFCAP_TSO4 &&
1906 ifp->if_hwassist & CSUM_TSO) {
1907 cap->ifcap |= CSUM_TSO;
1908 cap->tsomax = ifp->if_hw_tsomax;
1909 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1910 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1911 }
1912 }
1913 RTFREE(sro.ro_rt);
1914 }
1915 return (maxmtu);
1916}
1917#endif /* INET */
1918
1919#ifdef INET6
1920u_long
1921tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1922{
1923 struct route_in6 sro6;
1924 struct ifnet *ifp;
1925 u_long maxmtu = 0;
1926
1927 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1928
1929 bzero(&sro6, sizeof(sro6));
1930 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1931 sro6.ro_dst.sin6_family = AF_INET6;
1932 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1933 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1934 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1935 }
1936 if (sro6.ro_rt != NULL) {
1937 ifp = sro6.ro_rt->rt_ifp;
1938 if (sro6.ro_rt->rt_mtu == 0)
1939 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1940 else
1941 maxmtu = min(sro6.ro_rt->rt_mtu,
1942 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1943
1944 /* Report additional interface capabilities. */
1945 if (cap != NULL) {
1946 if (ifp->if_capenable & IFCAP_TSO6 &&
1947 ifp->if_hwassist & CSUM_TSO) {
1948 cap->ifcap |= CSUM_TSO;
1949 cap->tsomax = ifp->if_hw_tsomax;
1950 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1951 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1952 }
1953 }
1954 RTFREE(sro6.ro_rt);
1955 }
1956
1957 return (maxmtu);
1958}
1959#endif /* INET6 */
1960
1961#ifdef IPSEC
1962/* compute ESP/AH header size for TCP, including outer IP header. */
1963size_t
1964ipsec_hdrsiz_tcp(struct tcpcb *tp)
1965{
1966 struct inpcb *inp;
1967 struct mbuf *m;
1968 size_t hdrsiz;
1969 struct ip *ip;
1970#ifdef INET6
1971 struct ip6_hdr *ip6;
1972#endif
1973 struct tcphdr *th;
1974
1975 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1976 return (0);
1977 m = m_gethdr(M_NOWAIT, MT_DATA);
1978 if (!m)
1979 return (0);
1980
1981#ifdef INET6
1982 if ((inp->inp_vflag & INP_IPV6) != 0) {
1983 ip6 = mtod(m, struct ip6_hdr *);
1984 th = (struct tcphdr *)(ip6 + 1);
1985 m->m_pkthdr.len = m->m_len =
1986 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1987 tcpip_fillheaders(inp, ip6, th);
1988 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1989 } else
1990#endif /* INET6 */
1991 {
1992 ip = mtod(m, struct ip *);
1993 th = (struct tcphdr *)(ip + 1);
1994 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1995 tcpip_fillheaders(inp, ip, th);
1996 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1997 }
1998
1999 m_free(m);
2000 return (hdrsiz);
2001}
2002#endif /* IPSEC */
2003
2004#ifdef TCP_SIGNATURE
2005/*
2006 * Callback function invoked by m_apply() to digest TCP segment data
2007 * contained within an mbuf chain.
2008 */
2009static int
2010tcp_signature_apply(void *fstate, void *data, u_int len)
2011{
2012
2013 MD5Update(fstate, (u_char *)data, len);
2014 return (0);
2015}
2016
2017/*
2018 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2019 * search with the destination IP address, and a 'magic SPI' to be
2020 * determined by the application. This is hardcoded elsewhere to 1179
2021*/
2022struct secasvar *
2023tcp_get_sav(struct mbuf *m, u_int direction)
2024{
2025 union sockaddr_union dst;
2026 struct secasvar *sav;
2027 struct ip *ip;
2028#ifdef INET6
2029 struct ip6_hdr *ip6;
2030 char ip6buf[INET6_ADDRSTRLEN];
2031#endif
2032
2033 /* Extract the destination from the IP header in the mbuf. */
2034 bzero(&dst, sizeof(union sockaddr_union));
2035 ip = mtod(m, struct ip *);
2036#ifdef INET6
2037 ip6 = NULL; /* Make the compiler happy. */
2038#endif
2039 switch (ip->ip_v) {
2040#ifdef INET
2041 case IPVERSION:
2042 dst.sa.sa_len = sizeof(struct sockaddr_in);
2043 dst.sa.sa_family = AF_INET;
2044 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2045 ip->ip_src : ip->ip_dst;
2046 break;
2047#endif
2048#ifdef INET6
2049 case (IPV6_VERSION >> 4):
2050 ip6 = mtod(m, struct ip6_hdr *);
2051 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2052 dst.sa.sa_family = AF_INET6;
2053 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2054 ip6->ip6_src : ip6->ip6_dst;
2055 break;
2056#endif
2057 default:
2058 return (NULL);
2059 /* NOTREACHED */
2060 break;
2061 }
2062
2063 /* Look up an SADB entry which matches the address of the peer. */
2064 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2065 if (sav == NULL) {
2066 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2067 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2068#ifdef INET6
2069 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2070 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2071#endif
2072 "(unsupported)"));
2073 }
2074
2075 return (sav);
2076}
2077
2078/*
2079 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2080 *
2081 * Parameters:
2082 * m pointer to head of mbuf chain
2083 * len length of TCP segment data, excluding options
2084 * optlen length of TCP segment options
2085 * buf pointer to storage for computed MD5 digest
2086 * sav pointer to security assosiation
2087 *
2088 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2089 * When called from tcp_input(), we can be sure that th_sum has been
2090 * zeroed out and verified already.
2091 *
2092 * Releases reference to SADB key before return.
2093 *
2094 * Return 0 if successful, otherwise return -1.
2095 *
2096 */
2097int
2098tcp_signature_do_compute(struct mbuf *m, int len, int optlen,
2099 u_char *buf, struct secasvar *sav)
2100{
2101#ifdef INET
2102 struct ippseudo ippseudo;
2103#endif
2104 MD5_CTX ctx;
2105 int doff;
2106 struct ip *ip;
2107#ifdef INET
2108 struct ipovly *ipovly;
2109#endif
2110 struct tcphdr *th;
2111#ifdef INET6
2112 struct ip6_hdr *ip6;
2113 struct in6_addr in6;
2114 uint32_t plen;
2115 uint16_t nhdr;
2116#endif
2117 u_short savecsum;
2118
2119 KASSERT(m != NULL, ("NULL mbuf chain"));
2120 KASSERT(buf != NULL, ("NULL signature pointer"));
2121
2122 /* Extract the destination from the IP header in the mbuf. */
2123 ip = mtod(m, struct ip *);
2124#ifdef INET6
2125 ip6 = NULL; /* Make the compiler happy. */
2126#endif
2127
2128 MD5Init(&ctx);
2129 /*
2130 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2131 *
2132 * XXX The ippseudo header MUST be digested in network byte order,
2133 * or else we'll fail the regression test. Assume all fields we've
2134 * been doing arithmetic on have been in host byte order.
2135 * XXX One cannot depend on ipovly->ih_len here. When called from
2136 * tcp_output(), the underlying ip_len member has not yet been set.
2137 */
2138 switch (ip->ip_v) {
2139#ifdef INET
2140 case IPVERSION:
2141 ipovly = (struct ipovly *)ip;
2142 ippseudo.ippseudo_src = ipovly->ih_src;
2143 ippseudo.ippseudo_dst = ipovly->ih_dst;
2144 ippseudo.ippseudo_pad = 0;
2145 ippseudo.ippseudo_p = IPPROTO_TCP;
2146 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2147 optlen);
2148 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2149
2150 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2151 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2152 break;
2153#endif
2154#ifdef INET6
2155 /*
2156 * RFC 2385, 2.0 Proposal
2157 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2158 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2159 * extended next header value (to form 32 bits), and 32-bit segment
2160 * length.
2161 * Note: Upper-Layer Packet Length comes before Next Header.
2162 */
2163 case (IPV6_VERSION >> 4):
2164 in6 = ip6->ip6_src;
2165 in6_clearscope(&in6);
2166 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2167 in6 = ip6->ip6_dst;
2168 in6_clearscope(&in6);
2169 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2170 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2171 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2172 nhdr = 0;
2173 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2174 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2175 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2176 nhdr = IPPROTO_TCP;
2177 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2178
2179 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2180 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2181 break;
2182#endif
2183 default:
2184 KEY_FREESAV(&sav);
2185 return (-1);
2186 /* NOTREACHED */
2187 break;
2188 }
2189
2190
2191 /*
2192 * Step 2: Update MD5 hash with TCP header, excluding options.
2193 * The TCP checksum must be set to zero.
2194 */
2195 savecsum = th->th_sum;
2196 th->th_sum = 0;
2197 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2198 th->th_sum = savecsum;
2199
2200 /*
2201 * Step 3: Update MD5 hash with TCP segment data.
2202 * Use m_apply() to avoid an early m_pullup().
2203 */
2204 if (len > 0)
2205 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2206
2207 /*
2208 * Step 4: Update MD5 hash with shared secret.
2209 */
2210 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2211 MD5Final(buf, &ctx);
2212
2213 key_sa_recordxfer(sav, m);
2214 KEY_FREESAV(&sav);
2215 return (0);
2216}
2217
2218/*
2219 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2220 *
2221 * Return 0 if successful, otherwise return -1.
2222 */
2223int
2224tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2225 u_char *buf, u_int direction)
2226{
2227 struct secasvar *sav;
2228
2229 if ((sav = tcp_get_sav(m, direction)) == NULL)
2230 return (-1);
2231
2232 return (tcp_signature_do_compute(m, len, optlen, buf, sav));
2233}
2234
2235/*
2236 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2237 *
2238 * Parameters:
2239 * m pointer to head of mbuf chain
2240 * len length of TCP segment data, excluding options
2241 * optlen length of TCP segment options
2242 * buf pointer to storage for computed MD5 digest
2243 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2244 *
2245 * Return 1 if successful, otherwise return 0.
2246 */
2247int
2248tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2249 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2250{
2251 char tmpdigest[TCP_SIGLEN];
2252
2253 if (tcp_sig_checksigs == 0)
2254 return (1);
2255 if ((tcpbflag & TF_SIGNATURE) == 0) {
2256 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2257
2258 /*
2259 * If this socket is not expecting signature but
2260 * the segment contains signature just fail.
2261 */
2262 TCPSTAT_INC(tcps_sig_err_sigopt);
2263 TCPSTAT_INC(tcps_sig_rcvbadsig);
2264 return (0);
2265 }
2266
2267 /* Signature is not expected, and not present in segment. */
2268 return (1);
2269 }
2270
2271 /*
2272 * If this socket is expecting signature but the segment does not
2273 * contain any just fail.
2274 */
2275 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2276 TCPSTAT_INC(tcps_sig_err_nosigopt);
2277 TCPSTAT_INC(tcps_sig_rcvbadsig);
2278 return (0);
2279 }
2280 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2281 IPSEC_DIR_INBOUND) == -1) {
2282 TCPSTAT_INC(tcps_sig_err_buildsig);
2283 TCPSTAT_INC(tcps_sig_rcvbadsig);
2284 return (0);
2285 }
2286
2287 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2288 TCPSTAT_INC(tcps_sig_rcvbadsig);
2289 return (0);
2290 }
2291 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2292 return (1);
2293}
2294#endif /* TCP_SIGNATURE */
2295
2296static int
2297sysctl_drop(SYSCTL_HANDLER_ARGS)
2298{
2299 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2300 struct sockaddr_storage addrs[2];
2301 struct inpcb *inp;
2302 struct tcpcb *tp;
2303 struct tcptw *tw;
2304 struct sockaddr_in *fin, *lin;
2305#ifdef INET6
2306 struct sockaddr_in6 *fin6, *lin6;
2307#endif
2308 int error;
2309
2310 inp = NULL;
2311 fin = lin = NULL;
2312#ifdef INET6
2313 fin6 = lin6 = NULL;
2314#endif
2315 error = 0;
2316
2317 if (req->oldptr != NULL || req->oldlen != 0)
2318 return (EINVAL);
2319 if (req->newptr == NULL)
2320 return (EPERM);
2321 if (req->newlen < sizeof(addrs))
2322 return (ENOMEM);
2323 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2324 if (error)
2325 return (error);
2326
2327 switch (addrs[0].ss_family) {
2328#ifdef INET6
2329 case AF_INET6:
2330 fin6 = (struct sockaddr_in6 *)&addrs[0];
2331 lin6 = (struct sockaddr_in6 *)&addrs[1];
2332 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2333 lin6->sin6_len != sizeof(struct sockaddr_in6))
2334 return (EINVAL);
2335 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2336 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2337 return (EINVAL);
2338 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2339 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2340 fin = (struct sockaddr_in *)&addrs[0];
2341 lin = (struct sockaddr_in *)&addrs[1];
2342 break;
2343 }
2344 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2345 if (error)
2346 return (error);
2347 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2348 if (error)
2349 return (error);
2350 break;
2351#endif
2352#ifdef INET
2353 case AF_INET:
2354 fin = (struct sockaddr_in *)&addrs[0];
2355 lin = (struct sockaddr_in *)&addrs[1];
2356 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2357 lin->sin_len != sizeof(struct sockaddr_in))
2358 return (EINVAL);
2359 break;
2360#endif
2361 default:
2362 return (EINVAL);
2363 }
2364 INP_INFO_RLOCK(&V_tcbinfo);
2365 switch (addrs[0].ss_family) {
2366#ifdef INET6
2367 case AF_INET6:
2368 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2369 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2370 INPLOOKUP_WLOCKPCB, NULL);
2371 break;
2372#endif
2373#ifdef INET
2374 case AF_INET:
2375 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2376 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2377 break;
2378#endif
2379 }
2380 if (inp != NULL) {
2381 if (inp->inp_flags & INP_TIMEWAIT) {
2382 /*
2383 * XXXRW: There currently exists a state where an
2384 * inpcb is present, but its timewait state has been
2385 * discarded. For now, don't allow dropping of this
2386 * type of inpcb.
2387 */
2388 tw = intotw(inp);
2389 if (tw != NULL)
2390 tcp_twclose(tw, 0);
2391 else
2392 INP_WUNLOCK(inp);
2393 } else if (!(inp->inp_flags & INP_DROPPED) &&
2394 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2395 tp = intotcpcb(inp);
2396 tp = tcp_drop(tp, ECONNABORTED);
2397 if (tp != NULL)
2398 INP_WUNLOCK(inp);
2399 } else
2400 INP_WUNLOCK(inp);
2401 } else
2402 error = ESRCH;
2403 INP_INFO_RUNLOCK(&V_tcbinfo);
2404 return (error);
2405}
2406
2407SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2408 CTLFLAG_VNET | CTLTYPE_STRUCT | CTLFLAG_WR | CTLFLAG_SKIP, NULL,
2409 0, sysctl_drop, "", "Drop TCP connection");
2410
2411/*
2412 * Generate a standardized TCP log line for use throughout the
2413 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2414 * allow use in the interrupt context.
2415 *
2416 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2417 * NB: The function may return NULL if memory allocation failed.
2418 *
2419 * Due to header inclusion and ordering limitations the struct ip
2420 * and ip6_hdr pointers have to be passed as void pointers.
2421 */
2422char *
2423tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2424 const void *ip6hdr)
2425{
2426
2427 /* Is logging enabled? */
2428 if (tcp_log_in_vain == 0)
2429 return (NULL);
2430
2431 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2432}
2433
2434char *
2435tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2436 const void *ip6hdr)
2437{
2438
2439 /* Is logging enabled? */
2440 if (tcp_log_debug == 0)
2441 return (NULL);
2442
2443 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2444}
2445
2446static char *
2447tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2448 const void *ip6hdr)
2449{
2450 char *s, *sp;
2451 size_t size;
2452 struct ip *ip;
2453#ifdef INET6
2454 const struct ip6_hdr *ip6;
2455
2456 ip6 = (const struct ip6_hdr *)ip6hdr;
2457#endif /* INET6 */
2458 ip = (struct ip *)ip4hdr;
2459
2460 /*
2461 * The log line looks like this:
2462 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2463 */
2464 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2465 sizeof(PRINT_TH_FLAGS) + 1 +
2466#ifdef INET6
2467 2 * INET6_ADDRSTRLEN;
2468#else
2469 2 * INET_ADDRSTRLEN;
2470#endif /* INET6 */
2471
2472 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2473 if (s == NULL)
2474 return (NULL);
2475
2476 strcat(s, "TCP: [");
2477 sp = s + strlen(s);
2478
2479 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2480 inet_ntoa_r(inc->inc_faddr, sp);
2481 sp = s + strlen(s);
2482 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2483 sp = s + strlen(s);
2484 inet_ntoa_r(inc->inc_laddr, sp);
2485 sp = s + strlen(s);
2486 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2487#ifdef INET6
2488 } else if (inc) {
2489 ip6_sprintf(sp, &inc->inc6_faddr);
2490 sp = s + strlen(s);
2491 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2492 sp = s + strlen(s);
2493 ip6_sprintf(sp, &inc->inc6_laddr);
2494 sp = s + strlen(s);
2495 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2496 } else if (ip6 && th) {
2497 ip6_sprintf(sp, &ip6->ip6_src);
2498 sp = s + strlen(s);
2499 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2500 sp = s + strlen(s);
2501 ip6_sprintf(sp, &ip6->ip6_dst);
2502 sp = s + strlen(s);
2503 sprintf(sp, "]:%i", ntohs(th->th_dport));
2504#endif /* INET6 */
2505#ifdef INET
2506 } else if (ip && th) {
2507 inet_ntoa_r(ip->ip_src, sp);
2508 sp = s + strlen(s);
2509 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2510 sp = s + strlen(s);
2511 inet_ntoa_r(ip->ip_dst, sp);
2512 sp = s + strlen(s);
2513 sprintf(sp, "]:%i", ntohs(th->th_dport));
2514#endif /* INET */
2515 } else {
2516 free(s, M_TCPLOG);
2517 return (NULL);
2518 }
2519 sp = s + strlen(s);
2520 if (th)
2521 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2522 if (*(s + size - 1) != '\0')
2523 panic("%s: string too long", __func__);
2524 return (s);
2525}
2526
2527/*
2528 * A subroutine which makes it easy to track TCP state changes with DTrace.
2529 * This function shouldn't be called for t_state initializations that don't
2530 * correspond to actual TCP state transitions.
2531 */
2532void
2533tcp_state_change(struct tcpcb *tp, int newstate)
2534{
2535#if defined(KDTRACE_HOOKS)
2536 int pstate = tp->t_state;
2537#endif
2538
2539 tp->t_state = newstate;
2540 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2541}