34
35#include "opt_compat.h"
36#include "opt_inet.h"
37#include "opt_inet6.h"
38#include "opt_ipsec.h"
39#include "opt_tcpdebug.h"
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/callout.h>
44#include <sys/hhook.h>
45#include <sys/kernel.h>
46#include <sys/khelp.h>
47#include <sys/sysctl.h>
48#include <sys/jail.h>
49#include <sys/malloc.h>
50#include <sys/mbuf.h>
51#ifdef INET6
52#include <sys/domain.h>
53#endif
54#include <sys/priv.h>
55#include <sys/proc.h>
56#include <sys/sdt.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/protosw.h>
60#include <sys/random.h>
61
62#include <vm/uma.h>
63
64#include <net/route.h>
65#include <net/if.h>
66#include <net/if_var.h>
67#include <net/vnet.h>
68
69#include <netinet/cc.h>
70#include <netinet/in.h>
71#include <netinet/in_kdtrace.h>
72#include <netinet/in_pcb.h>
73#include <netinet/in_systm.h>
74#include <netinet/in_var.h>
75#include <netinet/ip.h>
76#include <netinet/ip_icmp.h>
77#include <netinet/ip_var.h>
78#ifdef INET6
79#include <netinet/ip6.h>
80#include <netinet6/in6_pcb.h>
81#include <netinet6/ip6_var.h>
82#include <netinet6/scope6_var.h>
83#include <netinet6/nd6.h>
84#endif
85
86#include <netinet/tcp_fsm.h>
87#include <netinet/tcp_seq.h>
88#include <netinet/tcp_timer.h>
89#include <netinet/tcp_var.h>
90#include <netinet/tcp_syncache.h>
91#ifdef INET6
92#include <netinet6/tcp6_var.h>
93#endif
94#include <netinet/tcpip.h>
95#ifdef 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 int error;
435
436 tcp_reass_destroy();
437 tcp_hc_destroy();
438 syncache_destroy();
439 tcp_tw_destroy();
440 in_pcbinfo_destroy(&V_tcbinfo);
441 uma_zdestroy(V_sack_hole_zone);
442 uma_zdestroy(V_tcpcb_zone);
443
444 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
445 if (error != 0) {
446 printf("%s: WARNING: unable to deregister helper hook "
447 "type=%d, id=%d: error %d returned\n", __func__,
448 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
449 }
450 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
451 if (error != 0) {
452 printf("%s: WARNING: unable to deregister helper hook "
453 "type=%d, id=%d: error %d returned\n", __func__,
454 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
455 }
456}
457#endif
458
459void
460tcp_fini(void *xtp)
461{
462
463}
464
465/*
466 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
467 * tcp_template used to store this data in mbufs, but we now recopy it out
468 * of the tcpcb each time to conserve mbufs.
469 */
470void
471tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
472{
473 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
474
475 INP_WLOCK_ASSERT(inp);
476
477#ifdef INET6
478 if ((inp->inp_vflag & INP_IPV6) != 0) {
479 struct ip6_hdr *ip6;
480
481 ip6 = (struct ip6_hdr *)ip_ptr;
482 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
483 (inp->inp_flow & IPV6_FLOWINFO_MASK);
484 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
485 (IPV6_VERSION & IPV6_VERSION_MASK);
486 ip6->ip6_nxt = IPPROTO_TCP;
487 ip6->ip6_plen = htons(sizeof(struct tcphdr));
488 ip6->ip6_src = inp->in6p_laddr;
489 ip6->ip6_dst = inp->in6p_faddr;
490 }
491#endif /* INET6 */
492#if defined(INET6) && defined(INET)
493 else
494#endif
495#ifdef INET
496 {
497 struct ip *ip;
498
499 ip = (struct ip *)ip_ptr;
500 ip->ip_v = IPVERSION;
501 ip->ip_hl = 5;
502 ip->ip_tos = inp->inp_ip_tos;
503 ip->ip_len = 0;
504 ip->ip_id = 0;
505 ip->ip_off = 0;
506 ip->ip_ttl = inp->inp_ip_ttl;
507 ip->ip_sum = 0;
508 ip->ip_p = IPPROTO_TCP;
509 ip->ip_src = inp->inp_laddr;
510 ip->ip_dst = inp->inp_faddr;
511 }
512#endif /* INET */
513 th->th_sport = inp->inp_lport;
514 th->th_dport = inp->inp_fport;
515 th->th_seq = 0;
516 th->th_ack = 0;
517 th->th_x2 = 0;
518 th->th_off = 5;
519 th->th_flags = 0;
520 th->th_win = 0;
521 th->th_urp = 0;
522 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
523}
524
525/*
526 * Create template to be used to send tcp packets on a connection.
527 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
528 * use for this function is in keepalives, which use tcp_respond.
529 */
530struct tcptemp *
531tcpip_maketemplate(struct inpcb *inp)
532{
533 struct tcptemp *t;
534
535 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
536 if (t == NULL)
537 return (NULL);
538 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
539 return (t);
540}
541
542/*
543 * Send a single message to the TCP at address specified by
544 * the given TCP/IP header. If m == NULL, then we make a copy
545 * of the tcpiphdr at ti and send directly to the addressed host.
546 * This is used to force keep alive messages out using the TCP
547 * template for a connection. If flags are given then we send
548 * a message back to the TCP which originated the * segment ti,
549 * and discard the mbuf containing it and any other attached mbufs.
550 *
551 * In any case the ack and sequence number of the transmitted
552 * segment are as specified by the parameters.
553 *
554 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
555 */
556void
557tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
558 tcp_seq ack, tcp_seq seq, int flags)
559{
560 int tlen;
561 int win = 0;
562 struct ip *ip;
563 struct tcphdr *nth;
564#ifdef INET6
565 struct ip6_hdr *ip6;
566 int isipv6;
567#endif /* INET6 */
568 int ipflags = 0;
569 struct inpcb *inp;
570
571 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
572
573#ifdef INET6
574 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
575 ip6 = ipgen;
576#endif /* INET6 */
577 ip = ipgen;
578
579 if (tp != NULL) {
580 inp = tp->t_inpcb;
581 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
582 INP_WLOCK_ASSERT(inp);
583 } else
584 inp = NULL;
585
586 if (tp != NULL) {
587 if (!(flags & TH_RST)) {
588 win = sbspace(&inp->inp_socket->so_rcv);
589 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
590 win = (long)TCP_MAXWIN << tp->rcv_scale;
591 }
592 }
593 if (m == NULL) {
594 m = m_gethdr(M_NOWAIT, MT_DATA);
595 if (m == NULL)
596 return;
597 tlen = 0;
598 m->m_data += max_linkhdr;
599#ifdef INET6
600 if (isipv6) {
601 bcopy((caddr_t)ip6, mtod(m, caddr_t),
602 sizeof(struct ip6_hdr));
603 ip6 = mtod(m, struct ip6_hdr *);
604 nth = (struct tcphdr *)(ip6 + 1);
605 } else
606#endif /* INET6 */
607 {
608 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
609 ip = mtod(m, struct ip *);
610 nth = (struct tcphdr *)(ip + 1);
611 }
612 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
613 flags = TH_ACK;
614 } else {
615 /*
616 * reuse the mbuf.
617 * XXX MRT We inherrit the FIB, which is lucky.
618 */
619 m_freem(m->m_next);
620 m->m_next = NULL;
621 m->m_data = (caddr_t)ipgen;
622 /* m_len is set later */
623 tlen = 0;
624#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
625#ifdef INET6
626 if (isipv6) {
627 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
628 nth = (struct tcphdr *)(ip6 + 1);
629 } else
630#endif /* INET6 */
631 {
632 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
633 nth = (struct tcphdr *)(ip + 1);
634 }
635 if (th != nth) {
636 /*
637 * this is usually a case when an extension header
638 * exists between the IPv6 header and the
639 * TCP header.
640 */
641 nth->th_sport = th->th_sport;
642 nth->th_dport = th->th_dport;
643 }
644 xchg(nth->th_dport, nth->th_sport, uint16_t);
645#undef xchg
646 }
647#ifdef INET6
648 if (isipv6) {
649 ip6->ip6_flow = 0;
650 ip6->ip6_vfc = IPV6_VERSION;
651 ip6->ip6_nxt = IPPROTO_TCP;
652 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
653 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
654 }
655#endif
656#if defined(INET) && defined(INET6)
657 else
658#endif
659#ifdef INET
660 {
661 tlen += sizeof (struct tcpiphdr);
662 ip->ip_len = htons(tlen);
663 ip->ip_ttl = V_ip_defttl;
664 if (V_path_mtu_discovery)
665 ip->ip_off |= htons(IP_DF);
666 }
667#endif
668 m->m_len = tlen;
669 m->m_pkthdr.len = tlen;
670 m->m_pkthdr.rcvif = NULL;
671#ifdef MAC
672 if (inp != NULL) {
673 /*
674 * Packet is associated with a socket, so allow the
675 * label of the response to reflect the socket label.
676 */
677 INP_WLOCK_ASSERT(inp);
678 mac_inpcb_create_mbuf(inp, m);
679 } else {
680 /*
681 * Packet is not associated with a socket, so possibly
682 * update the label in place.
683 */
684 mac_netinet_tcp_reply(m);
685 }
686#endif
687 nth->th_seq = htonl(seq);
688 nth->th_ack = htonl(ack);
689 nth->th_x2 = 0;
690 nth->th_off = sizeof (struct tcphdr) >> 2;
691 nth->th_flags = flags;
692 if (tp != NULL)
693 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
694 else
695 nth->th_win = htons((u_short)win);
696 nth->th_urp = 0;
697
698 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
699#ifdef INET6
700 if (isipv6) {
701 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
702 nth->th_sum = in6_cksum_pseudo(ip6,
703 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
704 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
705 NULL, NULL);
706 }
707#endif /* INET6 */
708#if defined(INET6) && defined(INET)
709 else
710#endif
711#ifdef INET
712 {
713 m->m_pkthdr.csum_flags = CSUM_TCP;
714 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
715 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
716 }
717#endif /* INET */
718#ifdef TCPDEBUG
719 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
720 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
721#endif
722 if (flags & TH_RST)
723 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
724 tp, nth);
725
726 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
727#ifdef INET6
728 if (isipv6)
729 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
730#endif /* INET6 */
731#if defined(INET) && defined(INET6)
732 else
733#endif
734#ifdef INET
735 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
736#endif
737}
738
739/*
740 * Create a new TCP control block, making an
741 * empty reassembly queue and hooking it to the argument
742 * protocol control block. The `inp' parameter must have
743 * come from the zone allocator set up in tcp_init().
744 */
745struct tcpcb *
746tcp_newtcpcb(struct inpcb *inp)
747{
748 struct tcpcb_mem *tm;
749 struct tcpcb *tp;
750#ifdef INET6
751 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
752#endif /* INET6 */
753
754 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
755 if (tm == NULL)
756 return (NULL);
757 tp = &tm->tcb;
758
759 /* Initialise cc_var struct for this tcpcb. */
760 tp->ccv = &tm->ccv;
761 tp->ccv->type = IPPROTO_TCP;
762 tp->ccv->ccvc.tcp = tp;
763
764 /*
765 * Use the current system default CC algorithm.
766 */
767 CC_LIST_RLOCK();
768 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
769 CC_ALGO(tp) = CC_DEFAULT();
770 CC_LIST_RUNLOCK();
771
772 if (CC_ALGO(tp)->cb_init != NULL)
773 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
774 uma_zfree(V_tcpcb_zone, tm);
775 return (NULL);
776 }
777
778 tp->osd = &tm->osd;
779 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
780 uma_zfree(V_tcpcb_zone, tm);
781 return (NULL);
782 }
783
784#ifdef VIMAGE
785 tp->t_vnet = inp->inp_vnet;
786#endif
787 tp->t_timers = &tm->tt;
788 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
789 tp->t_maxseg = tp->t_maxopd =
790#ifdef INET6
791 isipv6 ? V_tcp_v6mssdflt :
792#endif /* INET6 */
793 V_tcp_mssdflt;
794
795 /* Set up our timeouts. */
796 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
797 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
798 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
799 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
800 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
801
802 if (V_tcp_do_rfc1323)
803 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
804 if (V_tcp_do_sack)
805 tp->t_flags |= TF_SACK_PERMIT;
806 TAILQ_INIT(&tp->snd_holes);
807 tp->t_inpcb = inp; /* XXX */
808 /*
809 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
810 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
811 * reasonable initial retransmit time.
812 */
813 tp->t_srtt = TCPTV_SRTTBASE;
814 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
815 tp->t_rttmin = tcp_rexmit_min;
816 tp->t_rxtcur = TCPTV_RTOBASE;
817 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
818 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
819 tp->t_rcvtime = ticks;
820 /*
821 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
822 * because the socket may be bound to an IPv6 wildcard address,
823 * which may match an IPv4-mapped IPv6 address.
824 */
825 inp->inp_ip_ttl = V_ip_defttl;
826 inp->inp_ppcb = tp;
827 return (tp); /* XXX */
828}
829
830/*
831 * Switch the congestion control algorithm back to NewReno for any active
832 * control blocks using an algorithm which is about to go away.
833 * This ensures the CC framework can allow the unload to proceed without leaving
834 * any dangling pointers which would trigger a panic.
835 * Returning non-zero would inform the CC framework that something went wrong
836 * and it would be unsafe to allow the unload to proceed. However, there is no
837 * way for this to occur with this implementation so we always return zero.
838 */
839int
840tcp_ccalgounload(struct cc_algo *unload_algo)
841{
842 struct cc_algo *tmpalgo;
843 struct inpcb *inp;
844 struct tcpcb *tp;
845 VNET_ITERATOR_DECL(vnet_iter);
846
847 /*
848 * Check all active control blocks across all network stacks and change
849 * any that are using "unload_algo" back to NewReno. If "unload_algo"
850 * requires cleanup code to be run, call it.
851 */
852 VNET_LIST_RLOCK();
853 VNET_FOREACH(vnet_iter) {
854 CURVNET_SET(vnet_iter);
855 INP_INFO_RLOCK(&V_tcbinfo);
856 /*
857 * New connections already part way through being initialised
858 * with the CC algo we're removing will not race with this code
859 * because the INP_INFO_WLOCK is held during initialisation. We
860 * therefore don't enter the loop below until the connection
861 * list has stabilised.
862 */
863 LIST_FOREACH(inp, &V_tcb, inp_list) {
864 INP_WLOCK(inp);
865 /* Important to skip tcptw structs. */
866 if (!(inp->inp_flags & INP_TIMEWAIT) &&
867 (tp = intotcpcb(inp)) != NULL) {
868 /*
869 * By holding INP_WLOCK here, we are assured
870 * that the connection is not currently
871 * executing inside the CC module's functions
872 * i.e. it is safe to make the switch back to
873 * NewReno.
874 */
875 if (CC_ALGO(tp) == unload_algo) {
876 tmpalgo = CC_ALGO(tp);
877 /* NewReno does not require any init. */
878 CC_ALGO(tp) = &newreno_cc_algo;
879 if (tmpalgo->cb_destroy != NULL)
880 tmpalgo->cb_destroy(tp->ccv);
881 }
882 }
883 INP_WUNLOCK(inp);
884 }
885 INP_INFO_RUNLOCK(&V_tcbinfo);
886 CURVNET_RESTORE();
887 }
888 VNET_LIST_RUNLOCK();
889
890 return (0);
891}
892
893/*
894 * Drop a TCP connection, reporting
895 * the specified error. If connection is synchronized,
896 * then send a RST to peer.
897 */
898struct tcpcb *
899tcp_drop(struct tcpcb *tp, int errno)
900{
901 struct socket *so = tp->t_inpcb->inp_socket;
902
903 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
904 INP_WLOCK_ASSERT(tp->t_inpcb);
905
906 if (TCPS_HAVERCVDSYN(tp->t_state)) {
907 tcp_state_change(tp, TCPS_CLOSED);
908 (void) tcp_output(tp);
909 TCPSTAT_INC(tcps_drops);
910 } else
911 TCPSTAT_INC(tcps_conndrops);
912 if (errno == ETIMEDOUT && tp->t_softerror)
913 errno = tp->t_softerror;
914 so->so_error = errno;
915 return (tcp_close(tp));
916}
917
918void
919tcp_discardcb(struct tcpcb *tp)
920{
921 struct inpcb *inp = tp->t_inpcb;
922 struct socket *so = inp->inp_socket;
923#ifdef INET6
924 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
925#endif /* INET6 */
926
927 INP_WLOCK_ASSERT(inp);
928
929 /*
930 * Make sure that all of our timers are stopped before we delete the
931 * PCB.
932 *
933 * XXXRW: Really, we would like to use callout_drain() here in order
934 * to avoid races experienced in tcp_timer.c where a timer is already
935 * executing at this point. However, we can't, both because we're
936 * running in a context where we can't sleep, and also because we
937 * hold locks required by the timers. What we instead need to do is
938 * test to see if callout_drain() is required, and if so, defer some
939 * portion of the remainder of tcp_discardcb() to an asynchronous
940 * context that can callout_drain() and then continue. Some care
941 * will be required to ensure that no further processing takes place
942 * on the tcpcb, even though it hasn't been freed (a flag?).
943 */
944 callout_stop(&tp->t_timers->tt_rexmt);
945 callout_stop(&tp->t_timers->tt_persist);
946 callout_stop(&tp->t_timers->tt_keep);
947 callout_stop(&tp->t_timers->tt_2msl);
948 callout_stop(&tp->t_timers->tt_delack);
949
950 /*
951 * If we got enough samples through the srtt filter,
952 * save the rtt and rttvar in the routing entry.
953 * 'Enough' is arbitrarily defined as 4 rtt samples.
954 * 4 samples is enough for the srtt filter to converge
955 * to within enough % of the correct value; fewer samples
956 * and we could save a bogus rtt. The danger is not high
957 * as tcp quickly recovers from everything.
958 * XXX: Works very well but needs some more statistics!
959 */
960 if (tp->t_rttupdated >= 4) {
961 struct hc_metrics_lite metrics;
962 u_long ssthresh;
963
964 bzero(&metrics, sizeof(metrics));
965 /*
966 * Update the ssthresh always when the conditions below
967 * are satisfied. This gives us better new start value
968 * for the congestion avoidance for new connections.
969 * ssthresh is only set if packet loss occured on a session.
970 *
971 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
972 * being torn down. Ideally this code would not use 'so'.
973 */
974 ssthresh = tp->snd_ssthresh;
975 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
976 /*
977 * convert the limit from user data bytes to
978 * packets then to packet data bytes.
979 */
980 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
981 if (ssthresh < 2)
982 ssthresh = 2;
983 ssthresh *= (u_long)(tp->t_maxseg +
984#ifdef INET6
985 (isipv6 ? sizeof (struct ip6_hdr) +
986 sizeof (struct tcphdr) :
987#endif
988 sizeof (struct tcpiphdr)
989#ifdef INET6
990 )
991#endif
992 );
993 } else
994 ssthresh = 0;
995 metrics.rmx_ssthresh = ssthresh;
996
997 metrics.rmx_rtt = tp->t_srtt;
998 metrics.rmx_rttvar = tp->t_rttvar;
999 metrics.rmx_cwnd = tp->snd_cwnd;
1000 metrics.rmx_sendpipe = 0;
1001 metrics.rmx_recvpipe = 0;
1002
1003 tcp_hc_update(&inp->inp_inc, &metrics);
1004 }
1005
1006 /* free the reassembly queue, if any */
1007 tcp_reass_flush(tp);
1008
1009#ifdef TCP_OFFLOAD
1010 /* Disconnect offload device, if any. */
1011 if (tp->t_flags & TF_TOE)
1012 tcp_offload_detach(tp);
1013#endif
1014
1015 tcp_free_sackholes(tp);
1016
1017 /* Allow the CC algorithm to clean up after itself. */
1018 if (CC_ALGO(tp)->cb_destroy != NULL)
1019 CC_ALGO(tp)->cb_destroy(tp->ccv);
1020
1021 khelp_destroy_osd(tp->osd);
1022
1023 CC_ALGO(tp) = NULL;
1024 inp->inp_ppcb = NULL;
1025 tp->t_inpcb = NULL;
1026 uma_zfree(V_tcpcb_zone, tp);
1027}
1028
1029/*
1030 * Attempt to close a TCP control block, marking it as dropped, and freeing
1031 * the socket if we hold the only reference.
1032 */
1033struct tcpcb *
1034tcp_close(struct tcpcb *tp)
1035{
1036 struct inpcb *inp = tp->t_inpcb;
1037 struct socket *so;
1038
1039 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1040 INP_WLOCK_ASSERT(inp);
1041
1042#ifdef TCP_OFFLOAD
1043 if (tp->t_state == TCPS_LISTEN)
1044 tcp_offload_listen_stop(tp);
1045#endif
1046 in_pcbdrop(inp);
1047 TCPSTAT_INC(tcps_closed);
1048 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1049 so = inp->inp_socket;
1050 soisdisconnected(so);
1051 if (inp->inp_flags & INP_SOCKREF) {
1052 KASSERT(so->so_state & SS_PROTOREF,
1053 ("tcp_close: !SS_PROTOREF"));
1054 inp->inp_flags &= ~INP_SOCKREF;
1055 INP_WUNLOCK(inp);
1056 ACCEPT_LOCK();
1057 SOCK_LOCK(so);
1058 so->so_state &= ~SS_PROTOREF;
1059 sofree(so);
1060 return (NULL);
1061 }
1062 return (tp);
1063}
1064
1065void
1066tcp_drain(void)
1067{
1068 VNET_ITERATOR_DECL(vnet_iter);
1069
1070 if (!do_tcpdrain)
1071 return;
1072
1073 VNET_LIST_RLOCK_NOSLEEP();
1074 VNET_FOREACH(vnet_iter) {
1075 CURVNET_SET(vnet_iter);
1076 struct inpcb *inpb;
1077 struct tcpcb *tcpb;
1078
1079 /*
1080 * Walk the tcpbs, if existing, and flush the reassembly queue,
1081 * if there is one...
1082 * XXX: The "Net/3" implementation doesn't imply that the TCP
1083 * reassembly queue should be flushed, but in a situation
1084 * where we're really low on mbufs, this is potentially
1085 * useful.
1086 */
1087 INP_INFO_RLOCK(&V_tcbinfo);
1088 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1089 if (inpb->inp_flags & INP_TIMEWAIT)
1090 continue;
1091 INP_WLOCK(inpb);
1092 if ((tcpb = intotcpcb(inpb)) != NULL) {
1093 tcp_reass_flush(tcpb);
1094 tcp_clean_sackreport(tcpb);
1095 }
1096 INP_WUNLOCK(inpb);
1097 }
1098 INP_INFO_RUNLOCK(&V_tcbinfo);
1099 CURVNET_RESTORE();
1100 }
1101 VNET_LIST_RUNLOCK_NOSLEEP();
1102}
1103
1104/*
1105 * Notify a tcp user of an asynchronous error;
1106 * store error as soft error, but wake up user
1107 * (for now, won't do anything until can select for soft error).
1108 *
1109 * Do not wake up user since there currently is no mechanism for
1110 * reporting soft errors (yet - a kqueue filter may be added).
1111 */
1112static struct inpcb *
1113tcp_notify(struct inpcb *inp, int error)
1114{
1115 struct tcpcb *tp;
1116
1117 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1118 INP_WLOCK_ASSERT(inp);
1119
1120 if ((inp->inp_flags & INP_TIMEWAIT) ||
1121 (inp->inp_flags & INP_DROPPED))
1122 return (inp);
1123
1124 tp = intotcpcb(inp);
1125 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1126
1127 /*
1128 * Ignore some errors if we are hooked up.
1129 * If connection hasn't completed, has retransmitted several times,
1130 * and receives a second error, give up now. This is better
1131 * than waiting a long time to establish a connection that
1132 * can never complete.
1133 */
1134 if (tp->t_state == TCPS_ESTABLISHED &&
1135 (error == EHOSTUNREACH || error == ENETUNREACH ||
1136 error == EHOSTDOWN)) {
1137 return (inp);
1138 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1139 tp->t_softerror) {
1140 tp = tcp_drop(tp, error);
1141 if (tp != NULL)
1142 return (inp);
1143 else
1144 return (NULL);
1145 } else {
1146 tp->t_softerror = error;
1147 return (inp);
1148 }
1149#if 0
1150 wakeup( &so->so_timeo);
1151 sorwakeup(so);
1152 sowwakeup(so);
1153#endif
1154}
1155
1156static int
1157tcp_pcblist(SYSCTL_HANDLER_ARGS)
1158{
1159 int error, i, m, n, pcb_count;
1160 struct inpcb *inp, **inp_list;
1161 inp_gen_t gencnt;
1162 struct xinpgen xig;
1163
1164 /*
1165 * The process of preparing the TCB list is too time-consuming and
1166 * resource-intensive to repeat twice on every request.
1167 */
1168 if (req->oldptr == NULL) {
1169 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1170 n += imax(n / 8, 10);
1171 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1172 return (0);
1173 }
1174
1175 if (req->newptr != NULL)
1176 return (EPERM);
1177
1178 /*
1179 * OK, now we're committed to doing something.
1180 */
1181 INP_INFO_RLOCK(&V_tcbinfo);
1182 gencnt = V_tcbinfo.ipi_gencnt;
1183 n = V_tcbinfo.ipi_count;
1184 INP_INFO_RUNLOCK(&V_tcbinfo);
1185
1186 m = syncache_pcbcount();
1187
1188 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1189 + (n + m) * sizeof(struct xtcpcb));
1190 if (error != 0)
1191 return (error);
1192
1193 xig.xig_len = sizeof xig;
1194 xig.xig_count = n + m;
1195 xig.xig_gen = gencnt;
1196 xig.xig_sogen = so_gencnt;
1197 error = SYSCTL_OUT(req, &xig, sizeof xig);
1198 if (error)
1199 return (error);
1200
1201 error = syncache_pcblist(req, m, &pcb_count);
1202 if (error)
1203 return (error);
1204
1205 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1206 if (inp_list == NULL)
1207 return (ENOMEM);
1208
1209 INP_INFO_RLOCK(&V_tcbinfo);
1210 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1211 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1212 INP_WLOCK(inp);
1213 if (inp->inp_gencnt <= gencnt) {
1214 /*
1215 * XXX: This use of cr_cansee(), introduced with
1216 * TCP state changes, is not quite right, but for
1217 * now, better than nothing.
1218 */
1219 if (inp->inp_flags & INP_TIMEWAIT) {
1220 if (intotw(inp) != NULL)
1221 error = cr_cansee(req->td->td_ucred,
1222 intotw(inp)->tw_cred);
1223 else
1224 error = EINVAL; /* Skip this inp. */
1225 } else
1226 error = cr_canseeinpcb(req->td->td_ucred, inp);
1227 if (error == 0) {
1228 in_pcbref(inp);
1229 inp_list[i++] = inp;
1230 }
1231 }
1232 INP_WUNLOCK(inp);
1233 }
1234 INP_INFO_RUNLOCK(&V_tcbinfo);
1235 n = i;
1236
1237 error = 0;
1238 for (i = 0; i < n; i++) {
1239 inp = inp_list[i];
1240 INP_RLOCK(inp);
1241 if (inp->inp_gencnt <= gencnt) {
1242 struct xtcpcb xt;
1243 void *inp_ppcb;
1244
1245 bzero(&xt, sizeof(xt));
1246 xt.xt_len = sizeof xt;
1247 /* XXX should avoid extra copy */
1248 bcopy(inp, &xt.xt_inp, sizeof *inp);
1249 inp_ppcb = inp->inp_ppcb;
1250 if (inp_ppcb == NULL)
1251 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1252 else if (inp->inp_flags & INP_TIMEWAIT) {
1253 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1254 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1255 } else {
1256 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1257 if (xt.xt_tp.t_timers)
1258 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1259 }
1260 if (inp->inp_socket != NULL)
1261 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1262 else {
1263 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1264 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1265 }
1266 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1267 INP_RUNLOCK(inp);
1268 error = SYSCTL_OUT(req, &xt, sizeof xt);
1269 } else
1270 INP_RUNLOCK(inp);
1271 }
1272 INP_INFO_WLOCK(&V_tcbinfo);
1273 for (i = 0; i < n; i++) {
1274 inp = inp_list[i];
1275 INP_RLOCK(inp);
1276 if (!in_pcbrele_rlocked(inp))
1277 INP_RUNLOCK(inp);
1278 }
1279 INP_INFO_WUNLOCK(&V_tcbinfo);
1280
1281 if (!error) {
1282 /*
1283 * Give the user an updated idea of our state.
1284 * If the generation differs from what we told
1285 * her before, she knows that something happened
1286 * while we were processing this request, and it
1287 * might be necessary to retry.
1288 */
1289 INP_INFO_RLOCK(&V_tcbinfo);
1290 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1291 xig.xig_sogen = so_gencnt;
1292 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1293 INP_INFO_RUNLOCK(&V_tcbinfo);
1294 error = SYSCTL_OUT(req, &xig, sizeof xig);
1295 }
1296 free(inp_list, M_TEMP);
1297 return (error);
1298}
1299
1300SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1301 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1302 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1303
1304#ifdef INET
1305static int
1306tcp_getcred(SYSCTL_HANDLER_ARGS)
1307{
1308 struct xucred xuc;
1309 struct sockaddr_in addrs[2];
1310 struct inpcb *inp;
1311 int error;
1312
1313 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1314 if (error)
1315 return (error);
1316 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1317 if (error)
1318 return (error);
1319 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1320 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1321 if (inp != NULL) {
1322 if (inp->inp_socket == NULL)
1323 error = ENOENT;
1324 if (error == 0)
1325 error = cr_canseeinpcb(req->td->td_ucred, inp);
1326 if (error == 0)
1327 cru2x(inp->inp_cred, &xuc);
1328 INP_RUNLOCK(inp);
1329 } else
1330 error = ENOENT;
1331 if (error == 0)
1332 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1333 return (error);
1334}
1335
1336SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1337 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1338 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1339#endif /* INET */
1340
1341#ifdef INET6
1342static int
1343tcp6_getcred(SYSCTL_HANDLER_ARGS)
1344{
1345 struct xucred xuc;
1346 struct sockaddr_in6 addrs[2];
1347 struct inpcb *inp;
1348 int error;
1349#ifdef INET
1350 int mapped = 0;
1351#endif
1352
1353 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1354 if (error)
1355 return (error);
1356 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1357 if (error)
1358 return (error);
1359 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1360 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1361 return (error);
1362 }
1363 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1364#ifdef INET
1365 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1366 mapped = 1;
1367 else
1368#endif
1369 return (EINVAL);
1370 }
1371
1372#ifdef INET
1373 if (mapped == 1)
1374 inp = in_pcblookup(&V_tcbinfo,
1375 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1376 addrs[1].sin6_port,
1377 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1378 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1379 else
1380#endif
1381 inp = in6_pcblookup(&V_tcbinfo,
1382 &addrs[1].sin6_addr, addrs[1].sin6_port,
1383 &addrs[0].sin6_addr, addrs[0].sin6_port,
1384 INPLOOKUP_RLOCKPCB, NULL);
1385 if (inp != NULL) {
1386 if (inp->inp_socket == NULL)
1387 error = ENOENT;
1388 if (error == 0)
1389 error = cr_canseeinpcb(req->td->td_ucred, inp);
1390 if (error == 0)
1391 cru2x(inp->inp_cred, &xuc);
1392 INP_RUNLOCK(inp);
1393 } else
1394 error = ENOENT;
1395 if (error == 0)
1396 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1397 return (error);
1398}
1399
1400SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1401 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1402 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1403#endif /* INET6 */
1404
1405
1406#ifdef INET
1407void
1408tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1409{
1410 struct ip *ip = vip;
1411 struct tcphdr *th;
1412 struct in_addr faddr;
1413 struct inpcb *inp;
1414 struct tcpcb *tp;
1415 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1416 struct icmp *icp;
1417 struct in_conninfo inc;
1418 tcp_seq icmp_tcp_seq;
1419 int mtu;
1420
1421 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1422 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1423 return;
1424
1425 if (cmd == PRC_MSGSIZE)
1426 notify = tcp_mtudisc_notify;
1427 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1428 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1429 notify = tcp_drop_syn_sent;
1430 /*
1431 * Redirects don't need to be handled up here.
1432 */
1433 else if (PRC_IS_REDIRECT(cmd))
1434 return;
1435 /*
1436 * Source quench is depreciated.
1437 */
1438 else if (cmd == PRC_QUENCH)
1439 return;
1440 /*
1441 * Hostdead is ugly because it goes linearly through all PCBs.
1442 * XXX: We never get this from ICMP, otherwise it makes an
1443 * excellent DoS attack on machines with many connections.
1444 */
1445 else if (cmd == PRC_HOSTDEAD)
1446 ip = NULL;
1447 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1448 return;
1449 if (ip != NULL) {
1450 icp = (struct icmp *)((caddr_t)ip
1451 - offsetof(struct icmp, icmp_ip));
1452 th = (struct tcphdr *)((caddr_t)ip
1453 + (ip->ip_hl << 2));
1454 INP_INFO_WLOCK(&V_tcbinfo);
1455 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1456 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1457 if (inp != NULL) {
1458 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1459 !(inp->inp_flags & INP_DROPPED) &&
1460 !(inp->inp_socket == NULL)) {
1461 icmp_tcp_seq = htonl(th->th_seq);
1462 tp = intotcpcb(inp);
1463 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1464 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1465 if (cmd == PRC_MSGSIZE) {
1466 /*
1467 * MTU discovery:
1468 * If we got a needfrag set the MTU
1469 * in the route to the suggested new
1470 * value (if given) and then notify.
1471 */
1472 bzero(&inc, sizeof(inc));
1473 inc.inc_faddr = faddr;
1474 inc.inc_fibnum =
1475 inp->inp_inc.inc_fibnum;
1476
1477 mtu = ntohs(icp->icmp_nextmtu);
1478 /*
1479 * If no alternative MTU was
1480 * proposed, try the next smaller
1481 * one.
1482 */
1483 if (!mtu)
1484 mtu = ip_next_mtu(
1485 ntohs(ip->ip_len), 1);
1486 if (mtu < V_tcp_minmss
1487 + sizeof(struct tcpiphdr))
1488 mtu = V_tcp_minmss
1489 + sizeof(struct tcpiphdr);
1490 /*
1491 * Only cache the MTU if it
1492 * is smaller than the interface
1493 * or route MTU. tcp_mtudisc()
1494 * will do right thing by itself.
1495 */
1496 if (mtu <= tcp_maxmtu(&inc, NULL))
1497 tcp_hc_updatemtu(&inc, mtu);
1498 tcp_mtudisc(inp, mtu);
1499 } else
1500 inp = (*notify)(inp,
1501 inetctlerrmap[cmd]);
1502 }
1503 }
1504 if (inp != NULL)
1505 INP_WUNLOCK(inp);
1506 } else {
1507 bzero(&inc, sizeof(inc));
1508 inc.inc_fport = th->th_dport;
1509 inc.inc_lport = th->th_sport;
1510 inc.inc_faddr = faddr;
1511 inc.inc_laddr = ip->ip_src;
1512 syncache_unreach(&inc, th);
1513 }
1514 INP_INFO_WUNLOCK(&V_tcbinfo);
1515 } else
1516 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1517}
1518#endif /* INET */
1519
1520#ifdef INET6
1521void
1522tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1523{
1524 struct tcphdr th;
1525 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1526 struct ip6_hdr *ip6;
1527 struct mbuf *m;
1528 struct ip6ctlparam *ip6cp = NULL;
1529 const struct sockaddr_in6 *sa6_src = NULL;
1530 int off;
1531 struct tcp_portonly {
1532 u_int16_t th_sport;
1533 u_int16_t th_dport;
1534 } *thp;
1535
1536 if (sa->sa_family != AF_INET6 ||
1537 sa->sa_len != sizeof(struct sockaddr_in6))
1538 return;
1539
1540 if (cmd == PRC_MSGSIZE)
1541 notify = tcp_mtudisc_notify;
1542 else if (!PRC_IS_REDIRECT(cmd) &&
1543 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1544 return;
1545 /* Source quench is depreciated. */
1546 else if (cmd == PRC_QUENCH)
1547 return;
1548
1549 /* if the parameter is from icmp6, decode it. */
1550 if (d != NULL) {
1551 ip6cp = (struct ip6ctlparam *)d;
1552 m = ip6cp->ip6c_m;
1553 ip6 = ip6cp->ip6c_ip6;
1554 off = ip6cp->ip6c_off;
1555 sa6_src = ip6cp->ip6c_src;
1556 } else {
1557 m = NULL;
1558 ip6 = NULL;
1559 off = 0; /* fool gcc */
1560 sa6_src = &sa6_any;
1561 }
1562
1563 if (ip6 != NULL) {
1564 struct in_conninfo inc;
1565 /*
1566 * XXX: We assume that when IPV6 is non NULL,
1567 * M and OFF are valid.
1568 */
1569
1570 /* check if we can safely examine src and dst ports */
1571 if (m->m_pkthdr.len < off + sizeof(*thp))
1572 return;
1573
1574 bzero(&th, sizeof(th));
1575 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1576
1577 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1578 (struct sockaddr *)ip6cp->ip6c_src,
1579 th.th_sport, cmd, NULL, notify);
1580
1581 bzero(&inc, sizeof(inc));
1582 inc.inc_fport = th.th_dport;
1583 inc.inc_lport = th.th_sport;
1584 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1585 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1586 inc.inc_flags |= INC_ISIPV6;
1587 INP_INFO_WLOCK(&V_tcbinfo);
1588 syncache_unreach(&inc, &th);
1589 INP_INFO_WUNLOCK(&V_tcbinfo);
1590 } else
1591 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1592 0, cmd, NULL, notify);
1593}
1594#endif /* INET6 */
1595
1596
1597/*
1598 * Following is where TCP initial sequence number generation occurs.
1599 *
1600 * There are two places where we must use initial sequence numbers:
1601 * 1. In SYN-ACK packets.
1602 * 2. In SYN packets.
1603 *
1604 * All ISNs for SYN-ACK packets are generated by the syncache. See
1605 * tcp_syncache.c for details.
1606 *
1607 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1608 * depends on this property. In addition, these ISNs should be
1609 * unguessable so as to prevent connection hijacking. To satisfy
1610 * the requirements of this situation, the algorithm outlined in
1611 * RFC 1948 is used, with only small modifications.
1612 *
1613 * Implementation details:
1614 *
1615 * Time is based off the system timer, and is corrected so that it
1616 * increases by one megabyte per second. This allows for proper
1617 * recycling on high speed LANs while still leaving over an hour
1618 * before rollover.
1619 *
1620 * As reading the *exact* system time is too expensive to be done
1621 * whenever setting up a TCP connection, we increment the time
1622 * offset in two ways. First, a small random positive increment
1623 * is added to isn_offset for each connection that is set up.
1624 * Second, the function tcp_isn_tick fires once per clock tick
1625 * and increments isn_offset as necessary so that sequence numbers
1626 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1627 * random positive increments serve only to ensure that the same
1628 * exact sequence number is never sent out twice (as could otherwise
1629 * happen when a port is recycled in less than the system tick
1630 * interval.)
1631 *
1632 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1633 * between seeding of isn_secret. This is normally set to zero,
1634 * as reseeding should not be necessary.
1635 *
1636 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1637 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1638 * general, this means holding an exclusive (write) lock.
1639 */
1640
1641#define ISN_BYTES_PER_SECOND 1048576
1642#define ISN_STATIC_INCREMENT 4096
1643#define ISN_RANDOM_INCREMENT (4096 - 1)
1644
1645static VNET_DEFINE(u_char, isn_secret[32]);
1646static VNET_DEFINE(int, isn_last);
1647static VNET_DEFINE(int, isn_last_reseed);
1648static VNET_DEFINE(u_int32_t, isn_offset);
1649static VNET_DEFINE(u_int32_t, isn_offset_old);
1650
1651#define V_isn_secret VNET(isn_secret)
1652#define V_isn_last VNET(isn_last)
1653#define V_isn_last_reseed VNET(isn_last_reseed)
1654#define V_isn_offset VNET(isn_offset)
1655#define V_isn_offset_old VNET(isn_offset_old)
1656
1657tcp_seq
1658tcp_new_isn(struct tcpcb *tp)
1659{
1660 MD5_CTX isn_ctx;
1661 u_int32_t md5_buffer[4];
1662 tcp_seq new_isn;
1663 u_int32_t projected_offset;
1664
1665 INP_WLOCK_ASSERT(tp->t_inpcb);
1666
1667 ISN_LOCK();
1668 /* Seed if this is the first use, reseed if requested. */
1669 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1670 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1671 < (u_int)ticks))) {
1672 read_random(&V_isn_secret, sizeof(V_isn_secret));
1673 V_isn_last_reseed = ticks;
1674 }
1675
1676 /* Compute the md5 hash and return the ISN. */
1677 MD5Init(&isn_ctx);
1678 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1679 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1680#ifdef INET6
1681 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1682 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1683 sizeof(struct in6_addr));
1684 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1685 sizeof(struct in6_addr));
1686 } else
1687#endif
1688 {
1689 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1690 sizeof(struct in_addr));
1691 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1692 sizeof(struct in_addr));
1693 }
1694 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1695 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1696 new_isn = (tcp_seq) md5_buffer[0];
1697 V_isn_offset += ISN_STATIC_INCREMENT +
1698 (arc4random() & ISN_RANDOM_INCREMENT);
1699 if (ticks != V_isn_last) {
1700 projected_offset = V_isn_offset_old +
1701 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1702 if (SEQ_GT(projected_offset, V_isn_offset))
1703 V_isn_offset = projected_offset;
1704 V_isn_offset_old = V_isn_offset;
1705 V_isn_last = ticks;
1706 }
1707 new_isn += V_isn_offset;
1708 ISN_UNLOCK();
1709 return (new_isn);
1710}
1711
1712/*
1713 * When a specific ICMP unreachable message is received and the
1714 * connection state is SYN-SENT, drop the connection. This behavior
1715 * is controlled by the icmp_may_rst sysctl.
1716 */
1717struct inpcb *
1718tcp_drop_syn_sent(struct inpcb *inp, int errno)
1719{
1720 struct tcpcb *tp;
1721
1722 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1723 INP_WLOCK_ASSERT(inp);
1724
1725 if ((inp->inp_flags & INP_TIMEWAIT) ||
1726 (inp->inp_flags & INP_DROPPED))
1727 return (inp);
1728
1729 tp = intotcpcb(inp);
1730 if (tp->t_state != TCPS_SYN_SENT)
1731 return (inp);
1732
1733 tp = tcp_drop(tp, errno);
1734 if (tp != NULL)
1735 return (inp);
1736 else
1737 return (NULL);
1738}
1739
1740/*
1741 * When `need fragmentation' ICMP is received, update our idea of the MSS
1742 * based on the new value. Also nudge TCP to send something, since we
1743 * know the packet we just sent was dropped.
1744 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1745 */
1746static struct inpcb *
1747tcp_mtudisc_notify(struct inpcb *inp, int error)
1748{
1749
1750 return (tcp_mtudisc(inp, -1));
1751}
1752
1753struct inpcb *
1754tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1755{
1756 struct tcpcb *tp;
1757 struct socket *so;
1758
1759 INP_WLOCK_ASSERT(inp);
1760 if ((inp->inp_flags & INP_TIMEWAIT) ||
1761 (inp->inp_flags & INP_DROPPED))
1762 return (inp);
1763
1764 tp = intotcpcb(inp);
1765 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1766
1767 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1768
1769 so = inp->inp_socket;
1770 SOCKBUF_LOCK(&so->so_snd);
1771 /* If the mss is larger than the socket buffer, decrease the mss. */
1772 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1773 tp->t_maxseg = so->so_snd.sb_hiwat;
1774 SOCKBUF_UNLOCK(&so->so_snd);
1775
1776 TCPSTAT_INC(tcps_mturesent);
1777 tp->t_rtttime = 0;
1778 tp->snd_nxt = tp->snd_una;
1779 tcp_free_sackholes(tp);
1780 tp->snd_recover = tp->snd_max;
1781 if (tp->t_flags & TF_SACK_PERMIT)
1782 EXIT_FASTRECOVERY(tp->t_flags);
1783 tcp_output(tp);
1784 return (inp);
1785}
1786
1787#ifdef INET
1788/*
1789 * Look-up the routing entry to the peer of this inpcb. If no route
1790 * is found and it cannot be allocated, then return 0. This routine
1791 * is called by TCP routines that access the rmx structure and by
1792 * tcp_mss_update to get the peer/interface MTU.
1793 */
1794u_long
1795tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1796{
1797 struct route sro;
1798 struct sockaddr_in *dst;
1799 struct ifnet *ifp;
1800 u_long maxmtu = 0;
1801
1802 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1803
1804 bzero(&sro, sizeof(sro));
1805 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1806 dst = (struct sockaddr_in *)&sro.ro_dst;
1807 dst->sin_family = AF_INET;
1808 dst->sin_len = sizeof(*dst);
1809 dst->sin_addr = inc->inc_faddr;
1810 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1811 }
1812 if (sro.ro_rt != NULL) {
1813 ifp = sro.ro_rt->rt_ifp;
| 34
35#include "opt_compat.h"
36#include "opt_inet.h"
37#include "opt_inet6.h"
38#include "opt_ipsec.h"
39#include "opt_tcpdebug.h"
40
41#include <sys/param.h>
42#include <sys/systm.h>
43#include <sys/callout.h>
44#include <sys/hhook.h>
45#include <sys/kernel.h>
46#include <sys/khelp.h>
47#include <sys/sysctl.h>
48#include <sys/jail.h>
49#include <sys/malloc.h>
50#include <sys/mbuf.h>
51#ifdef INET6
52#include <sys/domain.h>
53#endif
54#include <sys/priv.h>
55#include <sys/proc.h>
56#include <sys/sdt.h>
57#include <sys/socket.h>
58#include <sys/socketvar.h>
59#include <sys/protosw.h>
60#include <sys/random.h>
61
62#include <vm/uma.h>
63
64#include <net/route.h>
65#include <net/if.h>
66#include <net/if_var.h>
67#include <net/vnet.h>
68
69#include <netinet/cc.h>
70#include <netinet/in.h>
71#include <netinet/in_kdtrace.h>
72#include <netinet/in_pcb.h>
73#include <netinet/in_systm.h>
74#include <netinet/in_var.h>
75#include <netinet/ip.h>
76#include <netinet/ip_icmp.h>
77#include <netinet/ip_var.h>
78#ifdef INET6
79#include <netinet/ip6.h>
80#include <netinet6/in6_pcb.h>
81#include <netinet6/ip6_var.h>
82#include <netinet6/scope6_var.h>
83#include <netinet6/nd6.h>
84#endif
85
86#include <netinet/tcp_fsm.h>
87#include <netinet/tcp_seq.h>
88#include <netinet/tcp_timer.h>
89#include <netinet/tcp_var.h>
90#include <netinet/tcp_syncache.h>
91#ifdef INET6
92#include <netinet6/tcp6_var.h>
93#endif
94#include <netinet/tcpip.h>
95#ifdef 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 int error;
435
436 tcp_reass_destroy();
437 tcp_hc_destroy();
438 syncache_destroy();
439 tcp_tw_destroy();
440 in_pcbinfo_destroy(&V_tcbinfo);
441 uma_zdestroy(V_sack_hole_zone);
442 uma_zdestroy(V_tcpcb_zone);
443
444 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_IN]);
445 if (error != 0) {
446 printf("%s: WARNING: unable to deregister helper hook "
447 "type=%d, id=%d: error %d returned\n", __func__,
448 HHOOK_TYPE_TCP, HHOOK_TCP_EST_IN, error);
449 }
450 error = hhook_head_deregister(V_tcp_hhh[HHOOK_TCP_EST_OUT]);
451 if (error != 0) {
452 printf("%s: WARNING: unable to deregister helper hook "
453 "type=%d, id=%d: error %d returned\n", __func__,
454 HHOOK_TYPE_TCP, HHOOK_TCP_EST_OUT, error);
455 }
456}
457#endif
458
459void
460tcp_fini(void *xtp)
461{
462
463}
464
465/*
466 * Fill in the IP and TCP headers for an outgoing packet, given the tcpcb.
467 * tcp_template used to store this data in mbufs, but we now recopy it out
468 * of the tcpcb each time to conserve mbufs.
469 */
470void
471tcpip_fillheaders(struct inpcb *inp, void *ip_ptr, void *tcp_ptr)
472{
473 struct tcphdr *th = (struct tcphdr *)tcp_ptr;
474
475 INP_WLOCK_ASSERT(inp);
476
477#ifdef INET6
478 if ((inp->inp_vflag & INP_IPV6) != 0) {
479 struct ip6_hdr *ip6;
480
481 ip6 = (struct ip6_hdr *)ip_ptr;
482 ip6->ip6_flow = (ip6->ip6_flow & ~IPV6_FLOWINFO_MASK) |
483 (inp->inp_flow & IPV6_FLOWINFO_MASK);
484 ip6->ip6_vfc = (ip6->ip6_vfc & ~IPV6_VERSION_MASK) |
485 (IPV6_VERSION & IPV6_VERSION_MASK);
486 ip6->ip6_nxt = IPPROTO_TCP;
487 ip6->ip6_plen = htons(sizeof(struct tcphdr));
488 ip6->ip6_src = inp->in6p_laddr;
489 ip6->ip6_dst = inp->in6p_faddr;
490 }
491#endif /* INET6 */
492#if defined(INET6) && defined(INET)
493 else
494#endif
495#ifdef INET
496 {
497 struct ip *ip;
498
499 ip = (struct ip *)ip_ptr;
500 ip->ip_v = IPVERSION;
501 ip->ip_hl = 5;
502 ip->ip_tos = inp->inp_ip_tos;
503 ip->ip_len = 0;
504 ip->ip_id = 0;
505 ip->ip_off = 0;
506 ip->ip_ttl = inp->inp_ip_ttl;
507 ip->ip_sum = 0;
508 ip->ip_p = IPPROTO_TCP;
509 ip->ip_src = inp->inp_laddr;
510 ip->ip_dst = inp->inp_faddr;
511 }
512#endif /* INET */
513 th->th_sport = inp->inp_lport;
514 th->th_dport = inp->inp_fport;
515 th->th_seq = 0;
516 th->th_ack = 0;
517 th->th_x2 = 0;
518 th->th_off = 5;
519 th->th_flags = 0;
520 th->th_win = 0;
521 th->th_urp = 0;
522 th->th_sum = 0; /* in_pseudo() is called later for ipv4 */
523}
524
525/*
526 * Create template to be used to send tcp packets on a connection.
527 * Allocates an mbuf and fills in a skeletal tcp/ip header. The only
528 * use for this function is in keepalives, which use tcp_respond.
529 */
530struct tcptemp *
531tcpip_maketemplate(struct inpcb *inp)
532{
533 struct tcptemp *t;
534
535 t = malloc(sizeof(*t), M_TEMP, M_NOWAIT);
536 if (t == NULL)
537 return (NULL);
538 tcpip_fillheaders(inp, (void *)&t->tt_ipgen, (void *)&t->tt_t);
539 return (t);
540}
541
542/*
543 * Send a single message to the TCP at address specified by
544 * the given TCP/IP header. If m == NULL, then we make a copy
545 * of the tcpiphdr at ti and send directly to the addressed host.
546 * This is used to force keep alive messages out using the TCP
547 * template for a connection. If flags are given then we send
548 * a message back to the TCP which originated the * segment ti,
549 * and discard the mbuf containing it and any other attached mbufs.
550 *
551 * In any case the ack and sequence number of the transmitted
552 * segment are as specified by the parameters.
553 *
554 * NOTE: If m != NULL, then ti must point to *inside* the mbuf.
555 */
556void
557tcp_respond(struct tcpcb *tp, void *ipgen, struct tcphdr *th, struct mbuf *m,
558 tcp_seq ack, tcp_seq seq, int flags)
559{
560 int tlen;
561 int win = 0;
562 struct ip *ip;
563 struct tcphdr *nth;
564#ifdef INET6
565 struct ip6_hdr *ip6;
566 int isipv6;
567#endif /* INET6 */
568 int ipflags = 0;
569 struct inpcb *inp;
570
571 KASSERT(tp != NULL || m != NULL, ("tcp_respond: tp and m both NULL"));
572
573#ifdef INET6
574 isipv6 = ((struct ip *)ipgen)->ip_v == (IPV6_VERSION >> 4);
575 ip6 = ipgen;
576#endif /* INET6 */
577 ip = ipgen;
578
579 if (tp != NULL) {
580 inp = tp->t_inpcb;
581 KASSERT(inp != NULL, ("tcp control block w/o inpcb"));
582 INP_WLOCK_ASSERT(inp);
583 } else
584 inp = NULL;
585
586 if (tp != NULL) {
587 if (!(flags & TH_RST)) {
588 win = sbspace(&inp->inp_socket->so_rcv);
589 if (win > (long)TCP_MAXWIN << tp->rcv_scale)
590 win = (long)TCP_MAXWIN << tp->rcv_scale;
591 }
592 }
593 if (m == NULL) {
594 m = m_gethdr(M_NOWAIT, MT_DATA);
595 if (m == NULL)
596 return;
597 tlen = 0;
598 m->m_data += max_linkhdr;
599#ifdef INET6
600 if (isipv6) {
601 bcopy((caddr_t)ip6, mtod(m, caddr_t),
602 sizeof(struct ip6_hdr));
603 ip6 = mtod(m, struct ip6_hdr *);
604 nth = (struct tcphdr *)(ip6 + 1);
605 } else
606#endif /* INET6 */
607 {
608 bcopy((caddr_t)ip, mtod(m, caddr_t), sizeof(struct ip));
609 ip = mtod(m, struct ip *);
610 nth = (struct tcphdr *)(ip + 1);
611 }
612 bcopy((caddr_t)th, (caddr_t)nth, sizeof(struct tcphdr));
613 flags = TH_ACK;
614 } else {
615 /*
616 * reuse the mbuf.
617 * XXX MRT We inherrit the FIB, which is lucky.
618 */
619 m_freem(m->m_next);
620 m->m_next = NULL;
621 m->m_data = (caddr_t)ipgen;
622 /* m_len is set later */
623 tlen = 0;
624#define xchg(a,b,type) { type t; t=a; a=b; b=t; }
625#ifdef INET6
626 if (isipv6) {
627 xchg(ip6->ip6_dst, ip6->ip6_src, struct in6_addr);
628 nth = (struct tcphdr *)(ip6 + 1);
629 } else
630#endif /* INET6 */
631 {
632 xchg(ip->ip_dst.s_addr, ip->ip_src.s_addr, uint32_t);
633 nth = (struct tcphdr *)(ip + 1);
634 }
635 if (th != nth) {
636 /*
637 * this is usually a case when an extension header
638 * exists between the IPv6 header and the
639 * TCP header.
640 */
641 nth->th_sport = th->th_sport;
642 nth->th_dport = th->th_dport;
643 }
644 xchg(nth->th_dport, nth->th_sport, uint16_t);
645#undef xchg
646 }
647#ifdef INET6
648 if (isipv6) {
649 ip6->ip6_flow = 0;
650 ip6->ip6_vfc = IPV6_VERSION;
651 ip6->ip6_nxt = IPPROTO_TCP;
652 tlen += sizeof (struct ip6_hdr) + sizeof (struct tcphdr);
653 ip6->ip6_plen = htons(tlen - sizeof(*ip6));
654 }
655#endif
656#if defined(INET) && defined(INET6)
657 else
658#endif
659#ifdef INET
660 {
661 tlen += sizeof (struct tcpiphdr);
662 ip->ip_len = htons(tlen);
663 ip->ip_ttl = V_ip_defttl;
664 if (V_path_mtu_discovery)
665 ip->ip_off |= htons(IP_DF);
666 }
667#endif
668 m->m_len = tlen;
669 m->m_pkthdr.len = tlen;
670 m->m_pkthdr.rcvif = NULL;
671#ifdef MAC
672 if (inp != NULL) {
673 /*
674 * Packet is associated with a socket, so allow the
675 * label of the response to reflect the socket label.
676 */
677 INP_WLOCK_ASSERT(inp);
678 mac_inpcb_create_mbuf(inp, m);
679 } else {
680 /*
681 * Packet is not associated with a socket, so possibly
682 * update the label in place.
683 */
684 mac_netinet_tcp_reply(m);
685 }
686#endif
687 nth->th_seq = htonl(seq);
688 nth->th_ack = htonl(ack);
689 nth->th_x2 = 0;
690 nth->th_off = sizeof (struct tcphdr) >> 2;
691 nth->th_flags = flags;
692 if (tp != NULL)
693 nth->th_win = htons((u_short) (win >> tp->rcv_scale));
694 else
695 nth->th_win = htons((u_short)win);
696 nth->th_urp = 0;
697
698 m->m_pkthdr.csum_data = offsetof(struct tcphdr, th_sum);
699#ifdef INET6
700 if (isipv6) {
701 m->m_pkthdr.csum_flags = CSUM_TCP_IPV6;
702 nth->th_sum = in6_cksum_pseudo(ip6,
703 tlen - sizeof(struct ip6_hdr), IPPROTO_TCP, 0);
704 ip6->ip6_hlim = in6_selecthlim(tp != NULL ? tp->t_inpcb :
705 NULL, NULL);
706 }
707#endif /* INET6 */
708#if defined(INET6) && defined(INET)
709 else
710#endif
711#ifdef INET
712 {
713 m->m_pkthdr.csum_flags = CSUM_TCP;
714 nth->th_sum = in_pseudo(ip->ip_src.s_addr, ip->ip_dst.s_addr,
715 htons((u_short)(tlen - sizeof(struct ip) + ip->ip_p)));
716 }
717#endif /* INET */
718#ifdef TCPDEBUG
719 if (tp == NULL || (inp->inp_socket->so_options & SO_DEBUG))
720 tcp_trace(TA_OUTPUT, 0, tp, mtod(m, void *), th, 0);
721#endif
722 if (flags & TH_RST)
723 TCP_PROBE5(accept__refused, NULL, NULL, mtod(m, const char *),
724 tp, nth);
725
726 TCP_PROBE5(send, NULL, tp, mtod(m, const char *), tp, nth);
727#ifdef INET6
728 if (isipv6)
729 (void) ip6_output(m, NULL, NULL, ipflags, NULL, NULL, inp);
730#endif /* INET6 */
731#if defined(INET) && defined(INET6)
732 else
733#endif
734#ifdef INET
735 (void) ip_output(m, NULL, NULL, ipflags, NULL, inp);
736#endif
737}
738
739/*
740 * Create a new TCP control block, making an
741 * empty reassembly queue and hooking it to the argument
742 * protocol control block. The `inp' parameter must have
743 * come from the zone allocator set up in tcp_init().
744 */
745struct tcpcb *
746tcp_newtcpcb(struct inpcb *inp)
747{
748 struct tcpcb_mem *tm;
749 struct tcpcb *tp;
750#ifdef INET6
751 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
752#endif /* INET6 */
753
754 tm = uma_zalloc(V_tcpcb_zone, M_NOWAIT | M_ZERO);
755 if (tm == NULL)
756 return (NULL);
757 tp = &tm->tcb;
758
759 /* Initialise cc_var struct for this tcpcb. */
760 tp->ccv = &tm->ccv;
761 tp->ccv->type = IPPROTO_TCP;
762 tp->ccv->ccvc.tcp = tp;
763
764 /*
765 * Use the current system default CC algorithm.
766 */
767 CC_LIST_RLOCK();
768 KASSERT(!STAILQ_EMPTY(&cc_list), ("cc_list is empty!"));
769 CC_ALGO(tp) = CC_DEFAULT();
770 CC_LIST_RUNLOCK();
771
772 if (CC_ALGO(tp)->cb_init != NULL)
773 if (CC_ALGO(tp)->cb_init(tp->ccv) > 0) {
774 uma_zfree(V_tcpcb_zone, tm);
775 return (NULL);
776 }
777
778 tp->osd = &tm->osd;
779 if (khelp_init_osd(HELPER_CLASS_TCP, tp->osd)) {
780 uma_zfree(V_tcpcb_zone, tm);
781 return (NULL);
782 }
783
784#ifdef VIMAGE
785 tp->t_vnet = inp->inp_vnet;
786#endif
787 tp->t_timers = &tm->tt;
788 /* LIST_INIT(&tp->t_segq); */ /* XXX covered by M_ZERO */
789 tp->t_maxseg = tp->t_maxopd =
790#ifdef INET6
791 isipv6 ? V_tcp_v6mssdflt :
792#endif /* INET6 */
793 V_tcp_mssdflt;
794
795 /* Set up our timeouts. */
796 callout_init(&tp->t_timers->tt_rexmt, CALLOUT_MPSAFE);
797 callout_init(&tp->t_timers->tt_persist, CALLOUT_MPSAFE);
798 callout_init(&tp->t_timers->tt_keep, CALLOUT_MPSAFE);
799 callout_init(&tp->t_timers->tt_2msl, CALLOUT_MPSAFE);
800 callout_init(&tp->t_timers->tt_delack, CALLOUT_MPSAFE);
801
802 if (V_tcp_do_rfc1323)
803 tp->t_flags = (TF_REQ_SCALE|TF_REQ_TSTMP);
804 if (V_tcp_do_sack)
805 tp->t_flags |= TF_SACK_PERMIT;
806 TAILQ_INIT(&tp->snd_holes);
807 tp->t_inpcb = inp; /* XXX */
808 /*
809 * Init srtt to TCPTV_SRTTBASE (0), so we can tell that we have no
810 * rtt estimate. Set rttvar so that srtt + 4 * rttvar gives
811 * reasonable initial retransmit time.
812 */
813 tp->t_srtt = TCPTV_SRTTBASE;
814 tp->t_rttvar = ((TCPTV_RTOBASE - TCPTV_SRTTBASE) << TCP_RTTVAR_SHIFT) / 4;
815 tp->t_rttmin = tcp_rexmit_min;
816 tp->t_rxtcur = TCPTV_RTOBASE;
817 tp->snd_cwnd = TCP_MAXWIN << TCP_MAX_WINSHIFT;
818 tp->snd_ssthresh = TCP_MAXWIN << TCP_MAX_WINSHIFT;
819 tp->t_rcvtime = ticks;
820 /*
821 * IPv4 TTL initialization is necessary for an IPv6 socket as well,
822 * because the socket may be bound to an IPv6 wildcard address,
823 * which may match an IPv4-mapped IPv6 address.
824 */
825 inp->inp_ip_ttl = V_ip_defttl;
826 inp->inp_ppcb = tp;
827 return (tp); /* XXX */
828}
829
830/*
831 * Switch the congestion control algorithm back to NewReno for any active
832 * control blocks using an algorithm which is about to go away.
833 * This ensures the CC framework can allow the unload to proceed without leaving
834 * any dangling pointers which would trigger a panic.
835 * Returning non-zero would inform the CC framework that something went wrong
836 * and it would be unsafe to allow the unload to proceed. However, there is no
837 * way for this to occur with this implementation so we always return zero.
838 */
839int
840tcp_ccalgounload(struct cc_algo *unload_algo)
841{
842 struct cc_algo *tmpalgo;
843 struct inpcb *inp;
844 struct tcpcb *tp;
845 VNET_ITERATOR_DECL(vnet_iter);
846
847 /*
848 * Check all active control blocks across all network stacks and change
849 * any that are using "unload_algo" back to NewReno. If "unload_algo"
850 * requires cleanup code to be run, call it.
851 */
852 VNET_LIST_RLOCK();
853 VNET_FOREACH(vnet_iter) {
854 CURVNET_SET(vnet_iter);
855 INP_INFO_RLOCK(&V_tcbinfo);
856 /*
857 * New connections already part way through being initialised
858 * with the CC algo we're removing will not race with this code
859 * because the INP_INFO_WLOCK is held during initialisation. We
860 * therefore don't enter the loop below until the connection
861 * list has stabilised.
862 */
863 LIST_FOREACH(inp, &V_tcb, inp_list) {
864 INP_WLOCK(inp);
865 /* Important to skip tcptw structs. */
866 if (!(inp->inp_flags & INP_TIMEWAIT) &&
867 (tp = intotcpcb(inp)) != NULL) {
868 /*
869 * By holding INP_WLOCK here, we are assured
870 * that the connection is not currently
871 * executing inside the CC module's functions
872 * i.e. it is safe to make the switch back to
873 * NewReno.
874 */
875 if (CC_ALGO(tp) == unload_algo) {
876 tmpalgo = CC_ALGO(tp);
877 /* NewReno does not require any init. */
878 CC_ALGO(tp) = &newreno_cc_algo;
879 if (tmpalgo->cb_destroy != NULL)
880 tmpalgo->cb_destroy(tp->ccv);
881 }
882 }
883 INP_WUNLOCK(inp);
884 }
885 INP_INFO_RUNLOCK(&V_tcbinfo);
886 CURVNET_RESTORE();
887 }
888 VNET_LIST_RUNLOCK();
889
890 return (0);
891}
892
893/*
894 * Drop a TCP connection, reporting
895 * the specified error. If connection is synchronized,
896 * then send a RST to peer.
897 */
898struct tcpcb *
899tcp_drop(struct tcpcb *tp, int errno)
900{
901 struct socket *so = tp->t_inpcb->inp_socket;
902
903 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
904 INP_WLOCK_ASSERT(tp->t_inpcb);
905
906 if (TCPS_HAVERCVDSYN(tp->t_state)) {
907 tcp_state_change(tp, TCPS_CLOSED);
908 (void) tcp_output(tp);
909 TCPSTAT_INC(tcps_drops);
910 } else
911 TCPSTAT_INC(tcps_conndrops);
912 if (errno == ETIMEDOUT && tp->t_softerror)
913 errno = tp->t_softerror;
914 so->so_error = errno;
915 return (tcp_close(tp));
916}
917
918void
919tcp_discardcb(struct tcpcb *tp)
920{
921 struct inpcb *inp = tp->t_inpcb;
922 struct socket *so = inp->inp_socket;
923#ifdef INET6
924 int isipv6 = (inp->inp_vflag & INP_IPV6) != 0;
925#endif /* INET6 */
926
927 INP_WLOCK_ASSERT(inp);
928
929 /*
930 * Make sure that all of our timers are stopped before we delete the
931 * PCB.
932 *
933 * XXXRW: Really, we would like to use callout_drain() here in order
934 * to avoid races experienced in tcp_timer.c where a timer is already
935 * executing at this point. However, we can't, both because we're
936 * running in a context where we can't sleep, and also because we
937 * hold locks required by the timers. What we instead need to do is
938 * test to see if callout_drain() is required, and if so, defer some
939 * portion of the remainder of tcp_discardcb() to an asynchronous
940 * context that can callout_drain() and then continue. Some care
941 * will be required to ensure that no further processing takes place
942 * on the tcpcb, even though it hasn't been freed (a flag?).
943 */
944 callout_stop(&tp->t_timers->tt_rexmt);
945 callout_stop(&tp->t_timers->tt_persist);
946 callout_stop(&tp->t_timers->tt_keep);
947 callout_stop(&tp->t_timers->tt_2msl);
948 callout_stop(&tp->t_timers->tt_delack);
949
950 /*
951 * If we got enough samples through the srtt filter,
952 * save the rtt and rttvar in the routing entry.
953 * 'Enough' is arbitrarily defined as 4 rtt samples.
954 * 4 samples is enough for the srtt filter to converge
955 * to within enough % of the correct value; fewer samples
956 * and we could save a bogus rtt. The danger is not high
957 * as tcp quickly recovers from everything.
958 * XXX: Works very well but needs some more statistics!
959 */
960 if (tp->t_rttupdated >= 4) {
961 struct hc_metrics_lite metrics;
962 u_long ssthresh;
963
964 bzero(&metrics, sizeof(metrics));
965 /*
966 * Update the ssthresh always when the conditions below
967 * are satisfied. This gives us better new start value
968 * for the congestion avoidance for new connections.
969 * ssthresh is only set if packet loss occured on a session.
970 *
971 * XXXRW: 'so' may be NULL here, and/or socket buffer may be
972 * being torn down. Ideally this code would not use 'so'.
973 */
974 ssthresh = tp->snd_ssthresh;
975 if (ssthresh != 0 && ssthresh < so->so_snd.sb_hiwat / 2) {
976 /*
977 * convert the limit from user data bytes to
978 * packets then to packet data bytes.
979 */
980 ssthresh = (ssthresh + tp->t_maxseg / 2) / tp->t_maxseg;
981 if (ssthresh < 2)
982 ssthresh = 2;
983 ssthresh *= (u_long)(tp->t_maxseg +
984#ifdef INET6
985 (isipv6 ? sizeof (struct ip6_hdr) +
986 sizeof (struct tcphdr) :
987#endif
988 sizeof (struct tcpiphdr)
989#ifdef INET6
990 )
991#endif
992 );
993 } else
994 ssthresh = 0;
995 metrics.rmx_ssthresh = ssthresh;
996
997 metrics.rmx_rtt = tp->t_srtt;
998 metrics.rmx_rttvar = tp->t_rttvar;
999 metrics.rmx_cwnd = tp->snd_cwnd;
1000 metrics.rmx_sendpipe = 0;
1001 metrics.rmx_recvpipe = 0;
1002
1003 tcp_hc_update(&inp->inp_inc, &metrics);
1004 }
1005
1006 /* free the reassembly queue, if any */
1007 tcp_reass_flush(tp);
1008
1009#ifdef TCP_OFFLOAD
1010 /* Disconnect offload device, if any. */
1011 if (tp->t_flags & TF_TOE)
1012 tcp_offload_detach(tp);
1013#endif
1014
1015 tcp_free_sackholes(tp);
1016
1017 /* Allow the CC algorithm to clean up after itself. */
1018 if (CC_ALGO(tp)->cb_destroy != NULL)
1019 CC_ALGO(tp)->cb_destroy(tp->ccv);
1020
1021 khelp_destroy_osd(tp->osd);
1022
1023 CC_ALGO(tp) = NULL;
1024 inp->inp_ppcb = NULL;
1025 tp->t_inpcb = NULL;
1026 uma_zfree(V_tcpcb_zone, tp);
1027}
1028
1029/*
1030 * Attempt to close a TCP control block, marking it as dropped, and freeing
1031 * the socket if we hold the only reference.
1032 */
1033struct tcpcb *
1034tcp_close(struct tcpcb *tp)
1035{
1036 struct inpcb *inp = tp->t_inpcb;
1037 struct socket *so;
1038
1039 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1040 INP_WLOCK_ASSERT(inp);
1041
1042#ifdef TCP_OFFLOAD
1043 if (tp->t_state == TCPS_LISTEN)
1044 tcp_offload_listen_stop(tp);
1045#endif
1046 in_pcbdrop(inp);
1047 TCPSTAT_INC(tcps_closed);
1048 KASSERT(inp->inp_socket != NULL, ("tcp_close: inp_socket NULL"));
1049 so = inp->inp_socket;
1050 soisdisconnected(so);
1051 if (inp->inp_flags & INP_SOCKREF) {
1052 KASSERT(so->so_state & SS_PROTOREF,
1053 ("tcp_close: !SS_PROTOREF"));
1054 inp->inp_flags &= ~INP_SOCKREF;
1055 INP_WUNLOCK(inp);
1056 ACCEPT_LOCK();
1057 SOCK_LOCK(so);
1058 so->so_state &= ~SS_PROTOREF;
1059 sofree(so);
1060 return (NULL);
1061 }
1062 return (tp);
1063}
1064
1065void
1066tcp_drain(void)
1067{
1068 VNET_ITERATOR_DECL(vnet_iter);
1069
1070 if (!do_tcpdrain)
1071 return;
1072
1073 VNET_LIST_RLOCK_NOSLEEP();
1074 VNET_FOREACH(vnet_iter) {
1075 CURVNET_SET(vnet_iter);
1076 struct inpcb *inpb;
1077 struct tcpcb *tcpb;
1078
1079 /*
1080 * Walk the tcpbs, if existing, and flush the reassembly queue,
1081 * if there is one...
1082 * XXX: The "Net/3" implementation doesn't imply that the TCP
1083 * reassembly queue should be flushed, but in a situation
1084 * where we're really low on mbufs, this is potentially
1085 * useful.
1086 */
1087 INP_INFO_RLOCK(&V_tcbinfo);
1088 LIST_FOREACH(inpb, V_tcbinfo.ipi_listhead, inp_list) {
1089 if (inpb->inp_flags & INP_TIMEWAIT)
1090 continue;
1091 INP_WLOCK(inpb);
1092 if ((tcpb = intotcpcb(inpb)) != NULL) {
1093 tcp_reass_flush(tcpb);
1094 tcp_clean_sackreport(tcpb);
1095 }
1096 INP_WUNLOCK(inpb);
1097 }
1098 INP_INFO_RUNLOCK(&V_tcbinfo);
1099 CURVNET_RESTORE();
1100 }
1101 VNET_LIST_RUNLOCK_NOSLEEP();
1102}
1103
1104/*
1105 * Notify a tcp user of an asynchronous error;
1106 * store error as soft error, but wake up user
1107 * (for now, won't do anything until can select for soft error).
1108 *
1109 * Do not wake up user since there currently is no mechanism for
1110 * reporting soft errors (yet - a kqueue filter may be added).
1111 */
1112static struct inpcb *
1113tcp_notify(struct inpcb *inp, int error)
1114{
1115 struct tcpcb *tp;
1116
1117 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1118 INP_WLOCK_ASSERT(inp);
1119
1120 if ((inp->inp_flags & INP_TIMEWAIT) ||
1121 (inp->inp_flags & INP_DROPPED))
1122 return (inp);
1123
1124 tp = intotcpcb(inp);
1125 KASSERT(tp != NULL, ("tcp_notify: tp == NULL"));
1126
1127 /*
1128 * Ignore some errors if we are hooked up.
1129 * If connection hasn't completed, has retransmitted several times,
1130 * and receives a second error, give up now. This is better
1131 * than waiting a long time to establish a connection that
1132 * can never complete.
1133 */
1134 if (tp->t_state == TCPS_ESTABLISHED &&
1135 (error == EHOSTUNREACH || error == ENETUNREACH ||
1136 error == EHOSTDOWN)) {
1137 return (inp);
1138 } else if (tp->t_state < TCPS_ESTABLISHED && tp->t_rxtshift > 3 &&
1139 tp->t_softerror) {
1140 tp = tcp_drop(tp, error);
1141 if (tp != NULL)
1142 return (inp);
1143 else
1144 return (NULL);
1145 } else {
1146 tp->t_softerror = error;
1147 return (inp);
1148 }
1149#if 0
1150 wakeup( &so->so_timeo);
1151 sorwakeup(so);
1152 sowwakeup(so);
1153#endif
1154}
1155
1156static int
1157tcp_pcblist(SYSCTL_HANDLER_ARGS)
1158{
1159 int error, i, m, n, pcb_count;
1160 struct inpcb *inp, **inp_list;
1161 inp_gen_t gencnt;
1162 struct xinpgen xig;
1163
1164 /*
1165 * The process of preparing the TCB list is too time-consuming and
1166 * resource-intensive to repeat twice on every request.
1167 */
1168 if (req->oldptr == NULL) {
1169 n = V_tcbinfo.ipi_count + syncache_pcbcount();
1170 n += imax(n / 8, 10);
1171 req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1172 return (0);
1173 }
1174
1175 if (req->newptr != NULL)
1176 return (EPERM);
1177
1178 /*
1179 * OK, now we're committed to doing something.
1180 */
1181 INP_INFO_RLOCK(&V_tcbinfo);
1182 gencnt = V_tcbinfo.ipi_gencnt;
1183 n = V_tcbinfo.ipi_count;
1184 INP_INFO_RUNLOCK(&V_tcbinfo);
1185
1186 m = syncache_pcbcount();
1187
1188 error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1189 + (n + m) * sizeof(struct xtcpcb));
1190 if (error != 0)
1191 return (error);
1192
1193 xig.xig_len = sizeof xig;
1194 xig.xig_count = n + m;
1195 xig.xig_gen = gencnt;
1196 xig.xig_sogen = so_gencnt;
1197 error = SYSCTL_OUT(req, &xig, sizeof xig);
1198 if (error)
1199 return (error);
1200
1201 error = syncache_pcblist(req, m, &pcb_count);
1202 if (error)
1203 return (error);
1204
1205 inp_list = malloc(n * sizeof *inp_list, M_TEMP, M_WAITOK);
1206 if (inp_list == NULL)
1207 return (ENOMEM);
1208
1209 INP_INFO_RLOCK(&V_tcbinfo);
1210 for (inp = LIST_FIRST(V_tcbinfo.ipi_listhead), i = 0;
1211 inp != NULL && i < n; inp = LIST_NEXT(inp, inp_list)) {
1212 INP_WLOCK(inp);
1213 if (inp->inp_gencnt <= gencnt) {
1214 /*
1215 * XXX: This use of cr_cansee(), introduced with
1216 * TCP state changes, is not quite right, but for
1217 * now, better than nothing.
1218 */
1219 if (inp->inp_flags & INP_TIMEWAIT) {
1220 if (intotw(inp) != NULL)
1221 error = cr_cansee(req->td->td_ucred,
1222 intotw(inp)->tw_cred);
1223 else
1224 error = EINVAL; /* Skip this inp. */
1225 } else
1226 error = cr_canseeinpcb(req->td->td_ucred, inp);
1227 if (error == 0) {
1228 in_pcbref(inp);
1229 inp_list[i++] = inp;
1230 }
1231 }
1232 INP_WUNLOCK(inp);
1233 }
1234 INP_INFO_RUNLOCK(&V_tcbinfo);
1235 n = i;
1236
1237 error = 0;
1238 for (i = 0; i < n; i++) {
1239 inp = inp_list[i];
1240 INP_RLOCK(inp);
1241 if (inp->inp_gencnt <= gencnt) {
1242 struct xtcpcb xt;
1243 void *inp_ppcb;
1244
1245 bzero(&xt, sizeof(xt));
1246 xt.xt_len = sizeof xt;
1247 /* XXX should avoid extra copy */
1248 bcopy(inp, &xt.xt_inp, sizeof *inp);
1249 inp_ppcb = inp->inp_ppcb;
1250 if (inp_ppcb == NULL)
1251 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1252 else if (inp->inp_flags & INP_TIMEWAIT) {
1253 bzero((char *) &xt.xt_tp, sizeof xt.xt_tp);
1254 xt.xt_tp.t_state = TCPS_TIME_WAIT;
1255 } else {
1256 bcopy(inp_ppcb, &xt.xt_tp, sizeof xt.xt_tp);
1257 if (xt.xt_tp.t_timers)
1258 tcp_timer_to_xtimer(&xt.xt_tp, xt.xt_tp.t_timers, &xt.xt_timer);
1259 }
1260 if (inp->inp_socket != NULL)
1261 sotoxsocket(inp->inp_socket, &xt.xt_socket);
1262 else {
1263 bzero(&xt.xt_socket, sizeof xt.xt_socket);
1264 xt.xt_socket.xso_protocol = IPPROTO_TCP;
1265 }
1266 xt.xt_inp.inp_gencnt = inp->inp_gencnt;
1267 INP_RUNLOCK(inp);
1268 error = SYSCTL_OUT(req, &xt, sizeof xt);
1269 } else
1270 INP_RUNLOCK(inp);
1271 }
1272 INP_INFO_WLOCK(&V_tcbinfo);
1273 for (i = 0; i < n; i++) {
1274 inp = inp_list[i];
1275 INP_RLOCK(inp);
1276 if (!in_pcbrele_rlocked(inp))
1277 INP_RUNLOCK(inp);
1278 }
1279 INP_INFO_WUNLOCK(&V_tcbinfo);
1280
1281 if (!error) {
1282 /*
1283 * Give the user an updated idea of our state.
1284 * If the generation differs from what we told
1285 * her before, she knows that something happened
1286 * while we were processing this request, and it
1287 * might be necessary to retry.
1288 */
1289 INP_INFO_RLOCK(&V_tcbinfo);
1290 xig.xig_gen = V_tcbinfo.ipi_gencnt;
1291 xig.xig_sogen = so_gencnt;
1292 xig.xig_count = V_tcbinfo.ipi_count + pcb_count;
1293 INP_INFO_RUNLOCK(&V_tcbinfo);
1294 error = SYSCTL_OUT(req, &xig, sizeof xig);
1295 }
1296 free(inp_list, M_TEMP);
1297 return (error);
1298}
1299
1300SYSCTL_PROC(_net_inet_tcp, TCPCTL_PCBLIST, pcblist,
1301 CTLTYPE_OPAQUE | CTLFLAG_RD, NULL, 0,
1302 tcp_pcblist, "S,xtcpcb", "List of active TCP connections");
1303
1304#ifdef INET
1305static int
1306tcp_getcred(SYSCTL_HANDLER_ARGS)
1307{
1308 struct xucred xuc;
1309 struct sockaddr_in addrs[2];
1310 struct inpcb *inp;
1311 int error;
1312
1313 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1314 if (error)
1315 return (error);
1316 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1317 if (error)
1318 return (error);
1319 inp = in_pcblookup(&V_tcbinfo, addrs[1].sin_addr, addrs[1].sin_port,
1320 addrs[0].sin_addr, addrs[0].sin_port, INPLOOKUP_RLOCKPCB, NULL);
1321 if (inp != NULL) {
1322 if (inp->inp_socket == NULL)
1323 error = ENOENT;
1324 if (error == 0)
1325 error = cr_canseeinpcb(req->td->td_ucred, inp);
1326 if (error == 0)
1327 cru2x(inp->inp_cred, &xuc);
1328 INP_RUNLOCK(inp);
1329 } else
1330 error = ENOENT;
1331 if (error == 0)
1332 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1333 return (error);
1334}
1335
1336SYSCTL_PROC(_net_inet_tcp, OID_AUTO, getcred,
1337 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1338 tcp_getcred, "S,xucred", "Get the xucred of a TCP connection");
1339#endif /* INET */
1340
1341#ifdef INET6
1342static int
1343tcp6_getcred(SYSCTL_HANDLER_ARGS)
1344{
1345 struct xucred xuc;
1346 struct sockaddr_in6 addrs[2];
1347 struct inpcb *inp;
1348 int error;
1349#ifdef INET
1350 int mapped = 0;
1351#endif
1352
1353 error = priv_check(req->td, PRIV_NETINET_GETCRED);
1354 if (error)
1355 return (error);
1356 error = SYSCTL_IN(req, addrs, sizeof(addrs));
1357 if (error)
1358 return (error);
1359 if ((error = sa6_embedscope(&addrs[0], V_ip6_use_defzone)) != 0 ||
1360 (error = sa6_embedscope(&addrs[1], V_ip6_use_defzone)) != 0) {
1361 return (error);
1362 }
1363 if (IN6_IS_ADDR_V4MAPPED(&addrs[0].sin6_addr)) {
1364#ifdef INET
1365 if (IN6_IS_ADDR_V4MAPPED(&addrs[1].sin6_addr))
1366 mapped = 1;
1367 else
1368#endif
1369 return (EINVAL);
1370 }
1371
1372#ifdef INET
1373 if (mapped == 1)
1374 inp = in_pcblookup(&V_tcbinfo,
1375 *(struct in_addr *)&addrs[1].sin6_addr.s6_addr[12],
1376 addrs[1].sin6_port,
1377 *(struct in_addr *)&addrs[0].sin6_addr.s6_addr[12],
1378 addrs[0].sin6_port, INPLOOKUP_RLOCKPCB, NULL);
1379 else
1380#endif
1381 inp = in6_pcblookup(&V_tcbinfo,
1382 &addrs[1].sin6_addr, addrs[1].sin6_port,
1383 &addrs[0].sin6_addr, addrs[0].sin6_port,
1384 INPLOOKUP_RLOCKPCB, NULL);
1385 if (inp != NULL) {
1386 if (inp->inp_socket == NULL)
1387 error = ENOENT;
1388 if (error == 0)
1389 error = cr_canseeinpcb(req->td->td_ucred, inp);
1390 if (error == 0)
1391 cru2x(inp->inp_cred, &xuc);
1392 INP_RUNLOCK(inp);
1393 } else
1394 error = ENOENT;
1395 if (error == 0)
1396 error = SYSCTL_OUT(req, &xuc, sizeof(struct xucred));
1397 return (error);
1398}
1399
1400SYSCTL_PROC(_net_inet6_tcp6, OID_AUTO, getcred,
1401 CTLTYPE_OPAQUE|CTLFLAG_RW|CTLFLAG_PRISON, 0, 0,
1402 tcp6_getcred, "S,xucred", "Get the xucred of a TCP6 connection");
1403#endif /* INET6 */
1404
1405
1406#ifdef INET
1407void
1408tcp_ctlinput(int cmd, struct sockaddr *sa, void *vip)
1409{
1410 struct ip *ip = vip;
1411 struct tcphdr *th;
1412 struct in_addr faddr;
1413 struct inpcb *inp;
1414 struct tcpcb *tp;
1415 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1416 struct icmp *icp;
1417 struct in_conninfo inc;
1418 tcp_seq icmp_tcp_seq;
1419 int mtu;
1420
1421 faddr = ((struct sockaddr_in *)sa)->sin_addr;
1422 if (sa->sa_family != AF_INET || faddr.s_addr == INADDR_ANY)
1423 return;
1424
1425 if (cmd == PRC_MSGSIZE)
1426 notify = tcp_mtudisc_notify;
1427 else if (V_icmp_may_rst && (cmd == PRC_UNREACH_ADMIN_PROHIB ||
1428 cmd == PRC_UNREACH_PORT || cmd == PRC_TIMXCEED_INTRANS) && ip)
1429 notify = tcp_drop_syn_sent;
1430 /*
1431 * Redirects don't need to be handled up here.
1432 */
1433 else if (PRC_IS_REDIRECT(cmd))
1434 return;
1435 /*
1436 * Source quench is depreciated.
1437 */
1438 else if (cmd == PRC_QUENCH)
1439 return;
1440 /*
1441 * Hostdead is ugly because it goes linearly through all PCBs.
1442 * XXX: We never get this from ICMP, otherwise it makes an
1443 * excellent DoS attack on machines with many connections.
1444 */
1445 else if (cmd == PRC_HOSTDEAD)
1446 ip = NULL;
1447 else if ((unsigned)cmd >= PRC_NCMDS || inetctlerrmap[cmd] == 0)
1448 return;
1449 if (ip != NULL) {
1450 icp = (struct icmp *)((caddr_t)ip
1451 - offsetof(struct icmp, icmp_ip));
1452 th = (struct tcphdr *)((caddr_t)ip
1453 + (ip->ip_hl << 2));
1454 INP_INFO_WLOCK(&V_tcbinfo);
1455 inp = in_pcblookup(&V_tcbinfo, faddr, th->th_dport,
1456 ip->ip_src, th->th_sport, INPLOOKUP_WLOCKPCB, NULL);
1457 if (inp != NULL) {
1458 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1459 !(inp->inp_flags & INP_DROPPED) &&
1460 !(inp->inp_socket == NULL)) {
1461 icmp_tcp_seq = htonl(th->th_seq);
1462 tp = intotcpcb(inp);
1463 if (SEQ_GEQ(icmp_tcp_seq, tp->snd_una) &&
1464 SEQ_LT(icmp_tcp_seq, tp->snd_max)) {
1465 if (cmd == PRC_MSGSIZE) {
1466 /*
1467 * MTU discovery:
1468 * If we got a needfrag set the MTU
1469 * in the route to the suggested new
1470 * value (if given) and then notify.
1471 */
1472 bzero(&inc, sizeof(inc));
1473 inc.inc_faddr = faddr;
1474 inc.inc_fibnum =
1475 inp->inp_inc.inc_fibnum;
1476
1477 mtu = ntohs(icp->icmp_nextmtu);
1478 /*
1479 * If no alternative MTU was
1480 * proposed, try the next smaller
1481 * one.
1482 */
1483 if (!mtu)
1484 mtu = ip_next_mtu(
1485 ntohs(ip->ip_len), 1);
1486 if (mtu < V_tcp_minmss
1487 + sizeof(struct tcpiphdr))
1488 mtu = V_tcp_minmss
1489 + sizeof(struct tcpiphdr);
1490 /*
1491 * Only cache the MTU if it
1492 * is smaller than the interface
1493 * or route MTU. tcp_mtudisc()
1494 * will do right thing by itself.
1495 */
1496 if (mtu <= tcp_maxmtu(&inc, NULL))
1497 tcp_hc_updatemtu(&inc, mtu);
1498 tcp_mtudisc(inp, mtu);
1499 } else
1500 inp = (*notify)(inp,
1501 inetctlerrmap[cmd]);
1502 }
1503 }
1504 if (inp != NULL)
1505 INP_WUNLOCK(inp);
1506 } else {
1507 bzero(&inc, sizeof(inc));
1508 inc.inc_fport = th->th_dport;
1509 inc.inc_lport = th->th_sport;
1510 inc.inc_faddr = faddr;
1511 inc.inc_laddr = ip->ip_src;
1512 syncache_unreach(&inc, th);
1513 }
1514 INP_INFO_WUNLOCK(&V_tcbinfo);
1515 } else
1516 in_pcbnotifyall(&V_tcbinfo, faddr, inetctlerrmap[cmd], notify);
1517}
1518#endif /* INET */
1519
1520#ifdef INET6
1521void
1522tcp6_ctlinput(int cmd, struct sockaddr *sa, void *d)
1523{
1524 struct tcphdr th;
1525 struct inpcb *(*notify)(struct inpcb *, int) = tcp_notify;
1526 struct ip6_hdr *ip6;
1527 struct mbuf *m;
1528 struct ip6ctlparam *ip6cp = NULL;
1529 const struct sockaddr_in6 *sa6_src = NULL;
1530 int off;
1531 struct tcp_portonly {
1532 u_int16_t th_sport;
1533 u_int16_t th_dport;
1534 } *thp;
1535
1536 if (sa->sa_family != AF_INET6 ||
1537 sa->sa_len != sizeof(struct sockaddr_in6))
1538 return;
1539
1540 if (cmd == PRC_MSGSIZE)
1541 notify = tcp_mtudisc_notify;
1542 else if (!PRC_IS_REDIRECT(cmd) &&
1543 ((unsigned)cmd >= PRC_NCMDS || inet6ctlerrmap[cmd] == 0))
1544 return;
1545 /* Source quench is depreciated. */
1546 else if (cmd == PRC_QUENCH)
1547 return;
1548
1549 /* if the parameter is from icmp6, decode it. */
1550 if (d != NULL) {
1551 ip6cp = (struct ip6ctlparam *)d;
1552 m = ip6cp->ip6c_m;
1553 ip6 = ip6cp->ip6c_ip6;
1554 off = ip6cp->ip6c_off;
1555 sa6_src = ip6cp->ip6c_src;
1556 } else {
1557 m = NULL;
1558 ip6 = NULL;
1559 off = 0; /* fool gcc */
1560 sa6_src = &sa6_any;
1561 }
1562
1563 if (ip6 != NULL) {
1564 struct in_conninfo inc;
1565 /*
1566 * XXX: We assume that when IPV6 is non NULL,
1567 * M and OFF are valid.
1568 */
1569
1570 /* check if we can safely examine src and dst ports */
1571 if (m->m_pkthdr.len < off + sizeof(*thp))
1572 return;
1573
1574 bzero(&th, sizeof(th));
1575 m_copydata(m, off, sizeof(*thp), (caddr_t)&th);
1576
1577 in6_pcbnotify(&V_tcbinfo, sa, th.th_dport,
1578 (struct sockaddr *)ip6cp->ip6c_src,
1579 th.th_sport, cmd, NULL, notify);
1580
1581 bzero(&inc, sizeof(inc));
1582 inc.inc_fport = th.th_dport;
1583 inc.inc_lport = th.th_sport;
1584 inc.inc6_faddr = ((struct sockaddr_in6 *)sa)->sin6_addr;
1585 inc.inc6_laddr = ip6cp->ip6c_src->sin6_addr;
1586 inc.inc_flags |= INC_ISIPV6;
1587 INP_INFO_WLOCK(&V_tcbinfo);
1588 syncache_unreach(&inc, &th);
1589 INP_INFO_WUNLOCK(&V_tcbinfo);
1590 } else
1591 in6_pcbnotify(&V_tcbinfo, sa, 0, (const struct sockaddr *)sa6_src,
1592 0, cmd, NULL, notify);
1593}
1594#endif /* INET6 */
1595
1596
1597/*
1598 * Following is where TCP initial sequence number generation occurs.
1599 *
1600 * There are two places where we must use initial sequence numbers:
1601 * 1. In SYN-ACK packets.
1602 * 2. In SYN packets.
1603 *
1604 * All ISNs for SYN-ACK packets are generated by the syncache. See
1605 * tcp_syncache.c for details.
1606 *
1607 * The ISNs in SYN packets must be monotonic; TIME_WAIT recycling
1608 * depends on this property. In addition, these ISNs should be
1609 * unguessable so as to prevent connection hijacking. To satisfy
1610 * the requirements of this situation, the algorithm outlined in
1611 * RFC 1948 is used, with only small modifications.
1612 *
1613 * Implementation details:
1614 *
1615 * Time is based off the system timer, and is corrected so that it
1616 * increases by one megabyte per second. This allows for proper
1617 * recycling on high speed LANs while still leaving over an hour
1618 * before rollover.
1619 *
1620 * As reading the *exact* system time is too expensive to be done
1621 * whenever setting up a TCP connection, we increment the time
1622 * offset in two ways. First, a small random positive increment
1623 * is added to isn_offset for each connection that is set up.
1624 * Second, the function tcp_isn_tick fires once per clock tick
1625 * and increments isn_offset as necessary so that sequence numbers
1626 * are incremented at approximately ISN_BYTES_PER_SECOND. The
1627 * random positive increments serve only to ensure that the same
1628 * exact sequence number is never sent out twice (as could otherwise
1629 * happen when a port is recycled in less than the system tick
1630 * interval.)
1631 *
1632 * net.inet.tcp.isn_reseed_interval controls the number of seconds
1633 * between seeding of isn_secret. This is normally set to zero,
1634 * as reseeding should not be necessary.
1635 *
1636 * Locking of the global variables isn_secret, isn_last_reseed, isn_offset,
1637 * isn_offset_old, and isn_ctx is performed using the TCP pcbinfo lock. In
1638 * general, this means holding an exclusive (write) lock.
1639 */
1640
1641#define ISN_BYTES_PER_SECOND 1048576
1642#define ISN_STATIC_INCREMENT 4096
1643#define ISN_RANDOM_INCREMENT (4096 - 1)
1644
1645static VNET_DEFINE(u_char, isn_secret[32]);
1646static VNET_DEFINE(int, isn_last);
1647static VNET_DEFINE(int, isn_last_reseed);
1648static VNET_DEFINE(u_int32_t, isn_offset);
1649static VNET_DEFINE(u_int32_t, isn_offset_old);
1650
1651#define V_isn_secret VNET(isn_secret)
1652#define V_isn_last VNET(isn_last)
1653#define V_isn_last_reseed VNET(isn_last_reseed)
1654#define V_isn_offset VNET(isn_offset)
1655#define V_isn_offset_old VNET(isn_offset_old)
1656
1657tcp_seq
1658tcp_new_isn(struct tcpcb *tp)
1659{
1660 MD5_CTX isn_ctx;
1661 u_int32_t md5_buffer[4];
1662 tcp_seq new_isn;
1663 u_int32_t projected_offset;
1664
1665 INP_WLOCK_ASSERT(tp->t_inpcb);
1666
1667 ISN_LOCK();
1668 /* Seed if this is the first use, reseed if requested. */
1669 if ((V_isn_last_reseed == 0) || ((V_tcp_isn_reseed_interval > 0) &&
1670 (((u_int)V_isn_last_reseed + (u_int)V_tcp_isn_reseed_interval*hz)
1671 < (u_int)ticks))) {
1672 read_random(&V_isn_secret, sizeof(V_isn_secret));
1673 V_isn_last_reseed = ticks;
1674 }
1675
1676 /* Compute the md5 hash and return the ISN. */
1677 MD5Init(&isn_ctx);
1678 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_fport, sizeof(u_short));
1679 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_lport, sizeof(u_short));
1680#ifdef INET6
1681 if ((tp->t_inpcb->inp_vflag & INP_IPV6) != 0) {
1682 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_faddr,
1683 sizeof(struct in6_addr));
1684 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->in6p_laddr,
1685 sizeof(struct in6_addr));
1686 } else
1687#endif
1688 {
1689 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_faddr,
1690 sizeof(struct in_addr));
1691 MD5Update(&isn_ctx, (u_char *) &tp->t_inpcb->inp_laddr,
1692 sizeof(struct in_addr));
1693 }
1694 MD5Update(&isn_ctx, (u_char *) &V_isn_secret, sizeof(V_isn_secret));
1695 MD5Final((u_char *) &md5_buffer, &isn_ctx);
1696 new_isn = (tcp_seq) md5_buffer[0];
1697 V_isn_offset += ISN_STATIC_INCREMENT +
1698 (arc4random() & ISN_RANDOM_INCREMENT);
1699 if (ticks != V_isn_last) {
1700 projected_offset = V_isn_offset_old +
1701 ISN_BYTES_PER_SECOND / hz * (ticks - V_isn_last);
1702 if (SEQ_GT(projected_offset, V_isn_offset))
1703 V_isn_offset = projected_offset;
1704 V_isn_offset_old = V_isn_offset;
1705 V_isn_last = ticks;
1706 }
1707 new_isn += V_isn_offset;
1708 ISN_UNLOCK();
1709 return (new_isn);
1710}
1711
1712/*
1713 * When a specific ICMP unreachable message is received and the
1714 * connection state is SYN-SENT, drop the connection. This behavior
1715 * is controlled by the icmp_may_rst sysctl.
1716 */
1717struct inpcb *
1718tcp_drop_syn_sent(struct inpcb *inp, int errno)
1719{
1720 struct tcpcb *tp;
1721
1722 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1723 INP_WLOCK_ASSERT(inp);
1724
1725 if ((inp->inp_flags & INP_TIMEWAIT) ||
1726 (inp->inp_flags & INP_DROPPED))
1727 return (inp);
1728
1729 tp = intotcpcb(inp);
1730 if (tp->t_state != TCPS_SYN_SENT)
1731 return (inp);
1732
1733 tp = tcp_drop(tp, errno);
1734 if (tp != NULL)
1735 return (inp);
1736 else
1737 return (NULL);
1738}
1739
1740/*
1741 * When `need fragmentation' ICMP is received, update our idea of the MSS
1742 * based on the new value. Also nudge TCP to send something, since we
1743 * know the packet we just sent was dropped.
1744 * This duplicates some code in the tcp_mss() function in tcp_input.c.
1745 */
1746static struct inpcb *
1747tcp_mtudisc_notify(struct inpcb *inp, int error)
1748{
1749
1750 return (tcp_mtudisc(inp, -1));
1751}
1752
1753struct inpcb *
1754tcp_mtudisc(struct inpcb *inp, int mtuoffer)
1755{
1756 struct tcpcb *tp;
1757 struct socket *so;
1758
1759 INP_WLOCK_ASSERT(inp);
1760 if ((inp->inp_flags & INP_TIMEWAIT) ||
1761 (inp->inp_flags & INP_DROPPED))
1762 return (inp);
1763
1764 tp = intotcpcb(inp);
1765 KASSERT(tp != NULL, ("tcp_mtudisc: tp == NULL"));
1766
1767 tcp_mss_update(tp, -1, mtuoffer, NULL, NULL);
1768
1769 so = inp->inp_socket;
1770 SOCKBUF_LOCK(&so->so_snd);
1771 /* If the mss is larger than the socket buffer, decrease the mss. */
1772 if (so->so_snd.sb_hiwat < tp->t_maxseg)
1773 tp->t_maxseg = so->so_snd.sb_hiwat;
1774 SOCKBUF_UNLOCK(&so->so_snd);
1775
1776 TCPSTAT_INC(tcps_mturesent);
1777 tp->t_rtttime = 0;
1778 tp->snd_nxt = tp->snd_una;
1779 tcp_free_sackholes(tp);
1780 tp->snd_recover = tp->snd_max;
1781 if (tp->t_flags & TF_SACK_PERMIT)
1782 EXIT_FASTRECOVERY(tp->t_flags);
1783 tcp_output(tp);
1784 return (inp);
1785}
1786
1787#ifdef INET
1788/*
1789 * Look-up the routing entry to the peer of this inpcb. If no route
1790 * is found and it cannot be allocated, then return 0. This routine
1791 * is called by TCP routines that access the rmx structure and by
1792 * tcp_mss_update to get the peer/interface MTU.
1793 */
1794u_long
1795tcp_maxmtu(struct in_conninfo *inc, struct tcp_ifcap *cap)
1796{
1797 struct route sro;
1798 struct sockaddr_in *dst;
1799 struct ifnet *ifp;
1800 u_long maxmtu = 0;
1801
1802 KASSERT(inc != NULL, ("tcp_maxmtu with NULL in_conninfo pointer"));
1803
1804 bzero(&sro, sizeof(sro));
1805 if (inc->inc_faddr.s_addr != INADDR_ANY) {
1806 dst = (struct sockaddr_in *)&sro.ro_dst;
1807 dst->sin_family = AF_INET;
1808 dst->sin_len = sizeof(*dst);
1809 dst->sin_addr = inc->inc_faddr;
1810 in_rtalloc_ign(&sro, 0, inc->inc_fibnum);
1811 }
1812 if (sro.ro_rt != NULL) {
1813 ifp = sro.ro_rt->rt_ifp;
|
1855 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1856
1857 /* Report additional interface capabilities. */
1858 if (cap != NULL) {
1859 if (ifp->if_capenable & IFCAP_TSO6 &&
1860 ifp->if_hwassist & CSUM_TSO)
1861 cap->ifcap |= CSUM_TSO;
1862 cap->tsomax = ifp->if_hw_tsomax;
1863 }
1864 RTFREE(sro6.ro_rt);
1865 }
1866
1867 return (maxmtu);
1868}
1869#endif /* INET6 */
1870
1871#ifdef IPSEC
1872/* compute ESP/AH header size for TCP, including outer IP header. */
1873size_t
1874ipsec_hdrsiz_tcp(struct tcpcb *tp)
1875{
1876 struct inpcb *inp;
1877 struct mbuf *m;
1878 size_t hdrsiz;
1879 struct ip *ip;
1880#ifdef INET6
1881 struct ip6_hdr *ip6;
1882#endif
1883 struct tcphdr *th;
1884
1885 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1886 return (0);
1887 m = m_gethdr(M_NOWAIT, MT_DATA);
1888 if (!m)
1889 return (0);
1890
1891#ifdef INET6
1892 if ((inp->inp_vflag & INP_IPV6) != 0) {
1893 ip6 = mtod(m, struct ip6_hdr *);
1894 th = (struct tcphdr *)(ip6 + 1);
1895 m->m_pkthdr.len = m->m_len =
1896 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1897 tcpip_fillheaders(inp, ip6, th);
1898 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1899 } else
1900#endif /* INET6 */
1901 {
1902 ip = mtod(m, struct ip *);
1903 th = (struct tcphdr *)(ip + 1);
1904 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1905 tcpip_fillheaders(inp, ip, th);
1906 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1907 }
1908
1909 m_free(m);
1910 return (hdrsiz);
1911}
1912#endif /* IPSEC */
1913
1914#ifdef TCP_SIGNATURE
1915/*
1916 * Callback function invoked by m_apply() to digest TCP segment data
1917 * contained within an mbuf chain.
1918 */
1919static int
1920tcp_signature_apply(void *fstate, void *data, u_int len)
1921{
1922
1923 MD5Update(fstate, (u_char *)data, len);
1924 return (0);
1925}
1926
1927/*
1928 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1929 *
1930 * Parameters:
1931 * m pointer to head of mbuf chain
1932 * _unused
1933 * len length of TCP segment data, excluding options
1934 * optlen length of TCP segment options
1935 * buf pointer to storage for computed MD5 digest
1936 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1937 *
1938 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1939 * When called from tcp_input(), we can be sure that th_sum has been
1940 * zeroed out and verified already.
1941 *
1942 * Return 0 if successful, otherwise return -1.
1943 *
1944 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1945 * search with the destination IP address, and a 'magic SPI' to be
1946 * determined by the application. This is hardcoded elsewhere to 1179
1947 * right now. Another branch of this code exists which uses the SPD to
1948 * specify per-application flows but it is unstable.
1949 */
1950int
1951tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1952 u_char *buf, u_int direction)
1953{
1954 union sockaddr_union dst;
1955#ifdef INET
1956 struct ippseudo ippseudo;
1957#endif
1958 MD5_CTX ctx;
1959 int doff;
1960 struct ip *ip;
1961#ifdef INET
1962 struct ipovly *ipovly;
1963#endif
1964 struct secasvar *sav;
1965 struct tcphdr *th;
1966#ifdef INET6
1967 struct ip6_hdr *ip6;
1968 struct in6_addr in6;
1969 char ip6buf[INET6_ADDRSTRLEN];
1970 uint32_t plen;
1971 uint16_t nhdr;
1972#endif
1973 u_short savecsum;
1974
1975 KASSERT(m != NULL, ("NULL mbuf chain"));
1976 KASSERT(buf != NULL, ("NULL signature pointer"));
1977
1978 /* Extract the destination from the IP header in the mbuf. */
1979 bzero(&dst, sizeof(union sockaddr_union));
1980 ip = mtod(m, struct ip *);
1981#ifdef INET6
1982 ip6 = NULL; /* Make the compiler happy. */
1983#endif
1984 switch (ip->ip_v) {
1985#ifdef INET
1986 case IPVERSION:
1987 dst.sa.sa_len = sizeof(struct sockaddr_in);
1988 dst.sa.sa_family = AF_INET;
1989 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1990 ip->ip_src : ip->ip_dst;
1991 break;
1992#endif
1993#ifdef INET6
1994 case (IPV6_VERSION >> 4):
1995 ip6 = mtod(m, struct ip6_hdr *);
1996 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1997 dst.sa.sa_family = AF_INET6;
1998 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1999 ip6->ip6_src : ip6->ip6_dst;
2000 break;
2001#endif
2002 default:
2003 return (EINVAL);
2004 /* NOTREACHED */
2005 break;
2006 }
2007
2008 /* Look up an SADB entry which matches the address of the peer. */
2009 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2010 if (sav == NULL) {
2011 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2012 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2013#ifdef INET6
2014 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2015 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2016#endif
2017 "(unsupported)"));
2018 return (EINVAL);
2019 }
2020
2021 MD5Init(&ctx);
2022 /*
2023 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2024 *
2025 * XXX The ippseudo header MUST be digested in network byte order,
2026 * or else we'll fail the regression test. Assume all fields we've
2027 * been doing arithmetic on have been in host byte order.
2028 * XXX One cannot depend on ipovly->ih_len here. When called from
2029 * tcp_output(), the underlying ip_len member has not yet been set.
2030 */
2031 switch (ip->ip_v) {
2032#ifdef INET
2033 case IPVERSION:
2034 ipovly = (struct ipovly *)ip;
2035 ippseudo.ippseudo_src = ipovly->ih_src;
2036 ippseudo.ippseudo_dst = ipovly->ih_dst;
2037 ippseudo.ippseudo_pad = 0;
2038 ippseudo.ippseudo_p = IPPROTO_TCP;
2039 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2040 optlen);
2041 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2042
2043 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2044 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2045 break;
2046#endif
2047#ifdef INET6
2048 /*
2049 * RFC 2385, 2.0 Proposal
2050 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2051 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2052 * extended next header value (to form 32 bits), and 32-bit segment
2053 * length.
2054 * Note: Upper-Layer Packet Length comes before Next Header.
2055 */
2056 case (IPV6_VERSION >> 4):
2057 in6 = ip6->ip6_src;
2058 in6_clearscope(&in6);
2059 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2060 in6 = ip6->ip6_dst;
2061 in6_clearscope(&in6);
2062 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2063 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2064 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2065 nhdr = 0;
2066 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2067 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2068 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2069 nhdr = IPPROTO_TCP;
2070 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071
2072 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2073 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2074 break;
2075#endif
2076 default:
2077 return (EINVAL);
2078 /* NOTREACHED */
2079 break;
2080 }
2081
2082
2083 /*
2084 * Step 2: Update MD5 hash with TCP header, excluding options.
2085 * The TCP checksum must be set to zero.
2086 */
2087 savecsum = th->th_sum;
2088 th->th_sum = 0;
2089 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2090 th->th_sum = savecsum;
2091
2092 /*
2093 * Step 3: Update MD5 hash with TCP segment data.
2094 * Use m_apply() to avoid an early m_pullup().
2095 */
2096 if (len > 0)
2097 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2098
2099 /*
2100 * Step 4: Update MD5 hash with shared secret.
2101 */
2102 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2103 MD5Final(buf, &ctx);
2104
2105 key_sa_recordxfer(sav, m);
2106 KEY_FREESAV(&sav);
2107 return (0);
2108}
2109
2110/*
2111 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2112 *
2113 * Parameters:
2114 * m pointer to head of mbuf chain
2115 * len length of TCP segment data, excluding options
2116 * optlen length of TCP segment options
2117 * buf pointer to storage for computed MD5 digest
2118 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2119 *
2120 * Return 1 if successful, otherwise return 0.
2121 */
2122int
2123tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2124 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2125{
2126 char tmpdigest[TCP_SIGLEN];
2127
2128 if (tcp_sig_checksigs == 0)
2129 return (1);
2130 if ((tcpbflag & TF_SIGNATURE) == 0) {
2131 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2132
2133 /*
2134 * If this socket is not expecting signature but
2135 * the segment contains signature just fail.
2136 */
2137 TCPSTAT_INC(tcps_sig_err_sigopt);
2138 TCPSTAT_INC(tcps_sig_rcvbadsig);
2139 return (0);
2140 }
2141
2142 /* Signature is not expected, and not present in segment. */
2143 return (1);
2144 }
2145
2146 /*
2147 * If this socket is expecting signature but the segment does not
2148 * contain any just fail.
2149 */
2150 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2151 TCPSTAT_INC(tcps_sig_err_nosigopt);
2152 TCPSTAT_INC(tcps_sig_rcvbadsig);
2153 return (0);
2154 }
2155 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2156 IPSEC_DIR_INBOUND) == -1) {
2157 TCPSTAT_INC(tcps_sig_err_buildsig);
2158 TCPSTAT_INC(tcps_sig_rcvbadsig);
2159 return (0);
2160 }
2161
2162 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2163 TCPSTAT_INC(tcps_sig_rcvbadsig);
2164 return (0);
2165 }
2166 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2167 return (1);
2168}
2169#endif /* TCP_SIGNATURE */
2170
2171static int
2172sysctl_drop(SYSCTL_HANDLER_ARGS)
2173{
2174 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2175 struct sockaddr_storage addrs[2];
2176 struct inpcb *inp;
2177 struct tcpcb *tp;
2178 struct tcptw *tw;
2179 struct sockaddr_in *fin, *lin;
2180#ifdef INET6
2181 struct sockaddr_in6 *fin6, *lin6;
2182#endif
2183 int error;
2184
2185 inp = NULL;
2186 fin = lin = NULL;
2187#ifdef INET6
2188 fin6 = lin6 = NULL;
2189#endif
2190 error = 0;
2191
2192 if (req->oldptr != NULL || req->oldlen != 0)
2193 return (EINVAL);
2194 if (req->newptr == NULL)
2195 return (EPERM);
2196 if (req->newlen < sizeof(addrs))
2197 return (ENOMEM);
2198 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2199 if (error)
2200 return (error);
2201
2202 switch (addrs[0].ss_family) {
2203#ifdef INET6
2204 case AF_INET6:
2205 fin6 = (struct sockaddr_in6 *)&addrs[0];
2206 lin6 = (struct sockaddr_in6 *)&addrs[1];
2207 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2208 lin6->sin6_len != sizeof(struct sockaddr_in6))
2209 return (EINVAL);
2210 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2211 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2212 return (EINVAL);
2213 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2214 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2215 fin = (struct sockaddr_in *)&addrs[0];
2216 lin = (struct sockaddr_in *)&addrs[1];
2217 break;
2218 }
2219 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2220 if (error)
2221 return (error);
2222 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2223 if (error)
2224 return (error);
2225 break;
2226#endif
2227#ifdef INET
2228 case AF_INET:
2229 fin = (struct sockaddr_in *)&addrs[0];
2230 lin = (struct sockaddr_in *)&addrs[1];
2231 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2232 lin->sin_len != sizeof(struct sockaddr_in))
2233 return (EINVAL);
2234 break;
2235#endif
2236 default:
2237 return (EINVAL);
2238 }
2239 INP_INFO_WLOCK(&V_tcbinfo);
2240 switch (addrs[0].ss_family) {
2241#ifdef INET6
2242 case AF_INET6:
2243 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2244 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2245 INPLOOKUP_WLOCKPCB, NULL);
2246 break;
2247#endif
2248#ifdef INET
2249 case AF_INET:
2250 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2251 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2252 break;
2253#endif
2254 }
2255 if (inp != NULL) {
2256 if (inp->inp_flags & INP_TIMEWAIT) {
2257 /*
2258 * XXXRW: There currently exists a state where an
2259 * inpcb is present, but its timewait state has been
2260 * discarded. For now, don't allow dropping of this
2261 * type of inpcb.
2262 */
2263 tw = intotw(inp);
2264 if (tw != NULL)
2265 tcp_twclose(tw, 0);
2266 else
2267 INP_WUNLOCK(inp);
2268 } else if (!(inp->inp_flags & INP_DROPPED) &&
2269 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2270 tp = intotcpcb(inp);
2271 tp = tcp_drop(tp, ECONNABORTED);
2272 if (tp != NULL)
2273 INP_WUNLOCK(inp);
2274 } else
2275 INP_WUNLOCK(inp);
2276 } else
2277 error = ESRCH;
2278 INP_INFO_WUNLOCK(&V_tcbinfo);
2279 return (error);
2280}
2281
2282SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2283 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2284 0, sysctl_drop, "", "Drop TCP connection");
2285
2286/*
2287 * Generate a standardized TCP log line for use throughout the
2288 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2289 * allow use in the interrupt context.
2290 *
2291 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2292 * NB: The function may return NULL if memory allocation failed.
2293 *
2294 * Due to header inclusion and ordering limitations the struct ip
2295 * and ip6_hdr pointers have to be passed as void pointers.
2296 */
2297char *
2298tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2299 const void *ip6hdr)
2300{
2301
2302 /* Is logging enabled? */
2303 if (tcp_log_in_vain == 0)
2304 return (NULL);
2305
2306 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2307}
2308
2309char *
2310tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2311 const void *ip6hdr)
2312{
2313
2314 /* Is logging enabled? */
2315 if (tcp_log_debug == 0)
2316 return (NULL);
2317
2318 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2319}
2320
2321static char *
2322tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2323 const void *ip6hdr)
2324{
2325 char *s, *sp;
2326 size_t size;
2327 struct ip *ip;
2328#ifdef INET6
2329 const struct ip6_hdr *ip6;
2330
2331 ip6 = (const struct ip6_hdr *)ip6hdr;
2332#endif /* INET6 */
2333 ip = (struct ip *)ip4hdr;
2334
2335 /*
2336 * The log line looks like this:
2337 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2338 */
2339 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2340 sizeof(PRINT_TH_FLAGS) + 1 +
2341#ifdef INET6
2342 2 * INET6_ADDRSTRLEN;
2343#else
2344 2 * INET_ADDRSTRLEN;
2345#endif /* INET6 */
2346
2347 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2348 if (s == NULL)
2349 return (NULL);
2350
2351 strcat(s, "TCP: [");
2352 sp = s + strlen(s);
2353
2354 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2355 inet_ntoa_r(inc->inc_faddr, sp);
2356 sp = s + strlen(s);
2357 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2358 sp = s + strlen(s);
2359 inet_ntoa_r(inc->inc_laddr, sp);
2360 sp = s + strlen(s);
2361 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2362#ifdef INET6
2363 } else if (inc) {
2364 ip6_sprintf(sp, &inc->inc6_faddr);
2365 sp = s + strlen(s);
2366 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2367 sp = s + strlen(s);
2368 ip6_sprintf(sp, &inc->inc6_laddr);
2369 sp = s + strlen(s);
2370 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2371 } else if (ip6 && th) {
2372 ip6_sprintf(sp, &ip6->ip6_src);
2373 sp = s + strlen(s);
2374 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2375 sp = s + strlen(s);
2376 ip6_sprintf(sp, &ip6->ip6_dst);
2377 sp = s + strlen(s);
2378 sprintf(sp, "]:%i", ntohs(th->th_dport));
2379#endif /* INET6 */
2380#ifdef INET
2381 } else if (ip && th) {
2382 inet_ntoa_r(ip->ip_src, sp);
2383 sp = s + strlen(s);
2384 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2385 sp = s + strlen(s);
2386 inet_ntoa_r(ip->ip_dst, sp);
2387 sp = s + strlen(s);
2388 sprintf(sp, "]:%i", ntohs(th->th_dport));
2389#endif /* INET */
2390 } else {
2391 free(s, M_TCPLOG);
2392 return (NULL);
2393 }
2394 sp = s + strlen(s);
2395 if (th)
2396 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2397 if (*(s + size - 1) != '\0')
2398 panic("%s: string too long", __func__);
2399 return (s);
2400}
2401
2402/*
2403 * A subroutine which makes it easy to track TCP state changes with DTrace.
2404 * This function shouldn't be called for t_state initializations that don't
2405 * correspond to actual TCP state transitions.
2406 */
2407void
2408tcp_state_change(struct tcpcb *tp, int newstate)
2409{
2410#if defined(KDTRACE_HOOKS)
2411 int pstate = tp->t_state;
2412#endif
2413
2414 tp->t_state = newstate;
2415 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2416}
| 1855 IN6_LINKMTU(sro6.ro_rt->rt_ifp));
1856
1857 /* Report additional interface capabilities. */
1858 if (cap != NULL) {
1859 if (ifp->if_capenable & IFCAP_TSO6 &&
1860 ifp->if_hwassist & CSUM_TSO)
1861 cap->ifcap |= CSUM_TSO;
1862 cap->tsomax = ifp->if_hw_tsomax;
1863 }
1864 RTFREE(sro6.ro_rt);
1865 }
1866
1867 return (maxmtu);
1868}
1869#endif /* INET6 */
1870
1871#ifdef IPSEC
1872/* compute ESP/AH header size for TCP, including outer IP header. */
1873size_t
1874ipsec_hdrsiz_tcp(struct tcpcb *tp)
1875{
1876 struct inpcb *inp;
1877 struct mbuf *m;
1878 size_t hdrsiz;
1879 struct ip *ip;
1880#ifdef INET6
1881 struct ip6_hdr *ip6;
1882#endif
1883 struct tcphdr *th;
1884
1885 if ((tp == NULL) || ((inp = tp->t_inpcb) == NULL))
1886 return (0);
1887 m = m_gethdr(M_NOWAIT, MT_DATA);
1888 if (!m)
1889 return (0);
1890
1891#ifdef INET6
1892 if ((inp->inp_vflag & INP_IPV6) != 0) {
1893 ip6 = mtod(m, struct ip6_hdr *);
1894 th = (struct tcphdr *)(ip6 + 1);
1895 m->m_pkthdr.len = m->m_len =
1896 sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
1897 tcpip_fillheaders(inp, ip6, th);
1898 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1899 } else
1900#endif /* INET6 */
1901 {
1902 ip = mtod(m, struct ip *);
1903 th = (struct tcphdr *)(ip + 1);
1904 m->m_pkthdr.len = m->m_len = sizeof(struct tcpiphdr);
1905 tcpip_fillheaders(inp, ip, th);
1906 hdrsiz = ipsec_hdrsiz(m, IPSEC_DIR_OUTBOUND, inp);
1907 }
1908
1909 m_free(m);
1910 return (hdrsiz);
1911}
1912#endif /* IPSEC */
1913
1914#ifdef TCP_SIGNATURE
1915/*
1916 * Callback function invoked by m_apply() to digest TCP segment data
1917 * contained within an mbuf chain.
1918 */
1919static int
1920tcp_signature_apply(void *fstate, void *data, u_int len)
1921{
1922
1923 MD5Update(fstate, (u_char *)data, len);
1924 return (0);
1925}
1926
1927/*
1928 * Compute TCP-MD5 hash of a TCP segment. (RFC2385)
1929 *
1930 * Parameters:
1931 * m pointer to head of mbuf chain
1932 * _unused
1933 * len length of TCP segment data, excluding options
1934 * optlen length of TCP segment options
1935 * buf pointer to storage for computed MD5 digest
1936 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
1937 *
1938 * We do this over ip, tcphdr, segment data, and the key in the SADB.
1939 * When called from tcp_input(), we can be sure that th_sum has been
1940 * zeroed out and verified already.
1941 *
1942 * Return 0 if successful, otherwise return -1.
1943 *
1944 * XXX The key is retrieved from the system's PF_KEY SADB, by keying a
1945 * search with the destination IP address, and a 'magic SPI' to be
1946 * determined by the application. This is hardcoded elsewhere to 1179
1947 * right now. Another branch of this code exists which uses the SPD to
1948 * specify per-application flows but it is unstable.
1949 */
1950int
1951tcp_signature_compute(struct mbuf *m, int _unused, int len, int optlen,
1952 u_char *buf, u_int direction)
1953{
1954 union sockaddr_union dst;
1955#ifdef INET
1956 struct ippseudo ippseudo;
1957#endif
1958 MD5_CTX ctx;
1959 int doff;
1960 struct ip *ip;
1961#ifdef INET
1962 struct ipovly *ipovly;
1963#endif
1964 struct secasvar *sav;
1965 struct tcphdr *th;
1966#ifdef INET6
1967 struct ip6_hdr *ip6;
1968 struct in6_addr in6;
1969 char ip6buf[INET6_ADDRSTRLEN];
1970 uint32_t plen;
1971 uint16_t nhdr;
1972#endif
1973 u_short savecsum;
1974
1975 KASSERT(m != NULL, ("NULL mbuf chain"));
1976 KASSERT(buf != NULL, ("NULL signature pointer"));
1977
1978 /* Extract the destination from the IP header in the mbuf. */
1979 bzero(&dst, sizeof(union sockaddr_union));
1980 ip = mtod(m, struct ip *);
1981#ifdef INET6
1982 ip6 = NULL; /* Make the compiler happy. */
1983#endif
1984 switch (ip->ip_v) {
1985#ifdef INET
1986 case IPVERSION:
1987 dst.sa.sa_len = sizeof(struct sockaddr_in);
1988 dst.sa.sa_family = AF_INET;
1989 dst.sin.sin_addr = (direction == IPSEC_DIR_INBOUND) ?
1990 ip->ip_src : ip->ip_dst;
1991 break;
1992#endif
1993#ifdef INET6
1994 case (IPV6_VERSION >> 4):
1995 ip6 = mtod(m, struct ip6_hdr *);
1996 dst.sa.sa_len = sizeof(struct sockaddr_in6);
1997 dst.sa.sa_family = AF_INET6;
1998 dst.sin6.sin6_addr = (direction == IPSEC_DIR_INBOUND) ?
1999 ip6->ip6_src : ip6->ip6_dst;
2000 break;
2001#endif
2002 default:
2003 return (EINVAL);
2004 /* NOTREACHED */
2005 break;
2006 }
2007
2008 /* Look up an SADB entry which matches the address of the peer. */
2009 sav = KEY_ALLOCSA(&dst, IPPROTO_TCP, htonl(TCP_SIG_SPI));
2010 if (sav == NULL) {
2011 ipseclog((LOG_ERR, "%s: SADB lookup failed for %s\n", __func__,
2012 (ip->ip_v == IPVERSION) ? inet_ntoa(dst.sin.sin_addr) :
2013#ifdef INET6
2014 (ip->ip_v == (IPV6_VERSION >> 4)) ?
2015 ip6_sprintf(ip6buf, &dst.sin6.sin6_addr) :
2016#endif
2017 "(unsupported)"));
2018 return (EINVAL);
2019 }
2020
2021 MD5Init(&ctx);
2022 /*
2023 * Step 1: Update MD5 hash with IP(v6) pseudo-header.
2024 *
2025 * XXX The ippseudo header MUST be digested in network byte order,
2026 * or else we'll fail the regression test. Assume all fields we've
2027 * been doing arithmetic on have been in host byte order.
2028 * XXX One cannot depend on ipovly->ih_len here. When called from
2029 * tcp_output(), the underlying ip_len member has not yet been set.
2030 */
2031 switch (ip->ip_v) {
2032#ifdef INET
2033 case IPVERSION:
2034 ipovly = (struct ipovly *)ip;
2035 ippseudo.ippseudo_src = ipovly->ih_src;
2036 ippseudo.ippseudo_dst = ipovly->ih_dst;
2037 ippseudo.ippseudo_pad = 0;
2038 ippseudo.ippseudo_p = IPPROTO_TCP;
2039 ippseudo.ippseudo_len = htons(len + sizeof(struct tcphdr) +
2040 optlen);
2041 MD5Update(&ctx, (char *)&ippseudo, sizeof(struct ippseudo));
2042
2043 th = (struct tcphdr *)((u_char *)ip + sizeof(struct ip));
2044 doff = sizeof(struct ip) + sizeof(struct tcphdr) + optlen;
2045 break;
2046#endif
2047#ifdef INET6
2048 /*
2049 * RFC 2385, 2.0 Proposal
2050 * For IPv6, the pseudo-header is as described in RFC 2460, namely the
2051 * 128-bit source IPv6 address, 128-bit destination IPv6 address, zero-
2052 * extended next header value (to form 32 bits), and 32-bit segment
2053 * length.
2054 * Note: Upper-Layer Packet Length comes before Next Header.
2055 */
2056 case (IPV6_VERSION >> 4):
2057 in6 = ip6->ip6_src;
2058 in6_clearscope(&in6);
2059 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2060 in6 = ip6->ip6_dst;
2061 in6_clearscope(&in6);
2062 MD5Update(&ctx, (char *)&in6, sizeof(struct in6_addr));
2063 plen = htonl(len + sizeof(struct tcphdr) + optlen);
2064 MD5Update(&ctx, (char *)&plen, sizeof(uint32_t));
2065 nhdr = 0;
2066 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2067 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2068 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2069 nhdr = IPPROTO_TCP;
2070 MD5Update(&ctx, (char *)&nhdr, sizeof(uint8_t));
2071
2072 th = (struct tcphdr *)((u_char *)ip6 + sizeof(struct ip6_hdr));
2073 doff = sizeof(struct ip6_hdr) + sizeof(struct tcphdr) + optlen;
2074 break;
2075#endif
2076 default:
2077 return (EINVAL);
2078 /* NOTREACHED */
2079 break;
2080 }
2081
2082
2083 /*
2084 * Step 2: Update MD5 hash with TCP header, excluding options.
2085 * The TCP checksum must be set to zero.
2086 */
2087 savecsum = th->th_sum;
2088 th->th_sum = 0;
2089 MD5Update(&ctx, (char *)th, sizeof(struct tcphdr));
2090 th->th_sum = savecsum;
2091
2092 /*
2093 * Step 3: Update MD5 hash with TCP segment data.
2094 * Use m_apply() to avoid an early m_pullup().
2095 */
2096 if (len > 0)
2097 m_apply(m, doff, len, tcp_signature_apply, &ctx);
2098
2099 /*
2100 * Step 4: Update MD5 hash with shared secret.
2101 */
2102 MD5Update(&ctx, sav->key_auth->key_data, _KEYLEN(sav->key_auth));
2103 MD5Final(buf, &ctx);
2104
2105 key_sa_recordxfer(sav, m);
2106 KEY_FREESAV(&sav);
2107 return (0);
2108}
2109
2110/*
2111 * Verify the TCP-MD5 hash of a TCP segment. (RFC2385)
2112 *
2113 * Parameters:
2114 * m pointer to head of mbuf chain
2115 * len length of TCP segment data, excluding options
2116 * optlen length of TCP segment options
2117 * buf pointer to storage for computed MD5 digest
2118 * direction direction of flow (IPSEC_DIR_INBOUND or OUTBOUND)
2119 *
2120 * Return 1 if successful, otherwise return 0.
2121 */
2122int
2123tcp_signature_verify(struct mbuf *m, int off0, int tlen, int optlen,
2124 struct tcpopt *to, struct tcphdr *th, u_int tcpbflag)
2125{
2126 char tmpdigest[TCP_SIGLEN];
2127
2128 if (tcp_sig_checksigs == 0)
2129 return (1);
2130 if ((tcpbflag & TF_SIGNATURE) == 0) {
2131 if ((to->to_flags & TOF_SIGNATURE) != 0) {
2132
2133 /*
2134 * If this socket is not expecting signature but
2135 * the segment contains signature just fail.
2136 */
2137 TCPSTAT_INC(tcps_sig_err_sigopt);
2138 TCPSTAT_INC(tcps_sig_rcvbadsig);
2139 return (0);
2140 }
2141
2142 /* Signature is not expected, and not present in segment. */
2143 return (1);
2144 }
2145
2146 /*
2147 * If this socket is expecting signature but the segment does not
2148 * contain any just fail.
2149 */
2150 if ((to->to_flags & TOF_SIGNATURE) == 0) {
2151 TCPSTAT_INC(tcps_sig_err_nosigopt);
2152 TCPSTAT_INC(tcps_sig_rcvbadsig);
2153 return (0);
2154 }
2155 if (tcp_signature_compute(m, off0, tlen, optlen, &tmpdigest[0],
2156 IPSEC_DIR_INBOUND) == -1) {
2157 TCPSTAT_INC(tcps_sig_err_buildsig);
2158 TCPSTAT_INC(tcps_sig_rcvbadsig);
2159 return (0);
2160 }
2161
2162 if (bcmp(to->to_signature, &tmpdigest[0], TCP_SIGLEN) != 0) {
2163 TCPSTAT_INC(tcps_sig_rcvbadsig);
2164 return (0);
2165 }
2166 TCPSTAT_INC(tcps_sig_rcvgoodsig);
2167 return (1);
2168}
2169#endif /* TCP_SIGNATURE */
2170
2171static int
2172sysctl_drop(SYSCTL_HANDLER_ARGS)
2173{
2174 /* addrs[0] is a foreign socket, addrs[1] is a local one. */
2175 struct sockaddr_storage addrs[2];
2176 struct inpcb *inp;
2177 struct tcpcb *tp;
2178 struct tcptw *tw;
2179 struct sockaddr_in *fin, *lin;
2180#ifdef INET6
2181 struct sockaddr_in6 *fin6, *lin6;
2182#endif
2183 int error;
2184
2185 inp = NULL;
2186 fin = lin = NULL;
2187#ifdef INET6
2188 fin6 = lin6 = NULL;
2189#endif
2190 error = 0;
2191
2192 if (req->oldptr != NULL || req->oldlen != 0)
2193 return (EINVAL);
2194 if (req->newptr == NULL)
2195 return (EPERM);
2196 if (req->newlen < sizeof(addrs))
2197 return (ENOMEM);
2198 error = SYSCTL_IN(req, &addrs, sizeof(addrs));
2199 if (error)
2200 return (error);
2201
2202 switch (addrs[0].ss_family) {
2203#ifdef INET6
2204 case AF_INET6:
2205 fin6 = (struct sockaddr_in6 *)&addrs[0];
2206 lin6 = (struct sockaddr_in6 *)&addrs[1];
2207 if (fin6->sin6_len != sizeof(struct sockaddr_in6) ||
2208 lin6->sin6_len != sizeof(struct sockaddr_in6))
2209 return (EINVAL);
2210 if (IN6_IS_ADDR_V4MAPPED(&fin6->sin6_addr)) {
2211 if (!IN6_IS_ADDR_V4MAPPED(&lin6->sin6_addr))
2212 return (EINVAL);
2213 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[0]);
2214 in6_sin6_2_sin_in_sock((struct sockaddr *)&addrs[1]);
2215 fin = (struct sockaddr_in *)&addrs[0];
2216 lin = (struct sockaddr_in *)&addrs[1];
2217 break;
2218 }
2219 error = sa6_embedscope(fin6, V_ip6_use_defzone);
2220 if (error)
2221 return (error);
2222 error = sa6_embedscope(lin6, V_ip6_use_defzone);
2223 if (error)
2224 return (error);
2225 break;
2226#endif
2227#ifdef INET
2228 case AF_INET:
2229 fin = (struct sockaddr_in *)&addrs[0];
2230 lin = (struct sockaddr_in *)&addrs[1];
2231 if (fin->sin_len != sizeof(struct sockaddr_in) ||
2232 lin->sin_len != sizeof(struct sockaddr_in))
2233 return (EINVAL);
2234 break;
2235#endif
2236 default:
2237 return (EINVAL);
2238 }
2239 INP_INFO_WLOCK(&V_tcbinfo);
2240 switch (addrs[0].ss_family) {
2241#ifdef INET6
2242 case AF_INET6:
2243 inp = in6_pcblookup(&V_tcbinfo, &fin6->sin6_addr,
2244 fin6->sin6_port, &lin6->sin6_addr, lin6->sin6_port,
2245 INPLOOKUP_WLOCKPCB, NULL);
2246 break;
2247#endif
2248#ifdef INET
2249 case AF_INET:
2250 inp = in_pcblookup(&V_tcbinfo, fin->sin_addr, fin->sin_port,
2251 lin->sin_addr, lin->sin_port, INPLOOKUP_WLOCKPCB, NULL);
2252 break;
2253#endif
2254 }
2255 if (inp != NULL) {
2256 if (inp->inp_flags & INP_TIMEWAIT) {
2257 /*
2258 * XXXRW: There currently exists a state where an
2259 * inpcb is present, but its timewait state has been
2260 * discarded. For now, don't allow dropping of this
2261 * type of inpcb.
2262 */
2263 tw = intotw(inp);
2264 if (tw != NULL)
2265 tcp_twclose(tw, 0);
2266 else
2267 INP_WUNLOCK(inp);
2268 } else if (!(inp->inp_flags & INP_DROPPED) &&
2269 !(inp->inp_socket->so_options & SO_ACCEPTCONN)) {
2270 tp = intotcpcb(inp);
2271 tp = tcp_drop(tp, ECONNABORTED);
2272 if (tp != NULL)
2273 INP_WUNLOCK(inp);
2274 } else
2275 INP_WUNLOCK(inp);
2276 } else
2277 error = ESRCH;
2278 INP_INFO_WUNLOCK(&V_tcbinfo);
2279 return (error);
2280}
2281
2282SYSCTL_VNET_PROC(_net_inet_tcp, TCPCTL_DROP, drop,
2283 CTLTYPE_STRUCT|CTLFLAG_WR|CTLFLAG_SKIP, NULL,
2284 0, sysctl_drop, "", "Drop TCP connection");
2285
2286/*
2287 * Generate a standardized TCP log line for use throughout the
2288 * tcp subsystem. Memory allocation is done with M_NOWAIT to
2289 * allow use in the interrupt context.
2290 *
2291 * NB: The caller MUST free(s, M_TCPLOG) the returned string.
2292 * NB: The function may return NULL if memory allocation failed.
2293 *
2294 * Due to header inclusion and ordering limitations the struct ip
2295 * and ip6_hdr pointers have to be passed as void pointers.
2296 */
2297char *
2298tcp_log_vain(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2299 const void *ip6hdr)
2300{
2301
2302 /* Is logging enabled? */
2303 if (tcp_log_in_vain == 0)
2304 return (NULL);
2305
2306 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2307}
2308
2309char *
2310tcp_log_addrs(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2311 const void *ip6hdr)
2312{
2313
2314 /* Is logging enabled? */
2315 if (tcp_log_debug == 0)
2316 return (NULL);
2317
2318 return (tcp_log_addr(inc, th, ip4hdr, ip6hdr));
2319}
2320
2321static char *
2322tcp_log_addr(struct in_conninfo *inc, struct tcphdr *th, void *ip4hdr,
2323 const void *ip6hdr)
2324{
2325 char *s, *sp;
2326 size_t size;
2327 struct ip *ip;
2328#ifdef INET6
2329 const struct ip6_hdr *ip6;
2330
2331 ip6 = (const struct ip6_hdr *)ip6hdr;
2332#endif /* INET6 */
2333 ip = (struct ip *)ip4hdr;
2334
2335 /*
2336 * The log line looks like this:
2337 * "TCP: [1.2.3.4]:50332 to [1.2.3.4]:80 tcpflags 0x2<SYN>"
2338 */
2339 size = sizeof("TCP: []:12345 to []:12345 tcpflags 0x2<>") +
2340 sizeof(PRINT_TH_FLAGS) + 1 +
2341#ifdef INET6
2342 2 * INET6_ADDRSTRLEN;
2343#else
2344 2 * INET_ADDRSTRLEN;
2345#endif /* INET6 */
2346
2347 s = malloc(size, M_TCPLOG, M_ZERO|M_NOWAIT);
2348 if (s == NULL)
2349 return (NULL);
2350
2351 strcat(s, "TCP: [");
2352 sp = s + strlen(s);
2353
2354 if (inc && ((inc->inc_flags & INC_ISIPV6) == 0)) {
2355 inet_ntoa_r(inc->inc_faddr, sp);
2356 sp = s + strlen(s);
2357 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2358 sp = s + strlen(s);
2359 inet_ntoa_r(inc->inc_laddr, sp);
2360 sp = s + strlen(s);
2361 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2362#ifdef INET6
2363 } else if (inc) {
2364 ip6_sprintf(sp, &inc->inc6_faddr);
2365 sp = s + strlen(s);
2366 sprintf(sp, "]:%i to [", ntohs(inc->inc_fport));
2367 sp = s + strlen(s);
2368 ip6_sprintf(sp, &inc->inc6_laddr);
2369 sp = s + strlen(s);
2370 sprintf(sp, "]:%i", ntohs(inc->inc_lport));
2371 } else if (ip6 && th) {
2372 ip6_sprintf(sp, &ip6->ip6_src);
2373 sp = s + strlen(s);
2374 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2375 sp = s + strlen(s);
2376 ip6_sprintf(sp, &ip6->ip6_dst);
2377 sp = s + strlen(s);
2378 sprintf(sp, "]:%i", ntohs(th->th_dport));
2379#endif /* INET6 */
2380#ifdef INET
2381 } else if (ip && th) {
2382 inet_ntoa_r(ip->ip_src, sp);
2383 sp = s + strlen(s);
2384 sprintf(sp, "]:%i to [", ntohs(th->th_sport));
2385 sp = s + strlen(s);
2386 inet_ntoa_r(ip->ip_dst, sp);
2387 sp = s + strlen(s);
2388 sprintf(sp, "]:%i", ntohs(th->th_dport));
2389#endif /* INET */
2390 } else {
2391 free(s, M_TCPLOG);
2392 return (NULL);
2393 }
2394 sp = s + strlen(s);
2395 if (th)
2396 sprintf(sp, " tcpflags 0x%b", th->th_flags, PRINT_TH_FLAGS);
2397 if (*(s + size - 1) != '\0')
2398 panic("%s: string too long", __func__);
2399 return (s);
2400}
2401
2402/*
2403 * A subroutine which makes it easy to track TCP state changes with DTrace.
2404 * This function shouldn't be called for t_state initializations that don't
2405 * correspond to actual TCP state transitions.
2406 */
2407void
2408tcp_state_change(struct tcpcb *tp, int newstate)
2409{
2410#if defined(KDTRACE_HOOKS)
2411 int pstate = tp->t_state;
2412#endif
2413
2414 tp->t_state = newstate;
2415 TCP_PROBE6(state__change, NULL, tp, NULL, tp, NULL, pstate);
2416}
|