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: stable/10/sys/netinet/tcp_subr.c 314667 2017-03-04 13:03:31Z avg $");
34
35 #include "opt_compat.h"
36 #include "opt_inet.h"
37 #include "opt_inet6.h"
38 #include "opt_ipsec.h"
39 #include "opt_kdtrace.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/callout.h>
45 #include <sys/hhook.h>
46 #include <sys/kernel.h>
47 #include <sys/khelp.h>
48 #include <sys/sysctl.h>
49 #include <sys/jail.h>
50 #include <sys/malloc.h>
51 #include <sys/mbuf.h>
52 #ifdef INET6
53 #include <sys/domain.h>
54 #endif
55 #include <sys/priv.h>
56 #include <sys/proc.h>
57 #include <sys/sdt.h>
58 #include <sys/socket.h>
59 #include <sys/socketvar.h>
60 #include <sys/protosw.h>
61 #include <sys/random.h>
62
63 #include <vm/uma.h>
64
65 #include <net/route.h>
66 #include <net/if.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 #ifdef TCP_RFC7413
87 #include <netinet/tcp_fastopen.h>
88 #endif
89 #include <netinet/tcp_fsm.h>
90 #include <netinet/tcp_seq.h>
91 #include <netinet/tcp_timer.h>
92 #include <netinet/tcp_var.h>
93 #include <netinet/tcp_syncache.h>
94 #ifdef INET6
95 #include <netinet6/tcp6_var.h>
96 #endif
97 #include <netinet/tcpip.h>
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
123 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
124 #ifdef INET6
125 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
126 #endif
127
128 static int
sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)129 sysctl_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
144 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
145 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
150 static int
sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)151 sysctl_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
166 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
167 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 */
180 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
181 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
182 &VNET_NAME(tcp_minmss), 0,
183 "Minimum TCP Maximum Segment Size");
184
185 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
186 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc1323), 0,
188 "Enable rfc1323 (high performance TCP) extensions");
189
190 static int tcp_log_debug = 0;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
192 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
193
194 static int tcp_tcbhashsize = 0;
195 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
196 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
197
198 static int do_tcpdrain = 1;
199 SYSCTL_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
202 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
203 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
204
205 static VNET_DEFINE(int, icmp_may_rst) = 1;
206 #define V_icmp_may_rst VNET(icmp_may_rst)
207 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
208 &VNET_NAME(icmp_may_rst), 0,
209 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
210
211 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
212 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
213 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
214 &VNET_NAME(tcp_isn_reseed_interval), 0,
215 "Seconds between reseeding of ISN secret");
216
217 static int tcp_soreceive_stream = 0;
218 SYSCTL_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
222 static int tcp_sig_checksigs = 1;
223 SYSCTL_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
227 VNET_DEFINE(uma_zone_t, sack_hole_zone);
228 #define V_sack_hole_zone VNET(sack_hole_zone)
229
230 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
231
232 static struct inpcb *tcp_notify(struct inpcb *, int);
233 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
234 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
235 void *ip4hdr, const void *ip6hdr);
236 static void tcp_timer_discard(struct tcpcb *, uint32_t);
237
238 /*
239 * Target size of TCP PCB hash tables. Must be a power of two.
240 *
241 * Note that this can be overridden by the kernel environment
242 * variable net.inet.tcp.tcbhashsize
243 */
244 #ifndef TCBHASHSIZE
245 #define TCBHASHSIZE 0
246 #endif
247
248 /*
249 * XXX
250 * Callouts should be moved into struct tcp directly. They are currently
251 * separate because the tcpcb structure is exported to userland for sysctl
252 * parsing purposes, which do not know about callouts.
253 */
254 struct tcpcb_mem {
255 struct tcpcb tcb;
256 struct tcp_timer tt;
257 struct cc_var ccv;
258 struct osd osd;
259 };
260
261 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
262 #define V_tcpcb_zone VNET(tcpcb_zone)
263
264 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
265 static struct mtx isn_mtx;
266
267 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
268 #define ISN_LOCK() mtx_lock(&isn_mtx)
269 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
270
271 /*
272 * TCP initialization.
273 */
274 static void
tcp_zone_change(void * tag)275 tcp_zone_change(void *tag)
276 {
277
278 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
279 uma_zone_set_max(V_tcpcb_zone, maxsockets);
280 tcp_tw_zone_change();
281 }
282
283 static int
tcp_inpcb_init(void * mem,int size,int flags)284 tcp_inpcb_init(void *mem, int size, int flags)
285 {
286 struct inpcb *inp = mem;
287
288 INP_LOCK_INIT(inp, "inp", "tcpinp");
289 return (0);
290 }
291
292 /*
293 * Take a value and get the next power of 2 that doesn't overflow.
294 * Used to size the tcp_inpcb hash buckets.
295 */
296 static int
maketcp_hashsize(int size)297 maketcp_hashsize(int size)
298 {
299 int hashsize;
300
301 /*
302 * auto tune.
303 * get the next power of 2 higher than maxsockets.
304 */
305 hashsize = 1 << fls(size);
306 /* catch overflow, and just go one power of 2 smaller */
307 if (hashsize < size) {
308 hashsize = 1 << (fls(size) - 1);
309 }
310 return (hashsize);
311 }
312
313 void
tcp_init(void)314 tcp_init(void)
315 {
316 const char *tcbhash_tuneable;
317 int hashsize;
318
319 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
320
321 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
322 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 printf("%s: WARNING: unable to register helper hook\n", __func__);
324 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
325 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
326 printf("%s: WARNING: unable to register helper hook\n", __func__);
327
328 hashsize = TCBHASHSIZE;
329 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
330 if (hashsize == 0) {
331 /*
332 * Auto tune the hash size based on maxsockets.
333 * A perfect hash would have a 1:1 mapping
334 * (hashsize = maxsockets) however it's been
335 * suggested that O(2) average is better.
336 */
337 hashsize = maketcp_hashsize(maxsockets / 4);
338 /*
339 * Our historical default is 512,
340 * do not autotune lower than this.
341 */
342 if (hashsize < 512)
343 hashsize = 512;
344 if (bootverbose && IS_DEFAULT_VNET(curvnet))
345 printf("%s: %s auto tuned to %d\n", __func__,
346 tcbhash_tuneable, hashsize);
347 }
348 /*
349 * We require a hashsize to be a power of two.
350 * Previously if it was not a power of two we would just reset it
351 * back to 512, which could be a nasty surprise if you did not notice
352 * the error message.
353 * Instead what we do is clip it to the closest power of two lower
354 * than the specified hash value.
355 */
356 if (!powerof2(hashsize)) {
357 int oldhashsize = hashsize;
358
359 hashsize = maketcp_hashsize(hashsize);
360 /* prevent absurdly low value */
361 if (hashsize < 16)
362 hashsize = 16;
363 printf("%s: WARNING: TCB hash size not a power of 2, "
364 "clipped from %d to %d.\n", __func__, oldhashsize,
365 hashsize);
366 }
367 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
368 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
369 IPI_HASHFIELDS_4TUPLE);
370
371 /*
372 * These have to be type stable for the benefit of the timers.
373 */
374 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
375 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
376 uma_zone_set_max(V_tcpcb_zone, maxsockets);
377 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
378
379 tcp_tw_init();
380 syncache_init();
381 tcp_hc_init();
382
383 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
384 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
385 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
386
387 /* Skip initialization of globals for non-default instances. */
388 if (!IS_DEFAULT_VNET(curvnet))
389 return;
390
391 tcp_reass_global_init();
392
393 /* XXX virtualize those bellow? */
394 tcp_delacktime = TCPTV_DELACK;
395 tcp_keepinit = TCPTV_KEEP_INIT;
396 tcp_keepidle = TCPTV_KEEP_IDLE;
397 tcp_keepintvl = TCPTV_KEEPINTVL;
398 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
399 tcp_msl = TCPTV_MSL;
400 tcp_rexmit_min = TCPTV_MIN;
401 if (tcp_rexmit_min < 1)
402 tcp_rexmit_min = 1;
403 tcp_persmin = TCPTV_PERSMIN;
404 tcp_persmax = TCPTV_PERSMAX;
405 tcp_rexmit_slop = TCPTV_CPU_VAR;
406 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
407 tcp_tcbhashsize = hashsize;
408
409 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
410 if (tcp_soreceive_stream) {
411 #ifdef INET
412 tcp_usrreqs.pru_soreceive = soreceive_stream;
413 #endif
414 #ifdef INET6
415 tcp6_usrreqs.pru_soreceive = soreceive_stream;
416 #endif /* INET6 */
417 }
418
419 #ifdef INET6
420 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
421 #else /* INET6 */
422 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
423 #endif /* INET6 */
424 if (max_protohdr < TCP_MINPROTOHDR)
425 max_protohdr = TCP_MINPROTOHDR;
426 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
427 panic("tcp_init");
428 #undef TCP_MINPROTOHDR
429
430 ISN_LOCK_INIT();
431 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
432 SHUTDOWN_PRI_DEFAULT);
433 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
434 EVENTHANDLER_PRI_ANY);
435
436 #ifdef TCP_RFC7413
437 tcp_fastopen_init();
438 #endif
439 }
440
441 #ifdef VIMAGE
442 void
tcp_destroy(void)443 tcp_destroy(void)
444 {
445
446 #ifdef TCP_RFC7413
447 tcp_fastopen_destroy();
448 #endif
449 tcp_hc_destroy();
450 syncache_destroy();
451 tcp_tw_destroy();
452 in_pcbinfo_destroy(&V_tcbinfo);
453 uma_zdestroy(V_sack_hole_zone);
454 uma_zdestroy(V_tcpcb_zone);
455 }
456 #endif
457
458 void
tcp_fini(void * xtp)459 tcp_fini(void *xtp)
460 {
461
462 }
463
464 /*
465 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
466 * tcp_template used to store this data in mbufs, but we now recopy it out
467 * of the tcpcb each time to conserve mbufs.
468 */
469 void
tcpip_fillheaders(struct inpcb * inp,void * ip_ptr,void * tcp_ptr)470 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
471 {
472 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
473
474 INP_WLOCK_ASSERT(inp);
475
476 #ifdef INET6
477 if ((inp->inp_vflag & INP_IPV6) != 0) {
478 struct ip6_hdr *ip6;
479
480 ip6 = (struct ip6_hdr *)ip_ptr;
481 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
482 (inp->inp_flow & IPV6_FLOWINFO_MASK);
483 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
484 (IPV6_VERSION & IPV6_VERSION_MASK);
485 ip6->ip6_nxt = IPPROTO_TCP;
486 ip6->ip6_plen = htons(sizeof(struct tcphdr));
487 ip6->ip6_src = inp->in6p_laddr;
488 ip6->ip6_dst = inp->in6p_faddr;
489 }
490 #endif /* INET6 */
491 #if defined(INET6) && defined(INET)
492 else
493 #endif
494 #ifdef INET
495 {
496 struct ip *ip;
497
498 ip = (struct ip *)ip_ptr;
499 ip->ip_v = IPVERSION;
500 ip->ip_hl = 5;
501 ip->ip_tos = inp->inp_ip_tos;
502 ip->ip_len = 0;
503 ip->ip_id = 0;
504 ip->ip_off = 0;
505 ip->ip_ttl = inp->inp_ip_ttl;
506 ip->ip_sum = 0;
507 ip->ip_p = IPPROTO_TCP;
508 ip->ip_src = inp->inp_laddr;
509 ip->ip_dst = inp->inp_faddr;
510 }
511 #endif /* INET */
512 th->th_sport = inp->inp_lport;
513 th->th_dport = inp->inp_fport;
514 th->th_seq = 0;
515 th->th_ack = 0;
516 th->th_x2 = 0;
517 th->th_off = 5;
518 th->th_flags = 0;
519 th->th_win = 0;
520 th->th_urp = 0;
521 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
522 }
523
524 /*
525 * Create template to be used to send tcp packets on a connection.
526 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
527 * use for this function is in keepalives, which use tcp_respond.
528 */
529 struct tcptemp *
tcpip_maketemplate(struct inpcb * inp)530 tcpip_maketemplate(struct inpcb *inp)
531 {
532 struct tcptemp *t;
533
534 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
535 if (t == NULL)
536 return (NULL);
537 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
538 return (t);
539 }
540
541 /*
542 * Send a single message to the TCP at address specified by
543 * the given TCP/IP header. If m == NULL, then we make a copy
544 * of the tcpiphdr at ti and send directly to the addressed host.
545 * This is used to force keep alive messages out using the TCP
546 * template for a connection. If flags are given then we send
547 * a message back to the TCP which originated the * segment ti,
548 * and discard the mbuf containing it and any other attached mbufs.
549 *
550 * In any case the ack and sequence number of the transmitted
551 * segment are as specified by the parameters.
552 *
553 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
554 */
555 void
tcp_respond(struct tcpcb * tp,void * ipgen,struct tcphdr * th,struct mbuf * m,tcp_seq ack,tcp_seq seq,int flags)556 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
557 tcp_seq ack, tcp_seq seq, int flags)
558 {
559 struct tcpopt to;
560 struct inpcb *inp;
561 struct ip *ip;
562 struct mbuf *optm;
563 struct tcphdr *nth;
564 u_char *optp;
565 #ifdef INET6
566 struct ip6_hdr *ip6;
567 int isipv6;
568 #endif /* INET6 */
569 int optlen, tlen, win;
570 bool incl_opts;
571
572 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
573
574 #ifdef INET6
575 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
576 ip6 = ipgen;
577 #endif /* INET6 */
578 ip = ipgen;
579
580 if (tp != NULL) {
581 inp = tp->t_inpcb;
582 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
583 INP_WLOCK_ASSERT(inp);
584 } else
585 inp = NULL;
586
587 incl_opts = false;
588 win = 0;
589 if (tp != NULL) {
590 if (!(flags & TH_RST)) {
591 win = sbspace(&inp->inp_socket->so_rcv);
592 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
593 win = (long)TCP_MAXWIN << tp->rcv_scale;
594 }
595 if ((tp->t_flags & TF_NOOPT) == 0)
596 incl_opts = true;
597 }
598 if (m == NULL) {
599 m = m_gethdr(M_NOWAIT, MT_DATA);
600 if (m == NULL)
601 return;
602 m->m_data += max_linkhdr;
603 #ifdef INET6
604 if (isipv6) {
605 bcopy((caddr_t)ip6, mtod(m, caddr_t),
606 sizeof(struct ip6_hdr));
607 ip6 = mtod(m, struct ip6_hdr *);
608 nth = (struct tcphdr *)(ip6 + 1);
609 } else
610 #endif /* INET6 */
611 {
612 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
613 ip = mtod(m, struct ip *);
614 nth = (struct tcphdr *)(ip + 1);
615 }
616 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
617 flags = TH_ACK;
618 } else if (!M_WRITABLE(m)) {
619 struct mbuf *n;
620
621 /* Can't reuse 'm', allocate a new mbuf. */
622 n = m_gethdr(M_NOWAIT, MT_DATA);
623 if (n == NULL) {
624 m_freem(m);
625 return;
626 }
627
628 if (!m_dup_pkthdr(n, m, M_NOWAIT)) {
629 m_freem(m);
630 m_freem(n);
631 return;
632 }
633
634 n->m_data += max_linkhdr;
635 /* m_len is set later */
636 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
637 #ifdef INET6
638 if (isipv6) {
639 bcopy((caddr_t)ip6, mtod(n, caddr_t),
640 sizeof(struct ip6_hdr));
641 ip6 = mtod(n, struct ip6_hdr *);
642 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
643 nth = (struct tcphdr *)(ip6 + 1);
644 } else
645 #endif /* INET6 */
646 {
647 bcopy((caddr_t)ip, mtod(n, caddr_t), sizeof(struct ip));
648 ip = mtod(n, struct ip *);
649 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
650 nth = (struct tcphdr *)(ip + 1);
651 }
652 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
653 xchg(nth->th_dport, nth->th_sport, uint16_t);
654 th = nth;
655 m_freem(m);
656 m = n;
657 } else {
658 /*
659 * reuse the mbuf.
660 * XXX MRT We inherit the FIB, which is lucky.
661 */
662 m_freem(m->m_next);
663 m->m_next = NULL;
664 m->m_data = (caddr_t)ipgen;
665 /* m_len is set later */
666 #ifdef INET6
667 if (isipv6) {
668 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
669 nth = (struct tcphdr *)(ip6 + 1);
670 } else
671 #endif /* INET6 */
672 {
673 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
674 nth = (struct tcphdr *)(ip + 1);
675 }
676 if (th != nth) {
677 /*
678 * this is usually a case when an extension header
679 * exists between the IPv6 header and the
680 * TCP header.
681 */
682 nth->th_sport = th->th_sport;
683 nth->th_dport = th->th_dport;
684 }
685 xchg(nth->th_dport, nth->th_sport, uint16_t);
686 #undef xchg
687 }
688 tlen = 0;
689 #ifdef INET6
690 if (isipv6)
691 tlen = sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
692 #endif
693 #if defined(INET) && defined(INET6)
694 else
695 #endif
696 #ifdef INET
697 tlen = sizeof (struct tcpiphdr);
698 #endif
699 #ifdef INVARIANTS
700 m->m_len = 0;
701 KASSERT(M_TRAILINGSPACE(m) >= tlen,
702 ("Not enough trailing space for message (m=%p, need=%d, have=%ld)",
703 m, tlen, (long)M_TRAILINGSPACE(m)));
704 #endif
705 m->m_len = tlen;
706 to.to_flags = 0;
707 if (incl_opts) {
708 /* Make sure we have room. */
709 if (M_TRAILINGSPACE(m) < TCP_MAXOLEN) {
710 m->m_next = m_get(M_NOWAIT, MT_DATA);
711 if (m->m_next) {
712 optp = mtod(m->m_next, u_char *);
713 optm = m->m_next;
714 } else
715 incl_opts = false;
716 } else {
717 optp = (u_char *) (nth + 1);
718 optm = m;
719 }
720 }
721 if (incl_opts) {
722 /* Timestamps. */
723 if (tp->t_flags & TF_RCVD_TSTMP) {
724 to.to_tsval = tcp_ts_getticks() + tp->ts_offset;
725 to.to_tsecr = tp->ts_recent;
726 to.to_flags |= TOF_TS;
727 }
728 #ifdef TCP_SIGNATURE
729 /* TCP-MD5 (RFC2385). */
730 if (tp->t_flags & TF_SIGNATURE)
731 to.to_flags |= TOF_SIGNATURE;
732 #endif
733
734 /* Add the options. */
735 tlen += optlen = tcp_addoptions(&to, optp);
736
737 /* Update m_len in the correct mbuf. */
738 optm->m_len += optlen;
739 } else
740 optlen = 0;
741 #ifdef INET6
742 if (isipv6) {
743 ip6->ip6_flow = 0;
744 ip6->ip6_vfc = IPV6_VERSION;
745 ip6->ip6_nxt = IPPROTO_TCP;
746 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
747 }
748 #endif
749 #if defined(INET) && defined(INET6)
750 else
751 #endif
752 #ifdef INET
753 {
754 ip->ip_len = htons(tlen);
755 ip->ip_ttl = V_ip_defttl;
756 if (V_path_mtu_discovery)
757 ip->ip_off |= htons(IP_DF);
758 }
759 #endif
760 m->m_pkthdr.len = tlen;
761 m->m_pkthdr.rcvif = NULL;
762 #ifdef MAC
763 if (inp != NULL) {
764 /*
765 * Packet is associated with a socket, so allow the
766 * label of the response to reflect the socket label.
767 */
768 INP_WLOCK_ASSERT(inp);
769 mac_inpcb_create_mbuf(inp, m);
770 } else {
771 /*
772 * Packet is not associated with a socket, so possibly
773 * update the label in place.
774 */
775 mac_netinet_tcp_reply(m);
776 }
777 #endif
778 nth->th_seq = htonl(seq);
779 nth->th_ack = htonl(ack);
780 nth->th_x2 = 0;
781 nth->th_off = (sizeof (struct tcphdr) + optlen) >> 2;
782 nth->th_flags = flags;
783 if (tp != NULL)
784 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
785 else
786 nth->th_win = htons((u_short)win);
787 nth->th_urp = 0;
788
789 #ifdef TCP_SIGNATURE
790 if (to.to_flags & TOF_SIGNATURE) {
791 tcp_signature_compute(m, 0, 0, optlen, to.to_signature,
792 IPSEC_DIR_OUTBOUND);
793 }
794 #endif
795
796 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
797 #ifdef INET6
798 if (isipv6) {
799 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
800 nth->th_sum = in6_cksum_pseudo(ip6,
801 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
802 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
803 NULL, NULL);
804 }
805 #endif /* INET6 */
806 #if defined(INET6) && defined(INET)
807 else
808 #endif
809 #ifdef INET
810 {
811 m->m_pkthdr.csum_flags = CSUM_TCP;
812 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
813 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
814 }
815 #endif /* INET */
816 #ifdef TCPDEBUG
817 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
818 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
819 #endif
820 if (flags & TH_RST)
821 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
822 tp, nth);
823
824 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
825 #ifdef INET6
826 if (isipv6)
827 (void) ip6_output(m, NULL, NULL, 0, NULL, NULL, inp);
828 #endif /* INET6 */
829 #if defined(INET) && defined(INET6)
830 else
831 #endif
832 #ifdef INET
833 (void) ip_output(m, NULL, NULL, 0, NULL, inp);
834 #endif
835 }
836
837 /*
838 * Create a new TCP control block, making an
839 * empty reassembly queue and hooking it to the argument
840 * protocol control block. The `inp' parameter must have
841 * come from the zone allocator set up in tcp_init().
842 */
843 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp)844 tcp_newtcpcb(struct inpcb *inp)
845 {
846 struct tcpcb_mem *tm;
847 struct tcpcb *tp;
848 #ifdef INET6
849 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
850 #endif /* INET6 */
851
852 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
853 if (tm == NULL)
854 return (NULL);
855 tp = &tm->tcb;
856
857 /* Initialise cc_var struct for this tcpcb. */
858 tp->ccv = &tm->ccv;
859 tp->ccv->type = IPPROTO_TCP;
860 tp->ccv->ccvc.tcp = tp;
861
862 /*
863 * Use the current system default CC algorithm.
864 */
865 CC_LIST_RLOCK();
866 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
867 CC_ALGO(tp) = CC_DEFAULT();
868 CC_LIST_RUNLOCK();
869
870 if (CC_ALGO(tp)->cb_init != NULL)
871 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
872 uma_zfree(V_tcpcb_zone, tm);
873 return (NULL);
874 }
875
876 tp->osd = &tm->osd;
877 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
878 uma_zfree(V_tcpcb_zone, tm);
879 return (NULL);
880 }
881
882 #ifdef VIMAGE
883 tp->t_vnet = inp->inp_vnet;
884 #endif
885 tp->t_timers = &tm->tt;
886 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
887 tp->t_maxseg = tp->t_maxopd =
888 #ifdef INET6
889 isipv6 ? V_tcp_v6mssdflt :
890 #endif /* INET6 */
891 V_tcp_mssdflt;
892
893 /* Set up our timeouts. */
894 callout_init(&tp->t_timers->tt_rexmt, 1);
895 callout_init(&tp->t_timers->tt_persist, 1);
896 callout_init(&tp->t_timers->tt_keep, 1);
897 callout_init(&tp->t_timers->tt_2msl, 1);
898 callout_init(&tp->t_timers->tt_delack, 1);
899
900 if (V_tcp_do_rfc1323)
901 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
902 if (V_tcp_do_sack)
903 tp->t_flags |= TF_SACK_PERMIT;
904 TAILQ_INIT(&tp->snd_holes);
905 /*
906 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
907 * is called.
908 */
909 in_pcbref(inp); /* Reference for tcpcb */
910 tp->t_inpcb = inp;
911
912 /*
913 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
914 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
915 * reasonable initial retransmit time.
916 */
917 tp->t_srtt = TCPTV_SRTTBASE;
918 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
919 tp->t_rttmin = tcp_rexmit_min;
920 tp->t_rxtcur = TCPTV_RTOBASE;
921 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
922 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
923 tp->t_rcvtime = ticks;
924 /*
925 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
926 * because the socket may be bound to an IPv6 wildcard address,
927 * which may match an IPv4-mapped IPv6 address.
928 */
929 inp->inp_ip_ttl = V_ip_defttl;
930 inp->inp_ppcb = tp;
931 return (tp); /* XXX */
932 }
933
934 /*
935 * Switch the congestion control algorithm back to NewReno for any active
936 * control blocks using an algorithm which is about to go away.
937 * This ensures the CC framework can allow the unload to proceed without leaving
938 * any dangling pointers which would trigger a panic.
939 * Returning non-zero would inform the CC framework that something went wrong
940 * and it would be unsafe to allow the unload to proceed. However, there is no
941 * way for this to occur with this implementation so we always return zero.
942 */
943 int
tcp_ccalgounload(struct cc_algo * unload_algo)944 tcp_ccalgounload(struct cc_algo *unload_algo)
945 {
946 struct cc_algo *tmpalgo;
947 struct inpcb *inp;
948 struct tcpcb *tp;
949 VNET_ITERATOR_DECL(vnet_iter);
950
951 /*
952 * Check all active control blocks across all network stacks and change
953 * any that are using "unload_algo" back to NewReno. If "unload_algo"
954 * requires cleanup code to be run, call it.
955 */
956 VNET_LIST_RLOCK();
957 VNET_FOREACH(vnet_iter) {
958 CURVNET_SET(vnet_iter);
959 INP_INFO_WLOCK(&V_tcbinfo);
960 /*
961 * New connections already part way through being initialised
962 * with the CC algo we're removing will not race with this code
963 * because the INP_INFO_WLOCK is held during initialisation. We
964 * therefore don't enter the loop below until the connection
965 * list has stabilised.
966 */
967 LIST_FOREACH(inp, &V_tcb, inp_list) {
968 INP_WLOCK(inp);
969 /* Important to skip tcptw structs. */
970 if (!(inp->inp_flags & INP_TIMEWAIT) &&
971 (tp = intotcpcb(inp)) != NULL) {
972 /*
973 * By holding INP_WLOCK here, we are assured
974 * that the connection is not currently
975 * executing inside the CC module's functions
976 * i.e. it is safe to make the switch back to
977 * NewReno.
978 */
979 if (CC_ALGO(tp) == unload_algo) {
980 tmpalgo = CC_ALGO(tp);
981 /* NewReno does not require any init. */
982 CC_ALGO(tp) = &newreno_cc_algo;
983 if (tmpalgo->cb_destroy != NULL)
984 tmpalgo->cb_destroy(tp->ccv);
985 }
986 }
987 INP_WUNLOCK(inp);
988 }
989 INP_INFO_WUNLOCK(&V_tcbinfo);
990 CURVNET_RESTORE();
991 }
992 VNET_LIST_RUNLOCK();
993
994 return (0);
995 }
996
997 /*
998 * Drop a TCP connection, reporting
999 * the specified error. If connection is synchronized,
1000 * then send a RST to peer.
1001 */
1002 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)1003 tcp_drop(struct tcpcb *tp, int errno)
1004 {
1005 struct socket *so = tp->t_inpcb->inp_socket;
1006
1007 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1008 INP_WLOCK_ASSERT(tp->t_inpcb);
1009
1010 if (TCPS_HAVERCVDSYN(tp->t_state)) {
1011 tcp_state_change(tp, TCPS_CLOSED);
1012 (void) tcp_output(tp);
1013 TCPSTAT_INC(tcps_drops);
1014 } else
1015 TCPSTAT_INC(tcps_conndrops);
1016 if (errno == ETIMEDOUT && tp->t_softerror)
1017 errno = tp->t_softerror;
1018 so->so_error = errno;
1019 return (tcp_close(tp));
1020 }
1021
1022 void
tcp_discardcb(struct tcpcb * tp)1023 tcp_discardcb(struct tcpcb *tp)
1024 {
1025 struct inpcb *inp = tp->t_inpcb;
1026 struct socket *so = inp->inp_socket;
1027 #ifdef INET6
1028 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
1029 #endif /* INET6 */
1030 int released;
1031
1032 INP_WLOCK_ASSERT(inp);
1033
1034 /*
1035 * Make sure that all of our timers are stopped before we delete the
1036 * PCB.
1037 *
1038 * If stopping a timer fails, we schedule a discard function in same
1039 * callout, and the last discard function called will take care of
1040 * deleting the tcpcb.
1041 */
1042 tcp_timer_stop(tp, TT_REXMT);
1043 tcp_timer_stop(tp, TT_PERSIST);
1044 tcp_timer_stop(tp, TT_KEEP);
1045 tcp_timer_stop(tp, TT_2MSL);
1046 tcp_timer_stop(tp, TT_DELACK);
1047
1048 /*
1049 * If we got enough samples through the srtt filter,
1050 * save the rtt and rttvar in the routing entry.
1051 * 'Enough' is arbitrarily defined as 4 rtt samples.
1052 * 4 samples is enough for the srtt filter to converge
1053 * to within enough % of the correct value; fewer samples
1054 * and we could save a bogus rtt. The danger is not high
1055 * as tcp quickly recovers from everything.
1056 * XXX: Works very well but needs some more statistics!
1057 */
1058 if (tp->t_rttupdated >= 4) {
1059 struct hc_metrics_lite metrics;
1060 u_long ssthresh;
1061
1062 bzero(&metrics, sizeof(metrics));
1063 /*
1064 * Update the ssthresh always when the conditions below
1065 * are satisfied. This gives us better new start value
1066 * for the congestion avoidance for new connections.
1067 * ssthresh is only set if packet loss occured on a session.
1068 *
1069 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
1070 * being torn down. Ideally this code would not use 'so'.
1071 */
1072 ssthresh = tp->snd_ssthresh;
1073 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
1074 /*
1075 * convert the limit from user data bytes to
1076 * packets then to packet data bytes.
1077 */
1078 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
1079 if (ssthresh < 2)
1080 ssthresh = 2;
1081 ssthresh *= (u_long)(tp->t_maxseg +
1082 #ifdef INET6
1083 (isipv6 ? sizeof (struct ip6_hdr) +
1084 sizeof (struct tcphdr) :
1085 #endif
1086 sizeof (struct tcpiphdr)
1087 #ifdef INET6
1088 )
1089 #endif
1090 );
1091 } else
1092 ssthresh = 0;
1093 metrics.rmx_ssthresh = ssthresh;
1094
1095 metrics.rmx_rtt = tp->t_srtt;
1096 metrics.rmx_rttvar = tp->t_rttvar;
1097 metrics.rmx_cwnd = tp->snd_cwnd;
1098 metrics.rmx_sendpipe = 0;
1099 metrics.rmx_recvpipe = 0;
1100
1101 tcp_hc_update(&inp->inp_inc, &metrics);
1102 }
1103
1104 /* free the reassembly queue, if any */
1105 tcp_reass_flush(tp);
1106
1107 #ifdef TCP_OFFLOAD
1108 /* Disconnect offload device, if any. */
1109 if (tp->t_flags & TF_TOE)
1110 tcp_offload_detach(tp);
1111 #endif
1112
1113 tcp_free_sackholes(tp);
1114
1115 /* Allow the CC algorithm to clean up after itself. */
1116 if (CC_ALGO(tp)->cb_destroy != NULL)
1117 CC_ALGO(tp)->cb_destroy(tp->ccv);
1118
1119 khelp_destroy_osd(tp->osd);
1120
1121 CC_ALGO(tp) = NULL;
1122 inp->inp_ppcb = NULL;
1123 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1124 /* We own the last reference on tcpcb, let's free it. */
1125 tp->t_inpcb = NULL;
1126 uma_zfree(V_tcpcb_zone, tp);
1127 released = in_pcbrele_wlocked(inp);
1128 KASSERT(!released, ("%s: inp %p should not have been released "
1129 "here", __func__, inp));
1130 }
1131 }
1132
1133 void
tcp_timer_2msl_discard(void * xtp)1134 tcp_timer_2msl_discard(void *xtp)
1135 {
1136
1137 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1138 }
1139
1140 void
tcp_timer_keep_discard(void * xtp)1141 tcp_timer_keep_discard(void *xtp)
1142 {
1143
1144 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1145 }
1146
1147 void
tcp_timer_persist_discard(void * xtp)1148 tcp_timer_persist_discard(void *xtp)
1149 {
1150
1151 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1152 }
1153
1154 void
tcp_timer_rexmt_discard(void * xtp)1155 tcp_timer_rexmt_discard(void *xtp)
1156 {
1157
1158 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1159 }
1160
1161 void
tcp_timer_delack_discard(void * xtp)1162 tcp_timer_delack_discard(void *xtp)
1163 {
1164
1165 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1166 }
1167
1168 void
tcp_timer_discard(struct tcpcb * tp,uint32_t timer_type)1169 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1170 {
1171 struct inpcb *inp;
1172
1173 CURVNET_SET(tp->t_vnet);
1174 INP_INFO_RLOCK(&V_tcbinfo);
1175 inp = tp->t_inpcb;
1176 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1177 __func__, tp));
1178 INP_WLOCK(inp);
1179 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1180 ("%s: tcpcb has to be stopped here", __func__));
1181 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1182 ("%s: discard callout should be running", __func__));
1183 tp->t_timers->tt_flags &= ~timer_type;
1184 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1185 /* We own the last reference on this tcpcb, let's free it. */
1186 tp->t_inpcb = NULL;
1187 uma_zfree(V_tcpcb_zone, tp);
1188 if (in_pcbrele_wlocked(inp)) {
1189 INP_INFO_RUNLOCK(&V_tcbinfo);
1190 CURVNET_RESTORE();
1191 return;
1192 }
1193 }
1194 INP_WUNLOCK(inp);
1195 INP_INFO_RUNLOCK(&V_tcbinfo);
1196 CURVNET_RESTORE();
1197 }
1198
1199 /*
1200 * Attempt to close a TCP control block, marking it as dropped, and freeing
1201 * the socket if we hold the only reference.
1202 */
1203 struct tcpcb *
tcp_close(struct tcpcb * tp)1204 tcp_close(struct tcpcb *tp)
1205 {
1206 struct inpcb *inp = tp->t_inpcb;
1207 struct socket *so;
1208
1209 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1210 INP_WLOCK_ASSERT(inp);
1211
1212 #ifdef TCP_OFFLOAD
1213 if (tp->t_state == TCPS_LISTEN)
1214 tcp_offload_listen_stop(tp);
1215 #endif
1216 #ifdef TCP_RFC7413
1217 /*
1218 * This releases the TFO pending counter resource for TFO listen
1219 * sockets as well as passively-created TFO sockets that transition
1220 * from SYN_RECEIVED to CLOSED.
1221 */
1222 if (tp->t_tfo_pending) {
1223 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
1224 tp->t_tfo_pending = NULL;
1225 }
1226 #endif
1227 in_pcbdrop(inp);
1228 TCPSTAT_INC(tcps_closed);
1229 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1230 so = inp->inp_socket;
1231 soisdisconnected(so);
1232 if (inp->inp_flags & INP_SOCKREF) {
1233 KASSERT(so->so_state & SS_PROTOREF,
1234 ("tcp_close: !SS_PROTOREF"));
1235 inp->inp_flags &= ~INP_SOCKREF;
1236 INP_WUNLOCK(inp);
1237 ACCEPT_LOCK();
1238 SOCK_LOCK(so);
1239 so->so_state &= ~SS_PROTOREF;
1240 sofree(so);
1241 return (NULL);
1242 }
1243 return (tp);
1244 }
1245
1246 void
tcp_drain(void)1247 tcp_drain(void)
1248 {
1249 VNET_ITERATOR_DECL(vnet_iter);
1250
1251 if (!do_tcpdrain)
1252 return;
1253
1254 VNET_LIST_RLOCK_NOSLEEP();
1255 VNET_FOREACH(vnet_iter) {
1256 CURVNET_SET(vnet_iter);
1257 struct inpcb *inpb;
1258 struct tcpcb *tcpb;
1259
1260 /*
1261 * Walk the tcpbs, if existing, and flush the reassembly queue,
1262 * if there is one...
1263 * XXX: The "Net/3" implementation doesn't imply that the TCP
1264 * reassembly queue should be flushed, but in a situation
1265 * where we're really low on mbufs, this is potentially
1266 * useful.
1267 */
1268 INP_INFO_WLOCK(&V_tcbinfo);
1269 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1270 if (inpb->inp_flags & INP_TIMEWAIT)
1271 continue;
1272 INP_WLOCK(inpb);
1273 if ((tcpb = intotcpcb(inpb)) != NULL) {
1274 tcp_reass_flush(tcpb);
1275 tcp_clean_sackreport(tcpb);
1276 }
1277 INP_WUNLOCK(inpb);
1278 }
1279 INP_INFO_WUNLOCK(&V_tcbinfo);
1280 CURVNET_RESTORE();
1281 }
1282 VNET_LIST_RUNLOCK_NOSLEEP();
1283 }
1284
1285 /*
1286 * Notify a tcp user of an asynchronous error;
1287 * store error as soft error, but wake up user
1288 * (for now, won't do anything until can select for soft error).
1289 *
1290 * Do not wake up user since there currently is no mechanism for
1291 * reporting soft errors (yet - a kqueue filter may be added).
1292 */
1293 static struct inpcb *
tcp_notify(struct inpcb * inp,int error)1294 tcp_notify(struct inpcb *inp, int error)
1295 {
1296 struct tcpcb *tp;
1297
1298 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
1299 INP_WLOCK_ASSERT(inp);
1300
1301 if ((inp->inp_flags & INP_TIMEWAIT) ||
1302 (inp->inp_flags & INP_DROPPED))
1303 return (inp);
1304
1305 tp = intotcpcb(inp);
1306 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1307
1308 /*
1309 * Ignore some errors if we are hooked up.
1310 * If connection hasn't completed, has retransmitted several times,
1311 * and receives a second error, give up now. This is better
1312 * than waiting a long time to establish a connection that
1313 * can never complete.
1314 */
1315 if (tp->t_state == TCPS_ESTABLISHED &&
1316 (error == EHOSTUNREACH || error == ENETUNREACH ||
1317 error == EHOSTDOWN)) {
1318 return (inp);
1319 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1320 tp->t_softerror) {
1321 tp = tcp_drop(tp, error);
1322 if (tp != NULL)
1323 return (inp);
1324 else
1325 return (NULL);
1326 } else {
1327 tp->t_softerror = error;
1328 return (inp);
1329 }
1330 #if 0
1331 wakeup( &so->so_timeo);
1332 sorwakeup(so);
1333 sowwakeup(so);
1334 #endif
1335 }
1336
1337 static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)1338 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1339 {
1340 int error, i, m, n, pcb_count;
1341 struct inpcb *inp, **inp_list;
1342 inp_gen_t gencnt;
1343 struct xinpgen xig;
1344
1345 /*
1346 * The process of preparing the TCB list is too time-consuming and
1347 * resource-intensive to repeat twice on every request.
1348 */
1349 if (req->oldptr == NULL) {
1350 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1351 n += imax(n / 8, 10);
1352 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1353 return (0);
1354 }
1355
1356 if (req->newptr != NULL)
1357 return (EPERM);
1358
1359 /*
1360 * OK, now we're committed to doing something.
1361 */
1362 INP_LIST_RLOCK(&V_tcbinfo);
1363 gencnt = V_tcbinfo.ipi_gencnt;
1364 n = V_tcbinfo.ipi_count;
1365 INP_LIST_RUNLOCK(&V_tcbinfo);
1366
1367 m = syncache_pcbcount();
1368
1369 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1370 + (n + m) * sizeof(struct xtcpcb));
1371 if (error != 0)
1372 return (error);
1373
1374 xig.xig_len = sizeof xig;
1375 xig.xig_count = n + m;
1376 xig.xig_gen = gencnt;
1377 xig.xig_sogen = so_gencnt;
1378 error = SYSCTL_OUT(req, &xig, sizeof xig);
1379 if (error)
1380 return (error);
1381
1382 error = syncache_pcblist(req, m, &pcb_count);
1383 if (error)
1384 return (error);
1385
1386 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1387
1388 INP_INFO_WLOCK(&V_tcbinfo);
1389 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1390 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1391 INP_WLOCK(inp);
1392 if (inp->inp_gencnt <= gencnt) {
1393 /*
1394 * XXX: This use of cr_cansee(), introduced with
1395 * TCP state changes, is not quite right, but for
1396 * now, better than nothing.
1397 */
1398 if (inp->inp_flags & INP_TIMEWAIT) {
1399 if (intotw(inp) != NULL)
1400 error = cr_cansee(req->td->td_ucred,
1401 intotw(inp)->tw_cred);
1402 else
1403 error = EINVAL; /* Skip this inp. */
1404 } else
1405 error = cr_canseeinpcb(req->td->td_ucred, inp);
1406 if (error == 0) {
1407 in_pcbref(inp);
1408 inp_list[i++] = inp;
1409 }
1410 }
1411 INP_WUNLOCK(inp);
1412 }
1413 INP_INFO_WUNLOCK(&V_tcbinfo);
1414 n = i;
1415
1416 error = 0;
1417 for (i = 0; i < n; i++) {
1418 inp = inp_list[i];
1419 INP_RLOCK(inp);
1420 if (inp->inp_gencnt <= gencnt) {
1421 struct xtcpcb xt;
1422 void *inp_ppcb;
1423
1424 bzero(&xt, sizeof(xt));
1425 xt.xt_len = sizeof xt;
1426 /* XXX should avoid extra copy */
1427 bcopy(inp, &xt.xt_inp, sizeof *inp);
1428 inp_ppcb = inp->inp_ppcb;
1429 if (inp_ppcb == NULL)
1430 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1431 else if (inp->inp_flags & INP_TIMEWAIT) {
1432 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1433 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1434 } else {
1435 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1436 if (xt.xt_tp.t_timers)
1437 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1438 }
1439 if (inp->inp_socket != NULL)
1440 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1441 else {
1442 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1443 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1444 }
1445 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1446 INP_RUNLOCK(inp);
1447 error = SYSCTL_OUT(req, &xt, sizeof xt);
1448 } else
1449 INP_RUNLOCK(inp);
1450 }
1451 INP_INFO_RLOCK(&V_tcbinfo);
1452 for (i = 0; i < n; i++) {
1453 inp = inp_list[i];
1454 INP_RLOCK(inp);
1455 if (!in_pcbrele_rlocked(inp))
1456 INP_RUNLOCK(inp);
1457 }
1458 INP_INFO_RUNLOCK(&V_tcbinfo);
1459
1460 if (!error) {
1461 /*
1462 * Give the user an updated idea of our state.
1463 * If the generation differs from what we told
1464 * her before, she knows that something happened
1465 * while we were processing this request, and it
1466 * might be necessary to retry.
1467 */
1468 INP_LIST_RLOCK(&V_tcbinfo);
1469 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1470 xig.xig_sogen = so_gencnt;
1471 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1472 INP_LIST_RUNLOCK(&V_tcbinfo);
1473 error = SYSCTL_OUT(req, &xig, sizeof xig);
1474 }
1475 free(inp_list, M_TEMP);
1476 return (error);
1477 }
1478
1479 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1480 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1481 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1482
1483 #ifdef INET
1484 static int
tcp_getcred(SYSCTL_HANDLER_ARGS)1485 tcp_getcred(SYSCTL_HANDLER_ARGS)
1486 {
1487 struct xucred xuc;
1488 struct sockaddr_in addrs[2];
1489 struct inpcb *inp;
1490 int error;
1491
1492 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1493 if (error)
1494 return (error);
1495 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1496 if (error)
1497 return (error);
1498 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1499 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1500 if (inp != NULL) {
1501 if (inp->inp_socket == NULL)
1502 error = ENOENT;
1503 if (error == 0)
1504 error = cr_canseeinpcb(req->td->td_ucred, inp);
1505 if (error == 0)
1506 cru2x(inp->inp_cred, &xuc);
1507 INP_RUNLOCK(inp);
1508 } else
1509 error = ENOENT;
1510 if (error == 0)
1511 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1512 return (error);
1513 }
1514
1515 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1516 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1517 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1518 #endif /* INET */
1519
1520 #ifdef INET6
1521 static int
tcp6_getcred(SYSCTL_HANDLER_ARGS)1522 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1523 {
1524 struct xucred xuc;
1525 struct sockaddr_in6 addrs[2];
1526 struct inpcb *inp;
1527 int error;
1528 #ifdef INET
1529 int mapped = 0;
1530 #endif
1531
1532 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1533 if (error)
1534 return (error);
1535 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1536 if (error)
1537 return (error);
1538 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1539 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1540 return (error);
1541 }
1542 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1543 #ifdef INET
1544 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1545 mapped = 1;
1546 else
1547 #endif
1548 return (EINVAL);
1549 }
1550
1551 #ifdef INET
1552 if (mapped == 1)
1553 inp = in_pcblookup(&V_tcbinfo,
1554 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1555 addrs[1].sin6_port,
1556 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1557 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1558 else
1559 #endif
1560 inp = in6_pcblookup(&V_tcbinfo,
1561 &addrs[1].sin6_addr, addrs[1].sin6_port,
1562 &addrs[0].sin6_addr, addrs[0].sin6_port,
1563 INPLOOKUP_RLOCKPCB, NULL);
1564 if (inp != NULL) {
1565 if (inp->inp_socket == NULL)
1566 error = ENOENT;
1567 if (error == 0)
1568 error = cr_canseeinpcb(req->td->td_ucred, inp);
1569 if (error == 0)
1570 cru2x(inp->inp_cred, &xuc);
1571 INP_RUNLOCK(inp);
1572 } else
1573 error = ENOENT;
1574 if (error == 0)
1575 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1576 return (error);
1577 }
1578
1579 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1580 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1581 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1582 #endif /* INET6 */
1583
1584
1585 #ifdef INET
1586 void
tcp_ctlinput(int cmd,struct sockaddr * sa,void * vip)1587 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1588 {
1589 struct ip *ip = vip;
1590 struct tcphdr *th;
1591 struct in_addr faddr;
1592 struct inpcb *inp;
1593 struct tcpcb *tp;
1594 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1595 struct icmp *icp;
1596 struct in_conninfo inc;
1597 tcp_seq icmp_tcp_seq;
1598 int mtu;
1599
1600 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1601 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1602 return;
1603
1604 if (cmd == PRC_MSGSIZE)
1605 notify = tcp_mtudisc_notify;
1606 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1607 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1608 notify = tcp_drop_syn_sent;
1609 /*
1610 * Redirects don't need to be handled up here.
1611 */
1612 else if (PRC_IS_REDIRECT(cmd))
1613 return;
1614 /*
1615 * Source quench is depreciated.
1616 */
1617 else if (cmd == PRC_QUENCH)
1618 return;
1619 /*
1620 * Hostdead is ugly because it goes linearly through all PCBs.
1621 * XXX: We never get this from ICMP, otherwise it makes an
1622 * excellent DoS attack on machines with many connections.
1623 */
1624 else if (cmd == PRC_HOSTDEAD)
1625 ip = NULL;
1626 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1627 return;
1628 if (ip != NULL) {
1629 icp = (struct icmp *)((caddr_t)ip
1630 - offsetof(struct icmp, icmp_ip));
1631 th = (struct tcphdr *)((caddr_t)ip
1632 + (ip->ip_hl << 2));
1633 INP_INFO_RLOCK(&V_tcbinfo);
1634 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1635 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1636 if (inp != NULL) {
1637 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1638 !(inp->inp_flags & INP_DROPPED) &&
1639 !(inp->inp_socket == NULL)) {
1640 icmp_tcp_seq = htonl(th->th_seq);
1641 tp = intotcpcb(inp);
1642 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1643 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1644 if (cmd == PRC_MSGSIZE) {
1645 /*
1646 * MTU discovery:
1647 * If we got a needfrag set the MTU
1648 * in the route to the suggested new
1649 * value (if given) and then notify.
1650 */
1651 bzero(&inc, sizeof(inc));
1652 inc.inc_faddr = faddr;
1653 inc.inc_fibnum =
1654 inp->inp_inc.inc_fibnum;
1655
1656 mtu = ntohs(icp->icmp_nextmtu);
1657 /*
1658 * If no alternative MTU was
1659 * proposed, try the next smaller
1660 * one.
1661 */
1662 if (!mtu)
1663 mtu = ip_next_mtu(
1664 ntohs(ip->ip_len), 1);
1665 if (mtu < V_tcp_minmss
1666 + sizeof(struct tcpiphdr))
1667 mtu = V_tcp_minmss
1668 + sizeof(struct tcpiphdr);
1669 /*
1670 * Only cache the MTU if it
1671 * is smaller than the interface
1672 * or route MTU. tcp_mtudisc()
1673 * will do right thing by itself.
1674 */
1675 if (mtu <= tcp_maxmtu(&inc, NULL))
1676 tcp_hc_updatemtu(&inc, mtu);
1677 tcp_mtudisc(inp, mtu);
1678 } else
1679 inp = (*notify)(inp,
1680 inetctlerrmap[cmd]);
1681 }
1682 }
1683 if (inp != NULL)
1684 INP_WUNLOCK(inp);
1685 } else {
1686 bzero(&inc, sizeof(inc));
1687 inc.inc_fport = th->th_dport;
1688 inc.inc_lport = th->th_sport;
1689 inc.inc_faddr = faddr;
1690 inc.inc_laddr = ip->ip_src;
1691 syncache_unreach(&inc, th);
1692 }
1693 INP_INFO_RUNLOCK(&V_tcbinfo);
1694 } else
1695 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1696 }
1697 #endif /* INET */
1698
1699 #ifdef INET6
1700 void
tcp6_ctlinput(int cmd,struct sockaddr * sa,void * d)1701 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1702 {
1703 struct tcphdr th;
1704 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1705 struct ip6_hdr *ip6;
1706 struct mbuf *m;
1707 struct ip6ctlparam *ip6cp = NULL;
1708 const struct sockaddr_in6 *sa6_src = NULL;
1709 int off;
1710 struct tcp_portonly {
1711 u_int16_t th_sport;
1712 u_int16_t th_dport;
1713 } *thp;
1714
1715 if (sa->sa_family != AF_INET6 ||
1716 sa->sa_len != sizeof(struct sockaddr_in6))
1717 return;
1718
1719 if (cmd == PRC_MSGSIZE)
1720 notify = tcp_mtudisc_notify;
1721 else if (!PRC_IS_REDIRECT(cmd) &&
1722 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1723 return;
1724 /* Source quench is depreciated. */
1725 else if (cmd == PRC_QUENCH)
1726 return;
1727
1728 /* if the parameter is from icmp6, decode it. */
1729 if (d != NULL) {
1730 ip6cp = (struct ip6ctlparam *)d;
1731 m = ip6cp->ip6c_m;
1732 ip6 = ip6cp->ip6c_ip6;
1733 off = ip6cp->ip6c_off;
1734 sa6_src = ip6cp->ip6c_src;
1735 } else {
1736 m = NULL;
1737 ip6 = NULL;
1738 off = 0; /* fool gcc */
1739 sa6_src = &sa6_any;
1740 }
1741
1742 if (ip6 != NULL) {
1743 struct in_conninfo inc;
1744 /*
1745 * XXX: We assume that when IPV6 is non NULL,
1746 * M and OFF are valid.
1747 */
1748
1749 /* check if we can safely examine src and dst ports */
1750 if (m->m_pkthdr.len < off + sizeof(*thp))
1751 return;
1752
1753 bzero(&th, sizeof(th));
1754 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1755
1756 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1757 (struct sockaddr *)ip6cp->ip6c_src,
1758 th.th_sport, cmd, NULL, notify);
1759
1760 bzero(&inc, sizeof(inc));
1761 inc.inc_fport = th.th_dport;
1762 inc.inc_lport = th.th_sport;
1763 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1764 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1765 inc.inc_flags |= INC_ISIPV6;
1766 INP_INFO_RLOCK(&V_tcbinfo);
1767 syncache_unreach(&inc, &th);
1768 INP_INFO_RUNLOCK(&V_tcbinfo);
1769 } else
1770 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1771 0, cmd, NULL, notify);
1772 }
1773 #endif /* INET6 */
1774
1775
1776 /*
1777 * Following is where TCP initial sequence number generation occurs.
1778 *
1779 * There are two places where we must use initial sequence numbers:
1780 * 1. In SYN-ACK packets.
1781 * 2. In SYN packets.
1782 *
1783 * All ISNs for SYN-ACK packets are generated by the syncache. See
1784 * tcp_syncache.c for details.
1785 *
1786 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1787 * depends on this property. In addition, these ISNs should be
1788 * unguessable so as to prevent connection hijacking. To satisfy
1789 * the requirements of this situation, the algorithm outlined in
1790 * RFC 1948 is used, with only small modifications.
1791 *
1792 * Implementation details:
1793 *
1794 * Time is based off the system timer, and is corrected so that it
1795 * increases by one megabyte per second. This allows for proper
1796 * recycling on high speed LANs while still leaving over an hour
1797 * before rollover.
1798 *
1799 * As reading the *exact* system time is too expensive to be done
1800 * whenever setting up a TCP connection, we increment the time
1801 * offset in two ways. First, a small random positive increment
1802 * is added to isn_offset for each connection that is set up.
1803 * Second, the function tcp_isn_tick fires once per clock tick
1804 * and increments isn_offset as necessary so that sequence numbers
1805 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1806 * random positive increments serve only to ensure that the same
1807 * exact sequence number is never sent out twice (as could otherwise
1808 * happen when a port is recycled in less than the system tick
1809 * interval.)
1810 *
1811 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1812 * between seeding of isn_secret. This is normally set to zero,
1813 * as reseeding should not be necessary.
1814 *
1815 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1816 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1817 * general, this means holding an exclusive (write) lock.
1818 */
1819
1820 #define ISN_BYTES_PER_SECOND 1048576
1821 #define ISN_STATIC_INCREMENT 4096
1822 #define ISN_RANDOM_INCREMENT (4096 - 1)
1823
1824 static VNET_DEFINE(u_char, isn_secret[32]);
1825 static VNET_DEFINE(int, isn_last);
1826 static VNET_DEFINE(int, isn_last_reseed);
1827 static VNET_DEFINE(u_int32_t, isn_offset);
1828 static VNET_DEFINE(u_int32_t, isn_offset_old);
1829
1830 #define V_isn_secret VNET(isn_secret)
1831 #define V_isn_last VNET(isn_last)
1832 #define V_isn_last_reseed VNET(isn_last_reseed)
1833 #define V_isn_offset VNET(isn_offset)
1834 #define V_isn_offset_old VNET(isn_offset_old)
1835
1836 tcp_seq
tcp_new_isn(struct tcpcb * tp)1837 tcp_new_isn(struct tcpcb *tp)
1838 {
1839 MD5_CTX isn_ctx;
1840 u_int32_t md5_buffer[4];
1841 tcp_seq new_isn;
1842 u_int32_t projected_offset;
1843
1844 INP_WLOCK_ASSERT(tp->t_inpcb);
1845
1846 ISN_LOCK();
1847 /* Seed if this is the first use, reseed if requested. */
1848 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1849 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1850 < (u_int)ticks))) {
1851 read_random(&V_isn_secret, sizeof(V_isn_secret));
1852 V_isn_last_reseed = ticks;
1853 }
1854
1855 /* Compute the md5 hash and return the ISN. */
1856 MD5Init(&isn_ctx);
1857 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1858 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1859 #ifdef INET6
1860 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1861 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1862 sizeof(struct in6_addr));
1863 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1864 sizeof(struct in6_addr));
1865 } else
1866 #endif
1867 {
1868 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1869 sizeof(struct in_addr));
1870 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1871 sizeof(struct in_addr));
1872 }
1873 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1874 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1875 new_isn = (tcp_seq) md5_buffer[0];
1876 V_isn_offset += ISN_STATIC_INCREMENT +
1877 (arc4random() & ISN_RANDOM_INCREMENT);
1878 if (ticks != V_isn_last) {
1879 projected_offset = V_isn_offset_old +
1880 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1881 if (SEQ_GT(projected_offset, V_isn_offset))
1882 V_isn_offset = projected_offset;
1883 V_isn_offset_old = V_isn_offset;
1884 V_isn_last = ticks;
1885 }
1886 new_isn += V_isn_offset;
1887 ISN_UNLOCK();
1888 return (new_isn);
1889 }
1890
1891 /*
1892 * When a specific ICMP unreachable message is received and the
1893 * connection state is SYN-SENT, drop the connection. This behavior
1894 * is controlled by the icmp_may_rst sysctl.
1895 */
1896 struct inpcb *
tcp_drop_syn_sent(struct inpcb * inp,int errno)1897 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1898 {
1899 struct tcpcb *tp;
1900
1901 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1902 INP_WLOCK_ASSERT(inp);
1903
1904 if ((inp->inp_flags & INP_TIMEWAIT) ||
1905 (inp->inp_flags & INP_DROPPED))
1906 return (inp);
1907
1908 tp = intotcpcb(inp);
1909 if (tp->t_state != TCPS_SYN_SENT)
1910 return (inp);
1911
1912 tp = tcp_drop(tp, errno);
1913 if (tp != NULL)
1914 return (inp);
1915 else
1916 return (NULL);
1917 }
1918
1919 /*
1920 * When `need fragmentation' ICMP is received, update our idea of the MSS
1921 * based on the new value. Also nudge TCP to send something, since we
1922 * know the packet we just sent was dropped.
1923 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1924 */
1925 static struct inpcb *
tcp_mtudisc_notify(struct inpcb * inp,int error)1926 tcp_mtudisc_notify(struct inpcb *inp, int error)
1927 {
1928
1929 return (tcp_mtudisc(inp, -1));
1930 }
1931
1932 struct inpcb *
tcp_mtudisc(struct inpcb * inp,int mtuoffer)1933 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1934 {
1935 struct tcpcb *tp;
1936 struct socket *so;
1937
1938 INP_WLOCK_ASSERT(inp);
1939 if ((inp->inp_flags & INP_TIMEWAIT) ||
1940 (inp->inp_flags & INP_DROPPED))
1941 return (inp);
1942
1943 tp = intotcpcb(inp);
1944 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1945
1946 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1947
1948 so = inp->inp_socket;
1949 SOCKBUF_LOCK(&so->so_snd);
1950 /* If the mss is larger than the socket buffer, decrease the mss. */
1951 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1952 tp->t_maxseg = so->so_snd.sb_hiwat;
1953 SOCKBUF_UNLOCK(&so->so_snd);
1954
1955 TCPSTAT_INC(tcps_mturesent);
1956 tp->t_rtttime = 0;
1957 tp->snd_nxt = tp->snd_una;
1958 tcp_free_sackholes(tp);
1959 tp->snd_recover = tp->snd_max;
1960 if (tp->t_flags & TF_SACK_PERMIT)
1961 EXIT_FASTRECOVERY(tp->t_flags);
1962 tcp_output(tp);
1963 return (inp);
1964 }
1965
1966 #ifdef INET
1967 /*
1968 * Look-up the routing entry to the peer of this inpcb. If no route
1969 * is found and it cannot be allocated, then return 0. This routine
1970 * is called by TCP routines that access the rmx structure and by
1971 * tcp_mss_update to get the peer/interface MTU.
1972 */
1973 u_long
tcp_maxmtu(struct in_conninfo * inc,struct tcp_ifcap * cap)1974 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1975 {
1976 struct route sro;
1977 struct sockaddr_in *dst;
1978 struct ifnet *ifp;
1979 u_long maxmtu = 0;
1980
1981 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1982
1983 bzero(&sro, sizeof(sro));
1984 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1985 dst = (struct sockaddr_in *)&sro.ro_dst;
1986 dst->sin_family = AF_INET;
1987 dst->sin_len = sizeof(*dst);
1988 dst->sin_addr = inc->inc_faddr;
1989 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1990 }
1991 if (sro.ro_rt != NULL) {
1992 ifp = sro.ro_rt->rt_ifp;
1993 if (sro.ro_rt->rt_mtu == 0)
1994 maxmtu = ifp->if_mtu;
1995 else
1996 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1997
1998 /* Report additional interface capabilities. */
1999 if (cap != NULL) {
2000 if (ifp->if_capenable & IFCAP_TSO4 &&
2001 ifp->if_hwassist & CSUM_TSO) {
2002 cap->ifcap |= CSUM_TSO;
2003 cap->tsomax = ifp->if_hw_tsomax;
2004 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2005 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2006 }
2007 }
2008 RTFREE(sro.ro_rt);
2009 }
2010 return (maxmtu);
2011 }
2012 #endif /* INET */
2013
2014 #ifdef INET6
2015 u_long
tcp_maxmtu6(struct in_conninfo * inc,struct tcp_ifcap * cap)2016 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
2017 {
2018 struct route_in6 sro6;
2019 struct ifnet *ifp;
2020 u_long maxmtu = 0;
2021
2022 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
2023
2024 bzero(&sro6, sizeof(sro6));
2025 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
2026 sro6.ro_dst.sin6_family = AF_INET6;
2027 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
2028 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
2029 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
2030 }
2031 if (sro6.ro_rt != NULL) {
2032 ifp = sro6.ro_rt->rt_ifp;
2033 if (sro6.ro_rt->rt_mtu == 0)
2034 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
2035 else
2036 maxmtu = min(sro6.ro_rt->rt_mtu,
2037 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
2038
2039 /* Report additional interface capabilities. */
2040 if (cap != NULL) {
2041 if (ifp->if_capenable & IFCAP_TSO6 &&
2042 ifp->if_hwassist & CSUM_TSO) {
2043 cap->ifcap |= CSUM_TSO;
2044 cap->tsomax = ifp->if_hw_tsomax;
2045 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
2046 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
2047 }
2048 }
2049 RTFREE(sro6.ro_rt);
2050 }
2051
2052 return (maxmtu);
2053 }
2054 #endif /* INET6 */
2055
2056 #ifdef IPSEC
2057 /* compute ESP/AH header size for TCP, including outer IP header. */
2058 size_t
ipsec_hdrsiz_tcp(struct tcpcb * tp)2059 ipsec_hdrsiz_tcp(struct tcpcb *tp)
2060 {
2061 struct inpcb *inp;
2062 struct mbuf *m;
2063 size_t hdrsiz;
2064 struct ip *ip;
2065 #ifdef INET6
2066 struct ip6_hdr *ip6;
2067 #endif
2068 struct tcphdr *th;
2069
2070 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL) ||
2071 (!key_havesp(IPSEC_DIR_OUTBOUND)))
2072 return (0);
2073 m = m_gethdr(M_NOWAIT, MT_DATA);
2074 if (!m)
2075 return (0);
2076
2077 #ifdef INET6
2078 if ((inp->inp_vflag & INP_IPV6) != 0) {
2079 ip6 = mtod(m, struct ip6_hdr *);
2080 th = (struct tcphdr *)(ip6 + 1);
2081 m->m_pkthdr.len = m->m_len =
2082 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
2083 tcpip_fillheaders(inp, ip6, th);
2084 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2085 } else
2086 #endif /* INET6 */
2087 {
2088 ip = mtod(m, struct ip *);
2089 th = (struct tcphdr *)(ip + 1);
2090 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
2091 tcpip_fillheaders(inp, ip, th);
2092 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
2093 }
2094
2095 m_free(m);
2096 return (hdrsiz);
2097 }
2098 #endif /* IPSEC */
2099
2100 #ifdef TCP_SIGNATURE
2101 /*
2102 * Callback function invoked by m_apply() to digest TCP segment data
2103 * contained within an mbuf chain.
2104 */
2105 static int
tcp_signature_apply(void * fstate,void * data,u_int len)2106 tcp_signature_apply(void *fstate, void *data, u_int len)
2107 {
2108
2109 MD5Update(fstate, (u_char *)data, len);
2110 return (0);
2111 }
2112
2113 /*
2114 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
2115 *
2116 * Parameters:
2117 * m pointer to head of mbuf chain
2118 * _unused
2119 * len length of TCP segment data, excluding options
2120 * optlen length of TCP segment options
2121 * buf pointer to storage for computed MD5 digest
2122 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2123 *
2124 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2125 * When called from tcp_input(), we can be sure that th_sum has been
2126 * zeroed out and verified already.
2127 *
2128 * Return 0 if successful, otherwise return -1.
2129 *
2130 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2131 * search with the destination IP address, and a 'magic SPI' to be
2132 * determined by the application. This is hardcoded elsewhere to 1179
2133 * right now. Another branch of this code exists which uses the SPD to
2134 * specify per-application flows but it is unstable.
2135 */
2136 int
tcp_signature_compute(struct mbuf * m,int _unused,int len,int optlen,u_char * buf,u_int direction)2137 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2138 u_char *buf, u_int direction)
2139 {
2140 union sockaddr_union dst;
2141 #ifdef INET
2142 struct ippseudo ippseudo;
2143 #endif
2144 MD5_CTX ctx;
2145 int doff;
2146 struct ip *ip;
2147 #ifdef INET
2148 struct ipovly *ipovly;
2149 #endif
2150 struct secasvar *sav;
2151 struct tcphdr *th;
2152 #ifdef INET6
2153 struct ip6_hdr *ip6;
2154 struct in6_addr in6;
2155 char ip6buf[INET6_ADDRSTRLEN];
2156 uint32_t plen;
2157 uint16_t nhdr;
2158 #endif
2159 u_short savecsum;
2160
2161 KASSERT(m != NULL, ("NULL mbuf chain"));
2162 KASSERT(buf != NULL, ("NULL signature pointer"));
2163
2164 /* Extract the destination from the IP header in the mbuf. */
2165 bzero(&dst, sizeof(union sockaddr_union));
2166 ip = mtod(m, struct ip *);
2167 #ifdef INET6
2168 ip6 = NULL; /* Make the compiler happy. */
2169 #endif
2170 switch (ip->ip_v) {
2171 #ifdef INET
2172 case IPVERSION:
2173 dst.sa.sa_len = sizeof(struct sockaddr_in);
2174 dst.sa.sa_family = AF_INET;
2175 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2176 ip->ip_src : ip->ip_dst;
2177 break;
2178 #endif
2179 #ifdef INET6
2180 case (IPV6_VERSION >> 4):
2181 ip6 = mtod(m, struct ip6_hdr *);
2182 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2183 dst.sa.sa_family = AF_INET6;
2184 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2185 ip6->ip6_src : ip6->ip6_dst;
2186 break;
2187 #endif
2188 default:
2189 return (EINVAL);
2190 /* NOTREACHED */
2191 break;
2192 }
2193
2194 /* Look up an SADB entry which matches the address of the peer. */
2195 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2196 if (sav == NULL) {
2197 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2198 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2199 #ifdef INET6
2200 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2201 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2202 #endif
2203 "(unsupported)"));
2204 return (EINVAL);
2205 }
2206
2207 MD5Init(&ctx);
2208 /*
2209 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2210 *
2211 * XXX The ippseudo header MUST be digested in network byte order,
2212 * or else we'll fail the regression test. Assume all fields we've
2213 * been doing arithmetic on have been in host byte order.
2214 * XXX One cannot depend on ipovly->ih_len here. When called from
2215 * tcp_output(), the underlying ip_len member has not yet been set.
2216 */
2217 switch (ip->ip_v) {
2218 #ifdef INET
2219 case IPVERSION:
2220 ipovly = (struct ipovly *)ip;
2221 ippseudo.ippseudo_src = ipovly->ih_src;
2222 ippseudo.ippseudo_dst = ipovly->ih_dst;
2223 ippseudo.ippseudo_pad = 0;
2224 ippseudo.ippseudo_p = IPPROTO_TCP;
2225 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2226 optlen);
2227 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2228
2229 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2230 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2231 break;
2232 #endif
2233 #ifdef INET6
2234 /*
2235 * RFC 2385, 2.0 Proposal
2236 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2237 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2238 * extended next header value (to form 32 bits), and 32-bit segment
2239 * length.
2240 * Note: Upper-Layer Packet Length comes before Next Header.
2241 */
2242 case (IPV6_VERSION >> 4):
2243 in6 = ip6->ip6_src;
2244 in6_clearscope(&in6);
2245 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2246 in6 = ip6->ip6_dst;
2247 in6_clearscope(&in6);
2248 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2249 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2250 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2251 nhdr = 0;
2252 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2253 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2254 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2255 nhdr = IPPROTO_TCP;
2256 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2257
2258 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2259 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2260 break;
2261 #endif
2262 default:
2263 return (EINVAL);
2264 /* NOTREACHED */
2265 break;
2266 }
2267
2268
2269 /*
2270 * Step 2: Update MD5 hash with TCP header, excluding options.
2271 * The TCP checksum must be set to zero.
2272 */
2273 savecsum = th->th_sum;
2274 th->th_sum = 0;
2275 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2276 th->th_sum = savecsum;
2277
2278 /*
2279 * Step 3: Update MD5 hash with TCP segment data.
2280 * Use m_apply() to avoid an early m_pullup().
2281 */
2282 if (len > 0)
2283 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2284
2285 /*
2286 * Step 4: Update MD5 hash with shared secret.
2287 */
2288 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2289 MD5Final(buf, &ctx);
2290
2291 key_sa_recordxfer(sav, m);
2292 KEY_FREESAV(&sav);
2293 return (0);
2294 }
2295
2296 /*
2297 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2298 *
2299 * Parameters:
2300 * m pointer to head of mbuf chain
2301 * len length of TCP segment data, excluding options
2302 * optlen length of TCP segment options
2303 * buf pointer to storage for computed MD5 digest
2304 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2305 *
2306 * Return 1 if successful, otherwise return 0.
2307 */
2308 int
tcp_signature_verify(struct mbuf * m,int off0,int tlen,int optlen,struct tcpopt * to,struct tcphdr * th,u_int tcpbflag)2309 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2310 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2311 {
2312 char tmpdigest[TCP_SIGLEN];
2313
2314 if (tcp_sig_checksigs == 0)
2315 return (1);
2316 if ((tcpbflag & TF_SIGNATURE) == 0) {
2317 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2318
2319 /*
2320 * If this socket is not expecting signature but
2321 * the segment contains signature just fail.
2322 */
2323 TCPSTAT_INC(tcps_sig_err_sigopt);
2324 TCPSTAT_INC(tcps_sig_rcvbadsig);
2325 return (0);
2326 }
2327
2328 /* Signature is not expected, and not present in segment. */
2329 return (1);
2330 }
2331
2332 /*
2333 * If this socket is expecting signature but the segment does not
2334 * contain any just fail.
2335 */
2336 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2337 TCPSTAT_INC(tcps_sig_err_nosigopt);
2338 TCPSTAT_INC(tcps_sig_rcvbadsig);
2339 return (0);
2340 }
2341 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2342 IPSEC_DIR_INBOUND) == -1) {
2343 TCPSTAT_INC(tcps_sig_err_buildsig);
2344 TCPSTAT_INC(tcps_sig_rcvbadsig);
2345 return (0);
2346 }
2347
2348 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2349 TCPSTAT_INC(tcps_sig_rcvbadsig);
2350 return (0);
2351 }
2352 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2353 return (1);
2354 }
2355 #endif /* TCP_SIGNATURE */
2356
2357 static int
sysctl_drop(SYSCTL_HANDLER_ARGS)2358 sysctl_drop(SYSCTL_HANDLER_ARGS)
2359 {
2360 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2361 struct sockaddr_storage addrs[2];
2362 struct inpcb *inp;
2363 struct tcpcb *tp;
2364 struct tcptw *tw;
2365 struct sockaddr_in *fin, *lin;
2366 #ifdef INET6
2367 struct sockaddr_in6 *fin6, *lin6;
2368 #endif
2369 int error;
2370
2371 inp = NULL;
2372 fin = lin = NULL;
2373 #ifdef INET6
2374 fin6 = lin6 = NULL;
2375 #endif
2376 error = 0;
2377
2378 if (req->oldptr != NULL || req->oldlen != 0)
2379 return (EINVAL);
2380 if (req->newptr == NULL)
2381 return (EPERM);
2382 if (req->newlen < sizeof(addrs))
2383 return (ENOMEM);
2384 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2385 if (error)
2386 return (error);
2387
2388 switch (addrs[0].ss_family) {
2389 #ifdef INET6
2390 case AF_INET6:
2391 fin6 = (struct sockaddr_in6 *)&addrs[0];
2392 lin6 = (struct sockaddr_in6 *)&addrs[1];
2393 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2394 lin6->sin6_len != sizeof(struct sockaddr_in6))
2395 return (EINVAL);
2396 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2397 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2398 return (EINVAL);
2399 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2400 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2401 fin = (struct sockaddr_in *)&addrs[0];
2402 lin = (struct sockaddr_in *)&addrs[1];
2403 break;
2404 }
2405 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2406 if (error)
2407 return (error);
2408 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2409 if (error)
2410 return (error);
2411 break;
2412 #endif
2413 #ifdef INET
2414 case AF_INET:
2415 fin = (struct sockaddr_in *)&addrs[0];
2416 lin = (struct sockaddr_in *)&addrs[1];
2417 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2418 lin->sin_len != sizeof(struct sockaddr_in))
2419 return (EINVAL);
2420 break;
2421 #endif
2422 default:
2423 return (EINVAL);
2424 }
2425 INP_INFO_RLOCK(&V_tcbinfo);
2426 switch (addrs[0].ss_family) {
2427 #ifdef INET6
2428 case AF_INET6:
2429 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2430 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2431 INPLOOKUP_WLOCKPCB, NULL);
2432 break;
2433 #endif
2434 #ifdef INET
2435 case AF_INET:
2436 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2437 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2438 break;
2439 #endif
2440 }
2441 if (inp != NULL) {
2442 if (inp->inp_flags & INP_TIMEWAIT) {
2443 /*
2444 * XXXRW: There currently exists a state where an
2445 * inpcb is present, but its timewait state has been
2446 * discarded. For now, don't allow dropping of this
2447 * type of inpcb.
2448 */
2449 tw = intotw(inp);
2450 if (tw != NULL)
2451 tcp_twclose(tw, 0);
2452 else
2453 INP_WUNLOCK(inp);
2454 } else if (!(inp->inp_flags & INP_DROPPED) &&
2455 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2456 tp = intotcpcb(inp);
2457 tp = tcp_drop(tp, ECONNABORTED);
2458 if (tp != NULL)
2459 INP_WUNLOCK(inp);
2460 } else
2461 INP_WUNLOCK(inp);
2462 } else
2463 error = ESRCH;
2464 INP_INFO_RUNLOCK(&V_tcbinfo);
2465 return (error);
2466 }
2467
2468 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2469 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2470 0, sysctl_drop, "", "Drop TCP connection");
2471
2472 /*
2473 * Generate a standardized TCP log line for use throughout the
2474 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2475 * allow use in the interrupt context.
2476 *
2477 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2478 * NB: The function may return NULL if memory allocation failed.
2479 *
2480 * Due to header inclusion and ordering limitations the struct ip
2481 * and ip6_hdr pointers have to be passed as void pointers.
2482 */
2483 char *
tcp_log_vain(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2484 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2485 const void *ip6hdr)
2486 {
2487
2488 /* Is logging enabled? */
2489 if (tcp_log_in_vain == 0)
2490 return (NULL);
2491
2492 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2493 }
2494
2495 char *
tcp_log_addrs(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2496 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2497 const void *ip6hdr)
2498 {
2499
2500 /* Is logging enabled? */
2501 if (tcp_log_debug == 0)
2502 return (NULL);
2503
2504 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2505 }
2506
2507 static char *
tcp_log_addr(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2508 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2509 const void *ip6hdr)
2510 {
2511 char *s, *sp;
2512 size_t size;
2513 struct ip *ip;
2514 #ifdef INET6
2515 const struct ip6_hdr *ip6;
2516
2517 ip6 = (const struct ip6_hdr *)ip6hdr;
2518 #endif /* INET6 */
2519 ip = (struct ip *)ip4hdr;
2520
2521 /*
2522 * The log line looks like this:
2523 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2524 */
2525 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2526 sizeof(PRINT_TH_FLAGS) + 1 +
2527 #ifdef INET6
2528 2 * INET6_ADDRSTRLEN;
2529 #else
2530 2 * INET_ADDRSTRLEN;
2531 #endif /* INET6 */
2532
2533 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2534 if (s == NULL)
2535 return (NULL);
2536
2537 strcat(s, "TCP: [");
2538 sp = s + strlen(s);
2539
2540 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2541 inet_ntoa_r(inc->inc_faddr, sp);
2542 sp = s + strlen(s);
2543 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2544 sp = s + strlen(s);
2545 inet_ntoa_r(inc->inc_laddr, sp);
2546 sp = s + strlen(s);
2547 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2548 #ifdef INET6
2549 } else if (inc) {
2550 ip6_sprintf(sp, &inc->inc6_faddr);
2551 sp = s + strlen(s);
2552 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2553 sp = s + strlen(s);
2554 ip6_sprintf(sp, &inc->inc6_laddr);
2555 sp = s + strlen(s);
2556 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2557 } else if (ip6 && th) {
2558 ip6_sprintf(sp, &ip6->ip6_src);
2559 sp = s + strlen(s);
2560 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2561 sp = s + strlen(s);
2562 ip6_sprintf(sp, &ip6->ip6_dst);
2563 sp = s + strlen(s);
2564 sprintf(sp, "]:%i", ntohs(th->th_dport));
2565 #endif /* INET6 */
2566 #ifdef INET
2567 } else if (ip && th) {
2568 inet_ntoa_r(ip->ip_src, sp);
2569 sp = s + strlen(s);
2570 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2571 sp = s + strlen(s);
2572 inet_ntoa_r(ip->ip_dst, sp);
2573 sp = s + strlen(s);
2574 sprintf(sp, "]:%i", ntohs(th->th_dport));
2575 #endif /* INET */
2576 } else {
2577 free(s, M_TCPLOG);
2578 return (NULL);
2579 }
2580 sp = s + strlen(s);
2581 if (th)
2582 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2583 if (*(s + size - 1) != '\0')
2584 panic("%s: string too long", __func__);
2585 return (s);
2586 }
2587
2588 /*
2589 * A subroutine which makes it easy to track TCP state changes with DTrace.
2590 * This function shouldn't be called for t_state initializations that don't
2591 * correspond to actual TCP state transitions.
2592 */
2593 void
tcp_state_change(struct tcpcb * tp,int newstate)2594 tcp_state_change(struct tcpcb *tp, int newstate)
2595 {
2596 #if defined(KDTRACE_HOOKS)
2597 int pstate = tp->t_state;
2598 #endif
2599
2600 tp->t_state = newstate;
2601 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2602 }
2603