69#include <netinet/in_pcb.h>
70#include <netinet/in_systm.h>
71#include <netinet/in_var.h>
72#include <netinet/ip.h>
73#include <netinet/ip_icmp.h>
74#include <netinet/ip_var.h>
75#ifdef INET6
76#include <netinet/ip6.h>
77#include <netinet6/in6_pcb.h>
78#include <netinet6/ip6_var.h>
79#include <netinet6/scope6_var.h>
80#include <netinet6/nd6.h>
81#endif
82
83#include <netinet/tcp_fsm.h>
84#include <netinet/tcp_seq.h>
85#include <netinet/tcp_timer.h>
86#include <netinet/tcp_var.h>
87#include <netinet/tcp_syncache.h>
88#ifdef INET6
89#include <netinet6/tcp6_var.h>
90#endif
91#include <netinet/tcpip.h>
92#ifdef TCPDEBUG
93#include <netinet/tcp_debug.h>
94#endif
95#ifdef INET6
96#include <netinet6/ip6protosw.h>
97#endif
98#ifdef TCP_OFFLOAD
99#include <netinet/tcp_offload.h>
100#endif
101
102#ifdef IPSEC
103#include <netipsec/ipsec.h>
104#include <netipsec/xform.h>
105#ifdef INET6
106#include <netipsec/ipsec6.h>
107#endif
108#include <netipsec/key.h>
109#include <sys/syslog.h>
110#endif /*IPSEC*/
111
112#include <machine/in_cksum.h>
113#include <sys/md5.h>
114
115#include <security/mac/mac_framework.h>
116
117VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
118#ifdef INET6
119VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
120#endif
121
122static int
123sysctl_net_inet_tcp_mss_check(SYSCTL_HANDLER_ARGS)
124{
125 int error, new;
126
127 new = V_tcp_mssdflt;
128 error = sysctl_handle_int(oidp, &new, 0, req);
129 if (error == 0 && req->newptr) {
130 if (new < TCP_MINMSS)
131 error = EINVAL;
132 else
133 V_tcp_mssdflt = new;
134 }
135 return (error);
136}
137
138SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_MSSDFLT, mssdflt,
139 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_mssdflt), 0,
140 &sysctl_net_inet_tcp_mss_check, "I",
141 "Default TCP Maximum Segment Size");
142
143#ifdef INET6
144static int
145sysctl_net_inet_tcp_mss_v6_check(SYSCTL_HANDLER_ARGS)
146{
147 int error, new;
148
149 new = V_tcp_v6mssdflt;
150 error = sysctl_handle_int(oidp, &new, 0, req);
151 if (error == 0 && req->newptr) {
152 if (new < TCP_MINMSS)
153 error = EINVAL;
154 else
155 V_tcp_v6mssdflt = new;
156 }
157 return (error);
158}
159
160SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_V6MSSDFLT, v6mssdflt,
161 CTLTYPE_INT|CTLFLAG_RW, &VNET_NAME(tcp_v6mssdflt), 0,
162 &sysctl_net_inet_tcp_mss_v6_check, "I",
163 "Default TCP Maximum Segment Size for IPv6");
164#endif /* INET6 */
165
166/*
167 * Minimum MSS we accept and use. This prevents DoS attacks where
168 * we are forced to a ridiculous low MSS like 20 and send hundreds
169 * of packets instead of one. The effect scales with the available
170 * bandwidth and quickly saturates the CPU and network interface
171 * with packet generation and sending. Set to zero to disable MINMSS
172 * checking. This setting prevents us from sending too small packets.
173 */
174VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
175SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
176 &VNET_NAME(tcp_minmss), 0,
177 "Minimum TCP Maximum Segment Size");
178
179VNET_DEFINE(int, tcp_do_rfc1323) = 1;
180SYSCTL_VNET_INT(_net_inet_tcp, TCPCTL_DO_RFC1323, rfc1323, CTLFLAG_RW,
181 &VNET_NAME(tcp_do_rfc1323), 0,
182 "Enable rfc1323 (high performance TCP) extensions");
183
184static int tcp_log_debug = 0;
185SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
186 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
187
188static int tcp_tcbhashsize = 0;
189SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
190 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
191
192static int do_tcpdrain = 1;
193SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_tcpdrain, CTLFLAG_RW, &do_tcpdrain, 0,
194 "Enable tcp_drain routine for extra help when low on mbufs");
195
196SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
197 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
198
199static VNET_DEFINE(int, icmp_may_rst) = 1;
200#define V_icmp_may_rst VNET(icmp_may_rst)
201SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, icmp_may_rst, CTLFLAG_RW,
202 &VNET_NAME(icmp_may_rst), 0,
203 "Certain ICMP unreachable messages may abort connections in SYN_SENT");
204
205static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
206#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
207SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, isn_reseed_interval, CTLFLAG_RW,
208 &VNET_NAME(tcp_isn_reseed_interval), 0,
209 "Seconds between reseeding of ISN secret");
210
211static int tcp_soreceive_stream = 0;
212SYSCTL_INT(_net_inet_tcp, OID_AUTO, soreceive_stream, CTLFLAG_RDTUN,
213 &tcp_soreceive_stream, 0, "Using soreceive_stream for TCP sockets");
214
215#ifdef TCP_SIGNATURE
216static int tcp_sig_checksigs = 1;
217SYSCTL_INT(_net_inet_tcp, OID_AUTO, signature_verify_input, CTLFLAG_RW,
218 &tcp_sig_checksigs, 0, "Verify RFC2385 digests on inbound traffic");
219#endif
220
221VNET_DEFINE(uma_zone_t, sack_hole_zone);
222#define V_sack_hole_zone VNET(sack_hole_zone)
223
224VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
225
226static struct inpcb *tcp_notify(struct inpcb *, int);
227static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
228static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
229 void *ip4hdr, const void *ip6hdr);
230
231/*
232 * Target size of TCP PCB hash tables. Must be a power of two.
233 *
234 * Note that this can be overridden by the kernel environment
235 * variable net.inet.tcp.tcbhashsize
236 */
237#ifndef TCBHASHSIZE
238#define TCBHASHSIZE 0
239#endif
240
241/*
242 * XXX
243 * Callouts should be moved into struct tcp directly. They are currently
244 * separate because the tcpcb structure is exported to userland for sysctl
245 * parsing purposes, which do not know about callouts.
246 */
247struct tcpcb_mem {
248 struct tcpcb tcb;
249 struct tcp_timer tt;
250 struct cc_var ccv;
251 struct osd osd;
252};
253
254static VNET_DEFINE(uma_zone_t, tcpcb_zone);
255#define V_tcpcb_zone VNET(tcpcb_zone)
256
257MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
258static struct mtx isn_mtx;
259
260#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
261#define ISN_LOCK() mtx_lock(&isn_mtx)
262#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
263
264/*
265 * TCP initialization.
266 */
267static void
268tcp_zone_change(void *tag)
269{
270
271 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
272 uma_zone_set_max(V_tcpcb_zone, maxsockets);
273 tcp_tw_zone_change();
274}
275
276static int
277tcp_inpcb_init(void *mem, int size, int flags)
278{
279 struct inpcb *inp = mem;
280
281 INP_LOCK_INIT(inp, "inp", "tcpinp");
282 return (0);
283}
284
285/*
286 * Take a value and get the next power of 2 that doesn't overflow.
287 * Used to size the tcp_inpcb hash buckets.
288 */
289static int
290maketcp_hashsize(int size)
291{
292 int hashsize;
293
294 /*
295 * auto tune.
296 * get the next power of 2 higher than maxsockets.
297 */
298 hashsize = 1 << fls(size);
299 /* catch overflow, and just go one power of 2 smaller */
300 if (hashsize < size) {
301 hashsize = 1 << (fls(size) - 1);
302 }
303 return (hashsize);
304}
305
306void
307tcp_init(void)
308{
309 const char *tcbhash_tuneable;
310 int hashsize;
311
312 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
313
314 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
315 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
316 printf("%s: WARNING: unable to register helper hook\n", __func__);
317 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
318 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 printf("%s: WARNING: unable to register helper hook\n", __func__);
320
321 hashsize = TCBHASHSIZE;
322 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
323 if (hashsize == 0) {
324 /*
325 * Auto tune the hash size based on maxsockets.
326 * A perfect hash would have a 1:1 mapping
327 * (hashsize = maxsockets) however it's been
328 * suggested that O(2) average is better.
329 */
330 hashsize = maketcp_hashsize(maxsockets / 4);
331 /*
332 * Our historical default is 512,
333 * do not autotune lower than this.
334 */
335 if (hashsize < 512)
336 hashsize = 512;
337 if (bootverbose)
338 printf("%s: %s auto tuned to %d\n", __func__,
339 tcbhash_tuneable, hashsize);
340 }
341 /*
342 * We require a hashsize to be a power of two.
343 * Previously if it was not a power of two we would just reset it
344 * back to 512, which could be a nasty surprise if you did not notice
345 * the error message.
346 * Instead what we do is clip it to the closest power of two lower
347 * than the specified hash value.
348 */
349 if (!powerof2(hashsize)) {
350 int oldhashsize = hashsize;
351
352 hashsize = maketcp_hashsize(hashsize);
353 /* prevent absurdly low value */
354 if (hashsize < 16)
355 hashsize = 16;
356 printf("%s: WARNING: TCB hash size not a power of 2, "
357 "clipped from %d to %d.\n", __func__, oldhashsize,
358 hashsize);
359 }
360 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
361 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
362 IPI_HASHFIELDS_4TUPLE);
363
364 /*
365 * These have to be type stable for the benefit of the timers.
366 */
367 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
368 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
369 uma_zone_set_max(V_tcpcb_zone, maxsockets);
370 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
371
372 tcp_tw_init();
373 syncache_init();
374 tcp_hc_init();
375 tcp_reass_init();
376
377 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
378 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
379 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
380
381 /* Skip initialization of globals for non-default instances. */
382 if (!IS_DEFAULT_VNET(curvnet))
383 return;
384
385 /* XXX virtualize those bellow? */
386 tcp_delacktime = TCPTV_DELACK;
387 tcp_keepinit = TCPTV_KEEP_INIT;
388 tcp_keepidle = TCPTV_KEEP_IDLE;
389 tcp_keepintvl = TCPTV_KEEPINTVL;
390 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
391 tcp_msl = TCPTV_MSL;
392 tcp_rexmit_min = TCPTV_MIN;
393 if (tcp_rexmit_min < 1)
394 tcp_rexmit_min = 1;
395 tcp_rexmit_slop = TCPTV_CPU_VAR;
396 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
397 tcp_tcbhashsize = hashsize;
398
399 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
400 if (tcp_soreceive_stream) {
401#ifdef INET
402 tcp_usrreqs.pru_soreceive = soreceive_stream;
403#endif
404#ifdef INET6
405 tcp6_usrreqs.pru_soreceive = soreceive_stream;
406#endif /* INET6 */
407 }
408
409#ifdef INET6
410#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
411#else /* INET6 */
412#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
413#endif /* INET6 */
414 if (max_protohdr < TCP_MINPROTOHDR)
415 max_protohdr = TCP_MINPROTOHDR;
416 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
417 panic("tcp_init");
418#undef TCP_MINPROTOHDR
419
420 ISN_LOCK_INIT();
421 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
422 SHUTDOWN_PRI_DEFAULT);
423 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
424 EVENTHANDLER_PRI_ANY);
425}
426
427#ifdef VIMAGE
428void
429tcp_destroy(void)
430{
431
432 tcp_reass_destroy();
433 tcp_hc_destroy();
434 syncache_destroy();
435 tcp_tw_destroy();
436 in_pcbinfo_destroy(&V_tcbinfo);
437 uma_zdestroy(V_sack_hole_zone);
438 uma_zdestroy(V_tcpcb_zone);
439}
440#endif
441
442void
443tcp_fini(void *xtp)
444{
445
446}
447
448/*
449 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
450 * tcp_template used to store this data in mbufs, but we now recopy it out
451 * of the tcpcb each time to conserve mbufs.
452 */
453void
454tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
455{
456 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
457
458 INP_WLOCK_ASSERT(inp);
459
460#ifdef INET6
461 if ((inp->inp_vflag & INP_IPV6) != 0) {
462 struct ip6_hdr *ip6;
463
464 ip6 = (struct ip6_hdr *)ip_ptr;
465 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
466 (inp->inp_flow & IPV6_FLOWINFO_MASK);
467 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
468 (IPV6_VERSION & IPV6_VERSION_MASK);
469 ip6->ip6_nxt = IPPROTO_TCP;
470 ip6->ip6_plen = htons(sizeof(struct tcphdr));
471 ip6->ip6_src = inp->in6p_laddr;
472 ip6->ip6_dst = inp->in6p_faddr;
473 }
474#endif /* INET6 */
475#if defined(INET6) && defined(INET)
476 else
477#endif
478#ifdef INET
479 {
480 struct ip *ip;
481
482 ip = (struct ip *)ip_ptr;
483 ip->ip_v = IPVERSION;
484 ip->ip_hl = 5;
485 ip->ip_tos = inp->inp_ip_tos;
486 ip->ip_len = 0;
487 ip->ip_id = 0;
488 ip->ip_off = 0;
489 ip->ip_ttl = inp->inp_ip_ttl;
490 ip->ip_sum = 0;
491 ip->ip_p = IPPROTO_TCP;
492 ip->ip_src = inp->inp_laddr;
493 ip->ip_dst = inp->inp_faddr;
494 }
495#endif /* INET */
496 th->th_sport = inp->inp_lport;
497 th->th_dport = inp->inp_fport;
498 th->th_seq = 0;
499 th->th_ack = 0;
500 th->th_x2 = 0;
501 th->th_off = 5;
502 th->th_flags = 0;
503 th->th_win = 0;
504 th->th_urp = 0;
505 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
506}
507
508/*
509 * Create template to be used to send tcp packets on a connection.
510 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
511 * use for this function is in keepalives, which use tcp_respond.
512 */
513struct tcptemp *
514tcpip_maketemplate(struct inpcb *inp)
515{
516 struct tcptemp *t;
517
518 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
519 if (t == NULL)
520 return (NULL);
521 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
522 return (t);
523}
524
525/*
526 * Send a single message to the TCP at address specified by
527 * the given TCP/IP header. If m == NULL, then we make a copy
528 * of the tcpiphdr at ti and send directly to the addressed host.
529 * This is used to force keep alive messages out using the TCP
530 * template for a connection. If flags are given then we send
531 * a message back to the TCP which originated the * segment ti,
532 * and discard the mbuf containing it and any other attached mbufs.
533 *
534 * In any case the ack and sequence number of the transmitted
535 * segment are as specified by the parameters.
536 *
537 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
538 */
539void
540tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
541 tcp_seq ack, tcp_seq seq, int flags)
542{
543 int tlen;
544 int win = 0;
545 struct ip *ip;
546 struct tcphdr *nth;
547#ifdef INET6
548 struct ip6_hdr *ip6;
549 int isipv6;
550#endif /* INET6 */
551 int ipflags = 0;
552 struct inpcb *inp;
553
554 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
555
556#ifdef INET6
557 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
558 ip6 = ipgen;
559#endif /* INET6 */
560 ip = ipgen;
561
562 if (tp != NULL) {
563 inp = tp->t_inpcb;
564 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
565 INP_WLOCK_ASSERT(inp);
566 } else
567 inp = NULL;
568
569 if (tp != NULL) {
570 if (!(flags & TH_RST)) {
571 win = sbspace(&inp->inp_socket->so_rcv);
572 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
573 win = (long)TCP_MAXWIN << tp->rcv_scale;
574 }
575 }
576 if (m == NULL) {
577 m = m_gethdr(M_NOWAIT, MT_DATA);
578 if (m == NULL)
579 return;
580 tlen = 0;
581 m->m_data += max_linkhdr;
582#ifdef INET6
583 if (isipv6) {
584 bcopy((caddr_t)ip6, mtod(m, caddr_t),
585 sizeof(struct ip6_hdr));
586 ip6 = mtod(m, struct ip6_hdr *);
587 nth = (struct tcphdr *)(ip6 + 1);
588 } else
589#endif /* INET6 */
590 {
591 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
592 ip = mtod(m, struct ip *);
593 nth = (struct tcphdr *)(ip + 1);
594 }
595 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
596 flags = TH_ACK;
597 } else {
598 /*
599 * reuse the mbuf.
600 * XXX MRT We inherrit the FIB, which is lucky.
601 */
602 m_freem(m->m_next);
603 m->m_next = NULL;
604 m->m_data = (caddr_t)ipgen;
605 /* m_len is set later */
606 tlen = 0;
607#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
608#ifdef INET6
609 if (isipv6) {
610 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
611 nth = (struct tcphdr *)(ip6 + 1);
612 } else
613#endif /* INET6 */
614 {
615 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
616 nth = (struct tcphdr *)(ip + 1);
617 }
618 if (th != nth) {
619 /*
620 * this is usually a case when an extension header
621 * exists between the IPv6 header and the
622 * TCP header.
623 */
624 nth->th_sport = th->th_sport;
625 nth->th_dport = th->th_dport;
626 }
627 xchg(nth->th_dport, nth->th_sport, uint16_t);
628#undef xchg
629 }
630#ifdef INET6
631 if (isipv6) {
632 ip6->ip6_flow = 0;
633 ip6->ip6_vfc = IPV6_VERSION;
634 ip6->ip6_nxt = IPPROTO_TCP;
| 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
120VNET_DEFINE(int, tcp_mssdflt) = TCP_MSS;
121#ifdef INET6
122VNET_DEFINE(int, tcp_v6mssdflt) = TCP6_MSS;
123#endif
124
125static int
126sysctl_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
141SYSCTL_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
147static int
148sysctl_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
163SYSCTL_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 */
177VNET_DEFINE(int, tcp_minmss) = TCP_MINMSS;
178SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, minmss, CTLFLAG_RW,
179 &VNET_NAME(tcp_minmss), 0,
180 "Minimum TCP Maximum Segment Size");
181
182VNET_DEFINE(int, tcp_do_rfc1323) = 1;
183SYSCTL_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
187static int tcp_log_debug = 0;
188SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_debug, CTLFLAG_RW,
189 &tcp_log_debug, 0, "Log errors caused by incoming TCP segments");
190
191static int tcp_tcbhashsize = 0;
192SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcbhashsize, CTLFLAG_RDTUN,
193 &tcp_tcbhashsize, 0, "Size of TCP control-block hashtable");
194
195static int do_tcpdrain = 1;
196SYSCTL_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
199SYSCTL_VNET_UINT(_net_inet_tcp, OID_AUTO, pcbcount, CTLFLAG_RD,
200 &VNET_NAME(tcbinfo.ipi_count), 0, "Number of active PCBs");
201
202static VNET_DEFINE(int, icmp_may_rst) = 1;
203#define V_icmp_may_rst VNET(icmp_may_rst)
204SYSCTL_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
208static VNET_DEFINE(int, tcp_isn_reseed_interval) = 0;
209#define V_tcp_isn_reseed_interval VNET(tcp_isn_reseed_interval)
210SYSCTL_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
214static int tcp_soreceive_stream = 0;
215SYSCTL_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
219static int tcp_sig_checksigs = 1;
220SYSCTL_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
224VNET_DEFINE(uma_zone_t, sack_hole_zone);
225#define V_sack_hole_zone VNET(sack_hole_zone)
226
227VNET_DEFINE(struct hhook_head *, tcp_hhh[HHOOK_TCP_LAST+1]);
228
229static struct inpcb *tcp_notify(struct inpcb *, int);
230static struct inpcb *tcp_mtudisc_notify(struct inpcb *, int);
231static char * tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th,
232 void *ip4hdr, const void *ip6hdr);
233
234/*
235 * Target size of TCP PCB hash tables. Must be a power of two.
236 *
237 * Note that this can be overridden by the kernel environment
238 * variable net.inet.tcp.tcbhashsize
239 */
240#ifndef TCBHASHSIZE
241#define TCBHASHSIZE 0
242#endif
243
244/*
245 * XXX
246 * Callouts should be moved into struct tcp directly. They are currently
247 * separate because the tcpcb structure is exported to userland for sysctl
248 * parsing purposes, which do not know about callouts.
249 */
250struct tcpcb_mem {
251 struct tcpcb tcb;
252 struct tcp_timer tt;
253 struct cc_var ccv;
254 struct osd osd;
255};
256
257static VNET_DEFINE(uma_zone_t, tcpcb_zone);
258#define V_tcpcb_zone VNET(tcpcb_zone)
259
260MALLOC_DEFINE(M_TCPLOG, "tcplog", "TCP address and flags print buffers");
261static struct mtx isn_mtx;
262
263#define ISN_LOCK_INIT() mtx_init(&isn_mtx, "isn_mtx", NULL, MTX_DEF)
264#define ISN_LOCK() mtx_lock(&isn_mtx)
265#define ISN_UNLOCK() mtx_unlock(&isn_mtx)
266
267/*
268 * TCP initialization.
269 */
270static void
271tcp_zone_change(void *tag)
272{
273
274 uma_zone_set_max(V_tcbinfo.ipi_zone, maxsockets);
275 uma_zone_set_max(V_tcpcb_zone, maxsockets);
276 tcp_tw_zone_change();
277}
278
279static int
280tcp_inpcb_init(void *mem, int size, int flags)
281{
282 struct inpcb *inp = mem;
283
284 INP_LOCK_INIT(inp, "inp", "tcpinp");
285 return (0);
286}
287
288/*
289 * Take a value and get the next power of 2 that doesn't overflow.
290 * Used to size the tcp_inpcb hash buckets.
291 */
292static int
293maketcp_hashsize(int size)
294{
295 int hashsize;
296
297 /*
298 * auto tune.
299 * get the next power of 2 higher than maxsockets.
300 */
301 hashsize = 1 << fls(size);
302 /* catch overflow, and just go one power of 2 smaller */
303 if (hashsize < size) {
304 hashsize = 1 << (fls(size) - 1);
305 }
306 return (hashsize);
307}
308
309void
310tcp_init(void)
311{
312 const char *tcbhash_tuneable;
313 int hashsize;
314
315 tcbhash_tuneable = "net.inet.tcp.tcbhashsize";
316
317 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN,
318 &V_tcp_hhh[HHOOK_TCP_EST_IN], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
319 printf("%s: WARNING: unable to register helper hook\n", __func__);
320 if (hhook_head_register(HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT,
321 &V_tcp_hhh[HHOOK_TCP_EST_OUT], HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
322 printf("%s: WARNING: unable to register helper hook\n", __func__);
323
324 hashsize = TCBHASHSIZE;
325 TUNABLE_INT_FETCH(tcbhash_tuneable, &hashsize);
326 if (hashsize == 0) {
327 /*
328 * Auto tune the hash size based on maxsockets.
329 * A perfect hash would have a 1:1 mapping
330 * (hashsize = maxsockets) however it's been
331 * suggested that O(2) average is better.
332 */
333 hashsize = maketcp_hashsize(maxsockets / 4);
334 /*
335 * Our historical default is 512,
336 * do not autotune lower than this.
337 */
338 if (hashsize < 512)
339 hashsize = 512;
340 if (bootverbose)
341 printf("%s: %s auto tuned to %d\n", __func__,
342 tcbhash_tuneable, hashsize);
343 }
344 /*
345 * We require a hashsize to be a power of two.
346 * Previously if it was not a power of two we would just reset it
347 * back to 512, which could be a nasty surprise if you did not notice
348 * the error message.
349 * Instead what we do is clip it to the closest power of two lower
350 * than the specified hash value.
351 */
352 if (!powerof2(hashsize)) {
353 int oldhashsize = hashsize;
354
355 hashsize = maketcp_hashsize(hashsize);
356 /* prevent absurdly low value */
357 if (hashsize < 16)
358 hashsize = 16;
359 printf("%s: WARNING: TCB hash size not a power of 2, "
360 "clipped from %d to %d.\n", __func__, oldhashsize,
361 hashsize);
362 }
363 in_pcbinfo_init(&V_tcbinfo, "tcp", &V_tcb, hashsize, hashsize,
364 "tcp_inpcb", tcp_inpcb_init, NULL, UMA_ZONE_NOFREE,
365 IPI_HASHFIELDS_4TUPLE);
366
367 /*
368 * These have to be type stable for the benefit of the timers.
369 */
370 V_tcpcb_zone = uma_zcreate("tcpcb", sizeof(struct tcpcb_mem),
371 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
372 uma_zone_set_max(V_tcpcb_zone, maxsockets);
373 uma_zone_set_warning(V_tcpcb_zone, "kern.ipc.maxsockets limit reached");
374
375 tcp_tw_init();
376 syncache_init();
377 tcp_hc_init();
378 tcp_reass_init();
379
380 TUNABLE_INT_FETCH("net.inet.tcp.sack.enable", &V_tcp_do_sack);
381 V_sack_hole_zone = uma_zcreate("sackhole", sizeof(struct sackhole),
382 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
383
384 /* Skip initialization of globals for non-default instances. */
385 if (!IS_DEFAULT_VNET(curvnet))
386 return;
387
388 /* XXX virtualize those bellow? */
389 tcp_delacktime = TCPTV_DELACK;
390 tcp_keepinit = TCPTV_KEEP_INIT;
391 tcp_keepidle = TCPTV_KEEP_IDLE;
392 tcp_keepintvl = TCPTV_KEEPINTVL;
393 tcp_maxpersistidle = TCPTV_KEEP_IDLE;
394 tcp_msl = TCPTV_MSL;
395 tcp_rexmit_min = TCPTV_MIN;
396 if (tcp_rexmit_min < 1)
397 tcp_rexmit_min = 1;
398 tcp_rexmit_slop = TCPTV_CPU_VAR;
399 tcp_finwait2_timeout = TCPTV_FINWAIT2_TIMEOUT;
400 tcp_tcbhashsize = hashsize;
401
402 TUNABLE_INT_FETCH("net.inet.tcp.soreceive_stream", &tcp_soreceive_stream);
403 if (tcp_soreceive_stream) {
404#ifdef INET
405 tcp_usrreqs.pru_soreceive = soreceive_stream;
406#endif
407#ifdef INET6
408 tcp6_usrreqs.pru_soreceive = soreceive_stream;
409#endif /* INET6 */
410 }
411
412#ifdef INET6
413#define TCP_MINPROTOHDR (sizeof(struct ip6_hdr) + sizeof(struct tcphdr))
414#else /* INET6 */
415#define TCP_MINPROTOHDR (sizeof(struct tcpiphdr))
416#endif /* INET6 */
417 if (max_protohdr < TCP_MINPROTOHDR)
418 max_protohdr = TCP_MINPROTOHDR;
419 if (max_linkhdr + TCP_MINPROTOHDR > MHLEN)
420 panic("tcp_init");
421#undef TCP_MINPROTOHDR
422
423 ISN_LOCK_INIT();
424 EVENTHANDLER_REGISTER(shutdown_pre_sync, tcp_fini, NULL,
425 SHUTDOWN_PRI_DEFAULT);
426 EVENTHANDLER_REGISTER(maxsockets_change, tcp_zone_change, NULL,
427 EVENTHANDLER_PRI_ANY);
428}
429
430#ifdef VIMAGE
431void
432tcp_destroy(void)
433{
434
435 tcp_reass_destroy();
436 tcp_hc_destroy();
437 syncache_destroy();
438 tcp_tw_destroy();
439 in_pcbinfo_destroy(&V_tcbinfo);
440 uma_zdestroy(V_sack_hole_zone);
441 uma_zdestroy(V_tcpcb_zone);
442}
443#endif
444
445void
446tcp_fini(void *xtp)
447{
448
449}
450
451/*
452 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
453 * tcp_template used to store this data in mbufs, but we now recopy it out
454 * of the tcpcb each time to conserve mbufs.
455 */
456void
457tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
458{
459 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
460
461 INP_WLOCK_ASSERT(inp);
462
463#ifdef INET6
464 if ((inp->inp_vflag & INP_IPV6) != 0) {
465 struct ip6_hdr *ip6;
466
467 ip6 = (struct ip6_hdr *)ip_ptr;
468 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
469 (inp->inp_flow & IPV6_FLOWINFO_MASK);
470 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
471 (IPV6_VERSION & IPV6_VERSION_MASK);
472 ip6->ip6_nxt = IPPROTO_TCP;
473 ip6->ip6_plen = htons(sizeof(struct tcphdr));
474 ip6->ip6_src = inp->in6p_laddr;
475 ip6->ip6_dst = inp->in6p_faddr;
476 }
477#endif /* INET6 */
478#if defined(INET6) && defined(INET)
479 else
480#endif
481#ifdef INET
482 {
483 struct ip *ip;
484
485 ip = (struct ip *)ip_ptr;
486 ip->ip_v = IPVERSION;
487 ip->ip_hl = 5;
488 ip->ip_tos = inp->inp_ip_tos;
489 ip->ip_len = 0;
490 ip->ip_id = 0;
491 ip->ip_off = 0;
492 ip->ip_ttl = inp->inp_ip_ttl;
493 ip->ip_sum = 0;
494 ip->ip_p = IPPROTO_TCP;
495 ip->ip_src = inp->inp_laddr;
496 ip->ip_dst = inp->inp_faddr;
497 }
498#endif /* INET */
499 th->th_sport = inp->inp_lport;
500 th->th_dport = inp->inp_fport;
501 th->th_seq = 0;
502 th->th_ack = 0;
503 th->th_x2 = 0;
504 th->th_off = 5;
505 th->th_flags = 0;
506 th->th_win = 0;
507 th->th_urp = 0;
508 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
509}
510
511/*
512 * Create template to be used to send tcp packets on a connection.
513 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
514 * use for this function is in keepalives, which use tcp_respond.
515 */
516struct tcptemp *
517tcpip_maketemplate(struct inpcb *inp)
518{
519 struct tcptemp *t;
520
521 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
522 if (t == NULL)
523 return (NULL);
524 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
525 return (t);
526}
527
528/*
529 * Send a single message to the TCP at address specified by
530 * the given TCP/IP header. If m == NULL, then we make a copy
531 * of the tcpiphdr at ti and send directly to the addressed host.
532 * This is used to force keep alive messages out using the TCP
533 * template for a connection. If flags are given then we send
534 * a message back to the TCP which originated the * segment ti,
535 * and discard the mbuf containing it and any other attached mbufs.
536 *
537 * In any case the ack and sequence number of the transmitted
538 * segment are as specified by the parameters.
539 *
540 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
541 */
542void
543tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
544 tcp_seq ack, tcp_seq seq, int flags)
545{
546 int tlen;
547 int win = 0;
548 struct ip *ip;
549 struct tcphdr *nth;
550#ifdef INET6
551 struct ip6_hdr *ip6;
552 int isipv6;
553#endif /* INET6 */
554 int ipflags = 0;
555 struct inpcb *inp;
556
557 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
558
559#ifdef INET6
560 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
561 ip6 = ipgen;
562#endif /* INET6 */
563 ip = ipgen;
564
565 if (tp != NULL) {
566 inp = tp->t_inpcb;
567 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
568 INP_WLOCK_ASSERT(inp);
569 } else
570 inp = NULL;
571
572 if (tp != NULL) {
573 if (!(flags & TH_RST)) {
574 win = sbspace(&inp->inp_socket->so_rcv);
575 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
576 win = (long)TCP_MAXWIN << tp->rcv_scale;
577 }
578 }
579 if (m == NULL) {
580 m = m_gethdr(M_NOWAIT, MT_DATA);
581 if (m == NULL)
582 return;
583 tlen = 0;
584 m->m_data += max_linkhdr;
585#ifdef INET6
586 if (isipv6) {
587 bcopy((caddr_t)ip6, mtod(m, caddr_t),
588 sizeof(struct ip6_hdr));
589 ip6 = mtod(m, struct ip6_hdr *);
590 nth = (struct tcphdr *)(ip6 + 1);
591 } else
592#endif /* INET6 */
593 {
594 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
595 ip = mtod(m, struct ip *);
596 nth = (struct tcphdr *)(ip + 1);
597 }
598 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
599 flags = TH_ACK;
600 } else {
601 /*
602 * reuse the mbuf.
603 * XXX MRT We inherrit the FIB, which is lucky.
604 */
605 m_freem(m->m_next);
606 m->m_next = NULL;
607 m->m_data = (caddr_t)ipgen;
608 /* m_len is set later */
609 tlen = 0;
610#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
611#ifdef INET6
612 if (isipv6) {
613 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
614 nth = (struct tcphdr *)(ip6 + 1);
615 } else
616#endif /* INET6 */
617 {
618 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
619 nth = (struct tcphdr *)(ip + 1);
620 }
621 if (th != nth) {
622 /*
623 * this is usually a case when an extension header
624 * exists between the IPv6 header and the
625 * TCP header.
626 */
627 nth->th_sport = th->th_sport;
628 nth->th_dport = th->th_dport;
629 }
630 xchg(nth->th_dport, nth->th_sport, uint16_t);
631#undef xchg
632 }
633#ifdef INET6
634 if (isipv6) {
635 ip6->ip6_flow = 0;
636 ip6->ip6_vfc = IPV6_VERSION;
637 ip6->ip6_nxt = IPPROTO_TCP;
|
886 (void) tcp_output(tp);
887 TCPSTAT_INC(tcps_drops);
888 } else
889 TCPSTAT_INC(tcps_conndrops);
890 if (errno == ETIMEDOUT && tp->t_softerror)
891 errno = tp->t_softerror;
892 so->so_error = errno;
893 return (tcp_close(tp));
894}
895
896void
897tcp_discardcb(struct tcpcb *tp)
898{
899 struct inpcb *inp = tp->t_inpcb;
900 struct socket *so = inp->inp_socket;
901#ifdef INET6
902 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
903#endif /* INET6 */
904
905 INP_WLOCK_ASSERT(inp);
906
907 /*
908 * Make sure that all of our timers are stopped before we delete the
909 * PCB.
910 *
911 * XXXRW: Really, we would like to use callout_drain() here in order
912 * to avoid races experienced in tcp_timer.c where a timer is already
913 * executing at this point. However, we can't, both because we're
914 * running in a context where we can't sleep, and also because we
915 * hold locks required by the timers. What we instead need to do is
916 * test to see if callout_drain() is required, and if so, defer some
917 * portion of the remainder of tcp_discardcb() to an asynchronous
918 * context that can callout_drain() and then continue. Some care
919 * will be required to ensure that no further processing takes place
920 * on the tcpcb, even though it hasn't been freed (a flag?).
921 */
922 callout_stop(&tp->t_timers->tt_rexmt);
923 callout_stop(&tp->t_timers->tt_persist);
924 callout_stop(&tp->t_timers->tt_keep);
925 callout_stop(&tp->t_timers->tt_2msl);
926 callout_stop(&tp->t_timers->tt_delack);
927
928 /*
929 * If we got enough samples through the srtt filter,
930 * save the rtt and rttvar in the routing entry.
931 * 'Enough' is arbitrarily defined as 4 rtt samples.
932 * 4 samples is enough for the srtt filter to converge
933 * to within enough % of the correct value; fewer samples
934 * and we could save a bogus rtt. The danger is not high
935 * as tcp quickly recovers from everything.
936 * XXX: Works very well but needs some more statistics!
937 */
938 if (tp->t_rttupdated >= 4) {
939 struct hc_metrics_lite metrics;
940 u_long ssthresh;
941
942 bzero(&metrics, sizeof(metrics));
943 /*
944 * Update the ssthresh always when the conditions below
945 * are satisfied. This gives us better new start value
946 * for the congestion avoidance for new connections.
947 * ssthresh is only set if packet loss occured on a session.
948 *
949 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
950 * being torn down. Ideally this code would not use 'so'.
951 */
952 ssthresh = tp->snd_ssthresh;
953 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
954 /*
955 * convert the limit from user data bytes to
956 * packets then to packet data bytes.
957 */
958 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
959 if (ssthresh < 2)
960 ssthresh = 2;
961 ssthresh *= (u_long)(tp->t_maxseg +
962#ifdef INET6
963 (isipv6 ? sizeof (struct ip6_hdr) +
964 sizeof (struct tcphdr) :
965#endif
966 sizeof (struct tcpiphdr)
967#ifdef INET6
968 )
969#endif
970 );
971 } else
972 ssthresh = 0;
973 metrics.rmx_ssthresh = ssthresh;
974
975 metrics.rmx_rtt = tp->t_srtt;
976 metrics.rmx_rttvar = tp->t_rttvar;
977 metrics.rmx_cwnd = tp->snd_cwnd;
978 metrics.rmx_sendpipe = 0;
979 metrics.rmx_recvpipe = 0;
980
981 tcp_hc_update(&inp->inp_inc, &metrics);
982 }
983
984 /* free the reassembly queue, if any */
985 tcp_reass_flush(tp);
986
987#ifdef TCP_OFFLOAD
988 /* Disconnect offload device, if any. */
989 if (tp->t_flags & TF_TOE)
990 tcp_offload_detach(tp);
991#endif
992
993 tcp_free_sackholes(tp);
994
995 /* Allow the CC algorithm to clean up after itself. */
996 if (CC_ALGO(tp)->cb_destroy != NULL)
997 CC_ALGO(tp)->cb_destroy(tp->ccv);
998
999 khelp_destroy_osd(tp->osd);
1000
1001 CC_ALGO(tp) = NULL;
1002 inp->inp_ppcb = NULL;
1003 tp->t_inpcb = NULL;
1004 uma_zfree(V_tcpcb_zone, tp);
1005}
1006
1007/*
1008 * Attempt to close a TCP control block, marking it as dropped, and freeing
1009 * the socket if we hold the only reference.
1010 */
1011struct tcpcb *
1012tcp_close(struct tcpcb *tp)
1013{
1014 struct inpcb *inp = tp->t_inpcb;
1015 struct socket *so;
1016
1017 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1018 INP_WLOCK_ASSERT(inp);
1019
1020#ifdef TCP_OFFLOAD
1021 if (tp->t_state == TCPS_LISTEN)
1022 tcp_offload_listen_stop(tp);
1023#endif
1024 in_pcbdrop(inp);
1025 TCPSTAT_INC(tcps_closed);
1026 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1027 so = inp->inp_socket;
1028 soisdisconnected(so);
1029 if (inp->inp_flags & INP_SOCKREF) {
1030 KASSERT(so->so_state & SS_PROTOREF,
1031 ("tcp_close: !SS_PROTOREF"));
1032 inp->inp_flags &= ~INP_SOCKREF;
1033 INP_WUNLOCK(inp);
1034 ACCEPT_LOCK();
1035 SOCK_LOCK(so);
1036 so->so_state &= ~SS_PROTOREF;
1037 sofree(so);
1038 return (NULL);
1039 }
1040 return (tp);
1041}
1042
1043void
1044tcp_drain(void)
1045{
1046 VNET_ITERATOR_DECL(vnet_iter);
1047
1048 if (!do_tcpdrain)
1049 return;
1050
1051 VNET_LIST_RLOCK_NOSLEEP();
1052 VNET_FOREACH(vnet_iter) {
1053 CURVNET_SET(vnet_iter);
1054 struct inpcb *inpb;
1055 struct tcpcb *tcpb;
1056
1057 /*
1058 * Walk the tcpbs, if existing, and flush the reassembly queue,
1059 * if there is one...
1060 * XXX: The "Net/3" implementation doesn't imply that the TCP
1061 * reassembly queue should be flushed, but in a situation
1062 * where we're really low on mbufs, this is potentially
1063 * useful.
1064 */
1065 INP_INFO_RLOCK(&V_tcbinfo);
1066 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1067 if (inpb->inp_flags & INP_TIMEWAIT)
1068 continue;
1069 INP_WLOCK(inpb);
1070 if ((tcpb = intotcpcb(inpb)) != NULL) {
1071 tcp_reass_flush(tcpb);
1072 tcp_clean_sackreport(tcpb);
1073 }
1074 INP_WUNLOCK(inpb);
1075 }
1076 INP_INFO_RUNLOCK(&V_tcbinfo);
1077 CURVNET_RESTORE();
1078 }
1079 VNET_LIST_RUNLOCK_NOSLEEP();
1080}
1081
1082/*
1083 * Notify a tcp user of an asynchronous error;
1084 * store error as soft error, but wake up user
1085 * (for now, won't do anything until can select for soft error).
1086 *
1087 * Do not wake up user since there currently is no mechanism for
1088 * reporting soft errors (yet - a kqueue filter may be added).
1089 */
1090static struct inpcb *
1091tcp_notify(struct inpcb *inp, int error)
1092{
1093 struct tcpcb *tp;
1094
1095 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1096 INP_WLOCK_ASSERT(inp);
1097
1098 if ((inp->inp_flags & INP_TIMEWAIT) ||
1099 (inp->inp_flags & INP_DROPPED))
1100 return (inp);
1101
1102 tp = intotcpcb(inp);
1103 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1104
1105 /*
1106 * Ignore some errors if we are hooked up.
1107 * If connection hasn't completed, has retransmitted several times,
1108 * and receives a second error, give up now. This is better
1109 * than waiting a long time to establish a connection that
1110 * can never complete.
1111 */
1112 if (tp->t_state == TCPS_ESTABLISHED &&
1113 (error == EHOSTUNREACH || error == ENETUNREACH ||
1114 error == EHOSTDOWN)) {
1115 return (inp);
1116 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1117 tp->t_softerror) {
1118 tp = tcp_drop(tp, error);
1119 if (tp != NULL)
1120 return (inp);
1121 else
1122 return (NULL);
1123 } else {
1124 tp->t_softerror = error;
1125 return (inp);
1126 }
1127#if 0
1128 wakeup( &so->so_timeo);
1129 sorwakeup(so);
1130 sowwakeup(so);
1131#endif
1132}
1133
1134static int
1135tcp_pcblist(SYSCTL_HANDLER_ARGS)
1136{
1137 int error, i, m, n, pcb_count;
1138 struct inpcb *inp, **inp_list;
1139 inp_gen_t gencnt;
1140 struct xinpgen xig;
1141
1142 /*
1143 * The process of preparing the TCB list is too time-consuming and
1144 * resource-intensive to repeat twice on every request.
1145 */
1146 if (req->oldptr == NULL) {
1147 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1148 n += imax(n / 8, 10);
1149 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1150 return (0);
1151 }
1152
1153 if (req->newptr != NULL)
1154 return (EPERM);
1155
1156 /*
1157 * OK, now we're committed to doing something.
1158 */
1159 INP_INFO_RLOCK(&V_tcbinfo);
1160 gencnt = V_tcbinfo.ipi_gencnt;
1161 n = V_tcbinfo.ipi_count;
1162 INP_INFO_RUNLOCK(&V_tcbinfo);
1163
1164 m = syncache_pcbcount();
1165
1166 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1167 + (n + m) * sizeof(struct xtcpcb));
1168 if (error != 0)
1169 return (error);
1170
1171 xig.xig_len = sizeof xig;
1172 xig.xig_count = n + m;
1173 xig.xig_gen = gencnt;
1174 xig.xig_sogen = so_gencnt;
1175 error = SYSCTL_OUT(req, &xig, sizeof xig);
1176 if (error)
1177 return (error);
1178
1179 error = syncache_pcblist(req, m, &pcb_count);
1180 if (error)
1181 return (error);
1182
1183 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1184 if (inp_list == NULL)
1185 return (ENOMEM);
1186
1187 INP_INFO_RLOCK(&V_tcbinfo);
1188 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1189 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1190 INP_WLOCK(inp);
1191 if (inp->inp_gencnt <= gencnt) {
1192 /*
1193 * XXX: This use of cr_cansee(), introduced with
1194 * TCP state changes, is not quite right, but for
1195 * now, better than nothing.
1196 */
1197 if (inp->inp_flags & INP_TIMEWAIT) {
1198 if (intotw(inp) != NULL)
1199 error = cr_cansee(req->td->td_ucred,
1200 intotw(inp)->tw_cred);
1201 else
1202 error = EINVAL; /* Skip this inp. */
1203 } else
1204 error = cr_canseeinpcb(req->td->td_ucred, inp);
1205 if (error == 0) {
1206 in_pcbref(inp);
1207 inp_list[i++] = inp;
1208 }
1209 }
1210 INP_WUNLOCK(inp);
1211 }
1212 INP_INFO_RUNLOCK(&V_tcbinfo);
1213 n = i;
1214
1215 error = 0;
1216 for (i = 0; i < n; i++) {
1217 inp = inp_list[i];
1218 INP_RLOCK(inp);
1219 if (inp->inp_gencnt <= gencnt) {
1220 struct xtcpcb xt;
1221 void *inp_ppcb;
1222
1223 bzero(&xt, sizeof(xt));
1224 xt.xt_len = sizeof xt;
1225 /* XXX should avoid extra copy */
1226 bcopy(inp, &xt.xt_inp, sizeof *inp);
1227 inp_ppcb = inp->inp_ppcb;
1228 if (inp_ppcb == NULL)
1229 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1230 else if (inp->inp_flags & INP_TIMEWAIT) {
1231 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1232 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1233 } else {
1234 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1235 if (xt.xt_tp.t_timers)
1236 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1237 }
1238 if (inp->inp_socket != NULL)
1239 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1240 else {
1241 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1242 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1243 }
1244 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1245 INP_RUNLOCK(inp);
1246 error = SYSCTL_OUT(req, &xt, sizeof xt);
1247 } else
1248 INP_RUNLOCK(inp);
1249 }
1250 INP_INFO_WLOCK(&V_tcbinfo);
1251 for (i = 0; i < n; i++) {
1252 inp = inp_list[i];
1253 INP_RLOCK(inp);
1254 if (!in_pcbrele_rlocked(inp))
1255 INP_RUNLOCK(inp);
1256 }
1257 INP_INFO_WUNLOCK(&V_tcbinfo);
1258
1259 if (!error) {
1260 /*
1261 * Give the user an updated idea of our state.
1262 * If the generation differs from what we told
1263 * her before, she knows that something happened
1264 * while we were processing this request, and it
1265 * might be necessary to retry.
1266 */
1267 INP_INFO_RLOCK(&V_tcbinfo);
1268 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1269 xig.xig_sogen = so_gencnt;
1270 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1271 INP_INFO_RUNLOCK(&V_tcbinfo);
1272 error = SYSCTL_OUT(req, &xig, sizeof xig);
1273 }
1274 free(inp_list, M_TEMP);
1275 return (error);
1276}
1277
1278SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1279 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1280 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1281
1282#ifdef INET
1283static int
1284tcp_getcred(SYSCTL_HANDLER_ARGS)
1285{
1286 struct xucred xuc;
1287 struct sockaddr_in addrs[2];
1288 struct inpcb *inp;
1289 int error;
1290
1291 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1292 if (error)
1293 return (error);
1294 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1295 if (error)
1296 return (error);
1297 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1298 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1299 if (inp != NULL) {
1300 if (inp->inp_socket == NULL)
1301 error = ENOENT;
1302 if (error == 0)
1303 error = cr_canseeinpcb(req->td->td_ucred, inp);
1304 if (error == 0)
1305 cru2x(inp->inp_cred, &xuc);
1306 INP_RUNLOCK(inp);
1307 } else
1308 error = ENOENT;
1309 if (error == 0)
1310 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1311 return (error);
1312}
1313
1314SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1315 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1316 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1317#endif /* INET */
1318
1319#ifdef INET6
1320static int
1321tcp6_getcred(SYSCTL_HANDLER_ARGS)
1322{
1323 struct xucred xuc;
1324 struct sockaddr_in6 addrs[2];
1325 struct inpcb *inp;
1326 int error;
1327#ifdef INET
1328 int mapped = 0;
1329#endif
1330
1331 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1332 if (error)
1333 return (error);
1334 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1335 if (error)
1336 return (error);
1337 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1338 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1339 return (error);
1340 }
1341 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1342#ifdef INET
1343 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1344 mapped = 1;
1345 else
1346#endif
1347 return (EINVAL);
1348 }
1349
1350#ifdef INET
1351 if (mapped == 1)
1352 inp = in_pcblookup(&V_tcbinfo,
1353 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1354 addrs[1].sin6_port,
1355 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1356 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1357 else
1358#endif
1359 inp = in6_pcblookup(&V_tcbinfo,
1360 &addrs[1].sin6_addr, addrs[1].sin6_port,
1361 &addrs[0].sin6_addr, addrs[0].sin6_port,
1362 INPLOOKUP_RLOCKPCB, NULL);
1363 if (inp != NULL) {
1364 if (inp->inp_socket == NULL)
1365 error = ENOENT;
1366 if (error == 0)
1367 error = cr_canseeinpcb(req->td->td_ucred, inp);
1368 if (error == 0)
1369 cru2x(inp->inp_cred, &xuc);
1370 INP_RUNLOCK(inp);
1371 } else
1372 error = ENOENT;
1373 if (error == 0)
1374 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1375 return (error);
1376}
1377
1378SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1379 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1380 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1381#endif /* INET6 */
1382
1383
1384#ifdef INET
1385void
1386tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1387{
1388 struct ip *ip = vip;
1389 struct tcphdr *th;
1390 struct in_addr faddr;
1391 struct inpcb *inp;
1392 struct tcpcb *tp;
1393 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1394 struct icmp *icp;
1395 struct in_conninfo inc;
1396 tcp_seq icmp_tcp_seq;
1397 int mtu;
1398
1399 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1400 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1401 return;
1402
1403 if (cmd == PRC_MSGSIZE)
1404 notify = tcp_mtudisc_notify;
1405 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1406 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1407 notify = tcp_drop_syn_sent;
1408 /*
1409 * Redirects don't need to be handled up here.
1410 */
1411 else if (PRC_IS_REDIRECT(cmd))
1412 return;
1413 /*
1414 * Source quench is depreciated.
1415 */
1416 else if (cmd == PRC_QUENCH)
1417 return;
1418 /*
1419 * Hostdead is ugly because it goes linearly through all PCBs.
1420 * XXX: We never get this from ICMP, otherwise it makes an
1421 * excellent DoS attack on machines with many connections.
1422 */
1423 else if (cmd == PRC_HOSTDEAD)
1424 ip = NULL;
1425 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1426 return;
1427 if (ip != NULL) {
1428 icp = (struct icmp *)((caddr_t)ip
1429 - offsetof(struct icmp, icmp_ip));
1430 th = (struct tcphdr *)((caddr_t)ip
1431 + (ip->ip_hl << 2));
1432 INP_INFO_WLOCK(&V_tcbinfo);
1433 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1434 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1435 if (inp != NULL) {
1436 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1437 !(inp->inp_flags & INP_DROPPED) &&
1438 !(inp->inp_socket == NULL)) {
1439 icmp_tcp_seq = htonl(th->th_seq);
1440 tp = intotcpcb(inp);
1441 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1442 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1443 if (cmd == PRC_MSGSIZE) {
1444 /*
1445 * MTU discovery:
1446 * If we got a needfrag set the MTU
1447 * in the route to the suggested new
1448 * value (if given) and then notify.
1449 */
1450 bzero(&inc, sizeof(inc));
1451 inc.inc_faddr = faddr;
1452 inc.inc_fibnum =
1453 inp->inp_inc.inc_fibnum;
1454
1455 mtu = ntohs(icp->icmp_nextmtu);
1456 /*
1457 * If no alternative MTU was
1458 * proposed, try the next smaller
1459 * one.
1460 */
1461 if (!mtu)
1462 mtu = ip_next_mtu(
1463 ntohs(ip->ip_len), 1);
1464 if (mtu < V_tcp_minmss
1465 + sizeof(struct tcpiphdr))
1466 mtu = V_tcp_minmss
1467 + sizeof(struct tcpiphdr);
1468 /*
1469 * Only cache the MTU if it
1470 * is smaller than the interface
1471 * or route MTU. tcp_mtudisc()
1472 * will do right thing by itself.
1473 */
1474 if (mtu <= tcp_maxmtu(&inc, NULL))
1475 tcp_hc_updatemtu(&inc, mtu);
1476 tcp_mtudisc(inp, mtu);
1477 } else
1478 inp = (*notify)(inp,
1479 inetctlerrmap[cmd]);
1480 }
1481 }
1482 if (inp != NULL)
1483 INP_WUNLOCK(inp);
1484 } else {
1485 bzero(&inc, sizeof(inc));
1486 inc.inc_fport = th->th_dport;
1487 inc.inc_lport = th->th_sport;
1488 inc.inc_faddr = faddr;
1489 inc.inc_laddr = ip->ip_src;
1490 syncache_unreach(&inc, th);
1491 }
1492 INP_INFO_WUNLOCK(&V_tcbinfo);
1493 } else
1494 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1495}
1496#endif /* INET */
1497
1498#ifdef INET6
1499void
1500tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1501{
1502 struct tcphdr th;
1503 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1504 struct ip6_hdr *ip6;
1505 struct mbuf *m;
1506 struct ip6ctlparam *ip6cp = NULL;
1507 const struct sockaddr_in6 *sa6_src = NULL;
1508 int off;
1509 struct tcp_portonly {
1510 u_int16_t th_sport;
1511 u_int16_t th_dport;
1512 } *thp;
1513
1514 if (sa->sa_family != AF_INET6 ||
1515 sa->sa_len != sizeof(struct sockaddr_in6))
1516 return;
1517
1518 if (cmd == PRC_MSGSIZE)
1519 notify = tcp_mtudisc_notify;
1520 else if (!PRC_IS_REDIRECT(cmd) &&
1521 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1522 return;
1523 /* Source quench is depreciated. */
1524 else if (cmd == PRC_QUENCH)
1525 return;
1526
1527 /* if the parameter is from icmp6, decode it. */
1528 if (d != NULL) {
1529 ip6cp = (struct ip6ctlparam *)d;
1530 m = ip6cp->ip6c_m;
1531 ip6 = ip6cp->ip6c_ip6;
1532 off = ip6cp->ip6c_off;
1533 sa6_src = ip6cp->ip6c_src;
1534 } else {
1535 m = NULL;
1536 ip6 = NULL;
1537 off = 0; /* fool gcc */
1538 sa6_src = &sa6_any;
1539 }
1540
1541 if (ip6 != NULL) {
1542 struct in_conninfo inc;
1543 /*
1544 * XXX: We assume that when IPV6 is non NULL,
1545 * M and OFF are valid.
1546 */
1547
1548 /* check if we can safely examine src and dst ports */
1549 if (m->m_pkthdr.len < off + sizeof(*thp))
1550 return;
1551
1552 bzero(&th, sizeof(th));
1553 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1554
1555 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1556 (struct sockaddr *)ip6cp->ip6c_src,
1557 th.th_sport, cmd, NULL, notify);
1558
1559 bzero(&inc, sizeof(inc));
1560 inc.inc_fport = th.th_dport;
1561 inc.inc_lport = th.th_sport;
1562 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1563 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1564 inc.inc_flags |= INC_ISIPV6;
1565 INP_INFO_WLOCK(&V_tcbinfo);
1566 syncache_unreach(&inc, &th);
1567 INP_INFO_WUNLOCK(&V_tcbinfo);
1568 } else
1569 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1570 0, cmd, NULL, notify);
1571}
1572#endif /* INET6 */
1573
1574
1575/*
1576 * Following is where TCP initial sequence number generation occurs.
1577 *
1578 * There are two places where we must use initial sequence numbers:
1579 * 1. In SYN-ACK packets.
1580 * 2. In SYN packets.
1581 *
1582 * All ISNs for SYN-ACK packets are generated by the syncache. See
1583 * tcp_syncache.c for details.
1584 *
1585 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1586 * depends on this property. In addition, these ISNs should be
1587 * unguessable so as to prevent connection hijacking. To satisfy
1588 * the requirements of this situation, the algorithm outlined in
1589 * RFC 1948 is used, with only small modifications.
1590 *
1591 * Implementation details:
1592 *
1593 * Time is based off the system timer, and is corrected so that it
1594 * increases by one megabyte per second. This allows for proper
1595 * recycling on high speed LANs while still leaving over an hour
1596 * before rollover.
1597 *
1598 * As reading the *exact* system time is too expensive to be done
1599 * whenever setting up a TCP connection, we increment the time
1600 * offset in two ways. First, a small random positive increment
1601 * is added to isn_offset for each connection that is set up.
1602 * Second, the function tcp_isn_tick fires once per clock tick
1603 * and increments isn_offset as necessary so that sequence numbers
1604 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1605 * random positive increments serve only to ensure that the same
1606 * exact sequence number is never sent out twice (as could otherwise
1607 * happen when a port is recycled in less than the system tick
1608 * interval.)
1609 *
1610 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1611 * between seeding of isn_secret. This is normally set to zero,
1612 * as reseeding should not be necessary.
1613 *
1614 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1615 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1616 * general, this means holding an exclusive (write) lock.
1617 */
1618
1619#define ISN_BYTES_PER_SECOND 1048576
1620#define ISN_STATIC_INCREMENT 4096
1621#define ISN_RANDOM_INCREMENT (4096 - 1)
1622
1623static VNET_DEFINE(u_char, isn_secret[32]);
1624static VNET_DEFINE(int, isn_last);
1625static VNET_DEFINE(int, isn_last_reseed);
1626static VNET_DEFINE(u_int32_t, isn_offset);
1627static VNET_DEFINE(u_int32_t, isn_offset_old);
1628
1629#define V_isn_secret VNET(isn_secret)
1630#define V_isn_last VNET(isn_last)
1631#define V_isn_last_reseed VNET(isn_last_reseed)
1632#define V_isn_offset VNET(isn_offset)
1633#define V_isn_offset_old VNET(isn_offset_old)
1634
1635tcp_seq
1636tcp_new_isn(struct tcpcb *tp)
1637{
1638 MD5_CTX isn_ctx;
1639 u_int32_t md5_buffer[4];
1640 tcp_seq new_isn;
1641 u_int32_t projected_offset;
1642
1643 INP_WLOCK_ASSERT(tp->t_inpcb);
1644
1645 ISN_LOCK();
1646 /* Seed if this is the first use, reseed if requested. */
1647 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1648 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1649 < (u_int)ticks))) {
1650 read_random(&V_isn_secret, sizeof(V_isn_secret));
1651 V_isn_last_reseed = ticks;
1652 }
1653
1654 /* Compute the md5 hash and return the ISN. */
1655 MD5Init(&isn_ctx);
1656 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1657 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1658#ifdef INET6
1659 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1660 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1661 sizeof(struct in6_addr));
1662 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1663 sizeof(struct in6_addr));
1664 } else
1665#endif
1666 {
1667 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1668 sizeof(struct in_addr));
1669 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1670 sizeof(struct in_addr));
1671 }
1672 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1673 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1674 new_isn = (tcp_seq) md5_buffer[0];
1675 V_isn_offset += ISN_STATIC_INCREMENT +
1676 (arc4random() & ISN_RANDOM_INCREMENT);
1677 if (ticks != V_isn_last) {
1678 projected_offset = V_isn_offset_old +
1679 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1680 if (SEQ_GT(projected_offset, V_isn_offset))
1681 V_isn_offset = projected_offset;
1682 V_isn_offset_old = V_isn_offset;
1683 V_isn_last = ticks;
1684 }
1685 new_isn += V_isn_offset;
1686 ISN_UNLOCK();
1687 return (new_isn);
1688}
1689
1690/*
1691 * When a specific ICMP unreachable message is received and the
1692 * connection state is SYN-SENT, drop the connection. This behavior
1693 * is controlled by the icmp_may_rst sysctl.
1694 */
1695struct inpcb *
1696tcp_drop_syn_sent(struct inpcb *inp, int errno)
1697{
1698 struct tcpcb *tp;
1699
1700 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1701 INP_WLOCK_ASSERT(inp);
1702
1703 if ((inp->inp_flags & INP_TIMEWAIT) ||
1704 (inp->inp_flags & INP_DROPPED))
1705 return (inp);
1706
1707 tp = intotcpcb(inp);
1708 if (tp->t_state != TCPS_SYN_SENT)
1709 return (inp);
1710
1711 tp = tcp_drop(tp, errno);
1712 if (tp != NULL)
1713 return (inp);
1714 else
1715 return (NULL);
1716}
1717
1718/*
1719 * When `need fragmentation' ICMP is received, update our idea of the MSS
1720 * based on the new value. Also nudge TCP to send something, since we
1721 * know the packet we just sent was dropped.
1722 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1723 */
1724static struct inpcb *
1725tcp_mtudisc_notify(struct inpcb *inp, int error)
1726{
1727
1728 return (tcp_mtudisc(inp, -1));
1729}
1730
1731struct inpcb *
1732tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1733{
1734 struct tcpcb *tp;
1735 struct socket *so;
1736
1737 INP_WLOCK_ASSERT(inp);
1738 if ((inp->inp_flags & INP_TIMEWAIT) ||
1739 (inp->inp_flags & INP_DROPPED))
1740 return (inp);
1741
1742 tp = intotcpcb(inp);
1743 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1744
1745 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1746
1747 so = inp->inp_socket;
1748 SOCKBUF_LOCK(&so->so_snd);
1749 /* If the mss is larger than the socket buffer, decrease the mss. */
1750 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1751 tp->t_maxseg = so->so_snd.sb_hiwat;
1752 SOCKBUF_UNLOCK(&so->so_snd);
1753
1754 TCPSTAT_INC(tcps_mturesent);
1755 tp->t_rtttime = 0;
1756 tp->snd_nxt = tp->snd_una;
1757 tcp_free_sackholes(tp);
1758 tp->snd_recover = tp->snd_max;
1759 if (tp->t_flags & TF_SACK_PERMIT)
1760 EXIT_FASTRECOVERY(tp->t_flags);
1761 tcp_output(tp);
1762 return (inp);
1763}
1764
1765#ifdef INET
1766/*
1767 * Look-up the routing entry to the peer of this inpcb. If no route
1768 * is found and it cannot be allocated, then return 0. This routine
1769 * is called by TCP routines that access the rmx structure and by
1770 * tcp_mss_update to get the peer/interface MTU.
1771 */
1772u_long
1773tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1774{
1775 struct route sro;
1776 struct sockaddr_in *dst;
1777 struct ifnet *ifp;
1778 u_long maxmtu = 0;
1779
1780 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1781
1782 bzero(&sro, sizeof(sro));
1783 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1784 dst = (struct sockaddr_in *)&sro.ro_dst;
1785 dst->sin_family = AF_INET;
1786 dst->sin_len = sizeof(*dst);
1787 dst->sin_addr = inc->inc_faddr;
1788 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1789 }
1790 if (sro.ro_rt != NULL) {
1791 ifp = sro.ro_rt->rt_ifp;
1792 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1793 maxmtu = ifp->if_mtu;
1794 else
1795 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1796
1797 /* Report additional interface capabilities. */
1798 if (cap != NULL) {
1799 if (ifp->if_capenable & IFCAP_TSO4 &&
1800 ifp->if_hwassist & CSUM_TSO)
1801 cap->ifcap |= CSUM_TSO;
1802 cap->tsomax = ifp->if_hw_tsomax;
1803 }
1804 RTFREE(sro.ro_rt);
1805 }
1806 return (maxmtu);
1807}
1808#endif /* INET */
1809
1810#ifdef INET6
1811u_long
1812tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1813{
1814 struct route_in6 sro6;
1815 struct ifnet *ifp;
1816 u_long maxmtu = 0;
1817
1818 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1819
1820 bzero(&sro6, sizeof(sro6));
1821 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1822 sro6.ro_dst.sin6_family = AF_INET6;
1823 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1824 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1825 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1826 }
1827 if (sro6.ro_rt != NULL) {
1828 ifp = sro6.ro_rt->rt_ifp;
1829 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1830 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1831 else
1832 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1833 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1834
1835 /* Report additional interface capabilities. */
1836 if (cap != NULL) {
1837 if (ifp->if_capenable & IFCAP_TSO6 &&
1838 ifp->if_hwassist & CSUM_TSO)
1839 cap->ifcap |= CSUM_TSO;
1840 cap->tsomax = ifp->if_hw_tsomax;
1841 }
1842 RTFREE(sro6.ro_rt);
1843 }
1844
1845 return (maxmtu);
1846}
1847#endif /* INET6 */
1848
1849#ifdef IPSEC
1850/* compute ESP/AH header size for TCP, including outer IP header. */
1851size_t
1852ipsec_hdrsiz_tcp(struct tcpcb *tp)
1853{
1854 struct inpcb *inp;
1855 struct mbuf *m;
1856 size_t hdrsiz;
1857 struct ip *ip;
1858#ifdef INET6
1859 struct ip6_hdr *ip6;
1860#endif
1861 struct tcphdr *th;
1862
1863 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1864 return (0);
1865 m = m_gethdr(M_NOWAIT, MT_DATA);
1866 if (!m)
1867 return (0);
1868
1869#ifdef INET6
1870 if ((inp->inp_vflag & INP_IPV6) != 0) {
1871 ip6 = mtod(m, struct ip6_hdr *);
1872 th = (struct tcphdr *)(ip6 + 1);
1873 m->m_pkthdr.len = m->m_len =
1874 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1875 tcpip_fillheaders(inp, ip6, th);
1876 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1877 } else
1878#endif /* INET6 */
1879 {
1880 ip = mtod(m, struct ip *);
1881 th = (struct tcphdr *)(ip + 1);
1882 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1883 tcpip_fillheaders(inp, ip, th);
1884 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1885 }
1886
1887 m_free(m);
1888 return (hdrsiz);
1889}
1890#endif /* IPSEC */
1891
1892#ifdef TCP_SIGNATURE
1893/*
1894 * Callback function invoked by m_apply() to digest TCP segment data
1895 * contained within an mbuf chain.
1896 */
1897static int
1898tcp_signature_apply(void *fstate, void *data, u_int len)
1899{
1900
1901 MD5Update(fstate, (u_char *)data, len);
1902 return (0);
1903}
1904
1905/*
1906 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1907 *
1908 * Parameters:
1909 * m pointer to head of mbuf chain
1910 * _unused
1911 * len length of TCP segment data, excluding options
1912 * optlen length of TCP segment options
1913 * buf pointer to storage for computed MD5 digest
1914 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1915 *
1916 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1917 * When called from tcp_input(), we can be sure that th_sum has been
1918 * zeroed out and verified already.
1919 *
1920 * Return 0 if successful, otherwise return -1.
1921 *
1922 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1923 * search with the destination IP address, and a 'magic SPI' to be
1924 * determined by the application. This is hardcoded elsewhere to 1179
1925 * right now. Another branch of this code exists which uses the SPD to
1926 * specify per-application flows but it is unstable.
1927 */
1928int
1929tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1930 u_char *buf, u_int direction)
1931{
1932 union sockaddr_union dst;
1933#ifdef INET
1934 struct ippseudo ippseudo;
1935#endif
1936 MD5_CTX ctx;
1937 int doff;
1938 struct ip *ip;
1939#ifdef INET
1940 struct ipovly *ipovly;
1941#endif
1942 struct secasvar *sav;
1943 struct tcphdr *th;
1944#ifdef INET6
1945 struct ip6_hdr *ip6;
1946 struct in6_addr in6;
1947 char ip6buf[INET6_ADDRSTRLEN];
1948 uint32_t plen;
1949 uint16_t nhdr;
1950#endif
1951 u_short savecsum;
1952
1953 KASSERT(m != NULL, ("NULL mbuf chain"));
1954 KASSERT(buf != NULL, ("NULL signature pointer"));
1955
1956 /* Extract the destination from the IP header in the mbuf. */
1957 bzero(&dst, sizeof(union sockaddr_union));
1958 ip = mtod(m, struct ip *);
1959#ifdef INET6
1960 ip6 = NULL; /* Make the compiler happy. */
1961#endif
1962 switch (ip->ip_v) {
1963#ifdef INET
1964 case IPVERSION:
1965 dst.sa.sa_len = sizeof(struct sockaddr_in);
1966 dst.sa.sa_family = AF_INET;
1967 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1968 ip->ip_src : ip->ip_dst;
1969 break;
1970#endif
1971#ifdef INET6
1972 case (IPV6_VERSION >> 4):
1973 ip6 = mtod(m, struct ip6_hdr *);
1974 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1975 dst.sa.sa_family = AF_INET6;
1976 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1977 ip6->ip6_src : ip6->ip6_dst;
1978 break;
1979#endif
1980 default:
1981 return (EINVAL);
1982 /* NOTREACHED */
1983 break;
1984 }
1985
1986 /* Look up an SADB entry which matches the address of the peer. */
1987 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1988 if (sav == NULL) {
1989 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1990 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1991#ifdef INET6
1992 (ip->ip_v == (IPV6_VERSION >> 4)) ?
1993 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
1994#endif
1995 "(unsupported)"));
1996 return (EINVAL);
1997 }
1998
1999 MD5Init(&ctx);
2000 /*
2001 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2002 *
2003 * XXX The ippseudo header MUST be digested in network byte order,
2004 * or else we'll fail the regression test. Assume all fields we've
2005 * been doing arithmetic on have been in host byte order.
2006 * XXX One cannot depend on ipovly->ih_len here. When called from
2007 * tcp_output(), the underlying ip_len member has not yet been set.
2008 */
2009 switch (ip->ip_v) {
2010#ifdef INET
2011 case IPVERSION:
2012 ipovly = (struct ipovly *)ip;
2013 ippseudo.ippseudo_src = ipovly->ih_src;
2014 ippseudo.ippseudo_dst = ipovly->ih_dst;
2015 ippseudo.ippseudo_pad = 0;
2016 ippseudo.ippseudo_p = IPPROTO_TCP;
2017 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2018 optlen);
2019 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2020
2021 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2022 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2023 break;
2024#endif
2025#ifdef INET6
2026 /*
2027 * RFC 2385, 2.0 Proposal
2028 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2029 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2030 * extended next header value (to form 32 bits), and 32-bit segment
2031 * length.
2032 * Note: Upper-Layer Packet Length comes before Next Header.
2033 */
2034 case (IPV6_VERSION >> 4):
2035 in6 = ip6->ip6_src;
2036 in6_clearscope(&in6);
2037 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2038 in6 = ip6->ip6_dst;
2039 in6_clearscope(&in6);
2040 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2041 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2042 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2043 nhdr = 0;
2044 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2045 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2046 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2047 nhdr = IPPROTO_TCP;
2048 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2049
2050 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2051 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2052 break;
2053#endif
2054 default:
2055 return (EINVAL);
2056 /* NOTREACHED */
2057 break;
2058 }
2059
2060
2061 /*
2062 * Step 2: Update MD5 hash with TCP header, excluding options.
2063 * The TCP checksum must be set to zero.
2064 */
2065 savecsum = th->th_sum;
2066 th->th_sum = 0;
2067 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2068 th->th_sum = savecsum;
2069
2070 /*
2071 * Step 3: Update MD5 hash with TCP segment data.
2072 * Use m_apply() to avoid an early m_pullup().
2073 */
2074 if (len > 0)
2075 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2076
2077 /*
2078 * Step 4: Update MD5 hash with shared secret.
2079 */
2080 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2081 MD5Final(buf, &ctx);
2082
2083 key_sa_recordxfer(sav, m);
2084 KEY_FREESAV(&sav);
2085 return (0);
2086}
2087
2088/*
2089 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2090 *
2091 * Parameters:
2092 * m pointer to head of mbuf chain
2093 * len length of TCP segment data, excluding options
2094 * optlen length of TCP segment options
2095 * buf pointer to storage for computed MD5 digest
2096 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2097 *
2098 * Return 1 if successful, otherwise return 0.
2099 */
2100int
2101tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2102 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2103{
2104 char tmpdigest[TCP_SIGLEN];
2105
2106 if (tcp_sig_checksigs == 0)
2107 return (1);
2108 if ((tcpbflag & TF_SIGNATURE) == 0) {
2109 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2110
2111 /*
2112 * If this socket is not expecting signature but
2113 * the segment contains signature just fail.
2114 */
2115 TCPSTAT_INC(tcps_sig_err_sigopt);
2116 TCPSTAT_INC(tcps_sig_rcvbadsig);
2117 return (0);
2118 }
2119
2120 /* Signature is not expected, and not present in segment. */
2121 return (1);
2122 }
2123
2124 /*
2125 * If this socket is expecting signature but the segment does not
2126 * contain any just fail.
2127 */
2128 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2129 TCPSTAT_INC(tcps_sig_err_nosigopt);
2130 TCPSTAT_INC(tcps_sig_rcvbadsig);
2131 return (0);
2132 }
2133 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2134 IPSEC_DIR_INBOUND) == -1) {
2135 TCPSTAT_INC(tcps_sig_err_buildsig);
2136 TCPSTAT_INC(tcps_sig_rcvbadsig);
2137 return (0);
2138 }
2139
2140 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2141 TCPSTAT_INC(tcps_sig_rcvbadsig);
2142 return (0);
2143 }
2144 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2145 return (1);
2146}
2147#endif /* TCP_SIGNATURE */
2148
2149static int
2150sysctl_drop(SYSCTL_HANDLER_ARGS)
2151{
2152 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2153 struct sockaddr_storage addrs[2];
2154 struct inpcb *inp;
2155 struct tcpcb *tp;
2156 struct tcptw *tw;
2157 struct sockaddr_in *fin, *lin;
2158#ifdef INET6
2159 struct sockaddr_in6 *fin6, *lin6;
2160#endif
2161 int error;
2162
2163 inp = NULL;
2164 fin = lin = NULL;
2165#ifdef INET6
2166 fin6 = lin6 = NULL;
2167#endif
2168 error = 0;
2169
2170 if (req->oldptr != NULL || req->oldlen != 0)
2171 return (EINVAL);
2172 if (req->newptr == NULL)
2173 return (EPERM);
2174 if (req->newlen < sizeof(addrs))
2175 return (ENOMEM);
2176 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2177 if (error)
2178 return (error);
2179
2180 switch (addrs[0].ss_family) {
2181#ifdef INET6
2182 case AF_INET6:
2183 fin6 = (struct sockaddr_in6 *)&addrs[0];
2184 lin6 = (struct sockaddr_in6 *)&addrs[1];
2185 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2186 lin6->sin6_len != sizeof(struct sockaddr_in6))
2187 return (EINVAL);
2188 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2189 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2190 return (EINVAL);
2191 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2192 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2193 fin = (struct sockaddr_in *)&addrs[0];
2194 lin = (struct sockaddr_in *)&addrs[1];
2195 break;
2196 }
2197 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2198 if (error)
2199 return (error);
2200 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2201 if (error)
2202 return (error);
2203 break;
2204#endif
2205#ifdef INET
2206 case AF_INET:
2207 fin = (struct sockaddr_in *)&addrs[0];
2208 lin = (struct sockaddr_in *)&addrs[1];
2209 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2210 lin->sin_len != sizeof(struct sockaddr_in))
2211 return (EINVAL);
2212 break;
2213#endif
2214 default:
2215 return (EINVAL);
2216 }
2217 INP_INFO_WLOCK(&V_tcbinfo);
2218 switch (addrs[0].ss_family) {
2219#ifdef INET6
2220 case AF_INET6:
2221 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2222 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2223 INPLOOKUP_WLOCKPCB, NULL);
2224 break;
2225#endif
2226#ifdef INET
2227 case AF_INET:
2228 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2229 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2230 break;
2231#endif
2232 }
2233 if (inp != NULL) {
2234 if (inp->inp_flags & INP_TIMEWAIT) {
2235 /*
2236 * XXXRW: There currently exists a state where an
2237 * inpcb is present, but its timewait state has been
2238 * discarded. For now, don't allow dropping of this
2239 * type of inpcb.
2240 */
2241 tw = intotw(inp);
2242 if (tw != NULL)
2243 tcp_twclose(tw, 0);
2244 else
2245 INP_WUNLOCK(inp);
2246 } else if (!(inp->inp_flags & INP_DROPPED) &&
2247 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2248 tp = intotcpcb(inp);
2249 tp = tcp_drop(tp, ECONNABORTED);
2250 if (tp != NULL)
2251 INP_WUNLOCK(inp);
2252 } else
2253 INP_WUNLOCK(inp);
2254 } else
2255 error = ESRCH;
2256 INP_INFO_WUNLOCK(&V_tcbinfo);
2257 return (error);
2258}
2259
2260SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2261 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2262 0, sysctl_drop, "", "Drop TCP connection");
2263
2264/*
2265 * Generate a standardized TCP log line for use throughout the
2266 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2267 * allow use in the interrupt context.
2268 *
2269 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2270 * NB: The function may return NULL if memory allocation failed.
2271 *
2272 * Due to header inclusion and ordering limitations the struct ip
2273 * and ip6_hdr pointers have to be passed as void pointers.
2274 */
2275char *
2276tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2277 const void *ip6hdr)
2278{
2279
2280 /* Is logging enabled? */
2281 if (tcp_log_in_vain == 0)
2282 return (NULL);
2283
2284 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2285}
2286
2287char *
2288tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2289 const void *ip6hdr)
2290{
2291
2292 /* Is logging enabled? */
2293 if (tcp_log_debug == 0)
2294 return (NULL);
2295
2296 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2297}
2298
2299static char *
2300tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2301 const void *ip6hdr)
2302{
2303 char *s, *sp;
2304 size_t size;
2305 struct ip *ip;
2306#ifdef INET6
2307 const struct ip6_hdr *ip6;
2308
2309 ip6 = (const struct ip6_hdr *)ip6hdr;
2310#endif /* INET6 */
2311 ip = (struct ip *)ip4hdr;
2312
2313 /*
2314 * The log line looks like this:
2315 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2316 */
2317 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2318 sizeof(PRINT_TH_FLAGS) + 1 +
2319#ifdef INET6
2320 2 * INET6_ADDRSTRLEN;
2321#else
2322 2 * INET_ADDRSTRLEN;
2323#endif /* INET6 */
2324
2325 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2326 if (s == NULL)
2327 return (NULL);
2328
2329 strcat(s, "TCP: [");
2330 sp = s + strlen(s);
2331
2332 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2333 inet_ntoa_r(inc->inc_faddr, sp);
2334 sp = s + strlen(s);
2335 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2336 sp = s + strlen(s);
2337 inet_ntoa_r(inc->inc_laddr, sp);
2338 sp = s + strlen(s);
2339 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2340#ifdef INET6
2341 } else if (inc) {
2342 ip6_sprintf(sp, &inc->inc6_faddr);
2343 sp = s + strlen(s);
2344 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2345 sp = s + strlen(s);
2346 ip6_sprintf(sp, &inc->inc6_laddr);
2347 sp = s + strlen(s);
2348 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2349 } else if (ip6 && th) {
2350 ip6_sprintf(sp, &ip6->ip6_src);
2351 sp = s + strlen(s);
2352 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2353 sp = s + strlen(s);
2354 ip6_sprintf(sp, &ip6->ip6_dst);
2355 sp = s + strlen(s);
2356 sprintf(sp, "]:%i", ntohs(th->th_dport));
2357#endif /* INET6 */
2358#ifdef INET
2359 } else if (ip && th) {
2360 inet_ntoa_r(ip->ip_src, sp);
2361 sp = s + strlen(s);
2362 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2363 sp = s + strlen(s);
2364 inet_ntoa_r(ip->ip_dst, sp);
2365 sp = s + strlen(s);
2366 sprintf(sp, "]:%i", ntohs(th->th_dport));
2367#endif /* INET */
2368 } else {
2369 free(s, M_TCPLOG);
2370 return (NULL);
2371 }
2372 sp = s + strlen(s);
2373 if (th)
2374 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2375 if (*(s + size - 1) != '\0')
2376 panic("%s: string too long", __func__);
2377 return (s);
2378}
| 893 (void) tcp_output(tp);
894 TCPSTAT_INC(tcps_drops);
895 } else
896 TCPSTAT_INC(tcps_conndrops);
897 if (errno == ETIMEDOUT && tp->t_softerror)
898 errno = tp->t_softerror;
899 so->so_error = errno;
900 return (tcp_close(tp));
901}
902
903void
904tcp_discardcb(struct tcpcb *tp)
905{
906 struct inpcb *inp = tp->t_inpcb;
907 struct socket *so = inp->inp_socket;
908#ifdef INET6
909 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
910#endif /* INET6 */
911
912 INP_WLOCK_ASSERT(inp);
913
914 /*
915 * Make sure that all of our timers are stopped before we delete the
916 * PCB.
917 *
918 * XXXRW: Really, we would like to use callout_drain() here in order
919 * to avoid races experienced in tcp_timer.c where a timer is already
920 * executing at this point. However, we can't, both because we're
921 * running in a context where we can't sleep, and also because we
922 * hold locks required by the timers. What we instead need to do is
923 * test to see if callout_drain() is required, and if so, defer some
924 * portion of the remainder of tcp_discardcb() to an asynchronous
925 * context that can callout_drain() and then continue. Some care
926 * will be required to ensure that no further processing takes place
927 * on the tcpcb, even though it hasn't been freed (a flag?).
928 */
929 callout_stop(&tp->t_timers->tt_rexmt);
930 callout_stop(&tp->t_timers->tt_persist);
931 callout_stop(&tp->t_timers->tt_keep);
932 callout_stop(&tp->t_timers->tt_2msl);
933 callout_stop(&tp->t_timers->tt_delack);
934
935 /*
936 * If we got enough samples through the srtt filter,
937 * save the rtt and rttvar in the routing entry.
938 * 'Enough' is arbitrarily defined as 4 rtt samples.
939 * 4 samples is enough for the srtt filter to converge
940 * to within enough % of the correct value; fewer samples
941 * and we could save a bogus rtt. The danger is not high
942 * as tcp quickly recovers from everything.
943 * XXX: Works very well but needs some more statistics!
944 */
945 if (tp->t_rttupdated >= 4) {
946 struct hc_metrics_lite metrics;
947 u_long ssthresh;
948
949 bzero(&metrics, sizeof(metrics));
950 /*
951 * Update the ssthresh always when the conditions below
952 * are satisfied. This gives us better new start value
953 * for the congestion avoidance for new connections.
954 * ssthresh is only set if packet loss occured on a session.
955 *
956 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
957 * being torn down. Ideally this code would not use 'so'.
958 */
959 ssthresh = tp->snd_ssthresh;
960 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
961 /*
962 * convert the limit from user data bytes to
963 * packets then to packet data bytes.
964 */
965 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
966 if (ssthresh < 2)
967 ssthresh = 2;
968 ssthresh *= (u_long)(tp->t_maxseg +
969#ifdef INET6
970 (isipv6 ? sizeof (struct ip6_hdr) +
971 sizeof (struct tcphdr) :
972#endif
973 sizeof (struct tcpiphdr)
974#ifdef INET6
975 )
976#endif
977 );
978 } else
979 ssthresh = 0;
980 metrics.rmx_ssthresh = ssthresh;
981
982 metrics.rmx_rtt = tp->t_srtt;
983 metrics.rmx_rttvar = tp->t_rttvar;
984 metrics.rmx_cwnd = tp->snd_cwnd;
985 metrics.rmx_sendpipe = 0;
986 metrics.rmx_recvpipe = 0;
987
988 tcp_hc_update(&inp->inp_inc, &metrics);
989 }
990
991 /* free the reassembly queue, if any */
992 tcp_reass_flush(tp);
993
994#ifdef TCP_OFFLOAD
995 /* Disconnect offload device, if any. */
996 if (tp->t_flags & TF_TOE)
997 tcp_offload_detach(tp);
998#endif
999
1000 tcp_free_sackholes(tp);
1001
1002 /* Allow the CC algorithm to clean up after itself. */
1003 if (CC_ALGO(tp)->cb_destroy != NULL)
1004 CC_ALGO(tp)->cb_destroy(tp->ccv);
1005
1006 khelp_destroy_osd(tp->osd);
1007
1008 CC_ALGO(tp) = NULL;
1009 inp->inp_ppcb = NULL;
1010 tp->t_inpcb = NULL;
1011 uma_zfree(V_tcpcb_zone, tp);
1012}
1013
1014/*
1015 * Attempt to close a TCP control block, marking it as dropped, and freeing
1016 * the socket if we hold the only reference.
1017 */
1018struct tcpcb *
1019tcp_close(struct tcpcb *tp)
1020{
1021 struct inpcb *inp = tp->t_inpcb;
1022 struct socket *so;
1023
1024 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1025 INP_WLOCK_ASSERT(inp);
1026
1027#ifdef TCP_OFFLOAD
1028 if (tp->t_state == TCPS_LISTEN)
1029 tcp_offload_listen_stop(tp);
1030#endif
1031 in_pcbdrop(inp);
1032 TCPSTAT_INC(tcps_closed);
1033 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1034 so = inp->inp_socket;
1035 soisdisconnected(so);
1036 if (inp->inp_flags & INP_SOCKREF) {
1037 KASSERT(so->so_state & SS_PROTOREF,
1038 ("tcp_close: !SS_PROTOREF"));
1039 inp->inp_flags &= ~INP_SOCKREF;
1040 INP_WUNLOCK(inp);
1041 ACCEPT_LOCK();
1042 SOCK_LOCK(so);
1043 so->so_state &= ~SS_PROTOREF;
1044 sofree(so);
1045 return (NULL);
1046 }
1047 return (tp);
1048}
1049
1050void
1051tcp_drain(void)
1052{
1053 VNET_ITERATOR_DECL(vnet_iter);
1054
1055 if (!do_tcpdrain)
1056 return;
1057
1058 VNET_LIST_RLOCK_NOSLEEP();
1059 VNET_FOREACH(vnet_iter) {
1060 CURVNET_SET(vnet_iter);
1061 struct inpcb *inpb;
1062 struct tcpcb *tcpb;
1063
1064 /*
1065 * Walk the tcpbs, if existing, and flush the reassembly queue,
1066 * if there is one...
1067 * XXX: The "Net/3" implementation doesn't imply that the TCP
1068 * reassembly queue should be flushed, but in a situation
1069 * where we're really low on mbufs, this is potentially
1070 * useful.
1071 */
1072 INP_INFO_RLOCK(&V_tcbinfo);
1073 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1074 if (inpb->inp_flags & INP_TIMEWAIT)
1075 continue;
1076 INP_WLOCK(inpb);
1077 if ((tcpb = intotcpcb(inpb)) != NULL) {
1078 tcp_reass_flush(tcpb);
1079 tcp_clean_sackreport(tcpb);
1080 }
1081 INP_WUNLOCK(inpb);
1082 }
1083 INP_INFO_RUNLOCK(&V_tcbinfo);
1084 CURVNET_RESTORE();
1085 }
1086 VNET_LIST_RUNLOCK_NOSLEEP();
1087}
1088
1089/*
1090 * Notify a tcp user of an asynchronous error;
1091 * store error as soft error, but wake up user
1092 * (for now, won't do anything until can select for soft error).
1093 *
1094 * Do not wake up user since there currently is no mechanism for
1095 * reporting soft errors (yet - a kqueue filter may be added).
1096 */
1097static struct inpcb *
1098tcp_notify(struct inpcb *inp, int error)
1099{
1100 struct tcpcb *tp;
1101
1102 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1103 INP_WLOCK_ASSERT(inp);
1104
1105 if ((inp->inp_flags & INP_TIMEWAIT) ||
1106 (inp->inp_flags & INP_DROPPED))
1107 return (inp);
1108
1109 tp = intotcpcb(inp);
1110 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1111
1112 /*
1113 * Ignore some errors if we are hooked up.
1114 * If connection hasn't completed, has retransmitted several times,
1115 * and receives a second error, give up now. This is better
1116 * than waiting a long time to establish a connection that
1117 * can never complete.
1118 */
1119 if (tp->t_state == TCPS_ESTABLISHED &&
1120 (error == EHOSTUNREACH || error == ENETUNREACH ||
1121 error == EHOSTDOWN)) {
1122 return (inp);
1123 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1124 tp->t_softerror) {
1125 tp = tcp_drop(tp, error);
1126 if (tp != NULL)
1127 return (inp);
1128 else
1129 return (NULL);
1130 } else {
1131 tp->t_softerror = error;
1132 return (inp);
1133 }
1134#if 0
1135 wakeup( &so->so_timeo);
1136 sorwakeup(so);
1137 sowwakeup(so);
1138#endif
1139}
1140
1141static int
1142tcp_pcblist(SYSCTL_HANDLER_ARGS)
1143{
1144 int error, i, m, n, pcb_count;
1145 struct inpcb *inp, **inp_list;
1146 inp_gen_t gencnt;
1147 struct xinpgen xig;
1148
1149 /*
1150 * The process of preparing the TCB list is too time-consuming and
1151 * resource-intensive to repeat twice on every request.
1152 */
1153 if (req->oldptr == NULL) {
1154 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1155 n += imax(n / 8, 10);
1156 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1157 return (0);
1158 }
1159
1160 if (req->newptr != NULL)
1161 return (EPERM);
1162
1163 /*
1164 * OK, now we're committed to doing something.
1165 */
1166 INP_INFO_RLOCK(&V_tcbinfo);
1167 gencnt = V_tcbinfo.ipi_gencnt;
1168 n = V_tcbinfo.ipi_count;
1169 INP_INFO_RUNLOCK(&V_tcbinfo);
1170
1171 m = syncache_pcbcount();
1172
1173 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1174 + (n + m) * sizeof(struct xtcpcb));
1175 if (error != 0)
1176 return (error);
1177
1178 xig.xig_len = sizeof xig;
1179 xig.xig_count = n + m;
1180 xig.xig_gen = gencnt;
1181 xig.xig_sogen = so_gencnt;
1182 error = SYSCTL_OUT(req, &xig, sizeof xig);
1183 if (error)
1184 return (error);
1185
1186 error = syncache_pcblist(req, m, &pcb_count);
1187 if (error)
1188 return (error);
1189
1190 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1191 if (inp_list == NULL)
1192 return (ENOMEM);
1193
1194 INP_INFO_RLOCK(&V_tcbinfo);
1195 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1196 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1197 INP_WLOCK(inp);
1198 if (inp->inp_gencnt <= gencnt) {
1199 /*
1200 * XXX: This use of cr_cansee(), introduced with
1201 * TCP state changes, is not quite right, but for
1202 * now, better than nothing.
1203 */
1204 if (inp->inp_flags & INP_TIMEWAIT) {
1205 if (intotw(inp) != NULL)
1206 error = cr_cansee(req->td->td_ucred,
1207 intotw(inp)->tw_cred);
1208 else
1209 error = EINVAL; /* Skip this inp. */
1210 } else
1211 error = cr_canseeinpcb(req->td->td_ucred, inp);
1212 if (error == 0) {
1213 in_pcbref(inp);
1214 inp_list[i++] = inp;
1215 }
1216 }
1217 INP_WUNLOCK(inp);
1218 }
1219 INP_INFO_RUNLOCK(&V_tcbinfo);
1220 n = i;
1221
1222 error = 0;
1223 for (i = 0; i < n; i++) {
1224 inp = inp_list[i];
1225 INP_RLOCK(inp);
1226 if (inp->inp_gencnt <= gencnt) {
1227 struct xtcpcb xt;
1228 void *inp_ppcb;
1229
1230 bzero(&xt, sizeof(xt));
1231 xt.xt_len = sizeof xt;
1232 /* XXX should avoid extra copy */
1233 bcopy(inp, &xt.xt_inp, sizeof *inp);
1234 inp_ppcb = inp->inp_ppcb;
1235 if (inp_ppcb == NULL)
1236 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1237 else if (inp->inp_flags & INP_TIMEWAIT) {
1238 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1239 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1240 } else {
1241 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1242 if (xt.xt_tp.t_timers)
1243 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1244 }
1245 if (inp->inp_socket != NULL)
1246 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1247 else {
1248 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1249 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1250 }
1251 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1252 INP_RUNLOCK(inp);
1253 error = SYSCTL_OUT(req, &xt, sizeof xt);
1254 } else
1255 INP_RUNLOCK(inp);
1256 }
1257 INP_INFO_WLOCK(&V_tcbinfo);
1258 for (i = 0; i < n; i++) {
1259 inp = inp_list[i];
1260 INP_RLOCK(inp);
1261 if (!in_pcbrele_rlocked(inp))
1262 INP_RUNLOCK(inp);
1263 }
1264 INP_INFO_WUNLOCK(&V_tcbinfo);
1265
1266 if (!error) {
1267 /*
1268 * Give the user an updated idea of our state.
1269 * If the generation differs from what we told
1270 * her before, she knows that something happened
1271 * while we were processing this request, and it
1272 * might be necessary to retry.
1273 */
1274 INP_INFO_RLOCK(&V_tcbinfo);
1275 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1276 xig.xig_sogen = so_gencnt;
1277 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1278 INP_INFO_RUNLOCK(&V_tcbinfo);
1279 error = SYSCTL_OUT(req, &xig, sizeof xig);
1280 }
1281 free(inp_list, M_TEMP);
1282 return (error);
1283}
1284
1285SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1286 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1287 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1288
1289#ifdef INET
1290static int
1291tcp_getcred(SYSCTL_HANDLER_ARGS)
1292{
1293 struct xucred xuc;
1294 struct sockaddr_in addrs[2];
1295 struct inpcb *inp;
1296 int error;
1297
1298 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1299 if (error)
1300 return (error);
1301 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1302 if (error)
1303 return (error);
1304 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1305 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1306 if (inp != NULL) {
1307 if (inp->inp_socket == NULL)
1308 error = ENOENT;
1309 if (error == 0)
1310 error = cr_canseeinpcb(req->td->td_ucred, inp);
1311 if (error == 0)
1312 cru2x(inp->inp_cred, &xuc);
1313 INP_RUNLOCK(inp);
1314 } else
1315 error = ENOENT;
1316 if (error == 0)
1317 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1318 return (error);
1319}
1320
1321SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1322 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1323 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1324#endif /* INET */
1325
1326#ifdef INET6
1327static int
1328tcp6_getcred(SYSCTL_HANDLER_ARGS)
1329{
1330 struct xucred xuc;
1331 struct sockaddr_in6 addrs[2];
1332 struct inpcb *inp;
1333 int error;
1334#ifdef INET
1335 int mapped = 0;
1336#endif
1337
1338 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1339 if (error)
1340 return (error);
1341 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1342 if (error)
1343 return (error);
1344 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1345 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1346 return (error);
1347 }
1348 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1349#ifdef INET
1350 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1351 mapped = 1;
1352 else
1353#endif
1354 return (EINVAL);
1355 }
1356
1357#ifdef INET
1358 if (mapped == 1)
1359 inp = in_pcblookup(&V_tcbinfo,
1360 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1361 addrs[1].sin6_port,
1362 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1363 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1364 else
1365#endif
1366 inp = in6_pcblookup(&V_tcbinfo,
1367 &addrs[1].sin6_addr, addrs[1].sin6_port,
1368 &addrs[0].sin6_addr, addrs[0].sin6_port,
1369 INPLOOKUP_RLOCKPCB, NULL);
1370 if (inp != NULL) {
1371 if (inp->inp_socket == NULL)
1372 error = ENOENT;
1373 if (error == 0)
1374 error = cr_canseeinpcb(req->td->td_ucred, inp);
1375 if (error == 0)
1376 cru2x(inp->inp_cred, &xuc);
1377 INP_RUNLOCK(inp);
1378 } else
1379 error = ENOENT;
1380 if (error == 0)
1381 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1382 return (error);
1383}
1384
1385SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1386 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1387 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1388#endif /* INET6 */
1389
1390
1391#ifdef INET
1392void
1393tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1394{
1395 struct ip *ip = vip;
1396 struct tcphdr *th;
1397 struct in_addr faddr;
1398 struct inpcb *inp;
1399 struct tcpcb *tp;
1400 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1401 struct icmp *icp;
1402 struct in_conninfo inc;
1403 tcp_seq icmp_tcp_seq;
1404 int mtu;
1405
1406 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1407 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1408 return;
1409
1410 if (cmd == PRC_MSGSIZE)
1411 notify = tcp_mtudisc_notify;
1412 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1413 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1414 notify = tcp_drop_syn_sent;
1415 /*
1416 * Redirects don't need to be handled up here.
1417 */
1418 else if (PRC_IS_REDIRECT(cmd))
1419 return;
1420 /*
1421 * Source quench is depreciated.
1422 */
1423 else if (cmd == PRC_QUENCH)
1424 return;
1425 /*
1426 * Hostdead is ugly because it goes linearly through all PCBs.
1427 * XXX: We never get this from ICMP, otherwise it makes an
1428 * excellent DoS attack on machines with many connections.
1429 */
1430 else if (cmd == PRC_HOSTDEAD)
1431 ip = NULL;
1432 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1433 return;
1434 if (ip != NULL) {
1435 icp = (struct icmp *)((caddr_t)ip
1436 - offsetof(struct icmp, icmp_ip));
1437 th = (struct tcphdr *)((caddr_t)ip
1438 + (ip->ip_hl << 2));
1439 INP_INFO_WLOCK(&V_tcbinfo);
1440 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1441 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1442 if (inp != NULL) {
1443 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1444 !(inp->inp_flags & INP_DROPPED) &&
1445 !(inp->inp_socket == NULL)) {
1446 icmp_tcp_seq = htonl(th->th_seq);
1447 tp = intotcpcb(inp);
1448 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1449 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1450 if (cmd == PRC_MSGSIZE) {
1451 /*
1452 * MTU discovery:
1453 * If we got a needfrag set the MTU
1454 * in the route to the suggested new
1455 * value (if given) and then notify.
1456 */
1457 bzero(&inc, sizeof(inc));
1458 inc.inc_faddr = faddr;
1459 inc.inc_fibnum =
1460 inp->inp_inc.inc_fibnum;
1461
1462 mtu = ntohs(icp->icmp_nextmtu);
1463 /*
1464 * If no alternative MTU was
1465 * proposed, try the next smaller
1466 * one.
1467 */
1468 if (!mtu)
1469 mtu = ip_next_mtu(
1470 ntohs(ip->ip_len), 1);
1471 if (mtu < V_tcp_minmss
1472 + sizeof(struct tcpiphdr))
1473 mtu = V_tcp_minmss
1474 + sizeof(struct tcpiphdr);
1475 /*
1476 * Only cache the MTU if it
1477 * is smaller than the interface
1478 * or route MTU. tcp_mtudisc()
1479 * will do right thing by itself.
1480 */
1481 if (mtu <= tcp_maxmtu(&inc, NULL))
1482 tcp_hc_updatemtu(&inc, mtu);
1483 tcp_mtudisc(inp, mtu);
1484 } else
1485 inp = (*notify)(inp,
1486 inetctlerrmap[cmd]);
1487 }
1488 }
1489 if (inp != NULL)
1490 INP_WUNLOCK(inp);
1491 } else {
1492 bzero(&inc, sizeof(inc));
1493 inc.inc_fport = th->th_dport;
1494 inc.inc_lport = th->th_sport;
1495 inc.inc_faddr = faddr;
1496 inc.inc_laddr = ip->ip_src;
1497 syncache_unreach(&inc, th);
1498 }
1499 INP_INFO_WUNLOCK(&V_tcbinfo);
1500 } else
1501 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1502}
1503#endif /* INET */
1504
1505#ifdef INET6
1506void
1507tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1508{
1509 struct tcphdr th;
1510 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1511 struct ip6_hdr *ip6;
1512 struct mbuf *m;
1513 struct ip6ctlparam *ip6cp = NULL;
1514 const struct sockaddr_in6 *sa6_src = NULL;
1515 int off;
1516 struct tcp_portonly {
1517 u_int16_t th_sport;
1518 u_int16_t th_dport;
1519 } *thp;
1520
1521 if (sa->sa_family != AF_INET6 ||
1522 sa->sa_len != sizeof(struct sockaddr_in6))
1523 return;
1524
1525 if (cmd == PRC_MSGSIZE)
1526 notify = tcp_mtudisc_notify;
1527 else if (!PRC_IS_REDIRECT(cmd) &&
1528 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1529 return;
1530 /* Source quench is depreciated. */
1531 else if (cmd == PRC_QUENCH)
1532 return;
1533
1534 /* if the parameter is from icmp6, decode it. */
1535 if (d != NULL) {
1536 ip6cp = (struct ip6ctlparam *)d;
1537 m = ip6cp->ip6c_m;
1538 ip6 = ip6cp->ip6c_ip6;
1539 off = ip6cp->ip6c_off;
1540 sa6_src = ip6cp->ip6c_src;
1541 } else {
1542 m = NULL;
1543 ip6 = NULL;
1544 off = 0; /* fool gcc */
1545 sa6_src = &sa6_any;
1546 }
1547
1548 if (ip6 != NULL) {
1549 struct in_conninfo inc;
1550 /*
1551 * XXX: We assume that when IPV6 is non NULL,
1552 * M and OFF are valid.
1553 */
1554
1555 /* check if we can safely examine src and dst ports */
1556 if (m->m_pkthdr.len < off + sizeof(*thp))
1557 return;
1558
1559 bzero(&th, sizeof(th));
1560 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1561
1562 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1563 (struct sockaddr *)ip6cp->ip6c_src,
1564 th.th_sport, cmd, NULL, notify);
1565
1566 bzero(&inc, sizeof(inc));
1567 inc.inc_fport = th.th_dport;
1568 inc.inc_lport = th.th_sport;
1569 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1570 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1571 inc.inc_flags |= INC_ISIPV6;
1572 INP_INFO_WLOCK(&V_tcbinfo);
1573 syncache_unreach(&inc, &th);
1574 INP_INFO_WUNLOCK(&V_tcbinfo);
1575 } else
1576 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1577 0, cmd, NULL, notify);
1578}
1579#endif /* INET6 */
1580
1581
1582/*
1583 * Following is where TCP initial sequence number generation occurs.
1584 *
1585 * There are two places where we must use initial sequence numbers:
1586 * 1. In SYN-ACK packets.
1587 * 2. In SYN packets.
1588 *
1589 * All ISNs for SYN-ACK packets are generated by the syncache. See
1590 * tcp_syncache.c for details.
1591 *
1592 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1593 * depends on this property. In addition, these ISNs should be
1594 * unguessable so as to prevent connection hijacking. To satisfy
1595 * the requirements of this situation, the algorithm outlined in
1596 * RFC 1948 is used, with only small modifications.
1597 *
1598 * Implementation details:
1599 *
1600 * Time is based off the system timer, and is corrected so that it
1601 * increases by one megabyte per second. This allows for proper
1602 * recycling on high speed LANs while still leaving over an hour
1603 * before rollover.
1604 *
1605 * As reading the *exact* system time is too expensive to be done
1606 * whenever setting up a TCP connection, we increment the time
1607 * offset in two ways. First, a small random positive increment
1608 * is added to isn_offset for each connection that is set up.
1609 * Second, the function tcp_isn_tick fires once per clock tick
1610 * and increments isn_offset as necessary so that sequence numbers
1611 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1612 * random positive increments serve only to ensure that the same
1613 * exact sequence number is never sent out twice (as could otherwise
1614 * happen when a port is recycled in less than the system tick
1615 * interval.)
1616 *
1617 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1618 * between seeding of isn_secret. This is normally set to zero,
1619 * as reseeding should not be necessary.
1620 *
1621 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1622 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1623 * general, this means holding an exclusive (write) lock.
1624 */
1625
1626#define ISN_BYTES_PER_SECOND 1048576
1627#define ISN_STATIC_INCREMENT 4096
1628#define ISN_RANDOM_INCREMENT (4096 - 1)
1629
1630static VNET_DEFINE(u_char, isn_secret[32]);
1631static VNET_DEFINE(int, isn_last);
1632static VNET_DEFINE(int, isn_last_reseed);
1633static VNET_DEFINE(u_int32_t, isn_offset);
1634static VNET_DEFINE(u_int32_t, isn_offset_old);
1635
1636#define V_isn_secret VNET(isn_secret)
1637#define V_isn_last VNET(isn_last)
1638#define V_isn_last_reseed VNET(isn_last_reseed)
1639#define V_isn_offset VNET(isn_offset)
1640#define V_isn_offset_old VNET(isn_offset_old)
1641
1642tcp_seq
1643tcp_new_isn(struct tcpcb *tp)
1644{
1645 MD5_CTX isn_ctx;
1646 u_int32_t md5_buffer[4];
1647 tcp_seq new_isn;
1648 u_int32_t projected_offset;
1649
1650 INP_WLOCK_ASSERT(tp->t_inpcb);
1651
1652 ISN_LOCK();
1653 /* Seed if this is the first use, reseed if requested. */
1654 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1655 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1656 < (u_int)ticks))) {
1657 read_random(&V_isn_secret, sizeof(V_isn_secret));
1658 V_isn_last_reseed = ticks;
1659 }
1660
1661 /* Compute the md5 hash and return the ISN. */
1662 MD5Init(&isn_ctx);
1663 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1664 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1665#ifdef INET6
1666 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1667 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1668 sizeof(struct in6_addr));
1669 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1670 sizeof(struct in6_addr));
1671 } else
1672#endif
1673 {
1674 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1675 sizeof(struct in_addr));
1676 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1677 sizeof(struct in_addr));
1678 }
1679 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1680 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1681 new_isn = (tcp_seq) md5_buffer[0];
1682 V_isn_offset += ISN_STATIC_INCREMENT +
1683 (arc4random() & ISN_RANDOM_INCREMENT);
1684 if (ticks != V_isn_last) {
1685 projected_offset = V_isn_offset_old +
1686 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1687 if (SEQ_GT(projected_offset, V_isn_offset))
1688 V_isn_offset = projected_offset;
1689 V_isn_offset_old = V_isn_offset;
1690 V_isn_last = ticks;
1691 }
1692 new_isn += V_isn_offset;
1693 ISN_UNLOCK();
1694 return (new_isn);
1695}
1696
1697/*
1698 * When a specific ICMP unreachable message is received and the
1699 * connection state is SYN-SENT, drop the connection. This behavior
1700 * is controlled by the icmp_may_rst sysctl.
1701 */
1702struct inpcb *
1703tcp_drop_syn_sent(struct inpcb *inp, int errno)
1704{
1705 struct tcpcb *tp;
1706
1707 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1708 INP_WLOCK_ASSERT(inp);
1709
1710 if ((inp->inp_flags & INP_TIMEWAIT) ||
1711 (inp->inp_flags & INP_DROPPED))
1712 return (inp);
1713
1714 tp = intotcpcb(inp);
1715 if (tp->t_state != TCPS_SYN_SENT)
1716 return (inp);
1717
1718 tp = tcp_drop(tp, errno);
1719 if (tp != NULL)
1720 return (inp);
1721 else
1722 return (NULL);
1723}
1724
1725/*
1726 * When `need fragmentation' ICMP is received, update our idea of the MSS
1727 * based on the new value. Also nudge TCP to send something, since we
1728 * know the packet we just sent was dropped.
1729 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1730 */
1731static struct inpcb *
1732tcp_mtudisc_notify(struct inpcb *inp, int error)
1733{
1734
1735 return (tcp_mtudisc(inp, -1));
1736}
1737
1738struct inpcb *
1739tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1740{
1741 struct tcpcb *tp;
1742 struct socket *so;
1743
1744 INP_WLOCK_ASSERT(inp);
1745 if ((inp->inp_flags & INP_TIMEWAIT) ||
1746 (inp->inp_flags & INP_DROPPED))
1747 return (inp);
1748
1749 tp = intotcpcb(inp);
1750 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1751
1752 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1753
1754 so = inp->inp_socket;
1755 SOCKBUF_LOCK(&so->so_snd);
1756 /* If the mss is larger than the socket buffer, decrease the mss. */
1757 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1758 tp->t_maxseg = so->so_snd.sb_hiwat;
1759 SOCKBUF_UNLOCK(&so->so_snd);
1760
1761 TCPSTAT_INC(tcps_mturesent);
1762 tp->t_rtttime = 0;
1763 tp->snd_nxt = tp->snd_una;
1764 tcp_free_sackholes(tp);
1765 tp->snd_recover = tp->snd_max;
1766 if (tp->t_flags & TF_SACK_PERMIT)
1767 EXIT_FASTRECOVERY(tp->t_flags);
1768 tcp_output(tp);
1769 return (inp);
1770}
1771
1772#ifdef INET
1773/*
1774 * Look-up the routing entry to the peer of this inpcb. If no route
1775 * is found and it cannot be allocated, then return 0. This routine
1776 * is called by TCP routines that access the rmx structure and by
1777 * tcp_mss_update to get the peer/interface MTU.
1778 */
1779u_long
1780tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1781{
1782 struct route sro;
1783 struct sockaddr_in *dst;
1784 struct ifnet *ifp;
1785 u_long maxmtu = 0;
1786
1787 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1788
1789 bzero(&sro, sizeof(sro));
1790 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1791 dst = (struct sockaddr_in *)&sro.ro_dst;
1792 dst->sin_family = AF_INET;
1793 dst->sin_len = sizeof(*dst);
1794 dst->sin_addr = inc->inc_faddr;
1795 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1796 }
1797 if (sro.ro_rt != NULL) {
1798 ifp = sro.ro_rt->rt_ifp;
1799 if (sro.ro_rt->rt_rmx.rmx_mtu == 0)
1800 maxmtu = ifp->if_mtu;
1801 else
1802 maxmtu = min(sro.ro_rt->rt_rmx.rmx_mtu, ifp->if_mtu);
1803
1804 /* Report additional interface capabilities. */
1805 if (cap != NULL) {
1806 if (ifp->if_capenable & IFCAP_TSO4 &&
1807 ifp->if_hwassist & CSUM_TSO)
1808 cap->ifcap |= CSUM_TSO;
1809 cap->tsomax = ifp->if_hw_tsomax;
1810 }
1811 RTFREE(sro.ro_rt);
1812 }
1813 return (maxmtu);
1814}
1815#endif /* INET */
1816
1817#ifdef INET6
1818u_long
1819tcp_maxmtu6(struct in_conninfo *inc, struct tcp_ifcap *cap)
1820{
1821 struct route_in6 sro6;
1822 struct ifnet *ifp;
1823 u_long maxmtu = 0;
1824
1825 KASSERT(inc != NULL, ("tcp_maxmtu6 with NULL in_conninfo pointer"));
1826
1827 bzero(&sro6, sizeof(sro6));
1828 if (!IN6_IS_ADDR_UNSPECIFIED(&inc->inc6_faddr)) {
1829 sro6.ro_dst.sin6_family = AF_INET6;
1830 sro6.ro_dst.sin6_len = sizeof(struct sockaddr_in6);
1831 sro6.ro_dst.sin6_addr = inc->inc6_faddr;
1832 in6_rtalloc_ign(&sro6, 0, inc->inc_fibnum);
1833 }
1834 if (sro6.ro_rt != NULL) {
1835 ifp = sro6.ro_rt->rt_ifp;
1836 if (sro6.ro_rt->rt_rmx.rmx_mtu == 0)
1837 maxmtu = IN6_LINKMTU(sro6.ro_rt->rt_ifp);
1838 else
1839 maxmtu = min(sro6.ro_rt->rt_rmx.rmx_mtu,
1840 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1841
1842 /* Report additional interface capabilities. */
1843 if (cap != NULL) {
1844 if (ifp->if_capenable & IFCAP_TSO6 &&
1845 ifp->if_hwassist & CSUM_TSO)
1846 cap->ifcap |= CSUM_TSO;
1847 cap->tsomax = ifp->if_hw_tsomax;
1848 }
1849 RTFREE(sro6.ro_rt);
1850 }
1851
1852 return (maxmtu);
1853}
1854#endif /* INET6 */
1855
1856#ifdef IPSEC
1857/* compute ESP/AH header size for TCP, including outer IP header. */
1858size_t
1859ipsec_hdrsiz_tcp(struct tcpcb *tp)
1860{
1861 struct inpcb *inp;
1862 struct mbuf *m;
1863 size_t hdrsiz;
1864 struct ip *ip;
1865#ifdef INET6
1866 struct ip6_hdr *ip6;
1867#endif
1868 struct tcphdr *th;
1869
1870 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1871 return (0);
1872 m = m_gethdr(M_NOWAIT, MT_DATA);
1873 if (!m)
1874 return (0);
1875
1876#ifdef INET6
1877 if ((inp->inp_vflag & INP_IPV6) != 0) {
1878 ip6 = mtod(m, struct ip6_hdr *);
1879 th = (struct tcphdr *)(ip6 + 1);
1880 m->m_pkthdr.len = m->m_len =
1881 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1882 tcpip_fillheaders(inp, ip6, th);
1883 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1884 } else
1885#endif /* INET6 */
1886 {
1887 ip = mtod(m, struct ip *);
1888 th = (struct tcphdr *)(ip + 1);
1889 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1890 tcpip_fillheaders(inp, ip, th);
1891 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1892 }
1893
1894 m_free(m);
1895 return (hdrsiz);
1896}
1897#endif /* IPSEC */
1898
1899#ifdef TCP_SIGNATURE
1900/*
1901 * Callback function invoked by m_apply() to digest TCP segment data
1902 * contained within an mbuf chain.
1903 */
1904static int
1905tcp_signature_apply(void *fstate, void *data, u_int len)
1906{
1907
1908 MD5Update(fstate, (u_char *)data, len);
1909 return (0);
1910}
1911
1912/*
1913 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1914 *
1915 * Parameters:
1916 * m pointer to head of mbuf chain
1917 * _unused
1918 * len length of TCP segment data, excluding options
1919 * optlen length of TCP segment options
1920 * buf pointer to storage for computed MD5 digest
1921 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1922 *
1923 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1924 * When called from tcp_input(), we can be sure that th_sum has been
1925 * zeroed out and verified already.
1926 *
1927 * Return 0 if successful, otherwise return -1.
1928 *
1929 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1930 * search with the destination IP address, and a 'magic SPI' to be
1931 * determined by the application. This is hardcoded elsewhere to 1179
1932 * right now. Another branch of this code exists which uses the SPD to
1933 * specify per-application flows but it is unstable.
1934 */
1935int
1936tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1937 u_char *buf, u_int direction)
1938{
1939 union sockaddr_union dst;
1940#ifdef INET
1941 struct ippseudo ippseudo;
1942#endif
1943 MD5_CTX ctx;
1944 int doff;
1945 struct ip *ip;
1946#ifdef INET
1947 struct ipovly *ipovly;
1948#endif
1949 struct secasvar *sav;
1950 struct tcphdr *th;
1951#ifdef INET6
1952 struct ip6_hdr *ip6;
1953 struct in6_addr in6;
1954 char ip6buf[INET6_ADDRSTRLEN];
1955 uint32_t plen;
1956 uint16_t nhdr;
1957#endif
1958 u_short savecsum;
1959
1960 KASSERT(m != NULL, ("NULL mbuf chain"));
1961 KASSERT(buf != NULL, ("NULL signature pointer"));
1962
1963 /* Extract the destination from the IP header in the mbuf. */
1964 bzero(&dst, sizeof(union sockaddr_union));
1965 ip = mtod(m, struct ip *);
1966#ifdef INET6
1967 ip6 = NULL; /* Make the compiler happy. */
1968#endif
1969 switch (ip->ip_v) {
1970#ifdef INET
1971 case IPVERSION:
1972 dst.sa.sa_len = sizeof(struct sockaddr_in);
1973 dst.sa.sa_family = AF_INET;
1974 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1975 ip->ip_src : ip->ip_dst;
1976 break;
1977#endif
1978#ifdef INET6
1979 case (IPV6_VERSION >> 4):
1980 ip6 = mtod(m, struct ip6_hdr *);
1981 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1982 dst.sa.sa_family = AF_INET6;
1983 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1984 ip6->ip6_src : ip6->ip6_dst;
1985 break;
1986#endif
1987 default:
1988 return (EINVAL);
1989 /* NOTREACHED */
1990 break;
1991 }
1992
1993 /* Look up an SADB entry which matches the address of the peer. */
1994 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
1995 if (sav == NULL) {
1996 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
1997 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
1998#ifdef INET6
1999 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2000 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2001#endif
2002 "(unsupported)"));
2003 return (EINVAL);
2004 }
2005
2006 MD5Init(&ctx);
2007 /*
2008 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2009 *
2010 * XXX The ippseudo header MUST be digested in network byte order,
2011 * or else we'll fail the regression test. Assume all fields we've
2012 * been doing arithmetic on have been in host byte order.
2013 * XXX One cannot depend on ipovly->ih_len here. When called from
2014 * tcp_output(), the underlying ip_len member has not yet been set.
2015 */
2016 switch (ip->ip_v) {
2017#ifdef INET
2018 case IPVERSION:
2019 ipovly = (struct ipovly *)ip;
2020 ippseudo.ippseudo_src = ipovly->ih_src;
2021 ippseudo.ippseudo_dst = ipovly->ih_dst;
2022 ippseudo.ippseudo_pad = 0;
2023 ippseudo.ippseudo_p = IPPROTO_TCP;
2024 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2025 optlen);
2026 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2027
2028 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2029 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2030 break;
2031#endif
2032#ifdef INET6
2033 /*
2034 * RFC 2385, 2.0 Proposal
2035 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2036 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2037 * extended next header value (to form 32 bits), and 32-bit segment
2038 * length.
2039 * Note: Upper-Layer Packet Length comes before Next Header.
2040 */
2041 case (IPV6_VERSION >> 4):
2042 in6 = ip6->ip6_src;
2043 in6_clearscope(&in6);
2044 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2045 in6 = ip6->ip6_dst;
2046 in6_clearscope(&in6);
2047 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2048 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2049 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2050 nhdr = 0;
2051 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2052 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2053 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2054 nhdr = IPPROTO_TCP;
2055 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2056
2057 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2058 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2059 break;
2060#endif
2061 default:
2062 return (EINVAL);
2063 /* NOTREACHED */
2064 break;
2065 }
2066
2067
2068 /*
2069 * Step 2: Update MD5 hash with TCP header, excluding options.
2070 * The TCP checksum must be set to zero.
2071 */
2072 savecsum = th->th_sum;
2073 th->th_sum = 0;
2074 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2075 th->th_sum = savecsum;
2076
2077 /*
2078 * Step 3: Update MD5 hash with TCP segment data.
2079 * Use m_apply() to avoid an early m_pullup().
2080 */
2081 if (len > 0)
2082 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2083
2084 /*
2085 * Step 4: Update MD5 hash with shared secret.
2086 */
2087 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2088 MD5Final(buf, &ctx);
2089
2090 key_sa_recordxfer(sav, m);
2091 KEY_FREESAV(&sav);
2092 return (0);
2093}
2094
2095/*
2096 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2097 *
2098 * Parameters:
2099 * m pointer to head of mbuf chain
2100 * len length of TCP segment data, excluding options
2101 * optlen length of TCP segment options
2102 * buf pointer to storage for computed MD5 digest
2103 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2104 *
2105 * Return 1 if successful, otherwise return 0.
2106 */
2107int
2108tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2109 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2110{
2111 char tmpdigest[TCP_SIGLEN];
2112
2113 if (tcp_sig_checksigs == 0)
2114 return (1);
2115 if ((tcpbflag & TF_SIGNATURE) == 0) {
2116 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2117
2118 /*
2119 * If this socket is not expecting signature but
2120 * the segment contains signature just fail.
2121 */
2122 TCPSTAT_INC(tcps_sig_err_sigopt);
2123 TCPSTAT_INC(tcps_sig_rcvbadsig);
2124 return (0);
2125 }
2126
2127 /* Signature is not expected, and not present in segment. */
2128 return (1);
2129 }
2130
2131 /*
2132 * If this socket is expecting signature but the segment does not
2133 * contain any just fail.
2134 */
2135 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2136 TCPSTAT_INC(tcps_sig_err_nosigopt);
2137 TCPSTAT_INC(tcps_sig_rcvbadsig);
2138 return (0);
2139 }
2140 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2141 IPSEC_DIR_INBOUND) == -1) {
2142 TCPSTAT_INC(tcps_sig_err_buildsig);
2143 TCPSTAT_INC(tcps_sig_rcvbadsig);
2144 return (0);
2145 }
2146
2147 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2148 TCPSTAT_INC(tcps_sig_rcvbadsig);
2149 return (0);
2150 }
2151 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2152 return (1);
2153}
2154#endif /* TCP_SIGNATURE */
2155
2156static int
2157sysctl_drop(SYSCTL_HANDLER_ARGS)
2158{
2159 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2160 struct sockaddr_storage addrs[2];
2161 struct inpcb *inp;
2162 struct tcpcb *tp;
2163 struct tcptw *tw;
2164 struct sockaddr_in *fin, *lin;
2165#ifdef INET6
2166 struct sockaddr_in6 *fin6, *lin6;
2167#endif
2168 int error;
2169
2170 inp = NULL;
2171 fin = lin = NULL;
2172#ifdef INET6
2173 fin6 = lin6 = NULL;
2174#endif
2175 error = 0;
2176
2177 if (req->oldptr != NULL || req->oldlen != 0)
2178 return (EINVAL);
2179 if (req->newptr == NULL)
2180 return (EPERM);
2181 if (req->newlen < sizeof(addrs))
2182 return (ENOMEM);
2183 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2184 if (error)
2185 return (error);
2186
2187 switch (addrs[0].ss_family) {
2188#ifdef INET6
2189 case AF_INET6:
2190 fin6 = (struct sockaddr_in6 *)&addrs[0];
2191 lin6 = (struct sockaddr_in6 *)&addrs[1];
2192 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2193 lin6->sin6_len != sizeof(struct sockaddr_in6))
2194 return (EINVAL);
2195 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2196 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2197 return (EINVAL);
2198 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2199 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2200 fin = (struct sockaddr_in *)&addrs[0];
2201 lin = (struct sockaddr_in *)&addrs[1];
2202 break;
2203 }
2204 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2205 if (error)
2206 return (error);
2207 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2208 if (error)
2209 return (error);
2210 break;
2211#endif
2212#ifdef INET
2213 case AF_INET:
2214 fin = (struct sockaddr_in *)&addrs[0];
2215 lin = (struct sockaddr_in *)&addrs[1];
2216 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2217 lin->sin_len != sizeof(struct sockaddr_in))
2218 return (EINVAL);
2219 break;
2220#endif
2221 default:
2222 return (EINVAL);
2223 }
2224 INP_INFO_WLOCK(&V_tcbinfo);
2225 switch (addrs[0].ss_family) {
2226#ifdef INET6
2227 case AF_INET6:
2228 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2229 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2230 INPLOOKUP_WLOCKPCB, NULL);
2231 break;
2232#endif
2233#ifdef INET
2234 case AF_INET:
2235 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2236 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2237 break;
2238#endif
2239 }
2240 if (inp != NULL) {
2241 if (inp->inp_flags & INP_TIMEWAIT) {
2242 /*
2243 * XXXRW: There currently exists a state where an
2244 * inpcb is present, but its timewait state has been
2245 * discarded. For now, don't allow dropping of this
2246 * type of inpcb.
2247 */
2248 tw = intotw(inp);
2249 if (tw != NULL)
2250 tcp_twclose(tw, 0);
2251 else
2252 INP_WUNLOCK(inp);
2253 } else if (!(inp->inp_flags & INP_DROPPED) &&
2254 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2255 tp = intotcpcb(inp);
2256 tp = tcp_drop(tp, ECONNABORTED);
2257 if (tp != NULL)
2258 INP_WUNLOCK(inp);
2259 } else
2260 INP_WUNLOCK(inp);
2261 } else
2262 error = ESRCH;
2263 INP_INFO_WUNLOCK(&V_tcbinfo);
2264 return (error);
2265}
2266
2267SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2268 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2269 0, sysctl_drop, "", "Drop TCP connection");
2270
2271/*
2272 * Generate a standardized TCP log line for use throughout the
2273 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2274 * allow use in the interrupt context.
2275 *
2276 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2277 * NB: The function may return NULL if memory allocation failed.
2278 *
2279 * Due to header inclusion and ordering limitations the struct ip
2280 * and ip6_hdr pointers have to be passed as void pointers.
2281 */
2282char *
2283tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2284 const void *ip6hdr)
2285{
2286
2287 /* Is logging enabled? */
2288 if (tcp_log_in_vain == 0)
2289 return (NULL);
2290
2291 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2292}
2293
2294char *
2295tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2296 const void *ip6hdr)
2297{
2298
2299 /* Is logging enabled? */
2300 if (tcp_log_debug == 0)
2301 return (NULL);
2302
2303 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2304}
2305
2306static char *
2307tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2308 const void *ip6hdr)
2309{
2310 char *s, *sp;
2311 size_t size;
2312 struct ip *ip;
2313#ifdef INET6
2314 const struct ip6_hdr *ip6;
2315
2316 ip6 = (const struct ip6_hdr *)ip6hdr;
2317#endif /* INET6 */
2318 ip = (struct ip *)ip4hdr;
2319
2320 /*
2321 * The log line looks like this:
2322 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2323 */
2324 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2325 sizeof(PRINT_TH_FLAGS) + 1 +
2326#ifdef INET6
2327 2 * INET6_ADDRSTRLEN;
2328#else
2329 2 * INET_ADDRSTRLEN;
2330#endif /* INET6 */
2331
2332 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2333 if (s == NULL)
2334 return (NULL);
2335
2336 strcat(s, "TCP: [");
2337 sp = s + strlen(s);
2338
2339 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2340 inet_ntoa_r(inc->inc_faddr, sp);
2341 sp = s + strlen(s);
2342 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2343 sp = s + strlen(s);
2344 inet_ntoa_r(inc->inc_laddr, sp);
2345 sp = s + strlen(s);
2346 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2347#ifdef INET6
2348 } else if (inc) {
2349 ip6_sprintf(sp, &inc->inc6_faddr);
2350 sp = s + strlen(s);
2351 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2352 sp = s + strlen(s);
2353 ip6_sprintf(sp, &inc->inc6_laddr);
2354 sp = s + strlen(s);
2355 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2356 } else if (ip6 && th) {
2357 ip6_sprintf(sp, &ip6->ip6_src);
2358 sp = s + strlen(s);
2359 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2360 sp = s + strlen(s);
2361 ip6_sprintf(sp, &ip6->ip6_dst);
2362 sp = s + strlen(s);
2363 sprintf(sp, "]:%i", ntohs(th->th_dport));
2364#endif /* INET6 */
2365#ifdef INET
2366 } else if (ip && th) {
2367 inet_ntoa_r(ip->ip_src, sp);
2368 sp = s + strlen(s);
2369 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2370 sp = s + strlen(s);
2371 inet_ntoa_r(ip->ip_dst, sp);
2372 sp = s + strlen(s);
2373 sprintf(sp, "]:%i", ntohs(th->th_dport));
2374#endif /* INET */
2375 } else {
2376 free(s, M_TCPLOG);
2377 return (NULL);
2378 }
2379 sp = s + strlen(s);
2380 if (th)
2381 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2382 if (*(s + size - 1) != '\0')
2383 panic("%s: string too long", __func__);
2384 return (s);
2385}
|