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("$FreeBSD$");
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 #include <netinet/tcp_fsm.h>
87 #include <netinet/tcp_seq.h>
88 #include <netinet/tcp_timer.h>
89 #include <netinet/tcp_var.h>
90 #include <netinet/tcp_syncache.h>
91 #ifdef INET6
92 #include <netinet6/tcp6_var.h>
93 #endif
94 #include <netinet/tcpip.h>
95 #ifdef TCPDEBUG
96 #include <netinet/tcp_debug.h>
97 #endif
98 #ifdef INET6
99 #include <netinet6/ip6protosw.h>
100 #endif
101 #ifdef TCP_OFFLOAD
102 #include <netinet/tcp_offload.h>
103 #endif
104
105 #ifdef IPSEC
106 #include <netipsec/ipsec.h>
107 #include <netipsec/xform.h>
108 #ifdef INET6
109 #include <netipsec/ipsec6.h>
110 #endif
111 #include <netipsec/key.h>
112 #include <sys/syslog.h>
113 #endif /*IPSEC*/
114
115 #include <machine/in_cksum.h>
116 #include <sys/md5.h>
117
118 #include <security/mac/mac_framework.h>
119
120 VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
121 #ifdef INET6
122 VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
123 #endif
124
125 static int
sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)126 sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
127 {
128 int error, new;
129
130 new = V_tcp_mssdflt;
131 error = sysctl_handle_int(oidp, &new, 0, req);
132 if (error == 0 && req->newptr) {
133 if (new < TCP_MINMSS)
134 error = EINVAL;
135 else
136 V_tcp_mssdflt = new;
137 }
138 return (error);
139 }
140
141 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
142 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
143 &sysctl_net_inet_tcp_mss_check, "I",
144 "Default TCP Maximum Segment Size");
145
146 #ifdef INET6
147 static int
sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)148 sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
149 {
150 int error, new;
151
152 new = V_tcp_v6mssdflt;
153 error = sysctl_handle_int(oidp, &new, 0, req);
154 if (error == 0 && req->newptr) {
155 if (new < TCP_MINMSS)
156 error = EINVAL;
157 else
158 V_tcp_v6mssdflt = new;
159 }
160 return (error);
161 }
162
163 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
164 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
165 &sysctl_net_inet_tcp_mss_v6_check, "I",
166 "Default TCP Maximum Segment Size for IPv6");
167 #endif /* INET6 */
168
169 /*
170 * Minimum MSS we accept and use. This prevents DoS attacks where
171 * we are forced to a ridiculous low MSS like 20 and send hundreds
172 * of packets instead of one. The effect scales with the available
173 * bandwidth and quickly saturates the CPU and network interface
174 * with packet generation and sending. Set to zero to disable MINMSS
175 * checking. This setting prevents us from sending too small packets.
176 */
177 VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
178 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
179 &VNET_NAME(tcp_minmss), 0,
180 "Minimum TCP Maximum Segment Size");
181
182 VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183 SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
184 &VNET_NAME(tcp_do_rfc1323), 0,
185 "Enable rfc1323 (high performance TCP) extensions");
186
187 static int tcp_log_debug = 0;
188 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
189 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
190
191 static int tcp_tcbhashsize = 0;
192 SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
193 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
194
195 static int do_tcpdrain = 1;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
197 "Enable tcp_drain routine for extra help when low on mbufs");
198
199 SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
200 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
201
202 static VNET_DEFINE(int, icmp_may_rst) = 1;
203 #define V_icmp_may_rst VNET(icmp_may_rst)
204 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
205 &VNET_NAME(icmp_may_rst), 0,
206 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
207
208 static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
209 #define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
210 SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
211 &VNET_NAME(tcp_isn_reseed_interval), 0,
212 "Seconds between reseeding of ISN secret");
213
214 static int tcp_soreceive_stream = 0;
215 SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
216 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
217
218 #ifdef TCP_SIGNATURE
219 static int tcp_sig_checksigs = 1;
220 SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
221 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
222 #endif
223
224 VNET_DEFINE(uma_zone_t, sack_hole_zone);
225 #define V_sack_hole_zone VNET(sack_hole_zone)
226
227 VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
228
229 static struct inpcb *tcp_notify(struct inpcb *, int);
230 static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231 static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
232 void *ip4hdr, const void *ip6hdr);
233 static void tcp_timer_discard(struct tcpcb *, uint32_t);
234
235 /*
236 * Target size of TCP PCB hash tables. Must be a power of two.
237 *
238 * Note that this can be overridden by the kernel environment
239 * variable net.inet.tcp.tcbhashsize
240 */
241 #ifndef TCBHASHSIZE
242 #define TCBHASHSIZE 0
243 #endif
244
245 /*
246 * XXX
247 * Callouts should be moved into struct tcp directly. They are currently
248 * separate because the tcpcb structure is exported to userland for sysctl
249 * parsing purposes, which do not know about callouts.
250 */
251 struct tcpcb_mem {
252 struct tcpcb tcb;
253 struct tcp_timer tt;
254 struct cc_var ccv;
255 struct osd osd;
256 };
257
258 static VNET_DEFINE(uma_zone_t, tcpcb_zone);
259 #define V_tcpcb_zone VNET(tcpcb_zone)
260
261 MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
262 static struct mtx isn_mtx;
263
264 #define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
265 #define ISN_LOCK() mtx_lock(&isn_mtx)
266 #define ISN_UNLOCK() mtx_unlock(&isn_mtx)
267
268 /*
269 * TCP initialization.
270 */
271 static void
tcp_zone_change(void * tag)272 tcp_zone_change(void *tag)
273 {
274
275 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
276 uma_zone_set_max(V_tcpcb_zone, maxsockets);
277 tcp_tw_zone_change();
278 }
279
280 static int
tcp_inpcb_init(void * mem,int size,int flags)281 tcp_inpcb_init(void *mem, int size, int flags)
282 {
283 struct inpcb *inp = mem;
284
285 INP_LOCK_INIT(inp, "inp", "tcpinp");
286 return (0);
287 }
288
289 /*
290 * Take a value and get the next power of 2 that doesn't overflow.
291 * Used to size the tcp_inpcb hash buckets.
292 */
293 static int
maketcp_hashsize(int size)294 maketcp_hashsize(int size)
295 {
296 int hashsize;
297
298 /*
299 * auto tune.
300 * get the next power of 2 higher than maxsockets.
301 */
302 hashsize = 1 << fls(size);
303 /* catch overflow, and just go one power of 2 smaller */
304 if (hashsize < size) {
305 hashsize = 1 << (fls(size) - 1);
306 }
307 return (hashsize);
308 }
309
310 void
tcp_init(void)311 tcp_init(void)
312 {
313 const char *tcbhash_tuneable;
314 int hashsize;
315
316 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
317
318 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
319 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
320 printf("%s: WARNING: unable to register helper hook\n", __func__);
321 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
322 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
323 printf("%s: WARNING: unable to register helper hook\n", __func__);
324
325 hashsize = TCBHASHSIZE;
326 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
327 if (hashsize == 0) {
328 /*
329 * Auto tune the hash size based on maxsockets.
330 * A perfect hash would have a 1:1 mapping
331 * (hashsize = maxsockets) however it's been
332 * suggested that O(2) average is better.
333 */
334 hashsize = maketcp_hashsize(maxsockets / 4);
335 /*
336 * Our historical default is 512,
337 * do not autotune lower than this.
338 */
339 if (hashsize < 512)
340 hashsize = 512;
341 if (bootverbose)
342 printf("%s: %s auto tuned to %d\n", __func__,
343 tcbhash_tuneable, hashsize);
344 }
345 /*
346 * We require a hashsize to be a power of two.
347 * Previously if it was not a power of two we would just reset it
348 * back to 512, which could be a nasty surprise if you did not notice
349 * the error message.
350 * Instead what we do is clip it to the closest power of two lower
351 * than the specified hash value.
352 */
353 if (!powerof2(hashsize)) {
354 int oldhashsize = hashsize;
355
356 hashsize = maketcp_hashsize(hashsize);
357 /* prevent absurdly low value */
358 if (hashsize < 16)
359 hashsize = 16;
360 printf("%s: WARNING: TCB hash size not a power of 2, "
361 "clipped from %d to %d.\n", __func__, oldhashsize,
362 hashsize);
363 }
364 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
365 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
366 IPI_HASHFIELDS_4TUPLE);
367
368 /*
369 * These have to be type stable for the benefit of the timers.
370 */
371 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
372 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
373 uma_zone_set_max(V_tcpcb_zone, maxsockets);
374 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
375
376 tcp_tw_init();
377 syncache_init();
378 tcp_hc_init();
379 tcp_reass_init();
380
381 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
382 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
383 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
384
385 /* Skip initialization of globals for non-default instances. */
386 if (!IS_DEFAULT_VNET(curvnet))
387 return;
388
389 /* XXX virtualize those bellow? */
390 tcp_delacktime = TCPTV_DELACK;
391 tcp_keepinit = TCPTV_KEEP_INIT;
392 tcp_keepidle = TCPTV_KEEP_IDLE;
393 tcp_keepintvl = TCPTV_KEEPINTVL;
394 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
395 tcp_msl = TCPTV_MSL;
396 tcp_rexmit_min = TCPTV_MIN;
397 if (tcp_rexmit_min < 1)
398 tcp_rexmit_min = 1;
399 tcp_rexmit_slop = TCPTV_CPU_VAR;
400 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
401 tcp_tcbhashsize = hashsize;
402
403 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
404 if (tcp_soreceive_stream) {
405 #ifdef INET
406 tcp_usrreqs.pru_soreceive = soreceive_stream;
407 #endif
408 #ifdef INET6
409 tcp6_usrreqs.pru_soreceive = soreceive_stream;
410 #endif /* INET6 */
411 }
412
413 #ifdef INET6
414 #define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
415 #else /* INET6 */
416 #define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
417 #endif /* INET6 */
418 if (max_protohdr < TCP_MINPROTOHDR)
419 max_protohdr = TCP_MINPROTOHDR;
420 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
421 panic("tcp_init");
422 #undef TCP_MINPROTOHDR
423
424 ISN_LOCK_INIT();
425 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
426 SHUTDOWN_PRI_DEFAULT);
427 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
428 EVENTHANDLER_PRI_ANY);
429 }
430
431 #ifdef VIMAGE
432 void
tcp_destroy(void)433 tcp_destroy(void)
434 {
435
436 tcp_reass_destroy();
437 tcp_hc_destroy();
438 syncache_destroy();
439 tcp_tw_destroy();
440 in_pcbinfo_destroy(&V_tcbinfo);
441 uma_zdestroy(V_sack_hole_zone);
442 uma_zdestroy(V_tcpcb_zone);
443 }
444 #endif
445
446 void
tcp_fini(void * xtp)447 tcp_fini(void *xtp)
448 {
449
450 }
451
452 /*
453 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
454 * tcp_template used to store this data in mbufs, but we now recopy it out
455 * of the tcpcb each time to conserve mbufs.
456 */
457 void
tcpip_fillheaders(struct inpcb * inp,void * ip_ptr,void * tcp_ptr)458 tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
459 {
460 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
461
462 INP_WLOCK_ASSERT(inp);
463
464 #ifdef INET6
465 if ((inp->inp_vflag & INP_IPV6) != 0) {
466 struct ip6_hdr *ip6;
467
468 ip6 = (struct ip6_hdr *)ip_ptr;
469 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
470 (inp->inp_flow & IPV6_FLOWINFO_MASK);
471 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
472 (IPV6_VERSION & IPV6_VERSION_MASK);
473 ip6->ip6_nxt = IPPROTO_TCP;
474 ip6->ip6_plen = htons(sizeof(struct tcphdr));
475 ip6->ip6_src = inp->in6p_laddr;
476 ip6->ip6_dst = inp->in6p_faddr;
477 }
478 #endif /* INET6 */
479 #if defined(INET6) && defined(INET)
480 else
481 #endif
482 #ifdef INET
483 {
484 struct ip *ip;
485
486 ip = (struct ip *)ip_ptr;
487 ip->ip_v = IPVERSION;
488 ip->ip_hl = 5;
489 ip->ip_tos = inp->inp_ip_tos;
490 ip->ip_len = 0;
491 ip->ip_id = 0;
492 ip->ip_off = 0;
493 ip->ip_ttl = inp->inp_ip_ttl;
494 ip->ip_sum = 0;
495 ip->ip_p = IPPROTO_TCP;
496 ip->ip_src = inp->inp_laddr;
497 ip->ip_dst = inp->inp_faddr;
498 }
499 #endif /* INET */
500 th->th_sport = inp->inp_lport;
501 th->th_dport = inp->inp_fport;
502 th->th_seq = 0;
503 th->th_ack = 0;
504 th->th_x2 = 0;
505 th->th_off = 5;
506 th->th_flags = 0;
507 th->th_win = 0;
508 th->th_urp = 0;
509 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
510 }
511
512 /*
513 * Create template to be used to send tcp packets on a connection.
514 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
515 * use for this function is in keepalives, which use tcp_respond.
516 */
517 struct tcptemp *
tcpip_maketemplate(struct inpcb * inp)518 tcpip_maketemplate(struct inpcb *inp)
519 {
520 struct tcptemp *t;
521
522 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
523 if (t == NULL)
524 return (NULL);
525 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
526 return (t);
527 }
528
529 /*
530 * Send a single message to the TCP at address specified by
531 * the given TCP/IP header. If m == NULL, then we make a copy
532 * of the tcpiphdr at ti and send directly to the addressed host.
533 * This is used to force keep alive messages out using the TCP
534 * template for a connection. If flags are given then we send
535 * a message back to the TCP which originated the * segment ti,
536 * and discard the mbuf containing it and any other attached mbufs.
537 *
538 * In any case the ack and sequence number of the transmitted
539 * segment are as specified by the parameters.
540 *
541 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
542 */
543 void
tcp_respond(struct tcpcb * tp,void * ipgen,struct tcphdr * th,struct mbuf * m,tcp_seq ack,tcp_seq seq,int flags)544 tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
545 tcp_seq ack, tcp_seq seq, int flags)
546 {
547 int tlen;
548 int win = 0;
549 struct ip *ip;
550 struct tcphdr *nth;
551 #ifdef INET6
552 struct ip6_hdr *ip6;
553 int isipv6;
554 #endif /* INET6 */
555 int ipflags = 0;
556 struct inpcb *inp;
557
558 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
559
560 #ifdef INET6
561 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
562 ip6 = ipgen;
563 #endif /* INET6 */
564 ip = ipgen;
565
566 if (tp != NULL) {
567 inp = tp->t_inpcb;
568 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
569 INP_WLOCK_ASSERT(inp);
570 } else
571 inp = NULL;
572
573 if (tp != NULL) {
574 if (!(flags & TH_RST)) {
575 win = sbspace(&inp->inp_socket->so_rcv);
576 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
577 win = (long)TCP_MAXWIN << tp->rcv_scale;
578 }
579 }
580 if (m == NULL) {
581 m = m_gethdr(M_NOWAIT, MT_DATA);
582 if (m == NULL)
583 return;
584 tlen = 0;
585 m->m_data += max_linkhdr;
586 #ifdef INET6
587 if (isipv6) {
588 bcopy((caddr_t)ip6, mtod(m, caddr_t),
589 sizeof(struct ip6_hdr));
590 ip6 = mtod(m, struct ip6_hdr *);
591 nth = (struct tcphdr *)(ip6 + 1);
592 } else
593 #endif /* INET6 */
594 {
595 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
596 ip = mtod(m, struct ip *);
597 nth = (struct tcphdr *)(ip + 1);
598 }
599 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
600 flags = TH_ACK;
601 } else {
602 /*
603 * reuse the mbuf.
604 * XXX MRT We inherrit the FIB, which is lucky.
605 */
606 m_freem(m->m_next);
607 m->m_next = NULL;
608 m->m_data = (caddr_t)ipgen;
609 /* m_len is set later */
610 tlen = 0;
611 #define xchg(a,b,type) { type t; t=a; a=b; b=t; }
612 #ifdef INET6
613 if (isipv6) {
614 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
615 nth = (struct tcphdr *)(ip6 + 1);
616 } else
617 #endif /* INET6 */
618 {
619 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
620 nth = (struct tcphdr *)(ip + 1);
621 }
622 if (th != nth) {
623 /*
624 * this is usually a case when an extension header
625 * exists between the IPv6 header and the
626 * TCP header.
627 */
628 nth->th_sport = th->th_sport;
629 nth->th_dport = th->th_dport;
630 }
631 xchg(nth->th_dport, nth->th_sport, uint16_t);
632 #undef xchg
633 }
634 #ifdef INET6
635 if (isipv6) {
636 ip6->ip6_flow = 0;
637 ip6->ip6_vfc = IPV6_VERSION;
638 ip6->ip6_nxt = IPPROTO_TCP;
639 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
640 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
641 }
642 #endif
643 #if defined(INET) && defined(INET6)
644 else
645 #endif
646 #ifdef INET
647 {
648 tlen += sizeof (struct tcpiphdr);
649 ip->ip_len = htons(tlen);
650 ip->ip_ttl = V_ip_defttl;
651 if (V_path_mtu_discovery)
652 ip->ip_off |= htons(IP_DF);
653 }
654 #endif
655 m->m_len = tlen;
656 m->m_pkthdr.len = tlen;
657 m->m_pkthdr.rcvif = NULL;
658 #ifdef MAC
659 if (inp != NULL) {
660 /*
661 * Packet is associated with a socket, so allow the
662 * label of the response to reflect the socket label.
663 */
664 INP_WLOCK_ASSERT(inp);
665 mac_inpcb_create_mbuf(inp, m);
666 } else {
667 /*
668 * Packet is not associated with a socket, so possibly
669 * update the label in place.
670 */
671 mac_netinet_tcp_reply(m);
672 }
673 #endif
674 nth->th_seq = htonl(seq);
675 nth->th_ack = htonl(ack);
676 nth->th_x2 = 0;
677 nth->th_off = sizeof (struct tcphdr) >> 2;
678 nth->th_flags = flags;
679 if (tp != NULL)
680 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
681 else
682 nth->th_win = htons((u_short)win);
683 nth->th_urp = 0;
684
685 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
686 #ifdef INET6
687 if (isipv6) {
688 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
689 nth->th_sum = in6_cksum_pseudo(ip6,
690 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
691 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
692 NULL, NULL);
693 }
694 #endif /* INET6 */
695 #if defined(INET6) && defined(INET)
696 else
697 #endif
698 #ifdef INET
699 {
700 m->m_pkthdr.csum_flags = CSUM_TCP;
701 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
702 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
703 }
704 #endif /* INET */
705 #ifdef TCPDEBUG
706 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
707 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
708 #endif
709 if (flags & TH_RST)
710 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
711 tp, nth);
712
713 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
714 #ifdef INET6
715 if (isipv6)
716 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
717 #endif /* INET6 */
718 #if defined(INET) && defined(INET6)
719 else
720 #endif
721 #ifdef INET
722 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
723 #endif
724 }
725
726 /*
727 * Create a new TCP control block, making an
728 * empty reassembly queue and hooking it to the argument
729 * protocol control block. The `inp' parameter must have
730 * come from the zone allocator set up in tcp_init().
731 */
732 struct tcpcb *
tcp_newtcpcb(struct inpcb * inp)733 tcp_newtcpcb(struct inpcb *inp)
734 {
735 struct tcpcb_mem *tm;
736 struct tcpcb *tp;
737 #ifdef INET6
738 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
739 #endif /* INET6 */
740
741 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
742 if (tm == NULL)
743 return (NULL);
744 tp = &tm->tcb;
745
746 /* Initialise cc_var struct for this tcpcb. */
747 tp->ccv = &tm->ccv;
748 tp->ccv->type = IPPROTO_TCP;
749 tp->ccv->ccvc.tcp = tp;
750
751 /*
752 * Use the current system default CC algorithm.
753 */
754 CC_LIST_RLOCK();
755 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
756 CC_ALGO(tp) = CC_DEFAULT();
757 CC_LIST_RUNLOCK();
758
759 if (CC_ALGO(tp)->cb_init != NULL)
760 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
761 uma_zfree(V_tcpcb_zone, tm);
762 return (NULL);
763 }
764
765 tp->osd = &tm->osd;
766 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
767 uma_zfree(V_tcpcb_zone, tm);
768 return (NULL);
769 }
770
771 #ifdef VIMAGE
772 tp->t_vnet = inp->inp_vnet;
773 #endif
774 tp->t_timers = &tm->tt;
775 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
776 tp->t_maxseg = tp->t_maxopd =
777 #ifdef INET6
778 isipv6 ? V_tcp_v6mssdflt :
779 #endif /* INET6 */
780 V_tcp_mssdflt;
781
782 /* Set up our timeouts. */
783 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
784 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
785 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
786 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
787 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
788
789 if (V_tcp_do_rfc1323)
790 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
791 if (V_tcp_do_sack)
792 tp->t_flags |= TF_SACK_PERMIT;
793 TAILQ_INIT(&tp->snd_holes);
794 /*
795 * The tcpcb will hold a reference on its inpcb until tcp_discardcb()
796 * is called.
797 */
798 in_pcbref(inp); /* Reference for tcpcb */
799 tp->t_inpcb = inp;
800
801 /*
802 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
803 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
804 * reasonable initial retransmit time.
805 */
806 tp->t_srtt = TCPTV_SRTTBASE;
807 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
808 tp->t_rttmin = tcp_rexmit_min;
809 tp->t_rxtcur = TCPTV_RTOBASE;
810 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
811 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
812 tp->t_rcvtime = ticks;
813 /*
814 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
815 * because the socket may be bound to an IPv6 wildcard address,
816 * which may match an IPv4-mapped IPv6 address.
817 */
818 inp->inp_ip_ttl = V_ip_defttl;
819 inp->inp_ppcb = tp;
820 return (tp); /* XXX */
821 }
822
823 /*
824 * Switch the congestion control algorithm back to NewReno for any active
825 * control blocks using an algorithm which is about to go away.
826 * This ensures the CC framework can allow the unload to proceed without leaving
827 * any dangling pointers which would trigger a panic.
828 * Returning non-zero would inform the CC framework that something went wrong
829 * and it would be unsafe to allow the unload to proceed. However, there is no
830 * way for this to occur with this implementation so we always return zero.
831 */
832 int
tcp_ccalgounload(struct cc_algo * unload_algo)833 tcp_ccalgounload(struct cc_algo *unload_algo)
834 {
835 struct cc_algo *tmpalgo;
836 struct inpcb *inp;
837 struct tcpcb *tp;
838 VNET_ITERATOR_DECL(vnet_iter);
839
840 /*
841 * Check all active control blocks across all network stacks and change
842 * any that are using "unload_algo" back to NewReno. If "unload_algo"
843 * requires cleanup code to be run, call it.
844 */
845 VNET_LIST_RLOCK();
846 VNET_FOREACH(vnet_iter) {
847 CURVNET_SET(vnet_iter);
848 INP_INFO_RLOCK(&V_tcbinfo);
849 /*
850 * New connections already part way through being initialised
851 * with the CC algo we're removing will not race with this code
852 * because the INP_INFO_WLOCK is held during initialisation. We
853 * therefore don't enter the loop below until the connection
854 * list has stabilised.
855 */
856 LIST_FOREACH(inp, &V_tcb, inp_list) {
857 INP_WLOCK(inp);
858 /* Important to skip tcptw structs. */
859 if (!(inp->inp_flags & INP_TIMEWAIT) &&
860 (tp = intotcpcb(inp)) != NULL) {
861 /*
862 * By holding INP_WLOCK here, we are assured
863 * that the connection is not currently
864 * executing inside the CC module's functions
865 * i.e. it is safe to make the switch back to
866 * NewReno.
867 */
868 if (CC_ALGO(tp) == unload_algo) {
869 tmpalgo = CC_ALGO(tp);
870 /* NewReno does not require any init. */
871 CC_ALGO(tp) = &newreno_cc_algo;
872 if (tmpalgo->cb_destroy != NULL)
873 tmpalgo->cb_destroy(tp->ccv);
874 }
875 }
876 INP_WUNLOCK(inp);
877 }
878 INP_INFO_RUNLOCK(&V_tcbinfo);
879 CURVNET_RESTORE();
880 }
881 VNET_LIST_RUNLOCK();
882
883 return (0);
884 }
885
886 /*
887 * Drop a TCP connection, reporting
888 * the specified error. If connection is synchronized,
889 * then send a RST to peer.
890 */
891 struct tcpcb *
tcp_drop(struct tcpcb * tp,int errno)892 tcp_drop(struct tcpcb *tp, int errno)
893 {
894 struct socket *so = tp->t_inpcb->inp_socket;
895
896 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
897 INP_WLOCK_ASSERT(tp->t_inpcb);
898
899 if (TCPS_HAVERCVDSYN(tp->t_state)) {
900 tcp_state_change(tp, TCPS_CLOSED);
901 (void) tcp_output(tp);
902 TCPSTAT_INC(tcps_drops);
903 } else
904 TCPSTAT_INC(tcps_conndrops);
905 if (errno == ETIMEDOUT && tp->t_softerror)
906 errno = tp->t_softerror;
907 so->so_error = errno;
908 return (tcp_close(tp));
909 }
910
911 void
tcp_discardcb(struct tcpcb * tp)912 tcp_discardcb(struct tcpcb *tp)
913 {
914 struct inpcb *inp = tp->t_inpcb;
915 struct socket *so = inp->inp_socket;
916 #ifdef INET6
917 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
918 #endif /* INET6 */
919 int released;
920
921 INP_WLOCK_ASSERT(inp);
922
923 /*
924 * Make sure that all of our timers are stopped before we delete the
925 * PCB.
926 *
927 * If stopping a timer fails, we schedule a discard function in same
928 * callout, and the last discard function called will take care of
929 * deleting the tcpcb.
930 */
931 tcp_timer_stop(tp, TT_REXMT);
932 tcp_timer_stop(tp, TT_PERSIST);
933 tcp_timer_stop(tp, TT_KEEP);
934 tcp_timer_stop(tp, TT_2MSL);
935 tcp_timer_stop(tp, TT_DELACK);
936
937 /*
938 * If we got enough samples through the srtt filter,
939 * save the rtt and rttvar in the routing entry.
940 * 'Enough' is arbitrarily defined as 4 rtt samples.
941 * 4 samples is enough for the srtt filter to converge
942 * to within enough % of the correct value; fewer samples
943 * and we could save a bogus rtt. The danger is not high
944 * as tcp quickly recovers from everything.
945 * XXX: Works very well but needs some more statistics!
946 */
947 if (tp->t_rttupdated >= 4) {
948 struct hc_metrics_lite metrics;
949 u_long ssthresh;
950
951 bzero(&metrics, sizeof(metrics));
952 /*
953 * Update the ssthresh always when the conditions below
954 * are satisfied. This gives us better new start value
955 * for the congestion avoidance for new connections.
956 * ssthresh is only set if packet loss occured on a session.
957 *
958 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
959 * being torn down. Ideally this code would not use 'so'.
960 */
961 ssthresh = tp->snd_ssthresh;
962 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
963 /*
964 * convert the limit from user data bytes to
965 * packets then to packet data bytes.
966 */
967 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
968 if (ssthresh < 2)
969 ssthresh = 2;
970 ssthresh *= (u_long)(tp->t_maxseg +
971 #ifdef INET6
972 (isipv6 ? sizeof (struct ip6_hdr) +
973 sizeof (struct tcphdr) :
974 #endif
975 sizeof (struct tcpiphdr)
976 #ifdef INET6
977 )
978 #endif
979 );
980 } else
981 ssthresh = 0;
982 metrics.rmx_ssthresh = ssthresh;
983
984 metrics.rmx_rtt = tp->t_srtt;
985 metrics.rmx_rttvar = tp->t_rttvar;
986 metrics.rmx_cwnd = tp->snd_cwnd;
987 metrics.rmx_sendpipe = 0;
988 metrics.rmx_recvpipe = 0;
989
990 tcp_hc_update(&inp->inp_inc, &metrics);
991 }
992
993 /* free the reassembly queue, if any */
994 tcp_reass_flush(tp);
995
996 #ifdef TCP_OFFLOAD
997 /* Disconnect offload device, if any. */
998 if (tp->t_flags & TF_TOE)
999 tcp_offload_detach(tp);
1000 #endif
1001
1002 tcp_free_sackholes(tp);
1003
1004 /* Allow the CC algorithm to clean up after itself. */
1005 if (CC_ALGO(tp)->cb_destroy != NULL)
1006 CC_ALGO(tp)->cb_destroy(tp->ccv);
1007
1008 khelp_destroy_osd(tp->osd);
1009
1010 CC_ALGO(tp) = NULL;
1011 inp->inp_ppcb = NULL;
1012 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1013 /* We own the last reference on tcpcb, let's free it. */
1014 tp->t_inpcb = NULL;
1015 uma_zfree(V_tcpcb_zone, tp);
1016 released = in_pcbrele_wlocked(inp);
1017 KASSERT(!released, ("%s: inp %p should not have been released "
1018 "here", __func__, inp));
1019 }
1020 }
1021
1022 void
tcp_timer_2msl_discard(void * xtp)1023 tcp_timer_2msl_discard(void *xtp)
1024 {
1025
1026 tcp_timer_discard((struct tcpcb *)xtp, TT_2MSL);
1027 }
1028
1029 void
tcp_timer_keep_discard(void * xtp)1030 tcp_timer_keep_discard(void *xtp)
1031 {
1032
1033 tcp_timer_discard((struct tcpcb *)xtp, TT_KEEP);
1034 }
1035
1036 void
tcp_timer_persist_discard(void * xtp)1037 tcp_timer_persist_discard(void *xtp)
1038 {
1039
1040 tcp_timer_discard((struct tcpcb *)xtp, TT_PERSIST);
1041 }
1042
1043 void
tcp_timer_rexmt_discard(void * xtp)1044 tcp_timer_rexmt_discard(void *xtp)
1045 {
1046
1047 tcp_timer_discard((struct tcpcb *)xtp, TT_REXMT);
1048 }
1049
1050 void
tcp_timer_delack_discard(void * xtp)1051 tcp_timer_delack_discard(void *xtp)
1052 {
1053
1054 tcp_timer_discard((struct tcpcb *)xtp, TT_DELACK);
1055 }
1056
1057 void
tcp_timer_discard(struct tcpcb * tp,uint32_t timer_type)1058 tcp_timer_discard(struct tcpcb *tp, uint32_t timer_type)
1059 {
1060 struct inpcb *inp;
1061
1062 CURVNET_SET(tp->t_vnet);
1063 INP_INFO_WLOCK(&V_tcbinfo);
1064 inp = tp->t_inpcb;
1065 KASSERT(inp != NULL, ("%s: tp %p tp->t_inpcb == NULL",
1066 __func__, tp));
1067 INP_WLOCK(inp);
1068 KASSERT((tp->t_timers->tt_flags & TT_STOPPED) != 0,
1069 ("%s: tcpcb has to be stopped here", __func__));
1070 KASSERT((tp->t_timers->tt_flags & timer_type) != 0,
1071 ("%s: discard callout should be running", __func__));
1072 tp->t_timers->tt_flags &= ~timer_type;
1073 if ((tp->t_timers->tt_flags & TT_MASK) == 0) {
1074 /* We own the last reference on this tcpcb, let's free it. */
1075 tp->t_inpcb = NULL;
1076 uma_zfree(V_tcpcb_zone, tp);
1077 if (in_pcbrele_wlocked(inp)) {
1078 INP_INFO_WUNLOCK(&V_tcbinfo);
1079 CURVNET_RESTORE();
1080 return;
1081 }
1082 }
1083 INP_WUNLOCK(inp);
1084 INP_INFO_WUNLOCK(&V_tcbinfo);
1085 CURVNET_RESTORE();
1086 }
1087
1088 /*
1089 * Attempt to close a TCP control block, marking it as dropped, and freeing
1090 * the socket if we hold the only reference.
1091 */
1092 struct tcpcb *
tcp_close(struct tcpcb * tp)1093 tcp_close(struct tcpcb *tp)
1094 {
1095 struct inpcb *inp = tp->t_inpcb;
1096 struct socket *so;
1097
1098 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1099 INP_WLOCK_ASSERT(inp);
1100
1101 #ifdef TCP_OFFLOAD
1102 if (tp->t_state == TCPS_LISTEN)
1103 tcp_offload_listen_stop(tp);
1104 #endif
1105 in_pcbdrop(inp);
1106 TCPSTAT_INC(tcps_closed);
1107 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1108 so = inp->inp_socket;
1109 soisdisconnected(so);
1110 if (inp->inp_flags & INP_SOCKREF) {
1111 KASSERT(so->so_state & SS_PROTOREF,
1112 ("tcp_close: !SS_PROTOREF"));
1113 inp->inp_flags &= ~INP_SOCKREF;
1114 INP_WUNLOCK(inp);
1115 ACCEPT_LOCK();
1116 SOCK_LOCK(so);
1117 so->so_state &= ~SS_PROTOREF;
1118 sofree(so);
1119 return (NULL);
1120 }
1121 return (tp);
1122 }
1123
1124 void
tcp_drain(void)1125 tcp_drain(void)
1126 {
1127 VNET_ITERATOR_DECL(vnet_iter);
1128
1129 if (!do_tcpdrain)
1130 return;
1131
1132 VNET_LIST_RLOCK_NOSLEEP();
1133 VNET_FOREACH(vnet_iter) {
1134 CURVNET_SET(vnet_iter);
1135 struct inpcb *inpb;
1136 struct tcpcb *tcpb;
1137
1138 /*
1139 * Walk the tcpbs, if existing, and flush the reassembly queue,
1140 * if there is one...
1141 * XXX: The "Net/3" implementation doesn't imply that the TCP
1142 * reassembly queue should be flushed, but in a situation
1143 * where we're really low on mbufs, this is potentially
1144 * useful.
1145 */
1146 INP_INFO_RLOCK(&V_tcbinfo);
1147 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1148 if (inpb->inp_flags & INP_TIMEWAIT)
1149 continue;
1150 INP_WLOCK(inpb);
1151 if ((tcpb = intotcpcb(inpb)) != NULL) {
1152 tcp_reass_flush(tcpb);
1153 tcp_clean_sackreport(tcpb);
1154 }
1155 INP_WUNLOCK(inpb);
1156 }
1157 INP_INFO_RUNLOCK(&V_tcbinfo);
1158 CURVNET_RESTORE();
1159 }
1160 VNET_LIST_RUNLOCK_NOSLEEP();
1161 }
1162
1163 /*
1164 * Notify a tcp user of an asynchronous error;
1165 * store error as soft error, but wake up user
1166 * (for now, won't do anything until can select for soft error).
1167 *
1168 * Do not wake up user since there currently is no mechanism for
1169 * reporting soft errors (yet - a kqueue filter may be added).
1170 */
1171 static struct inpcb *
tcp_notify(struct inpcb * inp,int error)1172 tcp_notify(struct inpcb *inp, int error)
1173 {
1174 struct tcpcb *tp;
1175
1176 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1177 INP_WLOCK_ASSERT(inp);
1178
1179 if ((inp->inp_flags & INP_TIMEWAIT) ||
1180 (inp->inp_flags & INP_DROPPED))
1181 return (inp);
1182
1183 tp = intotcpcb(inp);
1184 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1185
1186 /*
1187 * Ignore some errors if we are hooked up.
1188 * If connection hasn't completed, has retransmitted several times,
1189 * and receives a second error, give up now. This is better
1190 * than waiting a long time to establish a connection that
1191 * can never complete.
1192 */
1193 if (tp->t_state == TCPS_ESTABLISHED &&
1194 (error == EHOSTUNREACH || error == ENETUNREACH ||
1195 error == EHOSTDOWN)) {
1196 return (inp);
1197 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1198 tp->t_softerror) {
1199 tp = tcp_drop(tp, error);
1200 if (tp != NULL)
1201 return (inp);
1202 else
1203 return (NULL);
1204 } else {
1205 tp->t_softerror = error;
1206 return (inp);
1207 }
1208 #if 0
1209 wakeup( &so->so_timeo);
1210 sorwakeup(so);
1211 sowwakeup(so);
1212 #endif
1213 }
1214
1215 static int
tcp_pcblist(SYSCTL_HANDLER_ARGS)1216 tcp_pcblist(SYSCTL_HANDLER_ARGS)
1217 {
1218 int error, i, m, n, pcb_count;
1219 struct inpcb *inp, **inp_list;
1220 inp_gen_t gencnt;
1221 struct xinpgen xig;
1222
1223 /*
1224 * The process of preparing the TCB list is too time-consuming and
1225 * resource-intensive to repeat twice on every request.
1226 */
1227 if (req->oldptr == NULL) {
1228 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1229 n += imax(n / 8, 10);
1230 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1231 return (0);
1232 }
1233
1234 if (req->newptr != NULL)
1235 return (EPERM);
1236
1237 /*
1238 * OK, now we're committed to doing something.
1239 */
1240 INP_INFO_RLOCK(&V_tcbinfo);
1241 gencnt = V_tcbinfo.ipi_gencnt;
1242 n = V_tcbinfo.ipi_count;
1243 INP_INFO_RUNLOCK(&V_tcbinfo);
1244
1245 m = syncache_pcbcount();
1246
1247 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1248 + (n + m) * sizeof(struct xtcpcb));
1249 if (error != 0)
1250 return (error);
1251
1252 xig.xig_len = sizeof xig;
1253 xig.xig_count = n + m;
1254 xig.xig_gen = gencnt;
1255 xig.xig_sogen = so_gencnt;
1256 error = SYSCTL_OUT(req, &xig, sizeof xig);
1257 if (error)
1258 return (error);
1259
1260 error = syncache_pcblist(req, m, &pcb_count);
1261 if (error)
1262 return (error);
1263
1264 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1265 if (inp_list == NULL)
1266 return (ENOMEM);
1267
1268 INP_INFO_RLOCK(&V_tcbinfo);
1269 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1270 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1271 INP_WLOCK(inp);
1272 if (inp->inp_gencnt <= gencnt) {
1273 /*
1274 * XXX: This use of cr_cansee(), introduced with
1275 * TCP state changes, is not quite right, but for
1276 * now, better than nothing.
1277 */
1278 if (inp->inp_flags & INP_TIMEWAIT) {
1279 if (intotw(inp) != NULL)
1280 error = cr_cansee(req->td->td_ucred,
1281 intotw(inp)->tw_cred);
1282 else
1283 error = EINVAL; /* Skip this inp. */
1284 } else
1285 error = cr_canseeinpcb(req->td->td_ucred, inp);
1286 if (error == 0) {
1287 in_pcbref(inp);
1288 inp_list[i++] = inp;
1289 }
1290 }
1291 INP_WUNLOCK(inp);
1292 }
1293 INP_INFO_RUNLOCK(&V_tcbinfo);
1294 n = i;
1295
1296 error = 0;
1297 for (i = 0; i < n; i++) {
1298 inp = inp_list[i];
1299 INP_RLOCK(inp);
1300 if (inp->inp_gencnt <= gencnt) {
1301 struct xtcpcb xt;
1302 void *inp_ppcb;
1303
1304 bzero(&xt, sizeof(xt));
1305 xt.xt_len = sizeof xt;
1306 /* XXX should avoid extra copy */
1307 bcopy(inp, &xt.xt_inp, sizeof *inp);
1308 inp_ppcb = inp->inp_ppcb;
1309 if (inp_ppcb == NULL)
1310 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1311 else if (inp->inp_flags & INP_TIMEWAIT) {
1312 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1313 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1314 } else {
1315 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1316 if (xt.xt_tp.t_timers)
1317 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1318 }
1319 if (inp->inp_socket != NULL)
1320 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1321 else {
1322 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1323 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1324 }
1325 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1326 INP_RUNLOCK(inp);
1327 error = SYSCTL_OUT(req, &xt, sizeof xt);
1328 } else
1329 INP_RUNLOCK(inp);
1330 }
1331 INP_INFO_WLOCK(&V_tcbinfo);
1332 for (i = 0; i < n; i++) {
1333 inp = inp_list[i];
1334 INP_RLOCK(inp);
1335 if (!in_pcbrele_rlocked(inp))
1336 INP_RUNLOCK(inp);
1337 }
1338 INP_INFO_WUNLOCK(&V_tcbinfo);
1339
1340 if (!error) {
1341 /*
1342 * Give the user an updated idea of our state.
1343 * If the generation differs from what we told
1344 * her before, she knows that something happened
1345 * while we were processing this request, and it
1346 * might be necessary to retry.
1347 */
1348 INP_INFO_RLOCK(&V_tcbinfo);
1349 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1350 xig.xig_sogen = so_gencnt;
1351 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1352 INP_INFO_RUNLOCK(&V_tcbinfo);
1353 error = SYSCTL_OUT(req, &xig, sizeof xig);
1354 }
1355 free(inp_list, M_TEMP);
1356 return (error);
1357 }
1358
1359 SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1360 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1361 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1362
1363 #ifdef INET
1364 static int
tcp_getcred(SYSCTL_HANDLER_ARGS)1365 tcp_getcred(SYSCTL_HANDLER_ARGS)
1366 {
1367 struct xucred xuc;
1368 struct sockaddr_in addrs[2];
1369 struct inpcb *inp;
1370 int error;
1371
1372 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1373 if (error)
1374 return (error);
1375 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1376 if (error)
1377 return (error);
1378 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1379 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1380 if (inp != NULL) {
1381 if (inp->inp_socket == NULL)
1382 error = ENOENT;
1383 if (error == 0)
1384 error = cr_canseeinpcb(req->td->td_ucred, inp);
1385 if (error == 0)
1386 cru2x(inp->inp_cred, &xuc);
1387 INP_RUNLOCK(inp);
1388 } else
1389 error = ENOENT;
1390 if (error == 0)
1391 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1392 return (error);
1393 }
1394
1395 SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1396 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1397 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1398 #endif /* INET */
1399
1400 #ifdef INET6
1401 static int
tcp6_getcred(SYSCTL_HANDLER_ARGS)1402 tcp6_getcred(SYSCTL_HANDLER_ARGS)
1403 {
1404 struct xucred xuc;
1405 struct sockaddr_in6 addrs[2];
1406 struct inpcb *inp;
1407 int error;
1408 #ifdef INET
1409 int mapped = 0;
1410 #endif
1411
1412 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1413 if (error)
1414 return (error);
1415 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1416 if (error)
1417 return (error);
1418 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1419 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1420 return (error);
1421 }
1422 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1423 #ifdef INET
1424 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1425 mapped = 1;
1426 else
1427 #endif
1428 return (EINVAL);
1429 }
1430
1431 #ifdef INET
1432 if (mapped == 1)
1433 inp = in_pcblookup(&V_tcbinfo,
1434 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1435 addrs[1].sin6_port,
1436 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1437 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1438 else
1439 #endif
1440 inp = in6_pcblookup(&V_tcbinfo,
1441 &addrs[1].sin6_addr, addrs[1].sin6_port,
1442 &addrs[0].sin6_addr, addrs[0].sin6_port,
1443 INPLOOKUP_RLOCKPCB, NULL);
1444 if (inp != NULL) {
1445 if (inp->inp_socket == NULL)
1446 error = ENOENT;
1447 if (error == 0)
1448 error = cr_canseeinpcb(req->td->td_ucred, inp);
1449 if (error == 0)
1450 cru2x(inp->inp_cred, &xuc);
1451 INP_RUNLOCK(inp);
1452 } else
1453 error = ENOENT;
1454 if (error == 0)
1455 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1456 return (error);
1457 }
1458
1459 SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1460 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1461 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1462 #endif /* INET6 */
1463
1464
1465 #ifdef INET
1466 void
tcp_ctlinput(int cmd,struct sockaddr * sa,void * vip)1467 tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1468 {
1469 struct ip *ip = vip;
1470 struct tcphdr *th;
1471 struct in_addr faddr;
1472 struct inpcb *inp;
1473 struct tcpcb *tp;
1474 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1475 struct icmp *icp;
1476 struct in_conninfo inc;
1477 tcp_seq icmp_tcp_seq;
1478 int mtu;
1479
1480 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1481 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1482 return;
1483
1484 if (cmd == PRC_MSGSIZE)
1485 notify = tcp_mtudisc_notify;
1486 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1487 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1488 notify = tcp_drop_syn_sent;
1489 /*
1490 * Redirects don't need to be handled up here.
1491 */
1492 else if (PRC_IS_REDIRECT(cmd))
1493 return;
1494 /*
1495 * Source quench is depreciated.
1496 */
1497 else if (cmd == PRC_QUENCH)
1498 return;
1499 /*
1500 * Hostdead is ugly because it goes linearly through all PCBs.
1501 * XXX: We never get this from ICMP, otherwise it makes an
1502 * excellent DoS attack on machines with many connections.
1503 */
1504 else if (cmd == PRC_HOSTDEAD)
1505 ip = NULL;
1506 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1507 return;
1508 if (ip != NULL) {
1509 icp = (struct icmp *)((caddr_t)ip
1510 - offsetof(struct icmp, icmp_ip));
1511 th = (struct tcphdr *)((caddr_t)ip
1512 + (ip->ip_hl << 2));
1513 INP_INFO_WLOCK(&V_tcbinfo);
1514 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1515 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1516 if (inp != NULL) {
1517 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1518 !(inp->inp_flags & INP_DROPPED) &&
1519 !(inp->inp_socket == NULL)) {
1520 icmp_tcp_seq = htonl(th->th_seq);
1521 tp = intotcpcb(inp);
1522 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1523 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1524 if (cmd == PRC_MSGSIZE) {
1525 /*
1526 * MTU discovery:
1527 * If we got a needfrag set the MTU
1528 * in the route to the suggested new
1529 * value (if given) and then notify.
1530 */
1531 bzero(&inc, sizeof(inc));
1532 inc.inc_faddr = faddr;
1533 inc.inc_fibnum =
1534 inp->inp_inc.inc_fibnum;
1535
1536 mtu = ntohs(icp->icmp_nextmtu);
1537 /*
1538 * If no alternative MTU was
1539 * proposed, try the next smaller
1540 * one.
1541 */
1542 if (!mtu)
1543 mtu = ip_next_mtu(
1544 ntohs(ip->ip_len), 1);
1545 if (mtu < V_tcp_minmss
1546 + sizeof(struct tcpiphdr))
1547 mtu = V_tcp_minmss
1548 + sizeof(struct tcpiphdr);
1549 /*
1550 * Only cache the MTU if it
1551 * is smaller than the interface
1552 * or route MTU. tcp_mtudisc()
1553 * will do right thing by itself.
1554 */
1555 if (mtu <= tcp_maxmtu(&inc, NULL))
1556 tcp_hc_updatemtu(&inc, mtu);
1557 tcp_mtudisc(inp, mtu);
1558 } else
1559 inp = (*notify)(inp,
1560 inetctlerrmap[cmd]);
1561 }
1562 }
1563 if (inp != NULL)
1564 INP_WUNLOCK(inp);
1565 } else {
1566 bzero(&inc, sizeof(inc));
1567 inc.inc_fport = th->th_dport;
1568 inc.inc_lport = th->th_sport;
1569 inc.inc_faddr = faddr;
1570 inc.inc_laddr = ip->ip_src;
1571 syncache_unreach(&inc, th);
1572 }
1573 INP_INFO_WUNLOCK(&V_tcbinfo);
1574 } else
1575 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1576 }
1577 #endif /* INET */
1578
1579 #ifdef INET6
1580 void
tcp6_ctlinput(int cmd,struct sockaddr * sa,void * d)1581 tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1582 {
1583 struct tcphdr th;
1584 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1585 struct ip6_hdr *ip6;
1586 struct mbuf *m;
1587 struct ip6ctlparam *ip6cp = NULL;
1588 const struct sockaddr_in6 *sa6_src = NULL;
1589 int off;
1590 struct tcp_portonly {
1591 u_int16_t th_sport;
1592 u_int16_t th_dport;
1593 } *thp;
1594
1595 if (sa->sa_family != AF_INET6 ||
1596 sa->sa_len != sizeof(struct sockaddr_in6))
1597 return;
1598
1599 if (cmd == PRC_MSGSIZE)
1600 notify = tcp_mtudisc_notify;
1601 else if (!PRC_IS_REDIRECT(cmd) &&
1602 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1603 return;
1604 /* Source quench is depreciated. */
1605 else if (cmd == PRC_QUENCH)
1606 return;
1607
1608 /* if the parameter is from icmp6, decode it. */
1609 if (d != NULL) {
1610 ip6cp = (struct ip6ctlparam *)d;
1611 m = ip6cp->ip6c_m;
1612 ip6 = ip6cp->ip6c_ip6;
1613 off = ip6cp->ip6c_off;
1614 sa6_src = ip6cp->ip6c_src;
1615 } else {
1616 m = NULL;
1617 ip6 = NULL;
1618 off = 0; /* fool gcc */
1619 sa6_src = &sa6_any;
1620 }
1621
1622 if (ip6 != NULL) {
1623 struct in_conninfo inc;
1624 /*
1625 * XXX: We assume that when IPV6 is non NULL,
1626 * M and OFF are valid.
1627 */
1628
1629 /* check if we can safely examine src and dst ports */
1630 if (m->m_pkthdr.len < off + sizeof(*thp))
1631 return;
1632
1633 bzero(&th, sizeof(th));
1634 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1635
1636 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1637 (struct sockaddr *)ip6cp->ip6c_src,
1638 th.th_sport, cmd, NULL, notify);
1639
1640 bzero(&inc, sizeof(inc));
1641 inc.inc_fport = th.th_dport;
1642 inc.inc_lport = th.th_sport;
1643 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1644 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1645 inc.inc_flags |= INC_ISIPV6;
1646 INP_INFO_WLOCK(&V_tcbinfo);
1647 syncache_unreach(&inc, &th);
1648 INP_INFO_WUNLOCK(&V_tcbinfo);
1649 } else
1650 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1651 0, cmd, NULL, notify);
1652 }
1653 #endif /* INET6 */
1654
1655
1656 /*
1657 * Following is where TCP initial sequence number generation occurs.
1658 *
1659 * There are two places where we must use initial sequence numbers:
1660 * 1. In SYN-ACK packets.
1661 * 2. In SYN packets.
1662 *
1663 * All ISNs for SYN-ACK packets are generated by the syncache. See
1664 * tcp_syncache.c for details.
1665 *
1666 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1667 * depends on this property. In addition, these ISNs should be
1668 * unguessable so as to prevent connection hijacking. To satisfy
1669 * the requirements of this situation, the algorithm outlined in
1670 * RFC 1948 is used, with only small modifications.
1671 *
1672 * Implementation details:
1673 *
1674 * Time is based off the system timer, and is corrected so that it
1675 * increases by one megabyte per second. This allows for proper
1676 * recycling on high speed LANs while still leaving over an hour
1677 * before rollover.
1678 *
1679 * As reading the *exact* system time is too expensive to be done
1680 * whenever setting up a TCP connection, we increment the time
1681 * offset in two ways. First, a small random positive increment
1682 * is added to isn_offset for each connection that is set up.
1683 * Second, the function tcp_isn_tick fires once per clock tick
1684 * and increments isn_offset as necessary so that sequence numbers
1685 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1686 * random positive increments serve only to ensure that the same
1687 * exact sequence number is never sent out twice (as could otherwise
1688 * happen when a port is recycled in less than the system tick
1689 * interval.)
1690 *
1691 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1692 * between seeding of isn_secret. This is normally set to zero,
1693 * as reseeding should not be necessary.
1694 *
1695 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1696 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1697 * general, this means holding an exclusive (write) lock.
1698 */
1699
1700 #define ISN_BYTES_PER_SECOND 1048576
1701 #define ISN_STATIC_INCREMENT 4096
1702 #define ISN_RANDOM_INCREMENT (4096 - 1)
1703
1704 static VNET_DEFINE(u_char, isn_secret[32]);
1705 static VNET_DEFINE(int, isn_last);
1706 static VNET_DEFINE(int, isn_last_reseed);
1707 static VNET_DEFINE(u_int32_t, isn_offset);
1708 static VNET_DEFINE(u_int32_t, isn_offset_old);
1709
1710 #define V_isn_secret VNET(isn_secret)
1711 #define V_isn_last VNET(isn_last)
1712 #define V_isn_last_reseed VNET(isn_last_reseed)
1713 #define V_isn_offset VNET(isn_offset)
1714 #define V_isn_offset_old VNET(isn_offset_old)
1715
1716 tcp_seq
tcp_new_isn(struct tcpcb * tp)1717 tcp_new_isn(struct tcpcb *tp)
1718 {
1719 MD5_CTX isn_ctx;
1720 u_int32_t md5_buffer[4];
1721 tcp_seq new_isn;
1722 u_int32_t projected_offset;
1723
1724 INP_WLOCK_ASSERT(tp->t_inpcb);
1725
1726 ISN_LOCK();
1727 /* Seed if this is the first use, reseed if requested. */
1728 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1729 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1730 < (u_int)ticks))) {
1731 read_random(&V_isn_secret, sizeof(V_isn_secret));
1732 V_isn_last_reseed = ticks;
1733 }
1734
1735 /* Compute the md5 hash and return the ISN. */
1736 MD5Init(&isn_ctx);
1737 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1738 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1739 #ifdef INET6
1740 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1741 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1742 sizeof(struct in6_addr));
1743 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1744 sizeof(struct in6_addr));
1745 } else
1746 #endif
1747 {
1748 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1749 sizeof(struct in_addr));
1750 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1751 sizeof(struct in_addr));
1752 }
1753 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1754 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1755 new_isn = (tcp_seq) md5_buffer[0];
1756 V_isn_offset += ISN_STATIC_INCREMENT +
1757 (arc4random() & ISN_RANDOM_INCREMENT);
1758 if (ticks != V_isn_last) {
1759 projected_offset = V_isn_offset_old +
1760 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1761 if (SEQ_GT(projected_offset, V_isn_offset))
1762 V_isn_offset = projected_offset;
1763 V_isn_offset_old = V_isn_offset;
1764 V_isn_last = ticks;
1765 }
1766 new_isn += V_isn_offset;
1767 ISN_UNLOCK();
1768 return (new_isn);
1769 }
1770
1771 /*
1772 * When a specific ICMP unreachable message is received and the
1773 * connection state is SYN-SENT, drop the connection. This behavior
1774 * is controlled by the icmp_may_rst sysctl.
1775 */
1776 struct inpcb *
tcp_drop_syn_sent(struct inpcb * inp,int errno)1777 tcp_drop_syn_sent(struct inpcb *inp, int errno)
1778 {
1779 struct tcpcb *tp;
1780
1781 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1782 INP_WLOCK_ASSERT(inp);
1783
1784 if ((inp->inp_flags & INP_TIMEWAIT) ||
1785 (inp->inp_flags & INP_DROPPED))
1786 return (inp);
1787
1788 tp = intotcpcb(inp);
1789 if (tp->t_state != TCPS_SYN_SENT)
1790 return (inp);
1791
1792 tp = tcp_drop(tp, errno);
1793 if (tp != NULL)
1794 return (inp);
1795 else
1796 return (NULL);
1797 }
1798
1799 /*
1800 * When `need fragmentation' ICMP is received, update our idea of the MSS
1801 * based on the new value. Also nudge TCP to send something, since we
1802 * know the packet we just sent was dropped.
1803 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1804 */
1805 static struct inpcb *
tcp_mtudisc_notify(struct inpcb * inp,int error)1806 tcp_mtudisc_notify(struct inpcb *inp, int error)
1807 {
1808
1809 return (tcp_mtudisc(inp, -1));
1810 }
1811
1812 struct inpcb *
tcp_mtudisc(struct inpcb * inp,int mtuoffer)1813 tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1814 {
1815 struct tcpcb *tp;
1816 struct socket *so;
1817
1818 INP_WLOCK_ASSERT(inp);
1819 if ((inp->inp_flags & INP_TIMEWAIT) ||
1820 (inp->inp_flags & INP_DROPPED))
1821 return (inp);
1822
1823 tp = intotcpcb(inp);
1824 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1825
1826 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1827
1828 so = inp->inp_socket;
1829 SOCKBUF_LOCK(&so->so_snd);
1830 /* If the mss is larger than the socket buffer, decrease the mss. */
1831 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1832 tp->t_maxseg = so->so_snd.sb_hiwat;
1833 SOCKBUF_UNLOCK(&so->so_snd);
1834
1835 TCPSTAT_INC(tcps_mturesent);
1836 tp->t_rtttime = 0;
1837 tp->snd_nxt = tp->snd_una;
1838 tcp_free_sackholes(tp);
1839 tp->snd_recover = tp->snd_max;
1840 if (tp->t_flags & TF_SACK_PERMIT)
1841 EXIT_FASTRECOVERY(tp->t_flags);
1842 tcp_output(tp);
1843 return (inp);
1844 }
1845
1846 #ifdef INET
1847 /*
1848 * Look-up the routing entry to the peer of this inpcb. If no route
1849 * is found and it cannot be allocated, then return 0. This routine
1850 * is called by TCP routines that access the rmx structure and by
1851 * tcp_mss_update to get the peer/interface MTU.
1852 */
1853 u_long
tcp_maxmtu(struct in_conninfo * inc,struct tcp_ifcap * cap)1854 tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1855 {
1856 struct route sro;
1857 struct sockaddr_in *dst;
1858 struct ifnet *ifp;
1859 u_long maxmtu = 0;
1860
1861 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1862
1863 bzero(&sro, sizeof(sro));
1864 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1865 dst = (struct sockaddr_in *)&sro.ro_dst;
1866 dst->sin_family = AF_INET;
1867 dst->sin_len = sizeof(*dst);
1868 dst->sin_addr = inc->inc_faddr;
1869 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1870 }
1871 if (sro.ro_rt != NULL) {
1872 ifp = sro.ro_rt->rt_ifp;
1873 if (sro.ro_rt->rt_mtu == 0)
1874 maxmtu = ifp->if_mtu;
1875 else
1876 maxmtu = min(sro.ro_rt->rt_mtu, ifp->if_mtu);
1877
1878 /* Report additional interface capabilities. */
1879 if (cap != NULL) {
1880 if (ifp->if_capenable & IFCAP_TSO4 &&
1881 ifp->if_hwassist & CSUM_TSO) {
1882 cap->ifcap |= CSUM_TSO;
1883 cap->tsomax = ifp->if_hw_tsomax;
1884 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1885 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1886 }
1887 }
1888 RTFREE(sro.ro_rt);
1889 }
1890 return (maxmtu);
1891 }
1892 #endif /* INET */
1893
1894 #ifdef INET6
1895 u_long
tcp_maxmtu6(struct in_conninfo * inc,struct tcp_ifcap * cap)1896 tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1897 {
1898 struct route_in6 sro6;
1899 struct ifnet *ifp;
1900 u_long maxmtu = 0;
1901
1902 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1903
1904 bzero(&sro6, sizeof(sro6));
1905 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1906 sro6.ro_dst.sin6_family = AF_INET6;
1907 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1908 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1909 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1910 }
1911 if (sro6.ro_rt != NULL) {
1912 ifp = sro6.ro_rt->rt_ifp;
1913 if (sro6.ro_rt->rt_mtu == 0)
1914 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1915 else
1916 maxmtu = min(sro6.ro_rt->rt_mtu,
1917 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1918
1919 /* Report additional interface capabilities. */
1920 if (cap != NULL) {
1921 if (ifp->if_capenable & IFCAP_TSO6 &&
1922 ifp->if_hwassist & CSUM_TSO) {
1923 cap->ifcap |= CSUM_TSO;
1924 cap->tsomax = ifp->if_hw_tsomax;
1925 cap->tsomaxsegcount = ifp->if_hw_tsomaxsegcount;
1926 cap->tsomaxsegsize = ifp->if_hw_tsomaxsegsize;
1927 }
1928 }
1929 RTFREE(sro6.ro_rt);
1930 }
1931
1932 return (maxmtu);
1933 }
1934 #endif /* INET6 */
1935
1936 #ifdef IPSEC
1937 /* compute ESP/AH header size for TCP, including outer IP header. */
1938 size_t
ipsec_hdrsiz_tcp(struct tcpcb * tp)1939 ipsec_hdrsiz_tcp(struct tcpcb *tp)
1940 {
1941 struct inpcb *inp;
1942 struct mbuf *m;
1943 size_t hdrsiz;
1944 struct ip *ip;
1945 #ifdef INET6
1946 struct ip6_hdr *ip6;
1947 #endif
1948 struct tcphdr *th;
1949
1950 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1951 return (0);
1952 m = m_gethdr(M_NOWAIT, MT_DATA);
1953 if (!m)
1954 return (0);
1955
1956 #ifdef INET6
1957 if ((inp->inp_vflag & INP_IPV6) != 0) {
1958 ip6 = mtod(m, struct ip6_hdr *);
1959 th = (struct tcphdr *)(ip6 + 1);
1960 m->m_pkthdr.len = m->m_len =
1961 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1962 tcpip_fillheaders(inp, ip6, th);
1963 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1964 } else
1965 #endif /* INET6 */
1966 {
1967 ip = mtod(m, struct ip *);
1968 th = (struct tcphdr *)(ip + 1);
1969 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1970 tcpip_fillheaders(inp, ip, th);
1971 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1972 }
1973
1974 m_free(m);
1975 return (hdrsiz);
1976 }
1977 #endif /* IPSEC */
1978
1979 #ifdef TCP_SIGNATURE
1980 /*
1981 * Callback function invoked by m_apply() to digest TCP segment data
1982 * contained within an mbuf chain.
1983 */
1984 static int
tcp_signature_apply(void * fstate,void * data,u_int len)1985 tcp_signature_apply(void *fstate, void *data, u_int len)
1986 {
1987
1988 MD5Update(fstate, (u_char *)data, len);
1989 return (0);
1990 }
1991
1992 /*
1993 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1994 *
1995 * Parameters:
1996 * m pointer to head of mbuf chain
1997 * _unused
1998 * len length of TCP segment data, excluding options
1999 * optlen length of TCP segment options
2000 * buf pointer to storage for computed MD5 digest
2001 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2002 *
2003 * We do this over ip, tcphdr, segment data, and the key in the SADB.
2004 * When called from tcp_input(), we can be sure that th_sum has been
2005 * zeroed out and verified already.
2006 *
2007 * Return 0 if successful, otherwise return -1.
2008 *
2009 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
2010 * search with the destination IP address, and a 'magic SPI' to be
2011 * determined by the application. This is hardcoded elsewhere to 1179
2012 * right now. Another branch of this code exists which uses the SPD to
2013 * specify per-application flows but it is unstable.
2014 */
2015 int
tcp_signature_compute(struct mbuf * m,int _unused,int len,int optlen,u_char * buf,u_int direction)2016 tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
2017 u_char *buf, u_int direction)
2018 {
2019 union sockaddr_union dst;
2020 #ifdef INET
2021 struct ippseudo ippseudo;
2022 #endif
2023 MD5_CTX ctx;
2024 int doff;
2025 struct ip *ip;
2026 #ifdef INET
2027 struct ipovly *ipovly;
2028 #endif
2029 struct secasvar *sav;
2030 struct tcphdr *th;
2031 #ifdef INET6
2032 struct ip6_hdr *ip6;
2033 struct in6_addr in6;
2034 char ip6buf[INET6_ADDRSTRLEN];
2035 uint32_t plen;
2036 uint16_t nhdr;
2037 #endif
2038 u_short savecsum;
2039
2040 KASSERT(m != NULL, ("NULL mbuf chain"));
2041 KASSERT(buf != NULL, ("NULL signature pointer"));
2042
2043 /* Extract the destination from the IP header in the mbuf. */
2044 bzero(&dst, sizeof(union sockaddr_union));
2045 ip = mtod(m, struct ip *);
2046 #ifdef INET6
2047 ip6 = NULL; /* Make the compiler happy. */
2048 #endif
2049 switch (ip->ip_v) {
2050 #ifdef INET
2051 case IPVERSION:
2052 dst.sa.sa_len = sizeof(struct sockaddr_in);
2053 dst.sa.sa_family = AF_INET;
2054 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
2055 ip->ip_src : ip->ip_dst;
2056 break;
2057 #endif
2058 #ifdef INET6
2059 case (IPV6_VERSION >> 4):
2060 ip6 = mtod(m, struct ip6_hdr *);
2061 dst.sa.sa_len = sizeof(struct sockaddr_in6);
2062 dst.sa.sa_family = AF_INET6;
2063 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
2064 ip6->ip6_src : ip6->ip6_dst;
2065 break;
2066 #endif
2067 default:
2068 return (EINVAL);
2069 /* NOTREACHED */
2070 break;
2071 }
2072
2073 /* Look up an SADB entry which matches the address of the peer. */
2074 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2075 if (sav == NULL) {
2076 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2077 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2078 #ifdef INET6
2079 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2080 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2081 #endif
2082 "(unsupported)"));
2083 return (EINVAL);
2084 }
2085
2086 MD5Init(&ctx);
2087 /*
2088 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2089 *
2090 * XXX The ippseudo header MUST be digested in network byte order,
2091 * or else we'll fail the regression test. Assume all fields we've
2092 * been doing arithmetic on have been in host byte order.
2093 * XXX One cannot depend on ipovly->ih_len here. When called from
2094 * tcp_output(), the underlying ip_len member has not yet been set.
2095 */
2096 switch (ip->ip_v) {
2097 #ifdef INET
2098 case IPVERSION:
2099 ipovly = (struct ipovly *)ip;
2100 ippseudo.ippseudo_src = ipovly->ih_src;
2101 ippseudo.ippseudo_dst = ipovly->ih_dst;
2102 ippseudo.ippseudo_pad = 0;
2103 ippseudo.ippseudo_p = IPPROTO_TCP;
2104 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2105 optlen);
2106 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2107
2108 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2109 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2110 break;
2111 #endif
2112 #ifdef INET6
2113 /*
2114 * RFC 2385, 2.0 Proposal
2115 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2116 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2117 * extended next header value (to form 32 bits), and 32-bit segment
2118 * length.
2119 * Note: Upper-Layer Packet Length comes before Next Header.
2120 */
2121 case (IPV6_VERSION >> 4):
2122 in6 = ip6->ip6_src;
2123 in6_clearscope(&in6);
2124 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2125 in6 = ip6->ip6_dst;
2126 in6_clearscope(&in6);
2127 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2128 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2129 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2130 nhdr = 0;
2131 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2132 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2133 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2134 nhdr = IPPROTO_TCP;
2135 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2136
2137 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2138 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2139 break;
2140 #endif
2141 default:
2142 return (EINVAL);
2143 /* NOTREACHED */
2144 break;
2145 }
2146
2147
2148 /*
2149 * Step 2: Update MD5 hash with TCP header, excluding options.
2150 * The TCP checksum must be set to zero.
2151 */
2152 savecsum = th->th_sum;
2153 th->th_sum = 0;
2154 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2155 th->th_sum = savecsum;
2156
2157 /*
2158 * Step 3: Update MD5 hash with TCP segment data.
2159 * Use m_apply() to avoid an early m_pullup().
2160 */
2161 if (len > 0)
2162 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2163
2164 /*
2165 * Step 4: Update MD5 hash with shared secret.
2166 */
2167 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2168 MD5Final(buf, &ctx);
2169
2170 key_sa_recordxfer(sav, m);
2171 KEY_FREESAV(&sav);
2172 return (0);
2173 }
2174
2175 /*
2176 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2177 *
2178 * Parameters:
2179 * m pointer to head of mbuf chain
2180 * len length of TCP segment data, excluding options
2181 * optlen length of TCP segment options
2182 * buf pointer to storage for computed MD5 digest
2183 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2184 *
2185 * Return 1 if successful, otherwise return 0.
2186 */
2187 int
tcp_signature_verify(struct mbuf * m,int off0,int tlen,int optlen,struct tcpopt * to,struct tcphdr * th,u_int tcpbflag)2188 tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2189 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2190 {
2191 char tmpdigest[TCP_SIGLEN];
2192
2193 if (tcp_sig_checksigs == 0)
2194 return (1);
2195 if ((tcpbflag & TF_SIGNATURE) == 0) {
2196 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2197
2198 /*
2199 * If this socket is not expecting signature but
2200 * the segment contains signature just fail.
2201 */
2202 TCPSTAT_INC(tcps_sig_err_sigopt);
2203 TCPSTAT_INC(tcps_sig_rcvbadsig);
2204 return (0);
2205 }
2206
2207 /* Signature is not expected, and not present in segment. */
2208 return (1);
2209 }
2210
2211 /*
2212 * If this socket is expecting signature but the segment does not
2213 * contain any just fail.
2214 */
2215 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2216 TCPSTAT_INC(tcps_sig_err_nosigopt);
2217 TCPSTAT_INC(tcps_sig_rcvbadsig);
2218 return (0);
2219 }
2220 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2221 IPSEC_DIR_INBOUND) == -1) {
2222 TCPSTAT_INC(tcps_sig_err_buildsig);
2223 TCPSTAT_INC(tcps_sig_rcvbadsig);
2224 return (0);
2225 }
2226
2227 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2228 TCPSTAT_INC(tcps_sig_rcvbadsig);
2229 return (0);
2230 }
2231 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2232 return (1);
2233 }
2234 #endif /* TCP_SIGNATURE */
2235
2236 static int
sysctl_drop(SYSCTL_HANDLER_ARGS)2237 sysctl_drop(SYSCTL_HANDLER_ARGS)
2238 {
2239 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2240 struct sockaddr_storage addrs[2];
2241 struct inpcb *inp;
2242 struct tcpcb *tp;
2243 struct tcptw *tw;
2244 struct sockaddr_in *fin, *lin;
2245 #ifdef INET6
2246 struct sockaddr_in6 *fin6, *lin6;
2247 #endif
2248 int error;
2249
2250 inp = NULL;
2251 fin = lin = NULL;
2252 #ifdef INET6
2253 fin6 = lin6 = NULL;
2254 #endif
2255 error = 0;
2256
2257 if (req->oldptr != NULL || req->oldlen != 0)
2258 return (EINVAL);
2259 if (req->newptr == NULL)
2260 return (EPERM);
2261 if (req->newlen < sizeof(addrs))
2262 return (ENOMEM);
2263 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2264 if (error)
2265 return (error);
2266
2267 switch (addrs[0].ss_family) {
2268 #ifdef INET6
2269 case AF_INET6:
2270 fin6 = (struct sockaddr_in6 *)&addrs[0];
2271 lin6 = (struct sockaddr_in6 *)&addrs[1];
2272 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2273 lin6->sin6_len != sizeof(struct sockaddr_in6))
2274 return (EINVAL);
2275 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2276 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2277 return (EINVAL);
2278 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2279 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2280 fin = (struct sockaddr_in *)&addrs[0];
2281 lin = (struct sockaddr_in *)&addrs[1];
2282 break;
2283 }
2284 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2285 if (error)
2286 return (error);
2287 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2288 if (error)
2289 return (error);
2290 break;
2291 #endif
2292 #ifdef INET
2293 case AF_INET:
2294 fin = (struct sockaddr_in *)&addrs[0];
2295 lin = (struct sockaddr_in *)&addrs[1];
2296 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2297 lin->sin_len != sizeof(struct sockaddr_in))
2298 return (EINVAL);
2299 break;
2300 #endif
2301 default:
2302 return (EINVAL);
2303 }
2304 INP_INFO_WLOCK(&V_tcbinfo);
2305 switch (addrs[0].ss_family) {
2306 #ifdef INET6
2307 case AF_INET6:
2308 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2309 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2310 INPLOOKUP_WLOCKPCB, NULL);
2311 break;
2312 #endif
2313 #ifdef INET
2314 case AF_INET:
2315 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2316 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2317 break;
2318 #endif
2319 }
2320 if (inp != NULL) {
2321 if (inp->inp_flags & INP_TIMEWAIT) {
2322 /*
2323 * XXXRW: There currently exists a state where an
2324 * inpcb is present, but its timewait state has been
2325 * discarded. For now, don't allow dropping of this
2326 * type of inpcb.
2327 */
2328 tw = intotw(inp);
2329 if (tw != NULL)
2330 tcp_twclose(tw, 0);
2331 else
2332 INP_WUNLOCK(inp);
2333 } else if (!(inp->inp_flags & INP_DROPPED) &&
2334 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2335 tp = intotcpcb(inp);
2336 tp = tcp_drop(tp, ECONNABORTED);
2337 if (tp != NULL)
2338 INP_WUNLOCK(inp);
2339 } else
2340 INP_WUNLOCK(inp);
2341 } else
2342 error = ESRCH;
2343 INP_INFO_WUNLOCK(&V_tcbinfo);
2344 return (error);
2345 }
2346
2347 SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2348 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2349 0, sysctl_drop, "", "Drop TCP connection");
2350
2351 /*
2352 * Generate a standardized TCP log line for use throughout the
2353 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2354 * allow use in the interrupt context.
2355 *
2356 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2357 * NB: The function may return NULL if memory allocation failed.
2358 *
2359 * Due to header inclusion and ordering limitations the struct ip
2360 * and ip6_hdr pointers have to be passed as void pointers.
2361 */
2362 char *
tcp_log_vain(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2363 tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2364 const void *ip6hdr)
2365 {
2366
2367 /* Is logging enabled? */
2368 if (tcp_log_in_vain == 0)
2369 return (NULL);
2370
2371 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2372 }
2373
2374 char *
tcp_log_addrs(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2375 tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2376 const void *ip6hdr)
2377 {
2378
2379 /* Is logging enabled? */
2380 if (tcp_log_debug == 0)
2381 return (NULL);
2382
2383 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2384 }
2385
2386 static char *
tcp_log_addr(struct in_conninfo * inc,struct tcphdr * th,void * ip4hdr,const void * ip6hdr)2387 tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2388 const void *ip6hdr)
2389 {
2390 char *s, *sp;
2391 size_t size;
2392 struct ip *ip;
2393 #ifdef INET6
2394 const struct ip6_hdr *ip6;
2395
2396 ip6 = (const struct ip6_hdr *)ip6hdr;
2397 #endif /* INET6 */
2398 ip = (struct ip *)ip4hdr;
2399
2400 /*
2401 * The log line looks like this:
2402 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2403 */
2404 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2405 sizeof(PRINT_TH_FLAGS) + 1 +
2406 #ifdef INET6
2407 2 * INET6_ADDRSTRLEN;
2408 #else
2409 2 * INET_ADDRSTRLEN;
2410 #endif /* INET6 */
2411
2412 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2413 if (s == NULL)
2414 return (NULL);
2415
2416 strcat(s, "TCP: [");
2417 sp = s + strlen(s);
2418
2419 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2420 inet_ntoa_r(inc->inc_faddr, sp);
2421 sp = s + strlen(s);
2422 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2423 sp = s + strlen(s);
2424 inet_ntoa_r(inc->inc_laddr, sp);
2425 sp = s + strlen(s);
2426 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2427 #ifdef INET6
2428 } else if (inc) {
2429 ip6_sprintf(sp, &inc->inc6_faddr);
2430 sp = s + strlen(s);
2431 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2432 sp = s + strlen(s);
2433 ip6_sprintf(sp, &inc->inc6_laddr);
2434 sp = s + strlen(s);
2435 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2436 } else if (ip6 && th) {
2437 ip6_sprintf(sp, &ip6->ip6_src);
2438 sp = s + strlen(s);
2439 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2440 sp = s + strlen(s);
2441 ip6_sprintf(sp, &ip6->ip6_dst);
2442 sp = s + strlen(s);
2443 sprintf(sp, "]:%i", ntohs(th->th_dport));
2444 #endif /* INET6 */
2445 #ifdef INET
2446 } else if (ip && th) {
2447 inet_ntoa_r(ip->ip_src, sp);
2448 sp = s + strlen(s);
2449 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2450 sp = s + strlen(s);
2451 inet_ntoa_r(ip->ip_dst, sp);
2452 sp = s + strlen(s);
2453 sprintf(sp, "]:%i", ntohs(th->th_dport));
2454 #endif /* INET */
2455 } else {
2456 free(s, M_TCPLOG);
2457 return (NULL);
2458 }
2459 sp = s + strlen(s);
2460 if (th)
2461 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2462 if (*(s + size - 1) != '\0')
2463 panic("%s: string too long", __func__);
2464 return (s);
2465 }
2466
2467 /*
2468 * A subroutine which makes it easy to track TCP state changes with DTrace.
2469 * This function shouldn't be called for t_state initializations that don't
2470 * correspond to actual TCP state transitions.
2471 */
2472 void
tcp_state_change(struct tcpcb * tp,int newstate)2473 tcp_state_change(struct tcpcb *tp, int newstate)
2474 {
2475 #if defined(KDTRACE_HOOKS)
2476 int pstate = tp->t_state;
2477 #endif
2478
2479 tp->t_state = newstate;
2480 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2481 }
2482