1 /*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_subr.c 8.2 (Berkeley) 5/24/95
30 */
31
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD: releng/9.0/sys/netinet/tcp_subr.c 224010 2011-07-14 13:44:48Z bz $");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_tcpdebug.h"
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/callout.h>
44 #include <sys/hhook.h>
45 #include <sys/kernel.h>
46 #include <sys/khelp.h>
47 #include <sys/sysctl.h>
48 #include <sys/jail.h>
49 #include <sys/malloc.h>
50 #include <sys/mbuf.h>
51 #ifdef INET6
52 #include <sys/domain.h>
53 #endif
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/socket.h>
57 #include <sys/socketvar.h>
58 #include <sys/protosw.h>
59 #include <sys/random.h>
60
61 #include <vm/uma.h>
62
63 #include <net/route.h>
64 #include <net/if.h>
65 #include <net/vnet.h>
66
67 #include <netinet/cc.h>
68 #include <netinet/in.h>
69 #include <netinet/in_pcb.h>
70 #include <netinet/in_systm.h>
71 #include <netinet/in_var.h>
72 #include <netinet/ip.h>
73 #include <netinet/ip_icmp.h>
74 #include <netinet/ip_var.h>
75 #ifdef INET6
76 #include <netinet/ip6.h>
77 #include <netinet6/in6_pcb.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/scope6_var.h>
80 #include <netinet6/nd6.h>
81 #endif
82
83 #include <netinet/tcp_fsm.h>
84 #include <netinet/tcp_seq.h>
85 #include <netinet/tcp_timer.h>
86 #include <netinet/tcp_var.h>
87 #include <netinet/tcp_syncache.h>
88 #include <netinet/tcp_offload.h>
89 #ifdef INET6
90 #include <netinet6/tcp6_var.h>
91 #endif
92 #include <netinet/tcpip.h>
93 #ifdef TCPDEBUG
94 #include <netinet/tcp_debug.h>
95 #endif
96 #ifdef INET6
97 #include <netinet6/ip6protosw.h>
98 #endif
99
100 #ifdef IPSEC
101 #include <netipsec/ipsec.h>
102 #include <netipsec/xform.h>
103 #ifdef INET6
104 #include <netipsec/ipsec6.h>
105 #endif
106 #include <netipsec/key.h>
107 #include <sys/syslog.h>
108 #endif /*IPSEC*/
109
110 #include <machine/in_cksum.h>
111 #include <sys/md5.h>
112
113 #include <security/mac/mac_framework.h>
114
115 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
116 #ifdef INET6
117 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
118 #endif
119
120 static int
121 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
122 {
123 int error, new;
124
125 new = V_tcp_mssdflt;
126 error = sysctl_handle_int(oidp, &new, 0, req);
127 if (error == 0 && req->newptr) {
128 if (new < TCP_MINMSS)
129 error = EINVAL;
130 else
131 V_tcp_mssdflt = new;
132 }
133 return (error);
134 }
135
136 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
137 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
138 &sysctl_net_inet_tcp_mss_check, "I",
139 "Default TCP Maximum Segment Size");
140
141 #ifdef INET6
142 static int
143 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
144 {
145 int error, new;
146
147 new = V_tcp_v6mssdflt;
148 error = sysctl_handle_int(oidp, &new, 0, req);
149 if (error == 0 && req->newptr) {
150 if (new < TCP_MINMSS)
151 error = EINVAL;
152 else
153 V_tcp_v6mssdflt = new;
154 }
155 return (error);
156 }
157
158 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
159 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
160 &sysctl_net_inet_tcp_mss_v6_check, "I",
161 "Default TCP Maximum Segment Size for IPv6");
162 #endif /* INET6 */
163
164 /*
165 * Minimum MSS we accept and use. This prevents DoS attacks where
166 * we are forced to a ridiculous low MSS like 20 and send hundreds
167 * of packets instead of one. The effect scales with the available
168 * bandwidth and quickly saturates the CPU and network interface
169 * with packet generation and sending. Set to zero to disable MINMSS
170 * checking. This setting prevents us from sending too small packets.
171 */
172 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
173 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
174 &VNET_NAME(tcp_minmss), 0,
175 "Minmum TCP Maximum Segment Size");
176
177 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
178 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
179 &VNET_NAME(tcp_do_rfc1323), 0,
180 "Enable rfc1323 (high performance TCP) extensions");
181
182 static int tcp_log_debug = 0;
183 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
184 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
185
186 static int tcp_tcbhashsize = 0;
187 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
188 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
189
190 static int do_tcpdrain = 1;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
192 "Enable tcp_drain routine for extra help when low on mbufs");
193
194 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
195 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
196
197 static VNET_DEFINE(int, icmp_may_rst) = 1;
198 #define V_icmp_may_rst VNET(icmp_may_rst)
199 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
200 &VNET_NAME(icmp_may_rst), 0,
201 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
202
203 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
204 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
205 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
206 &VNET_NAME(tcp_isn_reseed_interval), 0,
207 "Seconds between reseeding of ISN secret");
208
209 static int tcp_soreceive_stream = 0;
210 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
211 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
212
213 #ifdef TCP_SIGNATURE
214 static int tcp_sig_checksigs = 1;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
216 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
217 #endif
218
219 VNET_DEFINE(uma_zone_t, sack_hole_zone);
220 #define V_sack_hole_zone VNET(sack_hole_zone)
221
222 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
223
224 static struct inpcb *tcp_notify(struct inpcb *, int);
225 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
226 void *ip4hdr, const void *ip6hdr);
227
228 /*
229 * Target size of TCP PCB hash tables. Must be a power of two.
230 *
231 * Note that this can be overridden by the kernel environment
232 * variable net.inet.tcp.tcbhashsize
233 */
234 #ifndef TCBHASHSIZE
235 #define TCBHASHSIZE 512
236 #endif
237
238 /*
239 * XXX
240 * Callouts should be moved into struct tcp directly. They are currently
241 * separate because the tcpcb structure is exported to userland for sysctl
242 * parsing purposes, which do not know about callouts.
243 */
244 struct tcpcb_mem {
245 struct tcpcb tcb;
246 struct tcp_timer tt;
247 struct cc_var ccv;
248 struct osd osd;
249 };
250
251 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
252 #define V_tcpcb_zone VNET(tcpcb_zone)
253
254 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
255 static struct mtx isn_mtx;
256
257 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
258 #define ISN_LOCK() mtx_lock(&isn_mtx)
259 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
260
261 /*
262 * TCP initialization.
263 */
264 static void
265 tcp_zone_change(void *tag)
266 {
267
268 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
269 uma_zone_set_max(V_tcpcb_zone, maxsockets);
270 tcp_tw_zone_change();
271 }
272
273 static int
274 tcp_inpcb_init(void *mem, int size, int flags)
275 {
276 struct inpcb *inp = mem;
277
278 INP_LOCK_INIT(inp, "inp", "tcpinp");
279 return (0);
280 }
281
282 void
283 tcp_init(void)
284 {
285 int hashsize;
286
287 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
288 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
289 printf("%s: WARNING: unable to register helper hook\n", __func__);
290 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
291 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
292 printf("%s: WARNING: unable to register helper hook\n", __func__);
293
294 hashsize = TCBHASHSIZE;
295 TUNABLE_INT_FETCH("net.inet.tcp.tcbhashsize", &hashsize);
296 if (!powerof2(hashsize)) {
297 printf("WARNING: TCB hash size not a power of 2\n");
298 hashsize = 512; /* safe default */
299 }
300 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
301 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
302 IPI_HASHFIELDS_4TUPLE);
303
304 /*
305 * These have to be type stable for the benefit of the timers.
306 */
307 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
308 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
309 uma_zone_set_max(V_tcpcb_zone, maxsockets);
310
311 tcp_tw_init();
312 syncache_init();
313 tcp_hc_init();
314 tcp_reass_init();
315
316 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
317 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
318 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
319
320 /* Skip initialization of globals for non-default instances. */
321 if (!IS_DEFAULT_VNET(curvnet))
322 return;
323
324 /* XXX virtualize those bellow? */
325 tcp_delacktime = TCPTV_DELACK;
326 tcp_keepinit = TCPTV_KEEP_INIT;
327 tcp_keepidle = TCPTV_KEEP_IDLE;
328 tcp_keepintvl = TCPTV_KEEPINTVL;
329 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
330 tcp_msl = TCPTV_MSL;
331 tcp_rexmit_min = TCPTV_MIN;
332 if (tcp_rexmit_min < 1)
333 tcp_rexmit_min = 1;
334 tcp_rexmit_slop = TCPTV_CPU_VAR;
335 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
336 tcp_tcbhashsize = hashsize;
337
338 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
339 if (tcp_soreceive_stream) {
340 #ifdef INET
341 tcp_usrreqs.pru_soreceive = soreceive_stream;
342 #endif
343 #ifdef INET6
344 tcp6_usrreqs.pru_soreceive = soreceive_stream;
345 #endif /* INET6 */
346 }
347
348 #ifdef INET6
349 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
350 #else /* INET6 */
351 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
352 #endif /* INET6 */
353 if (max_protohdr < TCP_MINPROTOHDR)
354 max_protohdr = TCP_MINPROTOHDR;
355 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
356 panic("tcp_init");
357 #undef TCP_MINPROTOHDR
358
359 ISN_LOCK_INIT();
360 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
361 SHUTDOWN_PRI_DEFAULT);
362 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
363 EVENTHANDLER_PRI_ANY);
364 }
365
366 #ifdef VIMAGE
367 void
368 tcp_destroy(void)
369 {
370
371 tcp_reass_destroy();
372 tcp_hc_destroy();
373 syncache_destroy();
374 tcp_tw_destroy();
375 in_pcbinfo_destroy(&V_tcbinfo);
376 uma_zdestroy(V_sack_hole_zone);
377 uma_zdestroy(V_tcpcb_zone);
378 }
379 #endif
380
381 void
382 tcp_fini(void *xtp)
383 {
384
385 }
386
387 /*
388 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
389 * tcp_template used to store this data in mbufs, but we now recopy it out
390 * of the tcpcb each time to conserve mbufs.
391 */
392 void
393 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
394 {
395 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
396
397 INP_WLOCK_ASSERT(inp);
398
399 #ifdef INET6
400 if ((inp->inp_vflag & INP_IPV6) != 0) {
401 struct ip6_hdr *ip6;
402
403 ip6 = (struct ip6_hdr *)ip_ptr;
404 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
405 (inp->inp_flow & IPV6_FLOWINFO_MASK);
406 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
407 (IPV6_VERSION & IPV6_VERSION_MASK);
408 ip6->ip6_nxt = IPPROTO_TCP;
409 ip6->ip6_plen = htons(sizeof(struct tcphdr));
410 ip6->ip6_src = inp->in6p_laddr;
411 ip6->ip6_dst = inp->in6p_faddr;
412 }
413 #endif /* INET6 */
414 #if defined(INET6) && defined(INET)
415 else
416 #endif
417 #ifdef INET
418 {
419 struct ip *ip;
420
421 ip = (struct ip *)ip_ptr;
422 ip->ip_v = IPVERSION;
423 ip->ip_hl = 5;
424 ip->ip_tos = inp->inp_ip_tos;
425 ip->ip_len = 0;
426 ip->ip_id = 0;
427 ip->ip_off = 0;
428 ip->ip_ttl = inp->inp_ip_ttl;
429 ip->ip_sum = 0;
430 ip->ip_p = IPPROTO_TCP;
431 ip->ip_src = inp->inp_laddr;
432 ip->ip_dst = inp->inp_faddr;
433 }
434 #endif /* INET */
435 th->th_sport = inp->inp_lport;
436 th->th_dport = inp->inp_fport;
437 th->th_seq = 0;
438 th->th_ack = 0;
439 th->th_x2 = 0;
440 th->th_off = 5;
441 th->th_flags = 0;
442 th->th_win = 0;
443 th->th_urp = 0;
444 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
445 }
446
447 /*
448 * Create template to be used to send tcp packets on a connection.
449 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
450 * use for this function is in keepalives, which use tcp_respond.
451 */
452 struct tcptemp *
453 tcpip_maketemplate(struct inpcb *inp)
454 {
455 struct tcptemp *t;
456
457 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
458 if (t == NULL)
459 return (NULL);
460 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
461 return (t);
462 }
463
464 /*
465 * Send a single message to the TCP at address specified by
466 * the given TCP/IP header. If m == NULL, then we make a copy
467 * of the tcpiphdr at ti and send directly to the addressed host.
468 * This is used to force keep alive messages out using the TCP
469 * template for a connection. If flags are given then we send
470 * a message back to the TCP which originated the * segment ti,
471 * and discard the mbuf containing it and any other attached mbufs.
472 *
473 * In any case the ack and sequence number of the transmitted
474 * segment are as specified by the parameters.
475 *
476 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
477 */
478 void
479 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
480 tcp_seq ack, tcp_seq seq, int flags)
481 {
482 int tlen;
483 int win = 0;
484 struct ip *ip;
485 struct tcphdr *nth;
486 #ifdef INET6
487 struct ip6_hdr *ip6;
488 int isipv6;
489 #endif /* INET6 */
490 int ipflags = 0;
491 struct inpcb *inp;
492
493 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
494
495 #ifdef INET6
496 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
497 ip6 = ipgen;
498 #endif /* INET6 */
499 ip = ipgen;
500
501 if (tp != NULL) {
502 inp = tp->t_inpcb;
503 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
504 INP_WLOCK_ASSERT(inp);
505 } else
506 inp = NULL;
507
508 if (tp != NULL) {
509 if (!(flags & TH_RST)) {
510 win = sbspace(&inp->inp_socket->so_rcv);
511 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
512 win = (long)TCP_MAXWIN << tp->rcv_scale;
513 }
514 }
515 if (m == NULL) {
516 m = m_gethdr(M_DONTWAIT, MT_DATA);
517 if (m == NULL)
518 return;
519 tlen = 0;
520 m->m_data += max_linkhdr;
521 #ifdef INET6
522 if (isipv6) {
523 bcopy((caddr_t)ip6, mtod(m, caddr_t),
524 sizeof(struct ip6_hdr));
525 ip6 = mtod(m, struct ip6_hdr *);
526 nth = (struct tcphdr *)(ip6 + 1);
527 } else
528 #endif /* INET6 */
529 {
530 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
531 ip = mtod(m, struct ip *);
532 nth = (struct tcphdr *)(ip + 1);
533 }
534 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
535 flags = TH_ACK;
536 } else {
537 /*
538 * reuse the mbuf.
539 * XXX MRT We inherrit the FIB, which is lucky.
540 */
541 m_freem(m->m_next);
542 m->m_next = NULL;
543 m->m_data = (caddr_t)ipgen;
544 m_addr_changed(m);
545 /* m_len is set later */
546 tlen = 0;
547 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
548 #ifdef INET6
549 if (isipv6) {
550 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
551 nth = (struct tcphdr *)(ip6 + 1);
552 } else
553 #endif /* INET6 */
554 {
555 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
556 nth = (struct tcphdr *)(ip + 1);
557 }
558 if (th != nth) {
559 /*
560 * this is usually a case when an extension header
561 * exists between the IPv6 header and the
562 * TCP header.
563 */
564 nth->th_sport = th->th_sport;
565 nth->th_dport = th->th_dport;
566 }
567 xchg(nth->th_dport, nth->th_sport, uint16_t);
568 #undef xchg
569 }
570 #ifdef INET6
571 if (isipv6) {
572 ip6->ip6_flow = 0;
573 ip6->ip6_vfc = IPV6_VERSION;
574 ip6->ip6_nxt = IPPROTO_TCP;
575 ip6->ip6_plen = htons((u_short)(sizeof (struct tcphdr) +
576 tlen));
577 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
578 }
579 #endif
580 #if defined(INET) && defined(INET6)
581 else
582 #endif
583 #ifdef INET
584 {
585 tlen += sizeof (struct tcpiphdr);
586 ip->ip_len = tlen;
587 ip->ip_ttl = V_ip_defttl;
588 if (V_path_mtu_discovery)
589 ip->ip_off |= IP_DF;
590 }
591 #endif
592 m->m_len = tlen;
593 m->m_pkthdr.len = tlen;
594 m->m_pkthdr.rcvif = NULL;
595 #ifdef MAC
596 if (inp != NULL) {
597 /*
598 * Packet is associated with a socket, so allow the
599 * label of the response to reflect the socket label.
600 */
601 INP_WLOCK_ASSERT(inp);
602 mac_inpcb_create_mbuf(inp, m);
603 } else {
604 /*
605 * Packet is not associated with a socket, so possibly
606 * update the label in place.
607 */
608 mac_netinet_tcp_reply(m);
609 }
610 #endif
611 nth->th_seq = htonl(seq);
612 nth->th_ack = htonl(ack);
613 nth->th_x2 = 0;
614 nth->th_off = sizeof (struct tcphdr) >> 2;
615 nth->th_flags = flags;
616 if (tp != NULL)
617 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
618 else
619 nth->th_win = htons((u_short)win);
620 nth->th_urp = 0;
621 #ifdef INET6
622 if (isipv6) {
623 nth->th_sum = 0;
624 nth->th_sum = in6_cksum(m, IPPROTO_TCP,
625 sizeof(struct ip6_hdr),
626 tlen - sizeof(struct ip6_hdr));
627 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
628 NULL, NULL);
629 }
630 #endif /* INET6 */
631 #if defined(INET6) && defined(INET)
632 else
633 #endif
634 #ifdef INET
635 {
636 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
637 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
638 m->m_pkthdr.csum_flags = CSUM_TCP;
639 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
640 }
641 #endif /* INET */
642 #ifdef TCPDEBUG
643 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
644 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
645 #endif
646 #ifdef INET6
647 if (isipv6)
648 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
649 #endif /* INET6 */
650 #if defined(INET) && defined(INET6)
651 else
652 #endif
653 #ifdef INET
654 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
655 #endif
656 }
657
658 /*
659 * Create a new TCP control block, making an
660 * empty reassembly queue and hooking it to the argument
661 * protocol control block. The `inp' parameter must have
662 * come from the zone allocator set up in tcp_init().
663 */
664 struct tcpcb *
665 tcp_newtcpcb(struct inpcb *inp)
666 {
667 struct tcpcb_mem *tm;
668 struct tcpcb *tp;
669 #ifdef INET6
670 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
671 #endif /* INET6 */
672
673 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
674 if (tm == NULL)
675 return (NULL);
676 tp = &tm->tcb;
677
678 /* Initialise cc_var struct for this tcpcb. */
679 tp->ccv = &tm->ccv;
680 tp->ccv->type = IPPROTO_TCP;
681 tp->ccv->ccvc.tcp = tp;
682
683 /*
684 * Use the current system default CC algorithm.
685 */
686 CC_LIST_RLOCK();
687 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
688 CC_ALGO(tp) = CC_DEFAULT();
689 CC_LIST_RUNLOCK();
690
691 if (CC_ALGO(tp)->cb_init != NULL)
692 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
693 uma_zfree(V_tcpcb_zone, tm);
694 return (NULL);
695 }
696
697 tp->osd = &tm->osd;
698 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
699 uma_zfree(V_tcpcb_zone, tm);
700 return (NULL);
701 }
702
703 #ifdef VIMAGE
704 tp->t_vnet = inp->inp_vnet;
705 #endif
706 tp->t_timers = &tm->tt;
707 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
708 tp->t_maxseg = tp->t_maxopd =
709 #ifdef INET6
710 isipv6 ? V_tcp_v6mssdflt :
711 #endif /* INET6 */
712 V_tcp_mssdflt;
713
714 /* Set up our timeouts. */
715 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
716 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
717 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
718 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
719 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
720
721 if (V_tcp_do_rfc1323)
722 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
723 if (V_tcp_do_sack)
724 tp->t_flags |= TF_SACK_PERMIT;
725 TAILQ_INIT(&tp->snd_holes);
726 tp->t_inpcb = inp; /* XXX */
727 /*
728 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
729 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
730 * reasonable initial retransmit time.
731 */
732 tp->t_srtt = TCPTV_SRTTBASE;
733 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
734 tp->t_rttmin = tcp_rexmit_min;
735 tp->t_rxtcur = TCPTV_RTOBASE;
736 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
737 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
738 tp->t_rcvtime = ticks;
739 /*
740 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
741 * because the socket may be bound to an IPv6 wildcard address,
742 * which may match an IPv4-mapped IPv6 address.
743 */
744 inp->inp_ip_ttl = V_ip_defttl;
745 inp->inp_ppcb = tp;
746 return (tp); /* XXX */
747 }
748
749 /*
750 * Switch the congestion control algorithm back to NewReno for any active
751 * control blocks using an algorithm which is about to go away.
752 * This ensures the CC framework can allow the unload to proceed without leaving
753 * any dangling pointers which would trigger a panic.
754 * Returning non-zero would inform the CC framework that something went wrong
755 * and it would be unsafe to allow the unload to proceed. However, there is no
756 * way for this to occur with this implementation so we always return zero.
757 */
758 int
759 tcp_ccalgounload(struct cc_algo *unload_algo)
760 {
761 struct cc_algo *tmpalgo;
762 struct inpcb *inp;
763 struct tcpcb *tp;
764 VNET_ITERATOR_DECL(vnet_iter);
765
766 /*
767 * Check all active control blocks across all network stacks and change
768 * any that are using "unload_algo" back to NewReno. If "unload_algo"
769 * requires cleanup code to be run, call it.
770 */
771 VNET_LIST_RLOCK();
772 VNET_FOREACH(vnet_iter) {
773 CURVNET_SET(vnet_iter);
774 INP_INFO_RLOCK(&V_tcbinfo);
775 /*
776 * New connections already part way through being initialised
777 * with the CC algo we're removing will not race with this code
778 * because the INP_INFO_WLOCK is held during initialisation. We
779 * therefore don't enter the loop below until the connection
780 * list has stabilised.
781 */
782 LIST_FOREACH(inp, &V_tcb, inp_list) {
783 INP_WLOCK(inp);
784 /* Important to skip tcptw structs. */
785 if (!(inp->inp_flags & INP_TIMEWAIT) &&
786 (tp = intotcpcb(inp)) != NULL) {
787 /*
788 * By holding INP_WLOCK here, we are assured
789 * that the connection is not currently
790 * executing inside the CC module's functions
791 * i.e. it is safe to make the switch back to
792 * NewReno.
793 */
794 if (CC_ALGO(tp) == unload_algo) {
795 tmpalgo = CC_ALGO(tp);
796 /* NewReno does not require any init. */
797 CC_ALGO(tp) = &newreno_cc_algo;
798 if (tmpalgo->cb_destroy != NULL)
799 tmpalgo->cb_destroy(tp->ccv);
800 }
801 }
802 INP_WUNLOCK(inp);
803 }
804 INP_INFO_RUNLOCK(&V_tcbinfo);
805 CURVNET_RESTORE();
806 }
807 VNET_LIST_RUNLOCK();
808
809 return (0);
810 }
811
812 /*
813 * Drop a TCP connection, reporting
814 * the specified error. If connection is synchronized,
815 * then send a RST to peer.
816 */
817 struct tcpcb *
818 tcp_drop(struct tcpcb *tp, int errno)
819 {
820 struct socket *so = tp->t_inpcb->inp_socket;
821
822 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
823 INP_WLOCK_ASSERT(tp->t_inpcb);
824
825 if (TCPS_HAVERCVDSYN(tp->t_state)) {
826 tp->t_state = TCPS_CLOSED;
827 (void) tcp_output_reset(tp);
828 TCPSTAT_INC(tcps_drops);
829 } else
830 TCPSTAT_INC(tcps_conndrops);
831 if (errno == ETIMEDOUT && tp->t_softerror)
832 errno = tp->t_softerror;
833 so->so_error = errno;
834 return (tcp_close(tp));
835 }
836
837 void
838 tcp_discardcb(struct tcpcb *tp)
839 {
840 struct inpcb *inp = tp->t_inpcb;
841 struct socket *so = inp->inp_socket;
842 #ifdef INET6
843 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
844 #endif /* INET6 */
845
846 INP_WLOCK_ASSERT(inp);
847
848 /*
849 * Make sure that all of our timers are stopped before we delete the
850 * PCB.
851 *
852 * XXXRW: Really, we would like to use callout_drain() here in order
853 * to avoid races experienced in tcp_timer.c where a timer is already
854 * executing at this point. However, we can't, both because we're
855 * running in a context where we can't sleep, and also because we
856 * hold locks required by the timers. What we instead need to do is
857 * test to see if callout_drain() is required, and if so, defer some
858 * portion of the remainder of tcp_discardcb() to an asynchronous
859 * context that can callout_drain() and then continue. Some care
860 * will be required to ensure that no further processing takes place
861 * on the tcpcb, even though it hasn't been freed (a flag?).
862 */
863 callout_stop(&tp->t_timers->tt_rexmt);
864 callout_stop(&tp->t_timers->tt_persist);
865 callout_stop(&tp->t_timers->tt_keep);
866 callout_stop(&tp->t_timers->tt_2msl);
867 callout_stop(&tp->t_timers->tt_delack);
868
869 /*
870 * If we got enough samples through the srtt filter,
871 * save the rtt and rttvar in the routing entry.
872 * 'Enough' is arbitrarily defined as 4 rtt samples.
873 * 4 samples is enough for the srtt filter to converge
874 * to within enough % of the correct value; fewer samples
875 * and we could save a bogus rtt. The danger is not high
876 * as tcp quickly recovers from everything.
877 * XXX: Works very well but needs some more statistics!
878 */
879 if (tp->t_rttupdated >= 4) {
880 struct hc_metrics_lite metrics;
881 u_long ssthresh;
882
883 bzero(&metrics, sizeof(metrics));
884 /*
885 * Update the ssthresh always when the conditions below
886 * are satisfied. This gives us better new start value
887 * for the congestion avoidance for new connections.
888 * ssthresh is only set if packet loss occured on a session.
889 *
890 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
891 * being torn down. Ideally this code would not use 'so'.
892 */
893 ssthresh = tp->snd_ssthresh;
894 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
895 /*
896 * convert the limit from user data bytes to
897 * packets then to packet data bytes.
898 */
899 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
900 if (ssthresh < 2)
901 ssthresh = 2;
902 ssthresh *= (u_long)(tp->t_maxseg +
903 #ifdef INET6
904 (isipv6 ? sizeof (struct ip6_hdr) +
905 sizeof (struct tcphdr) :
906 #endif
907 sizeof (struct tcpiphdr)
908 #ifdef INET6
909 )
910 #endif
911 );
912 } else
913 ssthresh = 0;
914 metrics.rmx_ssthresh = ssthresh;
915
916 metrics.rmx_rtt = tp->t_srtt;
917 metrics.rmx_rttvar = tp->t_rttvar;
918 metrics.rmx_cwnd = tp->snd_cwnd;
919 metrics.rmx_sendpipe = 0;
920 metrics.rmx_recvpipe = 0;
921
922 tcp_hc_update(&inp->inp_inc, &metrics);
923 }
924
925 /* free the reassembly queue, if any */
926 tcp_reass_flush(tp);
927 /* Disconnect offload device, if any. */
928 tcp_offload_detach(tp);
929
930 tcp_free_sackholes(tp);
931
932 /* Allow the CC algorithm to clean up after itself. */
933 if (CC_ALGO(tp)->cb_destroy != NULL)
934 CC_ALGO(tp)->cb_destroy(tp->ccv);
935
936 khelp_destroy_osd(tp->osd);
937
938 CC_ALGO(tp) = NULL;
939 inp->inp_ppcb = NULL;
940 tp->t_inpcb = NULL;
941 uma_zfree(V_tcpcb_zone, tp);
942 }
943
944 /*
945 * Attempt to close a TCP control block, marking it as dropped, and freeing
946 * the socket if we hold the only reference.
947 */
948 struct tcpcb *
949 tcp_close(struct tcpcb *tp)
950 {
951 struct inpcb *inp = tp->t_inpcb;
952 struct socket *so;
953
954 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
955 INP_WLOCK_ASSERT(inp);
956
957 /* Notify any offload devices of listener close */
958 if (tp->t_state == TCPS_LISTEN)
959 tcp_offload_listen_close(tp);
960 in_pcbdrop(inp);
961 TCPSTAT_INC(tcps_closed);
962 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
963 so = inp->inp_socket;
964 soisdisconnected(so);
965 if (inp->inp_flags & INP_SOCKREF) {
966 KASSERT(so->so_state & SS_PROTOREF,
967 ("tcp_close: !SS_PROTOREF"));
968 inp->inp_flags &= ~INP_SOCKREF;
969 INP_WUNLOCK(inp);
970 ACCEPT_LOCK();
971 SOCK_LOCK(so);
972 so->so_state &= ~SS_PROTOREF;
973 sofree(so);
974 return (NULL);
975 }
976 return (tp);
977 }
978
979 void
980 tcp_drain(void)
981 {
982 VNET_ITERATOR_DECL(vnet_iter);
983
984 if (!do_tcpdrain)
985 return;
986
987 VNET_LIST_RLOCK_NOSLEEP();
988 VNET_FOREACH(vnet_iter) {
989 CURVNET_SET(vnet_iter);
990 struct inpcb *inpb;
991 struct tcpcb *tcpb;
992
993 /*
994 * Walk the tcpbs, if existing, and flush the reassembly queue,
995 * if there is one...
996 * XXX: The "Net/3" implementation doesn't imply that the TCP
997 * reassembly queue should be flushed, but in a situation
998 * where we're really low on mbufs, this is potentially
999 * usefull.
1000 */
1001 INP_INFO_RLOCK(&V_tcbinfo);
1002 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1003 if (inpb->inp_flags & INP_TIMEWAIT)
1004 continue;
1005 INP_WLOCK(inpb);
1006 if ((tcpb = intotcpcb(inpb)) != NULL) {
1007 tcp_reass_flush(tcpb);
1008 tcp_clean_sackreport(tcpb);
1009 }
1010 INP_WUNLOCK(inpb);
1011 }
1012 INP_INFO_RUNLOCK(&V_tcbinfo);
1013 CURVNET_RESTORE();
1014 }
1015 VNET_LIST_RUNLOCK_NOSLEEP();
1016 }
1017
1018 /*
1019 * Notify a tcp user of an asynchronous error;
1020 * store error as soft error, but wake up user
1021 * (for now, won't do anything until can select for soft error).
1022 *
1023 * Do not wake up user since there currently is no mechanism for
1024 * reporting soft errors (yet - a kqueue filter may be added).
1025 */
1026 static struct inpcb *
1027 tcp_notify(struct inpcb *inp, int error)
1028 {
1029 struct tcpcb *tp;
1030
1031 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1032 INP_WLOCK_ASSERT(inp);
1033
1034 if ((inp->inp_flags & INP_TIMEWAIT) ||
1035 (inp->inp_flags & INP_DROPPED))
1036 return (inp);
1037
1038 tp = intotcpcb(inp);
1039 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1040
1041 /*
1042 * Ignore some errors if we are hooked up.
1043 * If connection hasn't completed, has retransmitted several times,
1044 * and receives a second error, give up now. This is better
1045 * than waiting a long time to establish a connection that
1046 * can never complete.
1047 */
1048 if (tp->t_state == TCPS_ESTABLISHED &&
1049 (error == EHOSTUNREACH || error == ENETUNREACH ||
1050 error == EHOSTDOWN)) {
1051 return (inp);
1052 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1053 tp->t_softerror) {
1054 tp = tcp_drop(tp, error);
1055 if (tp != NULL)
1056 return (inp);
1057 else
1058 return (NULL);
1059 } else {
1060 tp->t_softerror = error;
1061 return (inp);
1062 }
1063 #if 0
1064 wakeup( &so->so_timeo);
1065 sorwakeup(so);
1066 sowwakeup(so);
1067 #endif
1068 }
1069
1070 static int
1071 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1072 {
1073 int error, i, m, n, pcb_count;
1074 struct inpcb *inp, **inp_list;
1075 inp_gen_t gencnt;
1076 struct xinpgen xig;
1077
1078 /*
1079 * The process of preparing the TCB list is too time-consuming and
1080 * resource-intensive to repeat twice on every request.
1081 */
1082 if (req->oldptr == NULL) {
1083 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1084 n += imax(n / 8, 10);
1085 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1086 return (0);
1087 }
1088
1089 if (req->newptr != NULL)
1090 return (EPERM);
1091
1092 /*
1093 * OK, now we're committed to doing something.
1094 */
1095 INP_INFO_RLOCK(&V_tcbinfo);
1096 gencnt = V_tcbinfo.ipi_gencnt;
1097 n = V_tcbinfo.ipi_count;
1098 INP_INFO_RUNLOCK(&V_tcbinfo);
1099
1100 m = syncache_pcbcount();
1101
1102 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1103 + (n + m) * sizeof(struct xtcpcb));
1104 if (error != 0)
1105 return (error);
1106
1107 xig.xig_len = sizeof xig;
1108 xig.xig_count = n + m;
1109 xig.xig_gen = gencnt;
1110 xig.xig_sogen = so_gencnt;
1111 error = SYSCTL_OUT(req, &xig, sizeof xig);
1112 if (error)
1113 return (error);
1114
1115 error = syncache_pcblist(req, m, &pcb_count);
1116 if (error)
1117 return (error);
1118
1119 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1120 if (inp_list == NULL)
1121 return (ENOMEM);
1122
1123 INP_INFO_RLOCK(&V_tcbinfo);
1124 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1125 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1126 INP_WLOCK(inp);
1127 if (inp->inp_gencnt <= gencnt) {
1128 /*
1129 * XXX: This use of cr_cansee(), introduced with
1130 * TCP state changes, is not quite right, but for
1131 * now, better than nothing.
1132 */
1133 if (inp->inp_flags & INP_TIMEWAIT) {
1134 if (intotw(inp) != NULL)
1135 error = cr_cansee(req->td->td_ucred,
1136 intotw(inp)->tw_cred);
1137 else
1138 error = EINVAL; /* Skip this inp. */
1139 } else
1140 error = cr_canseeinpcb(req->td->td_ucred, inp);
1141 if (error == 0) {
1142 in_pcbref(inp);
1143 inp_list[i++] = inp;
1144 }
1145 }
1146 INP_WUNLOCK(inp);
1147 }
1148 INP_INFO_RUNLOCK(&V_tcbinfo);
1149 n = i;
1150
1151 error = 0;
1152 for (i = 0; i < n; i++) {
1153 inp = inp_list[i];
1154 INP_RLOCK(inp);
1155 if (inp->inp_gencnt <= gencnt) {
1156 struct xtcpcb xt;
1157 void *inp_ppcb;
1158
1159 bzero(&xt, sizeof(xt));
1160 xt.xt_len = sizeof xt;
1161 /* XXX should avoid extra copy */
1162 bcopy(inp, &xt.xt_inp, sizeof *inp);
1163 inp_ppcb = inp->inp_ppcb;
1164 if (inp_ppcb == NULL)
1165 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1166 else if (inp->inp_flags & INP_TIMEWAIT) {
1167 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1168 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1169 } else {
1170 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1171 if (xt.xt_tp.t_timers)
1172 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1173 }
1174 if (inp->inp_socket != NULL)
1175 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1176 else {
1177 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1178 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1179 }
1180 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1181 INP_RUNLOCK(inp);
1182 error = SYSCTL_OUT(req, &xt, sizeof xt);
1183 } else
1184 INP_RUNLOCK(inp);
1185 }
1186 INP_INFO_WLOCK(&V_tcbinfo);
1187 for (i = 0; i < n; i++) {
1188 inp = inp_list[i];
1189 INP_RLOCK(inp);
1190 if (!in_pcbrele_rlocked(inp))
1191 INP_RUNLOCK(inp);
1192 }
1193 INP_INFO_WUNLOCK(&V_tcbinfo);
1194
1195 if (!error) {
1196 /*
1197 * Give the user an updated idea of our state.
1198 * If the generation differs from what we told
1199 * her before, she knows that something happened
1200 * while we were processing this request, and it
1201 * might be necessary to retry.
1202 */
1203 INP_INFO_RLOCK(&V_tcbinfo);
1204 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1205 xig.xig_sogen = so_gencnt;
1206 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1207 INP_INFO_RUNLOCK(&V_tcbinfo);
1208 error = SYSCTL_OUT(req, &xig, sizeof xig);
1209 }
1210 free(inp_list, M_TEMP);
1211 return (error);
1212 }
1213
1214 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1215 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1216 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1217
1218 #ifdef INET
1219 static int
1220 tcp_getcred(SYSCTL_HANDLER_ARGS)
1221 {
1222 struct xucred xuc;
1223 struct sockaddr_in addrs[2];
1224 struct inpcb *inp;
1225 int error;
1226
1227 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1228 if (error)
1229 return (error);
1230 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1231 if (error)
1232 return (error);
1233 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1234 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1235 if (inp != NULL) {
1236 if (inp->inp_socket == NULL)
1237 error = ENOENT;
1238 if (error == 0)
1239 error = cr_canseeinpcb(req->td->td_ucred, inp);
1240 if (error == 0)
1241 cru2x(inp->inp_cred, &xuc);
1242 INP_RUNLOCK(inp);
1243 } else
1244 error = ENOENT;
1245 if (error == 0)
1246 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1247 return (error);
1248 }
1249
1250 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1251 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1252 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1253 #endif /* INET */
1254
1255 #ifdef INET6
1256 static int
1257 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1258 {
1259 struct xucred xuc;
1260 struct sockaddr_in6 addrs[2];
1261 struct inpcb *inp;
1262 int error;
1263 #ifdef INET
1264 int mapped = 0;
1265 #endif
1266
1267 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1268 if (error)
1269 return (error);
1270 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1271 if (error)
1272 return (error);
1273 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1274 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1275 return (error);
1276 }
1277 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1278 #ifdef INET
1279 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1280 mapped = 1;
1281 else
1282 #endif
1283 return (EINVAL);
1284 }
1285
1286 #ifdef INET
1287 if (mapped == 1)
1288 inp = in_pcblookup(&V_tcbinfo,
1289 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1290 addrs[1].sin6_port,
1291 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1292 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1293 else
1294 #endif
1295 inp = in6_pcblookup(&V_tcbinfo,
1296 &addrs[1].sin6_addr, addrs[1].sin6_port,
1297 &addrs[0].sin6_addr, addrs[0].sin6_port,
1298 INPLOOKUP_RLOCKPCB, NULL);
1299 if (inp != NULL) {
1300 if (inp->inp_socket == NULL)
1301 error = ENOENT;
1302 if (error == 0)
1303 error = cr_canseeinpcb(req->td->td_ucred, inp);
1304 if (error == 0)
1305 cru2x(inp->inp_cred, &xuc);
1306 INP_RUNLOCK(inp);
1307 } else
1308 error = ENOENT;
1309 if (error == 0)
1310 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1311 return (error);
1312 }
1313
1314 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1315 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1316 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1317 #endif /* INET6 */
1318
1319
1320 #ifdef INET
1321 void
1322 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1323 {
1324 struct ip *ip = vip;
1325 struct tcphdr *th;
1326 struct in_addr faddr;
1327 struct inpcb *inp;
1328 struct tcpcb *tp;
1329 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1330 struct icmp *icp;
1331 struct in_conninfo inc;
1332 tcp_seq icmp_tcp_seq;
1333 int mtu;
1334
1335 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1336 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1337 return;
1338
1339 if (cmd == PRC_MSGSIZE)
1340 notify = tcp_mtudisc;
1341 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1342 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1343 notify = tcp_drop_syn_sent;
1344 /*
1345 * Redirects don't need to be handled up here.
1346 */
1347 else if (PRC_IS_REDIRECT(cmd))
1348 return;
1349 /*
1350 * Source quench is depreciated.
1351 */
1352 else if (cmd == PRC_QUENCH)
1353 return;
1354 /*
1355 * Hostdead is ugly because it goes linearly through all PCBs.
1356 * XXX: We never get this from ICMP, otherwise it makes an
1357 * excellent DoS attack on machines with many connections.
1358 */
1359 else if (cmd == PRC_HOSTDEAD)
1360 ip = NULL;
1361 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1362 return;
1363 if (ip != NULL) {
1364 icp = (struct icmp *)((caddr_t)ip
1365 - offsetof(struct icmp, icmp_ip));
1366 th = (struct tcphdr *)((caddr_t)ip
1367 + (ip->ip_hl << 2));
1368 INP_INFO_WLOCK(&V_tcbinfo);
1369 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1370 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1371 if (inp != NULL) {
1372 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1373 !(inp->inp_flags & INP_DROPPED) &&
1374 !(inp->inp_socket == NULL)) {
1375 icmp_tcp_seq = htonl(th->th_seq);
1376 tp = intotcpcb(inp);
1377 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1378 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1379 if (cmd == PRC_MSGSIZE) {
1380 /*
1381 * MTU discovery:
1382 * If we got a needfrag set the MTU
1383 * in the route to the suggested new
1384 * value (if given) and then notify.
1385 */
1386 bzero(&inc, sizeof(inc));
1387 inc.inc_faddr = faddr;
1388 inc.inc_fibnum =
1389 inp->inp_inc.inc_fibnum;
1390
1391 mtu = ntohs(icp->icmp_nextmtu);
1392 /*
1393 * If no alternative MTU was
1394 * proposed, try the next smaller
1395 * one. ip->ip_len has already
1396 * been swapped in icmp_input().
1397 */
1398 if (!mtu)
1399 mtu = ip_next_mtu(ip->ip_len,
1400 1);
1401 if (mtu < V_tcp_minmss
1402 + sizeof(struct tcpiphdr))
1403 mtu = V_tcp_minmss
1404 + sizeof(struct tcpiphdr);
1405 /*
1406 * Only cache the MTU if it
1407 * is smaller than the interface
1408 * or route MTU. tcp_mtudisc()
1409 * will do right thing by itself.
1410 */
1411 if (mtu <= tcp_maxmtu(&inc, NULL))
1412 tcp_hc_updatemtu(&inc, mtu);
1413 }
1414
1415 inp = (*notify)(inp, inetctlerrmap[cmd]);
1416 }
1417 }
1418 if (inp != NULL)
1419 INP_WUNLOCK(inp);
1420 } else {
1421 bzero(&inc, sizeof(inc));
1422 inc.inc_fport = th->th_dport;
1423 inc.inc_lport = th->th_sport;
1424 inc.inc_faddr = faddr;
1425 inc.inc_laddr = ip->ip_src;
1426 syncache_unreach(&inc, th);
1427 }
1428 INP_INFO_WUNLOCK(&V_tcbinfo);
1429 } else
1430 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1431 }
1432 #endif /* INET */
1433
1434 #ifdef INET6
1435 void
1436 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1437 {
1438 struct tcphdr th;
1439 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1440 struct ip6_hdr *ip6;
1441 struct mbuf *m;
1442 struct ip6ctlparam *ip6cp = NULL;
1443 const struct sockaddr_in6 *sa6_src = NULL;
1444 int off;
1445 struct tcp_portonly {
1446 u_int16_t th_sport;
1447 u_int16_t th_dport;
1448 } *thp;
1449
1450 if (sa->sa_family != AF_INET6 ||
1451 sa->sa_len != sizeof(struct sockaddr_in6))
1452 return;
1453
1454 if (cmd == PRC_MSGSIZE)
1455 notify = tcp_mtudisc;
1456 else if (!PRC_IS_REDIRECT(cmd) &&
1457 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1458 return;
1459 /* Source quench is depreciated. */
1460 else if (cmd == PRC_QUENCH)
1461 return;
1462
1463 /* if the parameter is from icmp6, decode it. */
1464 if (d != NULL) {
1465 ip6cp = (struct ip6ctlparam *)d;
1466 m = ip6cp->ip6c_m;
1467 ip6 = ip6cp->ip6c_ip6;
1468 off = ip6cp->ip6c_off;
1469 sa6_src = ip6cp->ip6c_src;
1470 } else {
1471 m = NULL;
1472 ip6 = NULL;
1473 off = 0; /* fool gcc */
1474 sa6_src = &sa6_any;
1475 }
1476
1477 if (ip6 != NULL) {
1478 struct in_conninfo inc;
1479 /*
1480 * XXX: We assume that when IPV6 is non NULL,
1481 * M and OFF are valid.
1482 */
1483
1484 /* check if we can safely examine src and dst ports */
1485 if (m->m_pkthdr.len < off + sizeof(*thp))
1486 return;
1487
1488 bzero(&th, sizeof(th));
1489 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1490
1491 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1492 (struct sockaddr *)ip6cp->ip6c_src,
1493 th.th_sport, cmd, NULL, notify);
1494
1495 bzero(&inc, sizeof(inc));
1496 inc.inc_fport = th.th_dport;
1497 inc.inc_lport = th.th_sport;
1498 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1499 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1500 inc.inc_flags |= INC_ISIPV6;
1501 INP_INFO_WLOCK(&V_tcbinfo);
1502 syncache_unreach(&inc, &th);
1503 INP_INFO_WUNLOCK(&V_tcbinfo);
1504 } else
1505 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1506 0, cmd, NULL, notify);
1507 }
1508 #endif /* INET6 */
1509
1510
1511 /*
1512 * Following is where TCP initial sequence number generation occurs.
1513 *
1514 * There are two places where we must use initial sequence numbers:
1515 * 1. In SYN-ACK packets.
1516 * 2. In SYN packets.
1517 *
1518 * All ISNs for SYN-ACK packets are generated by the syncache. See
1519 * tcp_syncache.c for details.
1520 *
1521 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1522 * depends on this property. In addition, these ISNs should be
1523 * unguessable so as to prevent connection hijacking. To satisfy
1524 * the requirements of this situation, the algorithm outlined in
1525 * RFC 1948 is used, with only small modifications.
1526 *
1527 * Implementation details:
1528 *
1529 * Time is based off the system timer, and is corrected so that it
1530 * increases by one megabyte per second. This allows for proper
1531 * recycling on high speed LANs while still leaving over an hour
1532 * before rollover.
1533 *
1534 * As reading the *exact* system time is too expensive to be done
1535 * whenever setting up a TCP connection, we increment the time
1536 * offset in two ways. First, a small random positive increment
1537 * is added to isn_offset for each connection that is set up.
1538 * Second, the function tcp_isn_tick fires once per clock tick
1539 * and increments isn_offset as necessary so that sequence numbers
1540 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1541 * random positive increments serve only to ensure that the same
1542 * exact sequence number is never sent out twice (as could otherwise
1543 * happen when a port is recycled in less than the system tick
1544 * interval.)
1545 *
1546 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1547 * between seeding of isn_secret. This is normally set to zero,
1548 * as reseeding should not be necessary.
1549 *
1550 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1551 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1552 * general, this means holding an exclusive (write) lock.
1553 */
1554
1555 #define ISN_BYTES_PER_SECOND 1048576
1556 #define ISN_STATIC_INCREMENT 4096
1557 #define ISN_RANDOM_INCREMENT (4096 - 1)
1558
1559 static VNET_DEFINE(u_char, isn_secret[32]);
1560 static VNET_DEFINE(int, isn_last);
1561 static VNET_DEFINE(int, isn_last_reseed);
1562 static VNET_DEFINE(u_int32_t, isn_offset);
1563 static VNET_DEFINE(u_int32_t, isn_offset_old);
1564
1565 #define V_isn_secret VNET(isn_secret)
1566 #define V_isn_last VNET(isn_last)
1567 #define V_isn_last_reseed VNET(isn_last_reseed)
1568 #define V_isn_offset VNET(isn_offset)
1569 #define V_isn_offset_old VNET(isn_offset_old)
1570
1571 tcp_seq
1572 tcp_new_isn(struct tcpcb *tp)
1573 {
1574 MD5_CTX isn_ctx;
1575 u_int32_t md5_buffer[4];
1576 tcp_seq new_isn;
1577 u_int32_t projected_offset;
1578
1579 INP_WLOCK_ASSERT(tp->t_inpcb);
1580
1581 ISN_LOCK();
1582 /* Seed if this is the first use, reseed if requested. */
1583 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1584 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1585 < (u_int)ticks))) {
1586 read_random(&V_isn_secret, sizeof(V_isn_secret));
1587 V_isn_last_reseed = ticks;
1588 }
1589
1590 /* Compute the md5 hash and return the ISN. */
1591 MD5Init(&isn_ctx);
1592 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1593 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1594 #ifdef INET6
1595 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1596 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1597 sizeof(struct in6_addr));
1598 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1599 sizeof(struct in6_addr));
1600 } else
1601 #endif
1602 {
1603 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1604 sizeof(struct in_addr));
1605 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1606 sizeof(struct in_addr));
1607 }
1608 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1609 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1610 new_isn = (tcp_seq) md5_buffer[0];
1611 V_isn_offset += ISN_STATIC_INCREMENT +
1612 (arc4random() & ISN_RANDOM_INCREMENT);
1613 if (ticks != V_isn_last) {
1614 projected_offset = V_isn_offset_old +
1615 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1616 if (SEQ_GT(projected_offset, V_isn_offset))
1617 V_isn_offset = projected_offset;
1618 V_isn_offset_old = V_isn_offset;
1619 V_isn_last = ticks;
1620 }
1621 new_isn += V_isn_offset;
1622 ISN_UNLOCK();
1623 return (new_isn);
1624 }
1625
1626 /*
1627 * When a specific ICMP unreachable message is received and the
1628 * connection state is SYN-SENT, drop the connection. This behavior
1629 * is controlled by the icmp_may_rst sysctl.
1630 */
1631 struct inpcb *
1632 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1633 {
1634 struct tcpcb *tp;
1635
1636 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1637 INP_WLOCK_ASSERT(inp);
1638
1639 if ((inp->inp_flags & INP_TIMEWAIT) ||
1640 (inp->inp_flags & INP_DROPPED))
1641 return (inp);
1642
1643 tp = intotcpcb(inp);
1644 if (tp->t_state != TCPS_SYN_SENT)
1645 return (inp);
1646
1647 tp = tcp_drop(tp, errno);
1648 if (tp != NULL)
1649 return (inp);
1650 else
1651 return (NULL);
1652 }
1653
1654 /*
1655 * When `need fragmentation' ICMP is received, update our idea of the MSS
1656 * based on the new value in the route. Also nudge TCP to send something,
1657 * since we know the packet we just sent was dropped.
1658 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1659 */
1660 struct inpcb *
1661 tcp_mtudisc(struct inpcb *inp, int errno)
1662 {
1663 struct tcpcb *tp;
1664 struct socket *so;
1665
1666 INP_WLOCK_ASSERT(inp);
1667 if ((inp->inp_flags & INP_TIMEWAIT) ||
1668 (inp->inp_flags & INP_DROPPED))
1669 return (inp);
1670
1671 tp = intotcpcb(inp);
1672 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1673
1674 tcp_mss_update(tp, -1, NULL, NULL);
1675
1676 so = inp->inp_socket;
1677 SOCKBUF_LOCK(&so->so_snd);
1678 /* If the mss is larger than the socket buffer, decrease the mss. */
1679 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1680 tp->t_maxseg = so->so_snd.sb_hiwat;
1681 SOCKBUF_UNLOCK(&so->so_snd);
1682
1683 TCPSTAT_INC(tcps_mturesent);
1684 tp->t_rtttime = 0;
1685 tp->snd_nxt = tp->snd_una;
1686 tcp_free_sackholes(tp);
1687 tp->snd_recover = tp->snd_max;
1688 if (tp->t_flags & TF_SACK_PERMIT)
1689 EXIT_FASTRECOVERY(tp->t_flags);
1690 tcp_output_send(tp);
1691 return (inp);
1692 }
1693
1694 #ifdef INET
1695 /*
1696 * Look-up the routing entry to the peer of this inpcb. If no route
1697 * is found and it cannot be allocated, then return 0. This routine
1698 * is called by TCP routines that access the rmx structure and by
1699 * tcp_mss_update to get the peer/interface MTU.
1700 */
1701 u_long
1702 tcp_maxmtu(struct in_conninfo *inc, int *flags)
1703 {
1704 struct route sro;
1705 struct sockaddr_in *dst;
1706 struct ifnet *ifp;
1707 u_long maxmtu = 0;
1708
1709 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1710
1711 bzero(&sro, sizeof(sro));
1712 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1713 dst = (struct sockaddr_in *)&sro.ro_dst;
1714 dst->sin_family = AF_INET;
1715 dst->sin_len = sizeof(*dst);
1716 dst->sin_addr = inc->inc_faddr;
1717 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1718 }
1719 if (sro.ro_rt != NULL) {
1720 ifp = sro.ro_rt->rt_ifp;
1721 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1722 maxmtu = ifp->if_mtu;
1723 else
1724 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1725
1726 /* Report additional interface capabilities. */
1727 if (flags != NULL) {
1728 if (ifp->if_capenable & IFCAP_TSO4 &&
1729 ifp->if_hwassist & CSUM_TSO)
1730 *flags |= CSUM_TSO;
1731 }
1732 RTFREE(sro.ro_rt);
1733 }
1734 return (maxmtu);
1735 }
1736 #endif /* INET */
1737
1738 #ifdef INET6
1739 u_long
1740 tcp_maxmtu6(struct in_conninfo *inc, int *flags)
1741 {
1742 struct route_in6 sro6;
1743 struct ifnet *ifp;
1744 u_long maxmtu = 0;
1745
1746 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1747
1748 bzero(&sro6, sizeof(sro6));
1749 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1750 sro6.ro_dst.sin6_family = AF_INET6;
1751 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1752 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1753 rtalloc_ign((struct route *)&sro6, 0);
1754 }
1755 if (sro6.ro_rt != NULL) {
1756 ifp = sro6.ro_rt->rt_ifp;
1757 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1758 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1759 else
1760 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1761 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1762
1763 /* Report additional interface capabilities. */
1764 if (flags != NULL) {
1765 if (ifp->if_capenable & IFCAP_TSO6 &&
1766 ifp->if_hwassist & CSUM_TSO)
1767 *flags |= CSUM_TSO;
1768 }
1769 RTFREE(sro6.ro_rt);
1770 }
1771
1772 return (maxmtu);
1773 }
1774 #endif /* INET6 */
1775
1776 #ifdef IPSEC
1777 /* compute ESP/AH header size for TCP, including outer IP header. */
1778 size_t
1779 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1780 {
1781 struct inpcb *inp;
1782 struct mbuf *m;
1783 size_t hdrsiz;
1784 struct ip *ip;
1785 #ifdef INET6
1786 struct ip6_hdr *ip6;
1787 #endif
1788 struct tcphdr *th;
1789
1790 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1791 return (0);
1792 MGETHDR(m, M_DONTWAIT, MT_DATA);
1793 if (!m)
1794 return (0);
1795
1796 #ifdef INET6
1797 if ((inp->inp_vflag & INP_IPV6) != 0) {
1798 ip6 = mtod(m, struct ip6_hdr *);
1799 th = (struct tcphdr *)(ip6 + 1);
1800 m->m_pkthdr.len = m->m_len =
1801 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1802 tcpip_fillheaders(inp, ip6, th);
1803 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1804 } else
1805 #endif /* INET6 */
1806 {
1807 ip = mtod(m, struct ip *);
1808 th = (struct tcphdr *)(ip + 1);
1809 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1810 tcpip_fillheaders(inp, ip, th);
1811 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1812 }
1813
1814 m_free(m);
1815 return (hdrsiz);
1816 }
1817 #endif /* IPSEC */
1818
1819 #ifdef TCP_SIGNATURE
1820 /*
1821 * Callback function invoked by m_apply() to digest TCP segment data
1822 * contained within an mbuf chain.
1823 */
1824 static int
1825 tcp_signature_apply(void *fstate, void *data, u_int len)
1826 {
1827
1828 MD5Update(fstate, (u_char *)data, len);
1829 return (0);
1830 }
1831
1832 /*
1833 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1834 *
1835 * Parameters:
1836 * m pointer to head of mbuf chain
1837 * _unused
1838 * len length of TCP segment data, excluding options
1839 * optlen length of TCP segment options
1840 * buf pointer to storage for computed MD5 digest
1841 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1842 *
1843 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1844 * When called from tcp_input(), we can be sure that th_sum has been
1845 * zeroed out and verified already.
1846 *
1847 * Return 0 if successful, otherwise return -1.
1848 *
1849 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1850 * search with the destination IP address, and a 'magic SPI' to be
1851 * determined by the application. This is hardcoded elsewhere to 1179
1852 * right now. Another branch of this code exists which uses the SPD to
1853 * specify per-application flows but it is unstable.
1854 */
1855 int
1856 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1857 u_char *buf, u_int direction)
1858 {
1859 union sockaddr_union dst;
1860 #ifdef INET
1861 struct ippseudo ippseudo;
1862 #endif
1863 MD5_CTX ctx;
1864 int doff;
1865 struct ip *ip;
1866 #ifdef INET
1867 struct ipovly *ipovly;
1868 #endif
1869 struct secasvar *sav;
1870 struct tcphdr *th;
1871 #ifdef INET6
1872 struct ip6_hdr *ip6;
1873 struct in6_addr in6;
1874 char ip6buf[INET6_ADDRSTRLEN];
1875 uint32_t plen;
1876 uint16_t nhdr;
1877 #endif
1878 u_short savecsum;
1879
1880 KASSERT(m != NULL, ("NULL mbuf chain"));
1881 KASSERT(buf != NULL, ("NULL signature pointer"));
1882
1883 /* Extract the destination from the IP header in the mbuf. */
1884 bzero(&dst, sizeof(union sockaddr_union));
1885 ip = mtod(m, struct ip *);
1886 #ifdef INET6
1887 ip6 = NULL; /* Make the compiler happy. */
1888 #endif
1889 switch (ip->ip_v) {
1890 #ifdef INET
1891 case IPVERSION:
1892 dst.sa.sa_len = sizeof(struct sockaddr_in);
1893 dst.sa.sa_family = AF_INET;
1894 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1895 ip->ip_src : ip->ip_dst;
1896 break;
1897 #endif
1898 #ifdef INET6
1899 case (IPV6_VERSION >> 4):
1900 ip6 = mtod(m, struct ip6_hdr *);
1901 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1902 dst.sa.sa_family = AF_INET6;
1903 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1904 ip6->ip6_src : ip6->ip6_dst;
1905 break;
1906 #endif
1907 default:
1908 return (EINVAL);
1909 /* NOTREACHED */
1910 break;
1911 }
1912
1913 /* Look up an SADB entry which matches the address of the peer. */
1914 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1915 if (sav == NULL) {
1916 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1917 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1918 #ifdef INET6
1919 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1920 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1921 #endif
1922 "(unsupported)"));
1923 return (EINVAL);
1924 }
1925
1926 MD5Init(&ctx);
1927 /*
1928 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
1929 *
1930 * XXX The ippseudo header MUST be digested in network byte order,
1931 * or else we'll fail the regression test. Assume all fields we've
1932 * been doing arithmetic on have been in host byte order.
1933 * XXX One cannot depend on ipovly->ih_len here. When called from
1934 * tcp_output(), the underlying ip_len member has not yet been set.
1935 */
1936 switch (ip->ip_v) {
1937 #ifdef INET
1938 case IPVERSION:
1939 ipovly = (struct ipovly *)ip;
1940 ippseudo.ippseudo_src = ipovly->ih_src;
1941 ippseudo.ippseudo_dst = ipovly->ih_dst;
1942 ippseudo.ippseudo_pad = 0;
1943 ippseudo.ippseudo_p = IPPROTO_TCP;
1944 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
1945 optlen);
1946 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
1947
1948 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
1949 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
1950 break;
1951 #endif
1952 #ifdef INET6
1953 /*
1954 * RFC 2385, 2.0 Proposal
1955 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
1956 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
1957 * extended next header value (to form 32 bits), and 32-bit segment
1958 * length.
1959 * Note: Upper-Layer Packet Length comes before Next Header.
1960 */
1961 case (IPV6_VERSION >> 4):
1962 in6 = ip6->ip6_src;
1963 in6_clearscope(&in6);
1964 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1965 in6 = ip6->ip6_dst;
1966 in6_clearscope(&in6);
1967 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
1968 plen = htonl(len + sizeof(struct tcphdr) + optlen);
1969 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
1970 nhdr = 0;
1971 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1972 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1973 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1974 nhdr = IPPROTO_TCP;
1975 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
1976
1977 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
1978 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
1979 break;
1980 #endif
1981 default:
1982 return (EINVAL);
1983 /* NOTREACHED */
1984 break;
1985 }
1986
1987
1988 /*
1989 * Step 2: Update MD5 hash with TCP header, excluding options.
1990 * The TCP checksum must be set to zero.
1991 */
1992 savecsum = th->th_sum;
1993 th->th_sum = 0;
1994 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
1995 th->th_sum = savecsum;
1996
1997 /*
1998 * Step 3: Update MD5 hash with TCP segment data.
1999 * Use m_apply() to avoid an early m_pullup().
2000 */
2001 if (len > 0)
2002 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2003
2004 /*
2005 * Step 4: Update MD5 hash with shared secret.
2006 */
2007 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2008 MD5Final(buf, &ctx);
2009
2010 key_sa_recordxfer(sav, m);
2011 KEY_FREESAV(&sav);
2012 return (0);
2013 }
2014
2015 /*
2016 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2017 *
2018 * Parameters:
2019 * m pointer to head of mbuf chain
2020 * len length of TCP segment data, excluding options
2021 * optlen length of TCP segment options
2022 * buf pointer to storage for computed MD5 digest
2023 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2024 *
2025 * Return 1 if successful, otherwise return 0.
2026 */
2027 int
2028 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2029 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2030 {
2031 char tmpdigest[TCP_SIGLEN];
2032
2033 if (tcp_sig_checksigs == 0)
2034 return (1);
2035 if ((tcpbflag & TF_SIGNATURE) == 0) {
2036 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2037
2038 /*
2039 * If this socket is not expecting signature but
2040 * the segment contains signature just fail.
2041 */
2042 TCPSTAT_INC(tcps_sig_err_sigopt);
2043 TCPSTAT_INC(tcps_sig_rcvbadsig);
2044 return (0);
2045 }
2046
2047 /* Signature is not expected, and not present in segment. */
2048 return (1);
2049 }
2050
2051 /*
2052 * If this socket is expecting signature but the segment does not
2053 * contain any just fail.
2054 */
2055 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2056 TCPSTAT_INC(tcps_sig_err_nosigopt);
2057 TCPSTAT_INC(tcps_sig_rcvbadsig);
2058 return (0);
2059 }
2060 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2061 IPSEC_DIR_INBOUND) == -1) {
2062 TCPSTAT_INC(tcps_sig_err_buildsig);
2063 TCPSTAT_INC(tcps_sig_rcvbadsig);
2064 return (0);
2065 }
2066
2067 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2068 TCPSTAT_INC(tcps_sig_rcvbadsig);
2069 return (0);
2070 }
2071 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2072 return (1);
2073 }
2074 #endif /* TCP_SIGNATURE */
2075
2076 static int
2077 sysctl_drop(SYSCTL_HANDLER_ARGS)
2078 {
2079 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2080 struct sockaddr_storage addrs[2];
2081 struct inpcb *inp;
2082 struct tcpcb *tp;
2083 struct tcptw *tw;
2084 struct sockaddr_in *fin, *lin;
2085 #ifdef INET6
2086 struct sockaddr_in6 *fin6, *lin6;
2087 #endif
2088 int error;
2089
2090 inp = NULL;
2091 fin = lin = NULL;
2092 #ifdef INET6
2093 fin6 = lin6 = NULL;
2094 #endif
2095 error = 0;
2096
2097 if (req->oldptr != NULL || req->oldlen != 0)
2098 return (EINVAL);
2099 if (req->newptr == NULL)
2100 return (EPERM);
2101 if (req->newlen < sizeof(addrs))
2102 return (ENOMEM);
2103 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2104 if (error)
2105 return (error);
2106
2107 switch (addrs[0].ss_family) {
2108 #ifdef INET6
2109 case AF_INET6:
2110 fin6 = (struct sockaddr_in6 *)&addrs[0];
2111 lin6 = (struct sockaddr_in6 *)&addrs[1];
2112 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2113 lin6->sin6_len != sizeof(struct sockaddr_in6))
2114 return (EINVAL);
2115 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2116 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2117 return (EINVAL);
2118 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2119 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2120 fin = (struct sockaddr_in *)&addrs[0];
2121 lin = (struct sockaddr_in *)&addrs[1];
2122 break;
2123 }
2124 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2125 if (error)
2126 return (error);
2127 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2128 if (error)
2129 return (error);
2130 break;
2131 #endif
2132 #ifdef INET
2133 case AF_INET:
2134 fin = (struct sockaddr_in *)&addrs[0];
2135 lin = (struct sockaddr_in *)&addrs[1];
2136 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2137 lin->sin_len != sizeof(struct sockaddr_in))
2138 return (EINVAL);
2139 break;
2140 #endif
2141 default:
2142 return (EINVAL);
2143 }
2144 INP_INFO_WLOCK(&V_tcbinfo);
2145 switch (addrs[0].ss_family) {
2146 #ifdef INET6
2147 case AF_INET6:
2148 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2149 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2150 INPLOOKUP_WLOCKPCB, NULL);
2151 break;
2152 #endif
2153 #ifdef INET
2154 case AF_INET:
2155 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2156 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2157 break;
2158 #endif
2159 }
2160 if (inp != NULL) {
2161 if (inp->inp_flags & INP_TIMEWAIT) {
2162 /*
2163 * XXXRW: There currently exists a state where an
2164 * inpcb is present, but its timewait state has been
2165 * discarded. For now, don't allow dropping of this
2166 * type of inpcb.
2167 */
2168 tw = intotw(inp);
2169 if (tw != NULL)
2170 tcp_twclose(tw, 0);
2171 else
2172 INP_WUNLOCK(inp);
2173 } else if (!(inp->inp_flags & INP_DROPPED) &&
2174 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2175 tp = intotcpcb(inp);
2176 tp = tcp_drop(tp, ECONNABORTED);
2177 if (tp != NULL)
2178 INP_WUNLOCK(inp);
2179 } else
2180 INP_WUNLOCK(inp);
2181 } else
2182 error = ESRCH;
2183 INP_INFO_WUNLOCK(&V_tcbinfo);
2184 return (error);
2185 }
2186
2187 SYSCTL_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2188 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2189 0, sysctl_drop, "", "Drop TCP connection");
2190
2191 /*
2192 * Generate a standardized TCP log line for use throughout the
2193 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2194 * allow use in the interrupt context.
2195 *
2196 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2197 * NB: The function may return NULL if memory allocation failed.
2198 *
2199 * Due to header inclusion and ordering limitations the struct ip
2200 * and ip6_hdr pointers have to be passed as void pointers.
2201 */
2202 char *
2203 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2204 const void *ip6hdr)
2205 {
2206
2207 /* Is logging enabled? */
2208 if (tcp_log_in_vain == 0)
2209 return (NULL);
2210
2211 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2212 }
2213
2214 char *
2215 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2216 const void *ip6hdr)
2217 {
2218
2219 /* Is logging enabled? */
2220 if (tcp_log_debug == 0)
2221 return (NULL);
2222
2223 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2224 }
2225
2226 static char *
2227 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2228 const void *ip6hdr)
2229 {
2230 char *s, *sp;
2231 size_t size;
2232 struct ip *ip;
2233 #ifdef INET6
2234 const struct ip6_hdr *ip6;
2235
2236 ip6 = (const struct ip6_hdr *)ip6hdr;
2237 #endif /* INET6 */
2238 ip = (struct ip *)ip4hdr;
2239
2240 /*
2241 * The log line looks like this:
2242 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2243 */
2244 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2245 sizeof(PRINT_TH_FLAGS) + 1 +
2246 #ifdef INET6
2247 2 * INET6_ADDRSTRLEN;
2248 #else
2249 2 * INET_ADDRSTRLEN;
2250 #endif /* INET6 */
2251
2252 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2253 if (s == NULL)
2254 return (NULL);
2255
2256 strcat(s, "TCP: [");
2257 sp = s + strlen(s);
2258
2259 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2260 inet_ntoa_r(inc->inc_faddr, sp);
2261 sp = s + strlen(s);
2262 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2263 sp = s + strlen(s);
2264 inet_ntoa_r(inc->inc_laddr, sp);
2265 sp = s + strlen(s);
2266 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2267 #ifdef INET6
2268 } else if (inc) {
2269 ip6_sprintf(sp, &inc->inc6_faddr);
2270 sp = s + strlen(s);
2271 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2272 sp = s + strlen(s);
2273 ip6_sprintf(sp, &inc->inc6_laddr);
2274 sp = s + strlen(s);
2275 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2276 } else if (ip6 && th) {
2277 ip6_sprintf(sp, &ip6->ip6_src);
2278 sp = s + strlen(s);
2279 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2280 sp = s + strlen(s);
2281 ip6_sprintf(sp, &ip6->ip6_dst);
2282 sp = s + strlen(s);
2283 sprintf(sp, "]:%i", ntohs(th->th_dport));
2284 #endif /* INET6 */
2285 #ifdef INET
2286 } else if (ip && th) {
2287 inet_ntoa_r(ip->ip_src, sp);
2288 sp = s + strlen(s);
2289 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2290 sp = s + strlen(s);
2291 inet_ntoa_r(ip->ip_dst, sp);
2292 sp = s + strlen(s);
2293 sprintf(sp, "]:%i", ntohs(th->th_dport));
2294 #endif /* INET */
2295 } else {
2296 free(s, M_TCPLOG);
2297 return (NULL);
2298 }
2299 sp = s + strlen(s);
2300 if (th)
2301 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2302 if (*(s + size - 1) != '\0')
2303 panic("%s: string too long", __func__);
2304 return (s);
2305 }
Cache object: 589b4e694c00ea63dd964d376dc403e0
|