1/*-
2 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
3 * The Regents of the University of California. All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 4. Neither the name of the University nor the names of its contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27 * SUCH DAMAGE.
28 *
29 * @(#)tcp_input.c 8.12 (Berkeley) 5/24/95
30 */
31
32#include <sys/cdefs.h>
33__FBSDID("$FreeBSD: head/sys/netinet/tcp_input.c 196019 2009-08-01 19:26:27Z rwatson $");
34
35#include "opt_ipfw.h" /* for ipfw_fwd */
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/kernel.h>
43#include <sys/malloc.h>
44#include <sys/mbuf.h>
45#include <sys/proc.h> /* for proc0 declaration */
46#include <sys/protosw.h>
47#include <sys/signalvar.h>
48#include <sys/socket.h>
49#include <sys/socketvar.h>
50#include <sys/sysctl.h>
51#include <sys/syslog.h>
52#include <sys/systm.h>
53
54#include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
55
56#include <vm/uma.h>
57
58#include <net/if.h>
59#include <net/route.h>
60#include <net/vnet.h>
61
62#define TCPSTATES /* for logging */
63
64#include <netinet/in.h>
65#include <netinet/in_pcb.h>
66#include <netinet/in_systm.h>
67#include <netinet/in_var.h>
68#include <netinet/ip.h>
69#include <netinet/ip_icmp.h> /* required for icmp_var.h */
70#include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
71#include <netinet/ip_var.h>
72#include <netinet/ip_options.h>
73#include <netinet/ip6.h>
74#include <netinet/icmp6.h>
75#include <netinet6/in6_pcb.h>
76#include <netinet6/ip6_var.h>
77#include <netinet6/nd6.h>
78#include <netinet/tcp.h>
79#include <netinet/tcp_fsm.h>
80#include <netinet/tcp_seq.h>
81#include <netinet/tcp_timer.h>
82#include <netinet/tcp_var.h>
83#include <netinet6/tcp6_var.h>
84#include <netinet/tcpip.h>
85#include <netinet/tcp_syncache.h>
86#ifdef TCPDEBUG
87#include <netinet/tcp_debug.h>
88#endif /* TCPDEBUG */
89
90#ifdef IPSEC
91#include <netipsec/ipsec.h>
92#include <netipsec/ipsec6.h>
93#endif /*IPSEC*/
94
95#include <machine/in_cksum.h>
96
97#include <security/mac/mac_framework.h>
98
99static const int tcprexmtthresh = 3;
100
101VNET_DEFINE(struct tcpstat, tcpstat);
102VNET_DEFINE(int, blackhole);
103VNET_DEFINE(int, tcp_delack_enabled);
104VNET_DEFINE(int, drop_synfin);
105VNET_DEFINE(int, tcp_do_rfc3042);
106VNET_DEFINE(int, tcp_do_rfc3390);
107VNET_DEFINE(int, tcp_do_ecn);
108VNET_DEFINE(int, tcp_ecn_maxretries);
109VNET_DEFINE(int, tcp_insecure_rst);
110VNET_DEFINE(int, tcp_do_autorcvbuf);
111VNET_DEFINE(int, tcp_autorcvbuf_inc);
112VNET_DEFINE(int, tcp_autorcvbuf_max);
113VNET_DEFINE(int, tcp_do_rfc3465);
114VNET_DEFINE(int, tcp_abc_l_var);
115
116SYSCTL_VNET_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
117 &VNET_NAME(tcpstat), tcpstat,
118 "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
119
120int tcp_log_in_vain = 0;
121SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
122 &tcp_log_in_vain, 0,
123 "Log all incoming TCP segments to closed ports");
124
125SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
126 &VNET_NAME(blackhole), 0,
127 "Do not send RST on segments to closed ports");
128
129SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
130 &VNET_NAME(tcp_delack_enabled), 0,
131 "Delay ACK to try and piggyback it onto a data packet");
132
133SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
134 &VNET_NAME(drop_synfin), 0,
135 "Drop TCP packets with SYN+FIN set");
136
137SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
138 &VNET_NAME(tcp_do_rfc3042), 0,
139 "Enable RFC 3042 (Limited Transmit)");
140
141SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
142 &VNET_NAME(tcp_do_rfc3390), 0,
143 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
144
145SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_RW,
146 &VNET_NAME(tcp_do_rfc3465), 0,
147 "Enable RFC 3465 (Appropriate Byte Counting)");
148
149SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_RW,
150 &VNET_NAME(tcp_abc_l_var), 2,
151 "Cap the max cwnd increment during slow-start to this number of segments");
152
153SYSCTL_NODE(_net_inet_tcp, OID_AUTO, ecn, CTLFLAG_RW, 0, "TCP ECN");
154
155SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, enable, CTLFLAG_RW,
156 &VNET_NAME(tcp_do_ecn), 0,
157 "TCP ECN support");
158
159SYSCTL_VNET_INT(_net_inet_tcp_ecn, OID_AUTO, maxretries, CTLFLAG_RW,
160 &VNET_NAME(tcp_ecn_maxretries), 0,
161 "Max retries before giving up on ECN");
162
163SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
164 &VNET_NAME(tcp_insecure_rst), 0,
165 "Follow the old (insecure) criteria for accepting RST packets");
166
167SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
168 &VNET_NAME(tcp_do_autorcvbuf), 0,
169 "Enable automatic receive buffer sizing");
170
171SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
172 &VNET_NAME(tcp_autorcvbuf_inc), 0,
173 "Incrementor step size of automatic receive buffer");
174
175SYSCTL_VNET_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
176 &VNET_NAME(tcp_autorcvbuf_max), 0,
177 "Max size of automatic receive buffer");
178
179int tcp_read_locking = 1;
180SYSCTL_INT(_net_inet_tcp, OID_AUTO, read_locking, CTLFLAG_RW,
181 &tcp_read_locking, 0, "Enable read locking strategy");
182
183int tcp_rlock_atfirst;
184SYSCTL_INT(_net_inet_tcp, OID_AUTO, rlock_atfirst, CTLFLAG_RD,
185 &tcp_rlock_atfirst, 0, "");
186
187int tcp_wlock_atfirst;
188SYSCTL_INT(_net_inet_tcp, OID_AUTO, tcp_wlock_atfirst, CTLFLAG_RD,
189 &tcp_wlock_atfirst, 0, "");
190
191int tcp_wlock_upgraded;
192SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_upgraded, CTLFLAG_RD,
193 &tcp_wlock_upgraded, 0, "");
194
195int tcp_wlock_relocked;
196SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_relocked, CTLFLAG_RD,
197 &tcp_wlock_relocked, 0, "");
198
199int tcp_wlock_looped;
200SYSCTL_INT(_net_inet_tcp, OID_AUTO, wlock_looped, CTLFLAG_RD,
201 &tcp_wlock_looped, 0, "");
202
203VNET_DEFINE(struct inpcbhead, tcb);
204VNET_DEFINE(struct inpcbinfo, tcbinfo);
205#define tcb6 tcb /* for KAME src sync over BSD*'s */
206
207static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
208static void tcp_do_segment(struct mbuf *, struct tcphdr *,
209 struct socket *, struct tcpcb *, int, int, uint8_t,
210 int);
211static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
212 struct tcpcb *, int, int);
213static void tcp_pulloutofband(struct socket *,
214 struct tcphdr *, struct mbuf *, int);
215static void tcp_xmit_timer(struct tcpcb *, int);
216static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
217static void inline
218 tcp_congestion_exp(struct tcpcb *);
219
220static void inline
221tcp_congestion_exp(struct tcpcb *tp)
222{
223 u_int win;
224
225 win = min(tp->snd_wnd, tp->snd_cwnd) /
226 2 / tp->t_maxseg;
227 if (win < 2)
228 win = 2;
229 tp->snd_ssthresh = win * tp->t_maxseg;
230 ENTER_FASTRECOVERY(tp);
231 tp->snd_recover = tp->snd_max;
232 if (tp->t_flags & TF_ECN_PERMIT)
233 tp->t_flags |= TF_ECN_SND_CWR;
234}
235
236/* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
237#ifdef INET6
238#define ND6_HINT(tp) \
239do { \
240 if ((tp) && (tp)->t_inpcb && \
241 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
242 nd6_nud_hint(NULL, NULL, 0); \
243} while (0)
244#else
245#define ND6_HINT(tp)
246#endif
247
248/*
249 * Indicate whether this ack should be delayed. We can delay the ack if
250 * - there is no delayed ack timer in progress and
251 * - our last ack wasn't a 0-sized window. We never want to delay
252 * the ack that opens up a 0-sized window and
253 * - delayed acks are enabled or
254 * - this is a half-synchronized T/TCP connection.
255 */
256#define DELAY_ACK(tp) \
257 ((!tcp_timer_active(tp, TT_DELACK) && \
258 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
259 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
260
261/*
262 * TCP input handling is split into multiple parts:
263 * tcp6_input is a thin wrapper around tcp_input for the extended
264 * ip6_protox[] call format in ip6_input
265 * tcp_input handles primary segment validation, inpcb lookup and
266 * SYN processing on listen sockets
267 * tcp_do_segment processes the ACK and text of the segment for
268 * establishing, established and closing connections
269 */
270#ifdef INET6
271int
272tcp6_input(struct mbuf **mp, int *offp, int proto)
273{
274 struct mbuf *m = *mp;
275 struct in6_ifaddr *ia6;
276
277 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
278
279 /*
280 * draft-itojun-ipv6-tcp-to-anycast
281 * better place to put this in?
282 */
283 ia6 = ip6_getdstifaddr(m);
284 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
285 struct ip6_hdr *ip6;
286
287 ifa_free(&ia6->ia_ifa);
288 ip6 = mtod(m, struct ip6_hdr *);
289 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
290 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
291 return IPPROTO_DONE;
292 }
293
294 tcp_input(m, *offp);
295 return IPPROTO_DONE;
296}
297#endif
298
299void
300tcp_input(struct mbuf *m, int off0)
301{
302 struct tcphdr *th;
303 struct ip *ip = NULL;
304 struct ipovly *ipov;
305 struct inpcb *inp = NULL;
306 struct tcpcb *tp = NULL;
307 struct socket *so = NULL;
308 u_char *optp = NULL;
309 int optlen = 0;
310 int len, tlen, off;
311 int drop_hdrlen;
312 int thflags;
313 int rstreason = 0; /* For badport_bandlim accounting purposes */
314 uint8_t iptos;
315#ifdef IPFIREWALL_FORWARD
316 struct m_tag *fwd_tag;
317#endif
318#ifdef INET6
319 struct ip6_hdr *ip6 = NULL;
320 int isipv6;
321#else
322 const void *ip6 = NULL;
323 const int isipv6 = 0;
324#endif
325 struct tcpopt to; /* options in this segment */
326 char *s = NULL; /* address and port logging */
327 int ti_locked;
328#define TI_UNLOCKED 1
329#define TI_RLOCKED 2
330#define TI_WLOCKED 3
331
332#ifdef TCPDEBUG
333 /*
334 * The size of tcp_saveipgen must be the size of the max ip header,
335 * now IPv6.
336 */
337 u_char tcp_saveipgen[IP6_HDR_LEN];
338 struct tcphdr tcp_savetcp;
339 short ostate = 0;
340#endif
341
342#ifdef INET6
343 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
344#endif
345
346 to.to_flags = 0;
347 TCPSTAT_INC(tcps_rcvtotal);
348
349 if (isipv6) {
350#ifdef INET6
351 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
352 ip6 = mtod(m, struct ip6_hdr *);
353 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
354 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
355 TCPSTAT_INC(tcps_rcvbadsum);
356 goto drop;
357 }
358 th = (struct tcphdr *)((caddr_t)ip6 + off0);
359
360 /*
361 * Be proactive about unspecified IPv6 address in source.
362 * As we use all-zero to indicate unbounded/unconnected pcb,
363 * unspecified IPv6 address can be used to confuse us.
364 *
365 * Note that packets with unspecified IPv6 destination is
366 * already dropped in ip6_input.
367 */
368 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
369 /* XXX stat */
370 goto drop;
371 }
372#else
373 th = NULL; /* XXX: Avoid compiler warning. */
374#endif
375 } else {
376 /*
377 * Get IP and TCP header together in first mbuf.
378 * Note: IP leaves IP header in first mbuf.
379 */
380 if (off0 > sizeof (struct ip)) {
381 ip_stripoptions(m, (struct mbuf *)0);
382 off0 = sizeof(struct ip);
383 }
384 if (m->m_len < sizeof (struct tcpiphdr)) {
385 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
386 == NULL) {
387 TCPSTAT_INC(tcps_rcvshort);
388 return;
389 }
390 }
391 ip = mtod(m, struct ip *);
392 ipov = (struct ipovly *)ip;
393 th = (struct tcphdr *)((caddr_t)ip + off0);
394 tlen = ip->ip_len;
395
396 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
397 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
398 th->th_sum = m->m_pkthdr.csum_data;
399 else
400 th->th_sum = in_pseudo(ip->ip_src.s_addr,
401 ip->ip_dst.s_addr,
402 htonl(m->m_pkthdr.csum_data +
403 ip->ip_len +
404 IPPROTO_TCP));
405 th->th_sum ^= 0xffff;
406#ifdef TCPDEBUG
407 ipov->ih_len = (u_short)tlen;
408 ipov->ih_len = htons(ipov->ih_len);
409#endif
410 } else {
411 /*
412 * Checksum extended TCP header and data.
413 */
414 len = sizeof (struct ip) + tlen;
415 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
416 ipov->ih_len = (u_short)tlen;
417 ipov->ih_len = htons(ipov->ih_len);
418 th->th_sum = in_cksum(m, len);
419 }
420 if (th->th_sum) {
421 TCPSTAT_INC(tcps_rcvbadsum);
422 goto drop;
423 }
424 /* Re-initialization for later version check */
425 ip->ip_v = IPVERSION;
426 }
427
428#ifdef INET6
429 if (isipv6)
430 iptos = (ntohl(ip6->ip6_flow) >> 20) & 0xff;
431 else
432#endif
433 iptos = ip->ip_tos;
434
435 /*
436 * Check that TCP offset makes sense,
437 * pull out TCP options and adjust length. XXX
438 */
439 off = th->th_off << 2;
440 if (off < sizeof (struct tcphdr) || off > tlen) {
441 TCPSTAT_INC(tcps_rcvbadoff);
442 goto drop;
443 }
444 tlen -= off; /* tlen is used instead of ti->ti_len */
445 if (off > sizeof (struct tcphdr)) {
446 if (isipv6) {
447#ifdef INET6
448 IP6_EXTHDR_CHECK(m, off0, off, );
449 ip6 = mtod(m, struct ip6_hdr *);
450 th = (struct tcphdr *)((caddr_t)ip6 + off0);
451#endif
452 } else {
453 if (m->m_len < sizeof(struct ip) + off) {
454 if ((m = m_pullup(m, sizeof (struct ip) + off))
455 == NULL) {
456 TCPSTAT_INC(tcps_rcvshort);
457 return;
458 }
459 ip = mtod(m, struct ip *);
460 ipov = (struct ipovly *)ip;
461 th = (struct tcphdr *)((caddr_t)ip + off0);
462 }
463 }
464 optlen = off - sizeof (struct tcphdr);
465 optp = (u_char *)(th + 1);
466 }
467 thflags = th->th_flags;
468
469 /*
470 * Convert TCP protocol specific fields to host format.
471 */
472 th->th_seq = ntohl(th->th_seq);
473 th->th_ack = ntohl(th->th_ack);
474 th->th_win = ntohs(th->th_win);
475 th->th_urp = ntohs(th->th_urp);
476
477 /*
478 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
479 */
480 drop_hdrlen = off0 + off;
481
482 /*
483 * Locate pcb for segment, which requires a lock on tcbinfo.
484 * Optimisticaly acquire a global read lock rather than a write lock
485 * unless header flags necessarily imply a state change. There are
486 * two cases where we might discover later we need a write lock
487 * despite the flags: ACKs moving a connection out of the syncache,
488 * and ACKs for a connection in TIMEWAIT.
489 */
490 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
491 tcp_read_locking == 0) {
492 INP_INFO_WLOCK(&V_tcbinfo);
493 ti_locked = TI_WLOCKED;
494 tcp_wlock_atfirst++;
495 } else {
496 INP_INFO_RLOCK(&V_tcbinfo);
497 ti_locked = TI_RLOCKED;
498 tcp_rlock_atfirst++;
499 }
500
501findpcb:
502#ifdef INVARIANTS
503 if (ti_locked == TI_RLOCKED)
504 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
505 else if (ti_locked == TI_WLOCKED)
506 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
507 else
508 panic("%s: findpcb ti_locked %d\n", __func__, ti_locked);
509#endif
510
511#ifdef IPFIREWALL_FORWARD
512 /*
513 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
514 */
515 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
516
517 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
518 struct sockaddr_in *next_hop;
519
520 next_hop = (struct sockaddr_in *)(fwd_tag+1);
521 /*
522 * Transparently forwarded. Pretend to be the destination.
523 * already got one like this?
524 */
525 inp = in_pcblookup_hash(&V_tcbinfo,
526 ip->ip_src, th->th_sport,
527 ip->ip_dst, th->th_dport,
528 0, m->m_pkthdr.rcvif);
529 if (!inp) {
530 /* It's new. Try to find the ambushing socket. */
531 inp = in_pcblookup_hash(&V_tcbinfo,
532 ip->ip_src, th->th_sport,
533 next_hop->sin_addr,
534 next_hop->sin_port ?
535 ntohs(next_hop->sin_port) :
536 th->th_dport,
537 INPLOOKUP_WILDCARD,
538 m->m_pkthdr.rcvif);
539 }
540 /* Remove the tag from the packet. We don't need it anymore. */
541 m_tag_delete(m, fwd_tag);
542 } else
543#endif /* IPFIREWALL_FORWARD */
544 {
545 if (isipv6) {
546#ifdef INET6
547 inp = in6_pcblookup_hash(&V_tcbinfo,
548 &ip6->ip6_src, th->th_sport,
549 &ip6->ip6_dst, th->th_dport,
550 INPLOOKUP_WILDCARD,
551 m->m_pkthdr.rcvif);
552#endif
553 } else
554 inp = in_pcblookup_hash(&V_tcbinfo,
555 ip->ip_src, th->th_sport,
556 ip->ip_dst, th->th_dport,
557 INPLOOKUP_WILDCARD,
558 m->m_pkthdr.rcvif);
559 }
560
561 /*
562 * If the INPCB does not exist then all data in the incoming
563 * segment is discarded and an appropriate RST is sent back.
564 * XXX MRT Send RST using which routing table?
565 */
566 if (inp == NULL) {
567 /*
568 * Log communication attempts to ports that are not
569 * in use.
570 */
571 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
572 tcp_log_in_vain == 2) {
573 if ((s = tcp_log_addrs(NULL, th, (void *)ip, ip6)))
574 log(LOG_INFO, "%s; %s: Connection attempt "
575 "to closed port\n", s, __func__);
576 }
577 /*
578 * When blackholing do not respond with a RST but
579 * completely ignore the segment and drop it.
580 */
581 if ((V_blackhole == 1 && (thflags & TH_SYN)) ||
582 V_blackhole == 2)
583 goto dropunlock;
584
585 rstreason = BANDLIM_RST_CLOSEDPORT;
586 goto dropwithreset;
587 }
588 INP_WLOCK(inp);
589 if (!(inp->inp_flags & INP_HW_FLOWID)
590 && (m->m_flags & M_FLOWID)
591 && ((inp->inp_socket == NULL)
592 || !(inp->inp_socket->so_options & SO_ACCEPTCONN))) {
593 inp->inp_flags |= INP_HW_FLOWID;
594 inp->inp_flags &= ~INP_SW_FLOWID;
595 inp->inp_flowid = m->m_pkthdr.flowid;
596 }
597#ifdef IPSEC
598#ifdef INET6
599 if (isipv6 && ipsec6_in_reject(m, inp)) {
600 V_ipsec6stat.in_polvio++;
601 goto dropunlock;
602 } else
603#endif /* INET6 */
604 if (ipsec4_in_reject(m, inp) != 0) {
605 V_ipsec4stat.in_polvio++;
606 goto dropunlock;
607 }
608#endif /* IPSEC */
609
610 /*
611 * Check the minimum TTL for socket.
612 */
613 if (inp->inp_ip_minttl != 0) {
614#ifdef INET6
615 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
616 goto dropunlock;
617 else
618#endif
619 if (inp->inp_ip_minttl > ip->ip_ttl)
620 goto dropunlock;
621 }
622
623 /*
624 * A previous connection in TIMEWAIT state is supposed to catch stray
625 * or duplicate segments arriving late. If this segment was a
626 * legitimate new connection attempt the old INPCB gets removed and
627 * we can try again to find a listening socket.
628 *
629 * At this point, due to earlier optimism, we may hold a read lock on
630 * the inpcbinfo, rather than a write lock. If so, we need to
631 * upgrade, or if that fails, acquire a reference on the inpcb, drop
632 * all locks, acquire a global write lock, and then re-acquire the
633 * inpcb lock. We may at that point discover that another thread has
634 * tried to free the inpcb, in which case we need to loop back and
635 * try to find a new inpcb to deliver to.
636 */
637 if (inp->inp_flags & INP_TIMEWAIT) {
638 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
639 ("%s: INP_TIMEWAIT ti_locked %d", __func__, ti_locked));
640
641 if (ti_locked == TI_RLOCKED) {
642 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) {
643 in_pcbref(inp);
644 INP_WUNLOCK(inp);
645 INP_INFO_RUNLOCK(&V_tcbinfo);
646 INP_INFO_WLOCK(&V_tcbinfo);
647 ti_locked = TI_WLOCKED;
648 INP_WLOCK(inp);
649 if (in_pcbrele(inp)) {
650 tcp_wlock_looped++;
651 inp = NULL;
652 goto findpcb;
653 }
654 tcp_wlock_relocked++;
655 } else {
656 ti_locked = TI_WLOCKED;
657 tcp_wlock_upgraded++;
658 }
659 }
660 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
661
662 if (thflags & TH_SYN)
663 tcp_dooptions(&to, optp, optlen, TO_SYN);
664 /*
665 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
666 */
667 if (tcp_twcheck(inp, &to, th, m, tlen))
668 goto findpcb;
669 INP_INFO_WUNLOCK(&V_tcbinfo);
670 return;
671 }
672 /*
673 * The TCPCB may no longer exist if the connection is winding
674 * down or it is in the CLOSED state. Either way we drop the
675 * segment and send an appropriate response.
676 */
677 tp = intotcpcb(inp);
678 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
679 rstreason = BANDLIM_RST_CLOSEDPORT;
680 goto dropwithreset;
681 }
682
683 /*
684 * We've identified a valid inpcb, but it could be that we need an
685 * inpcbinfo write lock and have only a read lock. In this case,
686 * attempt to upgrade/relock using the same strategy as the TIMEWAIT
687 * case above.
688 */
689 if (tp->t_state != TCPS_ESTABLISHED ||
690 (thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
691 tcp_read_locking == 0) {
692 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
693 ("%s: upgrade check ti_locked %d", __func__, ti_locked));
694
695 if (ti_locked == TI_RLOCKED) {
696 if (rw_try_upgrade(&V_tcbinfo.ipi_lock) == 0) {
697 in_pcbref(inp);
698 INP_WUNLOCK(inp);
699 INP_INFO_RUNLOCK(&V_tcbinfo);
700 INP_INFO_WLOCK(&V_tcbinfo);
701 ti_locked = TI_WLOCKED;
702 INP_WLOCK(inp);
703 if (in_pcbrele(inp)) {
704 tcp_wlock_looped++;
705 inp = NULL;
706 goto findpcb;
707 }
708 tcp_wlock_relocked++;
709 } else {
710 ti_locked = TI_WLOCKED;
711 tcp_wlock_upgraded++;
712 }
713 }
714 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
715 }
716
717#ifdef MAC
718 INP_WLOCK_ASSERT(inp);
719 if (mac_inpcb_check_deliver(inp, m))
720 goto dropunlock;
721#endif
722 so = inp->inp_socket;
723 KASSERT(so != NULL, ("%s: so == NULL", __func__));
724#ifdef TCPDEBUG
725 if (so->so_options & SO_DEBUG) {
726 ostate = tp->t_state;
727 if (isipv6) {
728#ifdef INET6
729 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
730#endif
731 } else
732 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
733 tcp_savetcp = *th;
734 }
735#endif
736 /*
737 * When the socket is accepting connections (the INPCB is in LISTEN
738 * state) we look into the SYN cache if this is a new connection
739 * attempt or the completion of a previous one.
740 */
741 if (so->so_options & SO_ACCEPTCONN) {
742 struct in_conninfo inc;
743
744 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
745 "tp not listening", __func__));
746
747 bzero(&inc, sizeof(inc));
748#ifdef INET6
749 if (isipv6) {
750 inc.inc_flags |= INC_ISIPV6;
751 inc.inc6_faddr = ip6->ip6_src;
752 inc.inc6_laddr = ip6->ip6_dst;
753 } else
754#endif
755 {
756 inc.inc_faddr = ip->ip_src;
757 inc.inc_laddr = ip->ip_dst;
758 }
759 inc.inc_fport = th->th_sport;
760 inc.inc_lport = th->th_dport;
761 inc.inc_fibnum = so->so_fibnum;
762
763 /*
764 * Check for an existing connection attempt in syncache if
765 * the flag is only ACK. A successful lookup creates a new
766 * socket appended to the listen queue in SYN_RECEIVED state.
767 */
768 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
769 /*
770 * Parse the TCP options here because
771 * syncookies need access to the reflected
772 * timestamp.
773 */
774 tcp_dooptions(&to, optp, optlen, 0);
775 /*
776 * NB: syncache_expand() doesn't unlock
777 * inp and tcpinfo locks.
778 */
779 if (!syncache_expand(&inc, &to, th, &so, m)) {
780 /*
781 * No syncache entry or ACK was not
782 * for our SYN/ACK. Send a RST.
783 * NB: syncache did its own logging
784 * of the failure cause.
785 */
786 rstreason = BANDLIM_RST_OPENPORT;
787 goto dropwithreset;
788 }
789 if (so == NULL) {
790 /*
791 * We completed the 3-way handshake
792 * but could not allocate a socket
793 * either due to memory shortage,
794 * listen queue length limits or
795 * global socket limits. Send RST
796 * or wait and have the remote end
797 * retransmit the ACK for another
798 * try.
799 */
800 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
801 log(LOG_DEBUG, "%s; %s: Listen socket: "
802 "Socket allocation failed due to "
803 "limits or memory shortage, %s\n",
804 s, __func__,
805 V_tcp_sc_rst_sock_fail ?
806 "sending RST" : "try again");
807 if (V_tcp_sc_rst_sock_fail) {
808 rstreason = BANDLIM_UNLIMITED;
809 goto dropwithreset;
810 } else
811 goto dropunlock;
812 }
813 /*
814 * Socket is created in state SYN_RECEIVED.
815 * Unlock the listen socket, lock the newly
816 * created socket and update the tp variable.
817 */
818 INP_WUNLOCK(inp); /* listen socket */
819 inp = sotoinpcb(so);
820 INP_WLOCK(inp); /* new connection */
821 tp = intotcpcb(inp);
822 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
823 ("%s: ", __func__));
824 /*
825 * Process the segment and the data it
826 * contains. tcp_do_segment() consumes
827 * the mbuf chain and unlocks the inpcb.
828 */
829 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen,
830 iptos, ti_locked);
831 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
832 return;
833 }
834 /*
835 * Segment flag validation for new connection attempts:
836 *
837 * Our (SYN|ACK) response was rejected.
838 * Check with syncache and remove entry to prevent
839 * retransmits.
840 *
841 * NB: syncache_chkrst does its own logging of failure
842 * causes.
843 */
844 if (thflags & TH_RST) {
845 syncache_chkrst(&inc, th);
846 goto dropunlock;
847 }
848 /*
849 * We can't do anything without SYN.
850 */
851 if ((thflags & TH_SYN) == 0) {
852 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
853 log(LOG_DEBUG, "%s; %s: Listen socket: "
854 "SYN is missing, segment ignored\n",
855 s, __func__);
856 TCPSTAT_INC(tcps_badsyn);
857 goto dropunlock;
858 }
859 /*
860 * (SYN|ACK) is bogus on a listen socket.
861 */
862 if (thflags & TH_ACK) {
863 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
864 log(LOG_DEBUG, "%s; %s: Listen socket: "
865 "SYN|ACK invalid, segment rejected\n",
866 s, __func__);
867 syncache_badack(&inc); /* XXX: Not needed! */
868 TCPSTAT_INC(tcps_badsyn);
869 rstreason = BANDLIM_RST_OPENPORT;
870 goto dropwithreset;
871 }
872 /*
873 * If the drop_synfin option is enabled, drop all
874 * segments with both the SYN and FIN bits set.
875 * This prevents e.g. nmap from identifying the
876 * TCP/IP stack.
877 * XXX: Poor reasoning. nmap has other methods
878 * and is constantly refining its stack detection
879 * strategies.
880 * XXX: This is a violation of the TCP specification
881 * and was used by RFC1644.
882 */
883 if ((thflags & TH_FIN) && V_drop_synfin) {
884 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
885 log(LOG_DEBUG, "%s; %s: Listen socket: "
886 "SYN|FIN segment ignored (based on "
887 "sysctl setting)\n", s, __func__);
888 TCPSTAT_INC(tcps_badsyn);
889 goto dropunlock;
890 }
891 /*
892 * Segment's flags are (SYN) or (SYN|FIN).
893 *
894 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
895 * as they do not affect the state of the TCP FSM.
896 * The data pointed to by TH_URG and th_urp is ignored.
897 */
898 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
899 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
900 KASSERT(thflags & (TH_SYN),
901 ("%s: Listen socket: TH_SYN not set", __func__));
902#ifdef INET6
903 /*
904 * If deprecated address is forbidden,
905 * we do not accept SYN to deprecated interface
906 * address to prevent any new inbound connection from
907 * getting established.
908 * When we do not accept SYN, we send a TCP RST,
909 * with deprecated source address (instead of dropping
910 * it). We compromise it as it is much better for peer
911 * to send a RST, and RST will be the final packet
912 * for the exchange.
913 *
914 * If we do not forbid deprecated addresses, we accept
915 * the SYN packet. RFC2462 does not suggest dropping
916 * SYN in this case.
917 * If we decipher RFC2462 5.5.4, it says like this:
918 * 1. use of deprecated addr with existing
919 * communication is okay - "SHOULD continue to be
920 * used"
921 * 2. use of it with new communication:
922 * (2a) "SHOULD NOT be used if alternate address
923 * with sufficient scope is available"
924 * (2b) nothing mentioned otherwise.
925 * Here we fall into (2b) case as we have no choice in
926 * our source address selection - we must obey the peer.
927 *
928 * The wording in RFC2462 is confusing, and there are
929 * multiple description text for deprecated address
930 * handling - worse, they are not exactly the same.
931 * I believe 5.5.4 is the best one, so we follow 5.5.4.
932 */
933 if (isipv6 && !V_ip6_use_deprecated) {
934 struct in6_ifaddr *ia6;
935
936 ia6 = ip6_getdstifaddr(m);
937 if (ia6 != NULL &&
938 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
939 ifa_free(&ia6->ia_ifa);
940 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
941 log(LOG_DEBUG, "%s; %s: Listen socket: "
942 "Connection attempt to deprecated "
943 "IPv6 address rejected\n",
944 s, __func__);
945 rstreason = BANDLIM_RST_OPENPORT;
946 goto dropwithreset;
947 }
948 ifa_free(&ia6->ia_ifa);
949 }
950#endif
951 /*
952 * Basic sanity checks on incoming SYN requests:
953 * Don't respond if the destination is a link layer
954 * broadcast according to RFC1122 4.2.3.10, p. 104.
955 * If it is from this socket it must be forged.
956 * Don't respond if the source or destination is a
957 * global or subnet broad- or multicast address.
958 * Note that it is quite possible to receive unicast
959 * link-layer packets with a broadcast IP address. Use
960 * in_broadcast() to find them.
961 */
962 if (m->m_flags & (M_BCAST|M_MCAST)) {
963 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
964 log(LOG_DEBUG, "%s; %s: Listen socket: "
965 "Connection attempt from broad- or multicast "
966 "link layer address ignored\n", s, __func__);
967 goto dropunlock;
968 }
969 if (isipv6) {
970#ifdef INET6
971 if (th->th_dport == th->th_sport &&
972 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
973 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
974 log(LOG_DEBUG, "%s; %s: Listen socket: "
975 "Connection attempt to/from self "
976 "ignored\n", s, __func__);
977 goto dropunlock;
978 }
979 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
980 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
981 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
982 log(LOG_DEBUG, "%s; %s: Listen socket: "
983 "Connection attempt from/to multicast "
984 "address ignored\n", s, __func__);
985 goto dropunlock;
986 }
987#endif
988 } else {
989 if (th->th_dport == th->th_sport &&
990 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
991 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
992 log(LOG_DEBUG, "%s; %s: Listen socket: "
993 "Connection attempt from/to self "
994 "ignored\n", s, __func__);
995 goto dropunlock;
996 }
997 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
998 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
999 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1000 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1001 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1002 log(LOG_DEBUG, "%s; %s: Listen socket: "
1003 "Connection attempt from/to broad- "
1004 "or multicast address ignored\n",
1005 s, __func__);
1006 goto dropunlock;
1007 }
1008 }
1009 /*
1010 * SYN appears to be valid. Create compressed TCP state
1011 * for syncache.
1012 */
1013#ifdef TCPDEBUG
1014 if (so->so_options & SO_DEBUG)
1015 tcp_trace(TA_INPUT, ostate, tp,
1016 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1017#endif
1018 tcp_dooptions(&to, optp, optlen, TO_SYN);
1019 syncache_add(&inc, &to, th, inp, &so, m);
1020 /*
1021 * Entry added to syncache and mbuf consumed.
1022 * Everything already unlocked by syncache_add().
1023 */
1024 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1025 return;
1026 }
1027
1028 /*
1029 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1030 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1031 * the inpcb, and unlocks pcbinfo.
1032 */
1033 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen, iptos, ti_locked);
1034 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1035 return;
1036
1037dropwithreset:
1038 if (ti_locked == TI_RLOCKED)
1039 INP_INFO_RUNLOCK(&V_tcbinfo);
1040 else if (ti_locked == TI_WLOCKED)
1041 INP_INFO_WUNLOCK(&V_tcbinfo);
1042 else
1043 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
1044 ti_locked = TI_UNLOCKED;
1045
1046 if (inp != NULL) {
1047 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1048 INP_WUNLOCK(inp);
1049 } else
1050 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
1051 m = NULL; /* mbuf chain got consumed. */
1052 goto drop;
1053
1054dropunlock:
1055 if (ti_locked == TI_RLOCKED)
1056 INP_INFO_RUNLOCK(&V_tcbinfo);
1057 else if (ti_locked == TI_WLOCKED)
1058 INP_INFO_WUNLOCK(&V_tcbinfo);
1059 else
1060 panic("%s: dropunlock ti_locked %d", __func__, ti_locked);
1061 ti_locked = TI_UNLOCKED;
1062
1063 if (inp != NULL)
1064 INP_WUNLOCK(inp);
1065
1066drop:
1067 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
1068 if (s != NULL)
1069 free(s, M_TCPLOG);
1070 if (m != NULL)
1071 m_freem(m);
1072}
1073
1074static void
1075tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
1076 struct tcpcb *tp, int drop_hdrlen, int tlen, uint8_t iptos,
1077 int ti_locked)
1078{
1079 int thflags, acked, ourfinisacked, needoutput = 0;
1080 int rstreason, todrop, win;
1081 u_long tiwin;
1082 struct tcpopt to;
1083
1084#ifdef TCPDEBUG
1085 /*
1086 * The size of tcp_saveipgen must be the size of the max ip header,
1087 * now IPv6.
1088 */
1089 u_char tcp_saveipgen[IP6_HDR_LEN];
1090 struct tcphdr tcp_savetcp;
1091 short ostate = 0;
1092#endif
1093 thflags = th->th_flags;
1094
1095 /*
1096 * If this is either a state-changing packet or current state isn't
1097 * established, we require a write lock on tcbinfo. Otherwise, we
1098 * allow either a read lock or a write lock, as we may have acquired
1099 * a write lock due to a race.
1100 *
1101 * Require a global write lock for SYN/FIN/RST segments or
1102 * non-established connections; otherwise accept either a read or
1103 * write lock, as we may have conservatively acquired a write lock in
1104 * certain cases in tcp_input() (is this still true?). Currently we
1105 * will never enter with no lock, so we try to drop it quickly in the
1106 * common pure ack/pure data cases.
1107 */
1108 if ((thflags & (TH_SYN | TH_FIN | TH_RST)) != 0 ||
1109 tp->t_state != TCPS_ESTABLISHED) {
1110 KASSERT(ti_locked == TI_WLOCKED, ("%s ti_locked %d for "
1111 "SYN/FIN/RST/!EST", __func__, ti_locked));
1112 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1113 } else {
1114#ifdef INVARIANTS
1115 if (ti_locked == TI_RLOCKED)
1116 INP_INFO_RLOCK_ASSERT(&V_tcbinfo);
1117 else if (ti_locked == TI_WLOCKED)
1118 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1119 else
1120 panic("%s: ti_locked %d for EST", __func__,
1121 ti_locked);
1122#endif
1123 }
1124 INP_WLOCK_ASSERT(tp->t_inpcb);
1125 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1126 __func__));
1127 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1128 __func__));
1129
1130 /*
1131 * Segment received on connection.
1132 * Reset idle time and keep-alive timer.
1133 * XXX: This should be done after segment
1134 * validation to ignore broken/spoofed segs.
1135 */
1136 tp->t_rcvtime = ticks;
1137 if (TCPS_HAVEESTABLISHED(tp->t_state))
1138 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1139
1140 /*
1141 * Unscale the window into a 32-bit value.
1142 * For the SYN_SENT state the scale is zero.
1143 */
1144 tiwin = th->th_win << tp->snd_scale;
1145
1146 /*
1147 * TCP ECN processing.
1148 */
1149 if (tp->t_flags & TF_ECN_PERMIT) {
1150 switch (iptos & IPTOS_ECN_MASK) {
1151 case IPTOS_ECN_CE:
1152 tp->t_flags |= TF_ECN_SND_ECE;
1153 TCPSTAT_INC(tcps_ecn_ce);
1154 break;
1155 case IPTOS_ECN_ECT0:
1156 TCPSTAT_INC(tcps_ecn_ect0);
1157 break;
1158 case IPTOS_ECN_ECT1:
1159 TCPSTAT_INC(tcps_ecn_ect1);
1160 break;
1161 }
1162
1163 if (thflags & TH_CWR)
1164 tp->t_flags &= ~TF_ECN_SND_ECE;
1165
1166 /*
1167 * Congestion experienced.
1168 * Ignore if we are already trying to recover.
1169 */
1170 if ((thflags & TH_ECE) &&
1171 SEQ_LEQ(th->th_ack, tp->snd_recover)) {
1172 TCPSTAT_INC(tcps_ecn_rcwnd);
1173 tcp_congestion_exp(tp);
1174 }
1175 }
1176
1177 /*
1178 * Parse options on any incoming segment.
1179 */
1180 tcp_dooptions(&to, (u_char *)(th + 1),
1181 (th->th_off << 2) - sizeof(struct tcphdr),
1182 (thflags & TH_SYN) ? TO_SYN : 0);
1183
1184 /*
1185 * If echoed timestamp is later than the current time,
1186 * fall back to non RFC1323 RTT calculation. Normalize
1187 * timestamp if syncookies were used when this connection
1188 * was established.
1189 */
1190 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1191 to.to_tsecr -= tp->ts_offset;
1192 if (TSTMP_GT(to.to_tsecr, ticks))
1193 to.to_tsecr = 0;
1194 }
1195
1196 /*
1197 * Process options only when we get SYN/ACK back. The SYN case
1198 * for incoming connections is handled in tcp_syncache.
1199 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1200 * or <SYN,ACK>) segment itself is never scaled.
1201 * XXX this is traditional behavior, may need to be cleaned up.
1202 */
1203 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1204 if ((to.to_flags & TOF_SCALE) &&
1205 (tp->t_flags & TF_REQ_SCALE)) {
1206 tp->t_flags |= TF_RCVD_SCALE;
1207 tp->snd_scale = to.to_wscale;
1208 }
1209 /*
1210 * Initial send window. It will be updated with
1211 * the next incoming segment to the scaled value.
1212 */
1213 tp->snd_wnd = th->th_win;
1214 if (to.to_flags & TOF_TS) {
1215 tp->t_flags |= TF_RCVD_TSTMP;
1216 tp->ts_recent = to.to_tsval;
1217 tp->ts_recent_age = ticks;
1218 }
1219 if (to.to_flags & TOF_MSS)
1220 tcp_mss(tp, to.to_mss);
1221 if ((tp->t_flags & TF_SACK_PERMIT) &&
1222 (to.to_flags & TOF_SACKPERM) == 0)
1223 tp->t_flags &= ~TF_SACK_PERMIT;
1224 }
1225
1226 /*
1227 * Header prediction: check for the two common cases
1228 * of a uni-directional data xfer. If the packet has
1229 * no control flags, is in-sequence, the window didn't
1230 * change and we're not retransmitting, it's a
1231 * candidate. If the length is zero and the ack moved
1232 * forward, we're the sender side of the xfer. Just
1233 * free the data acked & wake any higher level process
1234 * that was blocked waiting for space. If the length
1235 * is non-zero and the ack didn't move, we're the
1236 * receiver side. If we're getting packets in-order
1237 * (the reassembly queue is empty), add the data to
1238 * the socket buffer and note that we need a delayed ack.
1239 * Make sure that the hidden state-flags are also off.
1240 * Since we check for TCPS_ESTABLISHED first, it can only
1241 * be TH_NEEDSYN.
1242 */
1243 if (tp->t_state == TCPS_ESTABLISHED &&
1244 th->th_seq == tp->rcv_nxt &&
1245 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1246 tp->snd_nxt == tp->snd_max &&
1247 tiwin && tiwin == tp->snd_wnd &&
1248 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1249 LIST_EMPTY(&tp->t_segq) &&
1250 ((to.to_flags & TOF_TS) == 0 ||
1251 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1252
1253 /*
1254 * If last ACK falls within this segment's sequence numbers,
1255 * record the timestamp.
1256 * NOTE that the test is modified according to the latest
1257 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1258 */
1259 if ((to.to_flags & TOF_TS) != 0 &&
1260 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1261 tp->ts_recent_age = ticks;
1262 tp->ts_recent = to.to_tsval;
1263 }
1264
1265 if (tlen == 0) {
1266 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1267 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1268 tp->snd_cwnd >= tp->snd_wnd &&
1269 ((!V_tcp_do_newreno &&
1270 !(tp->t_flags & TF_SACK_PERMIT) &&
1271 tp->t_dupacks < tcprexmtthresh) ||
1272 ((V_tcp_do_newreno ||
1273 (tp->t_flags & TF_SACK_PERMIT)) &&
1274 !IN_FASTRECOVERY(tp) &&
1275 (to.to_flags & TOF_SACK) == 0 &&
1276 TAILQ_EMPTY(&tp->snd_holes)))) {
1277 /*
1278 * This is a pure ack for outstanding data.
1279 */
1280 if (ti_locked == TI_RLOCKED)
1281 INP_INFO_RUNLOCK(&V_tcbinfo);
1282 else if (ti_locked == TI_WLOCKED)
1283 INP_INFO_WUNLOCK(&V_tcbinfo);
1284 else
1285 panic("%s: ti_locked %d on pure ACK",
1286 __func__, ti_locked);
1287 ti_locked = TI_UNLOCKED;
1288
1289 TCPSTAT_INC(tcps_predack);
1290
1291 /*
1292 * "bad retransmit" recovery.
1293 */
1294 if (tp->t_rxtshift == 1 &&
1295 (int)(ticks - tp->t_badrxtwin) < 0) {
1296 TCPSTAT_INC(tcps_sndrexmitbad);
1297 tp->snd_cwnd = tp->snd_cwnd_prev;
1298 tp->snd_ssthresh =
1299 tp->snd_ssthresh_prev;
1300 tp->snd_recover = tp->snd_recover_prev;
1301 if (tp->t_flags & TF_WASFRECOVERY)
1302 ENTER_FASTRECOVERY(tp);
1303 tp->snd_nxt = tp->snd_max;
1304 tp->t_badrxtwin = 0;
1305 }
1306
1307 /*
1308 * Recalculate the transmit timer / rtt.
1309 *
1310 * Some boxes send broken timestamp replies
1311 * during the SYN+ACK phase, ignore
1312 * timestamps of 0 or we could calculate a
1313 * huge RTT and blow up the retransmit timer.
1314 */
1315 if ((to.to_flags & TOF_TS) != 0 &&
1316 to.to_tsecr) {
1317 if (!tp->t_rttlow ||
1318 tp->t_rttlow > ticks - to.to_tsecr)
1319 tp->t_rttlow = ticks - to.to_tsecr;
1320 tcp_xmit_timer(tp,
1321 ticks - to.to_tsecr + 1);
1322 } else if (tp->t_rtttime &&
1323 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1324 if (!tp->t_rttlow ||
1325 tp->t_rttlow > ticks - tp->t_rtttime)
1326 tp->t_rttlow = ticks - tp->t_rtttime;
1327 tcp_xmit_timer(tp,
1328 ticks - tp->t_rtttime);
1329 }
1330 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1331 acked = th->th_ack - tp->snd_una;
1332 TCPSTAT_INC(tcps_rcvackpack);
1333 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1334 sbdrop(&so->so_snd, acked);
1335 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1336 SEQ_LEQ(th->th_ack, tp->snd_recover))
1337 tp->snd_recover = th->th_ack - 1;
1338 tp->snd_una = th->th_ack;
1339 /*
1340 * Pull snd_wl2 up to prevent seq wrap relative
1341 * to th_ack.
1342 */
1343 tp->snd_wl2 = th->th_ack;
1344 tp->t_dupacks = 0;
1345 m_freem(m);
1346 ND6_HINT(tp); /* Some progress has been made. */
1347
1348 /*
1349 * If all outstanding data are acked, stop
1350 * retransmit timer, otherwise restart timer
1351 * using current (possibly backed-off) value.
1352 * If process is waiting for space,
1353 * wakeup/selwakeup/signal. If data
1354 * are ready to send, let tcp_output
1355 * decide between more output or persist.
1356 */
1357#ifdef TCPDEBUG
1358 if (so->so_options & SO_DEBUG)
1359 tcp_trace(TA_INPUT, ostate, tp,
1360 (void *)tcp_saveipgen,
1361 &tcp_savetcp, 0);
1362#endif
1363 if (tp->snd_una == tp->snd_max)
1364 tcp_timer_activate(tp, TT_REXMT, 0);
1365 else if (!tcp_timer_active(tp, TT_PERSIST))
1366 tcp_timer_activate(tp, TT_REXMT,
1367 tp->t_rxtcur);
1368 sowwakeup(so);
1369 if (so->so_snd.sb_cc)
1370 (void) tcp_output(tp);
1371 goto check_delack;
1372 }
1373 } else if (th->th_ack == tp->snd_una &&
1374 tlen <= sbspace(&so->so_rcv)) {
1375 int newsize = 0; /* automatic sockbuf scaling */
1376
1377 /*
1378 * This is a pure, in-sequence data packet with
1379 * nothing on the reassembly queue and we have enough
1380 * buffer space to take it.
1381 */
1382 if (ti_locked == TI_RLOCKED)
1383 INP_INFO_RUNLOCK(&V_tcbinfo);
1384 else if (ti_locked == TI_WLOCKED)
1385 INP_INFO_WUNLOCK(&V_tcbinfo);
1386 else
1387 panic("%s: ti_locked %d on pure data "
1388 "segment", __func__, ti_locked);
1389 ti_locked = TI_UNLOCKED;
1390
1391 /* Clean receiver SACK report if present */
1392 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1393 tcp_clean_sackreport(tp);
1394 TCPSTAT_INC(tcps_preddat);
1395 tp->rcv_nxt += tlen;
1396 /*
1397 * Pull snd_wl1 up to prevent seq wrap relative to
1398 * th_seq.
1399 */
1400 tp->snd_wl1 = th->th_seq;
1401 /*
1402 * Pull rcv_up up to prevent seq wrap relative to
1403 * rcv_nxt.
1404 */
1405 tp->rcv_up = tp->rcv_nxt;
1406 TCPSTAT_INC(tcps_rcvpack);
1407 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1408 ND6_HINT(tp); /* Some progress has been made */
1409#ifdef TCPDEBUG
1410 if (so->so_options & SO_DEBUG)
1411 tcp_trace(TA_INPUT, ostate, tp,
1412 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1413#endif
1414 /*
1415 * Automatic sizing of receive socket buffer. Often the send
1416 * buffer size is not optimally adjusted to the actual network
1417 * conditions at hand (delay bandwidth product). Setting the
1418 * buffer size too small limits throughput on links with high
1419 * bandwidth and high delay (eg. trans-continental/oceanic links).
1420 *
1421 * On the receive side the socket buffer memory is only rarely
1422 * used to any significant extent. This allows us to be much
1423 * more aggressive in scaling the receive socket buffer. For
1424 * the case that the buffer space is actually used to a large
1425 * extent and we run out of kernel memory we can simply drop
1426 * the new segments; TCP on the sender will just retransmit it
1427 * later. Setting the buffer size too big may only consume too
1428 * much kernel memory if the application doesn't read() from
1429 * the socket or packet loss or reordering makes use of the
1430 * reassembly queue.
1431 *
1432 * The criteria to step up the receive buffer one notch are:
1433 * 1. the number of bytes received during the time it takes
1434 * one timestamp to be reflected back to us (the RTT);
1435 * 2. received bytes per RTT is within seven eighth of the
1436 * current socket buffer size;
1437 * 3. receive buffer size has not hit maximal automatic size;
1438 *
1439 * This algorithm does one step per RTT at most and only if
1440 * we receive a bulk stream w/o packet losses or reorderings.
1441 * Shrinking the buffer during idle times is not necessary as
1442 * it doesn't consume any memory when idle.
1443 *
1444 * TODO: Only step up if the application is actually serving
1445 * the buffer to better manage the socket buffer resources.
1446 */
1447 if (V_tcp_do_autorcvbuf &&
1448 to.to_tsecr &&
1449 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1450 if (to.to_tsecr > tp->rfbuf_ts &&
1451 to.to_tsecr - tp->rfbuf_ts < hz) {
1452 if (tp->rfbuf_cnt >
1453 (so->so_rcv.sb_hiwat / 8 * 7) &&
1454 so->so_rcv.sb_hiwat <
1455 V_tcp_autorcvbuf_max) {
1456 newsize =
1457 min(so->so_rcv.sb_hiwat +
1458 V_tcp_autorcvbuf_inc,
1459 V_tcp_autorcvbuf_max);
1460 }
1461 /* Start over with next RTT. */
1462 tp->rfbuf_ts = 0;
1463 tp->rfbuf_cnt = 0;
1464 } else
1465 tp->rfbuf_cnt += tlen; /* add up */
1466 }
1467
1468 /* Add data to socket buffer. */
1469 SOCKBUF_LOCK(&so->so_rcv);
1470 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1471 m_freem(m);
1472 } else {
1473 /*
1474 * Set new socket buffer size.
1475 * Give up when limit is reached.
1476 */
1477 if (newsize)
1478 if (!sbreserve_locked(&so->so_rcv,
1479 newsize, so, NULL))
1480 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1481 m_adj(m, drop_hdrlen); /* delayed header drop */
1482 sbappendstream_locked(&so->so_rcv, m);
1483 }
1484 /* NB: sorwakeup_locked() does an implicit unlock. */
1485 sorwakeup_locked(so);
1486 if (DELAY_ACK(tp)) {
1487 tp->t_flags |= TF_DELACK;
1488 } else {
1489 tp->t_flags |= TF_ACKNOW;
1490 tcp_output(tp);
1491 }
1492 goto check_delack;
1493 }
1494 }
1495
1496 /*
1497 * Calculate amount of space in receive window,
1498 * and then do TCP input processing.
1499 * Receive window is amount of space in rcv queue,
1500 * but not less than advertised window.
1501 */
1502 win = sbspace(&so->so_rcv);
1503 if (win < 0)
1504 win = 0;
1505 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1506
1507 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1508 tp->rfbuf_ts = 0;
1509 tp->rfbuf_cnt = 0;
1510
1511 switch (tp->t_state) {
1512
1513 /*
1514 * If the state is SYN_RECEIVED:
1515 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1516 */
1517 case TCPS_SYN_RECEIVED:
1518 if ((thflags & TH_ACK) &&
1519 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1520 SEQ_GT(th->th_ack, tp->snd_max))) {
1521 rstreason = BANDLIM_RST_OPENPORT;
1522 goto dropwithreset;
1523 }
1524 break;
1525
1526 /*
1527 * If the state is SYN_SENT:
1528 * if seg contains an ACK, but not for our SYN, drop the input.
1529 * if seg contains a RST, then drop the connection.
1530 * if seg does not contain SYN, then drop it.
1531 * Otherwise this is an acceptable SYN segment
1532 * initialize tp->rcv_nxt and tp->irs
1533 * if seg contains ack then advance tp->snd_una
1534 * if seg contains an ECE and ECN support is enabled, the stream
1535 * is ECN capable.
1536 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1537 * arrange for segment to be acked (eventually)
1538 * continue processing rest of data/controls, beginning with URG
1539 */
1540 case TCPS_SYN_SENT:
1541 if ((thflags & TH_ACK) &&
1542 (SEQ_LEQ(th->th_ack, tp->iss) ||
1543 SEQ_GT(th->th_ack, tp->snd_max))) {
1544 rstreason = BANDLIM_UNLIMITED;
1545 goto dropwithreset;
1546 }
1547 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1548 tp = tcp_drop(tp, ECONNREFUSED);
1549 if (thflags & TH_RST)
1550 goto drop;
1551 if (!(thflags & TH_SYN))
1552 goto drop;
1553
1554 tp->irs = th->th_seq;
1555 tcp_rcvseqinit(tp);
1556 if (thflags & TH_ACK) {
1557 TCPSTAT_INC(tcps_connects);
1558 soisconnected(so);
1559#ifdef MAC
1560 mac_socketpeer_set_from_mbuf(m, so);
1561#endif
1562 /* Do window scaling on this connection? */
1563 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1564 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1565 tp->rcv_scale = tp->request_r_scale;
1566 }
1567 tp->rcv_adv += tp->rcv_wnd;
1568 tp->snd_una++; /* SYN is acked */
1569 /*
1570 * If there's data, delay ACK; if there's also a FIN
1571 * ACKNOW will be turned on later.
1572 */
1573 if (DELAY_ACK(tp) && tlen != 0)
1574 tcp_timer_activate(tp, TT_DELACK,
1575 tcp_delacktime);
1576 else
1577 tp->t_flags |= TF_ACKNOW;
1578
1579 if ((thflags & TH_ECE) && V_tcp_do_ecn) {
1580 tp->t_flags |= TF_ECN_PERMIT;
1581 TCPSTAT_INC(tcps_ecn_shs);
1582 }
1583
1584 /*
1585 * Received <SYN,ACK> in SYN_SENT[*] state.
1586 * Transitions:
1587 * SYN_SENT --> ESTABLISHED
1588 * SYN_SENT* --> FIN_WAIT_1
1589 */
1590 tp->t_starttime = ticks;
1591 if (tp->t_flags & TF_NEEDFIN) {
1592 tp->t_state = TCPS_FIN_WAIT_1;
1593 tp->t_flags &= ~TF_NEEDFIN;
1594 thflags &= ~TH_SYN;
1595 } else {
1596 tp->t_state = TCPS_ESTABLISHED;
1597 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1598 }
1599 } else {
1600 /*
1601 * Received initial SYN in SYN-SENT[*] state =>
1602 * simultaneous open. If segment contains CC option
1603 * and there is a cached CC, apply TAO test.
1604 * If it succeeds, connection is * half-synchronized.
1605 * Otherwise, do 3-way handshake:
1606 * SYN-SENT -> SYN-RECEIVED
1607 * SYN-SENT* -> SYN-RECEIVED*
1608 * If there was no CC option, clear cached CC value.
1609 */
1610 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1611 tcp_timer_activate(tp, TT_REXMT, 0);
1612 tp->t_state = TCPS_SYN_RECEIVED;
1613 }
1614
1615 KASSERT(ti_locked == TI_WLOCKED, ("%s: trimthenstep6: "
1616 "ti_locked %d", __func__, ti_locked));
1617 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1618 INP_WLOCK_ASSERT(tp->t_inpcb);
1619
1620 /*
1621 * Advance th->th_seq to correspond to first data byte.
1622 * If data, trim to stay within window,
1623 * dropping FIN if necessary.
1624 */
1625 th->th_seq++;
1626 if (tlen > tp->rcv_wnd) {
1627 todrop = tlen - tp->rcv_wnd;
1628 m_adj(m, -todrop);
1629 tlen = tp->rcv_wnd;
1630 thflags &= ~TH_FIN;
1631 TCPSTAT_INC(tcps_rcvpackafterwin);
1632 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1633 }
1634 tp->snd_wl1 = th->th_seq - 1;
1635 tp->rcv_up = th->th_seq;
1636 /*
1637 * Client side of transaction: already sent SYN and data.
1638 * If the remote host used T/TCP to validate the SYN,
1639 * our data will be ACK'd; if so, enter normal data segment
1640 * processing in the middle of step 5, ack processing.
1641 * Otherwise, goto step 6.
1642 */
1643 if (thflags & TH_ACK)
1644 goto process_ACK;
1645
1646 goto step6;
1647
1648 /*
1649 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1650 * do normal processing.
1651 *
1652 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1653 */
1654 case TCPS_LAST_ACK:
1655 case TCPS_CLOSING:
1656 break; /* continue normal processing */
1657 }
1658
1659 /*
1660 * States other than LISTEN or SYN_SENT.
1661 * First check the RST flag and sequence number since reset segments
1662 * are exempt from the timestamp and connection count tests. This
1663 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1664 * below which allowed reset segments in half the sequence space
1665 * to fall though and be processed (which gives forged reset
1666 * segments with a random sequence number a 50 percent chance of
1667 * killing a connection).
1668 * Then check timestamp, if present.
1669 * Then check the connection count, if present.
1670 * Then check that at least some bytes of segment are within
1671 * receive window. If segment begins before rcv_nxt,
1672 * drop leading data (and SYN); if nothing left, just ack.
1673 *
1674 *
1675 * If the RST bit is set, check the sequence number to see
1676 * if this is a valid reset segment.
1677 * RFC 793 page 37:
1678 * In all states except SYN-SENT, all reset (RST) segments
1679 * are validated by checking their SEQ-fields. A reset is
1680 * valid if its sequence number is in the window.
1681 * Note: this does not take into account delayed ACKs, so
1682 * we should test against last_ack_sent instead of rcv_nxt.
1683 * The sequence number in the reset segment is normally an
1684 * echo of our outgoing acknowlegement numbers, but some hosts
1685 * send a reset with the sequence number at the rightmost edge
1686 * of our receive window, and we have to handle this case.
1687 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1688 * that brute force RST attacks are possible. To combat this,
1689 * we use a much stricter check while in the ESTABLISHED state,
1690 * only accepting RSTs where the sequence number is equal to
1691 * last_ack_sent. In all other states (the states in which a
1692 * RST is more likely), the more permissive check is used.
1693 * If we have multiple segments in flight, the initial reset
1694 * segment sequence numbers will be to the left of last_ack_sent,
1695 * but they will eventually catch up.
1696 * In any case, it never made sense to trim reset segments to
1697 * fit the receive window since RFC 1122 says:
1698 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1699 *
1700 * A TCP SHOULD allow a received RST segment to include data.
1701 *
1702 * DISCUSSION
1703 * It has been suggested that a RST segment could contain
1704 * ASCII text that encoded and explained the cause of the
1705 * RST. No standard has yet been established for such
1706 * data.
1707 *
1708 * If the reset segment passes the sequence number test examine
1709 * the state:
1710 * SYN_RECEIVED STATE:
1711 * If passive open, return to LISTEN state.
1712 * If active open, inform user that connection was refused.
1713 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1714 * Inform user that connection was reset, and close tcb.
1715 * CLOSING, LAST_ACK STATES:
1716 * Close the tcb.
1717 * TIME_WAIT STATE:
1718 * Drop the segment - see Stevens, vol. 2, p. 964 and
1719 * RFC 1337.
1720 */
1721 if (thflags & TH_RST) {
1722 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1723 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1724 switch (tp->t_state) {
1725
1726 case TCPS_SYN_RECEIVED:
1727 so->so_error = ECONNREFUSED;
1728 goto close;
1729
1730 case TCPS_ESTABLISHED:
1731 if (V_tcp_insecure_rst == 0 &&
1732 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
1733 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
1734 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1735 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
1736 TCPSTAT_INC(tcps_badrst);
1737 goto drop;
1738 }
1739 /* FALLTHROUGH */
1740 case TCPS_FIN_WAIT_1:
1741 case TCPS_FIN_WAIT_2:
1742 case TCPS_CLOSE_WAIT:
1743 so->so_error = ECONNRESET;
1744 close:
1745 KASSERT(ti_locked == TI_WLOCKED,
1746 ("tcp_do_segment: TH_RST 1 ti_locked %d",
1747 ti_locked));
1748 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1749
1750 tp->t_state = TCPS_CLOSED;
1751 TCPSTAT_INC(tcps_drops);
1752 tp = tcp_close(tp);
1753 break;
1754
1755 case TCPS_CLOSING:
1756 case TCPS_LAST_ACK:
1757 KASSERT(ti_locked == TI_WLOCKED,
1758 ("tcp_do_segment: TH_RST 2 ti_locked %d",
1759 ti_locked));
1760 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1761
1762 tp = tcp_close(tp);
1763 break;
1764 }
1765 }
1766 goto drop;
1767 }
1768
1769 /*
1770 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1771 * and it's less than ts_recent, drop it.
1772 */
1773 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1774 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1775
1776 /* Check to see if ts_recent is over 24 days old. */
1777 if (ticks - tp->ts_recent_age > TCP_PAWS_IDLE) {
1778 /*
1779 * Invalidate ts_recent. If this segment updates
1780 * ts_recent, the age will be reset later and ts_recent
1781 * will get a valid value. If it does not, setting
1782 * ts_recent to zero will at least satisfy the
1783 * requirement that zero be placed in the timestamp
1784 * echo reply when ts_recent isn't valid. The
1785 * age isn't reset until we get a valid ts_recent
1786 * because we don't want out-of-order segments to be
1787 * dropped when ts_recent is old.
1788 */
1789 tp->ts_recent = 0;
1790 } else {
1791 TCPSTAT_INC(tcps_rcvduppack);
1792 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
1793 TCPSTAT_INC(tcps_pawsdrop);
1794 if (tlen)
1795 goto dropafterack;
1796 goto drop;
1797 }
1798 }
1799
1800 /*
1801 * In the SYN-RECEIVED state, validate that the packet belongs to
1802 * this connection before trimming the data to fit the receive
1803 * window. Check the sequence number versus IRS since we know
1804 * the sequence numbers haven't wrapped. This is a partial fix
1805 * for the "LAND" DoS attack.
1806 */
1807 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1808 rstreason = BANDLIM_RST_OPENPORT;
1809 goto dropwithreset;
1810 }
1811
1812 todrop = tp->rcv_nxt - th->th_seq;
1813 if (todrop > 0) {
1814 /*
1815 * If this is a duplicate SYN for our current connection,
1816 * advance over it and pretend and it's not a SYN.
1817 */
1818 if (thflags & TH_SYN && th->th_seq == tp->irs) {
1819 thflags &= ~TH_SYN;
1820 th->th_seq++;
1821 if (th->th_urp > 1)
1822 th->th_urp--;
1823 else
1824 thflags &= ~TH_URG;
1825 todrop--;
1826 }
1827 /*
1828 * Following if statement from Stevens, vol. 2, p. 960.
1829 */
1830 if (todrop > tlen
1831 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1832 /*
1833 * Any valid FIN must be to the left of the window.
1834 * At this point the FIN must be a duplicate or out
1835 * of sequence; drop it.
1836 */
1837 thflags &= ~TH_FIN;
1838
1839 /*
1840 * Send an ACK to resynchronize and drop any data.
1841 * But keep on processing for RST or ACK.
1842 */
1843 tp->t_flags |= TF_ACKNOW;
1844 todrop = tlen;
1845 TCPSTAT_INC(tcps_rcvduppack);
1846 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
1847 } else {
1848 TCPSTAT_INC(tcps_rcvpartduppack);
1849 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
1850 }
1851 drop_hdrlen += todrop; /* drop from the top afterwards */
1852 th->th_seq += todrop;
1853 tlen -= todrop;
1854 if (th->th_urp > todrop)
1855 th->th_urp -= todrop;
1856 else {
1857 thflags &= ~TH_URG;
1858 th->th_urp = 0;
1859 }
1860 }
1861
1862 /*
1863 * If new data are received on a connection after the
1864 * user processes are gone, then RST the other end.
1865 */
1866 if ((so->so_state & SS_NOFDREF) &&
1867 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1868 char *s;
1869
1870 KASSERT(ti_locked == TI_WLOCKED, ("%s: SS_NOFDEREF && "
1871 "CLOSE_WAIT && tlen ti_locked %d", __func__, ti_locked));
1872 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1873
1874 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
1875 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
1876 "was closed, sending RST and removing tcpcb\n",
1877 s, __func__, tcpstates[tp->t_state], tlen);
1878 free(s, M_TCPLOG);
1879 }
1880 tp = tcp_close(tp);
1881 TCPSTAT_INC(tcps_rcvafterclose);
1882 rstreason = BANDLIM_UNLIMITED;
1883 goto dropwithreset;
1884 }
1885
1886 /*
1887 * If segment ends after window, drop trailing data
1888 * (and PUSH and FIN); if nothing left, just ACK.
1889 */
1890 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1891 if (todrop > 0) {
1892 TCPSTAT_INC(tcps_rcvpackafterwin);
1893 if (todrop >= tlen) {
1894 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
1895 /*
1896 * If window is closed can only take segments at
1897 * window edge, and have to drop data and PUSH from
1898 * incoming segments. Continue processing, but
1899 * remember to ack. Otherwise, drop segment
1900 * and ack.
1901 */
1902 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1903 tp->t_flags |= TF_ACKNOW;
1904 TCPSTAT_INC(tcps_rcvwinprobe);
1905 } else
1906 goto dropafterack;
1907 } else
1908 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
1909 m_adj(m, -todrop);
1910 tlen -= todrop;
1911 thflags &= ~(TH_PUSH|TH_FIN);
1912 }
1913
1914 /*
1915 * If last ACK falls within this segment's sequence numbers,
1916 * record its timestamp.
1917 * NOTE:
1918 * 1) That the test incorporates suggestions from the latest
1919 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1920 * 2) That updating only on newer timestamps interferes with
1921 * our earlier PAWS tests, so this check should be solely
1922 * predicated on the sequence space of this segment.
1923 * 3) That we modify the segment boundary check to be
1924 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1925 * instead of RFC1323's
1926 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1927 * This modified check allows us to overcome RFC1323's
1928 * limitations as described in Stevens TCP/IP Illustrated
1929 * Vol. 2 p.869. In such cases, we can still calculate the
1930 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1931 */
1932 if ((to.to_flags & TOF_TS) != 0 &&
1933 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1934 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1935 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1936 tp->ts_recent_age = ticks;
1937 tp->ts_recent = to.to_tsval;
1938 }
1939
1940 /*
1941 * If a SYN is in the window, then this is an
1942 * error and we send an RST and drop the connection.
1943 */
1944 if (thflags & TH_SYN) {
1945 KASSERT(ti_locked == TI_WLOCKED,
1946 ("tcp_do_segment: TH_SYN ti_locked %d", ti_locked));
1947 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
1948
1949 tp = tcp_drop(tp, ECONNRESET);
1950 rstreason = BANDLIM_UNLIMITED;
1951 goto drop;
1952 }
1953
1954 /*
1955 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1956 * flag is on (half-synchronized state), then queue data for
1957 * later processing; else drop segment and return.
1958 */
1959 if ((thflags & TH_ACK) == 0) {
1960 if (tp->t_state == TCPS_SYN_RECEIVED ||
1961 (tp->t_flags & TF_NEEDSYN))
1962 goto step6;
1963 else if (tp->t_flags & TF_ACKNOW)
1964 goto dropafterack;
1965 else
1966 goto drop;
1967 }
1968
1969 /*
1970 * Ack processing.
1971 */
1972 switch (tp->t_state) {
1973
1974 /*
1975 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1976 * ESTABLISHED state and continue processing.
1977 * The ACK was checked above.
1978 */
1979 case TCPS_SYN_RECEIVED:
1980
1981 TCPSTAT_INC(tcps_connects);
1982 soisconnected(so);
1983 /* Do window scaling? */
1984 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1985 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1986 tp->rcv_scale = tp->request_r_scale;
1987 tp->snd_wnd = tiwin;
1988 }
1989 /*
1990 * Make transitions:
1991 * SYN-RECEIVED -> ESTABLISHED
1992 * SYN-RECEIVED* -> FIN-WAIT-1
1993 */
1994 tp->t_starttime = ticks;
1995 if (tp->t_flags & TF_NEEDFIN) {
1996 tp->t_state = TCPS_FIN_WAIT_1;
1997 tp->t_flags &= ~TF_NEEDFIN;
1998 } else {
1999 tp->t_state = TCPS_ESTABLISHED;
2000 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
2001 }
2002 /*
2003 * If segment contains data or ACK, will call tcp_reass()
2004 * later; if not, do so now to pass queued data to user.
2005 */
2006 if (tlen == 0 && (thflags & TH_FIN) == 0)
2007 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
2008 (struct mbuf *)0);
2009 tp->snd_wl1 = th->th_seq - 1;
2010 /* FALLTHROUGH */
2011
2012 /*
2013 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2014 * ACKs. If the ack is in the range
2015 * tp->snd_una < th->th_ack <= tp->snd_max
2016 * then advance tp->snd_una to th->th_ack and drop
2017 * data from the retransmission queue. If this ACK reflects
2018 * more up to date window information we update our window information.
2019 */
2020 case TCPS_ESTABLISHED:
2021 case TCPS_FIN_WAIT_1:
2022 case TCPS_FIN_WAIT_2:
2023 case TCPS_CLOSE_WAIT:
2024 case TCPS_CLOSING:
2025 case TCPS_LAST_ACK:
2026 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2027 TCPSTAT_INC(tcps_rcvacktoomuch);
2028 goto dropafterack;
2029 }
2030 if ((tp->t_flags & TF_SACK_PERMIT) &&
2031 ((to.to_flags & TOF_SACK) ||
2032 !TAILQ_EMPTY(&tp->snd_holes)))
2033 tcp_sack_doack(tp, &to, th->th_ack);
2034 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2035 if (tlen == 0 && tiwin == tp->snd_wnd) {
2036 TCPSTAT_INC(tcps_rcvdupack);
2037 /*
2038 * If we have outstanding data (other than
2039 * a window probe), this is a completely
2040 * duplicate ack (ie, window info didn't
2041 * change), the ack is the biggest we've
2042 * seen and we've seen exactly our rexmt
2043 * threshhold of them, assume a packet
2044 * has been dropped and retransmit it.
2045 * Kludge snd_nxt & the congestion
2046 * window so we send only this one
2047 * packet.
2048 *
2049 * We know we're losing at the current
2050 * window size so do congestion avoidance
2051 * (set ssthresh to half the current window
2052 * and pull our congestion window back to
2053 * the new ssthresh).
2054 *
2055 * Dup acks mean that packets have left the
2056 * network (they're now cached at the receiver)
2057 * so bump cwnd by the amount in the receiver
2058 * to keep a constant cwnd packets in the
2059 * network.
2060 *
2061 * When using TCP ECN, notify the peer that
2062 * we reduced the cwnd.
2063 */
2064 if (!tcp_timer_active(tp, TT_REXMT) ||
2065 th->th_ack != tp->snd_una)
2066 tp->t_dupacks = 0;
2067 else if (++tp->t_dupacks > tcprexmtthresh ||
2068 ((V_tcp_do_newreno ||
2069 (tp->t_flags & TF_SACK_PERMIT)) &&
2070 IN_FASTRECOVERY(tp))) {
2071 if ((tp->t_flags & TF_SACK_PERMIT) &&
2072 IN_FASTRECOVERY(tp)) {
2073 int awnd;
2074
2075 /*
2076 * Compute the amount of data in flight first.
2077 * We can inject new data into the pipe iff
2078 * we have less than 1/2 the original window's
2079 * worth of data in flight.
2080 */
2081 awnd = (tp->snd_nxt - tp->snd_fack) +
2082 tp->sackhint.sack_bytes_rexmit;
2083 if (awnd < tp->snd_ssthresh) {
2084 tp->snd_cwnd += tp->t_maxseg;
2085 if (tp->snd_cwnd > tp->snd_ssthresh)
2086 tp->snd_cwnd = tp->snd_ssthresh;
2087 }
2088 } else
2089 tp->snd_cwnd += tp->t_maxseg;
2090 (void) tcp_output(tp);
2091 goto drop;
2092 } else if (tp->t_dupacks == tcprexmtthresh) {
2093 tcp_seq onxt = tp->snd_nxt;
2094
2095 /*
2096 * If we're doing sack, check to
2097 * see if we're already in sack
2098 * recovery. If we're not doing sack,
2099 * check to see if we're in newreno
2100 * recovery.
2101 */
2102 if (tp->t_flags & TF_SACK_PERMIT) {
2103 if (IN_FASTRECOVERY(tp)) {
2104 tp->t_dupacks = 0;
2105 break;
2106 }
2107 } else if (V_tcp_do_newreno ||
2108 V_tcp_do_ecn) {
2109 if (SEQ_LEQ(th->th_ack,
2110 tp->snd_recover)) {
2111 tp->t_dupacks = 0;
2112 break;
2113 }
2114 }
2115 tcp_congestion_exp(tp);
2116 tcp_timer_activate(tp, TT_REXMT, 0);
2117 tp->t_rtttime = 0;
2118 if (tp->t_flags & TF_SACK_PERMIT) {
2119 TCPSTAT_INC(
2120 tcps_sack_recovery_episode);
2121 tp->sack_newdata = tp->snd_nxt;
2122 tp->snd_cwnd = tp->t_maxseg;
2123 (void) tcp_output(tp);
2124 goto drop;
2125 }
2126 tp->snd_nxt = th->th_ack;
2127 tp->snd_cwnd = tp->t_maxseg;
2128 (void) tcp_output(tp);
2129 KASSERT(tp->snd_limited <= 2,
2130 ("%s: tp->snd_limited too big",
2131 __func__));
2132 tp->snd_cwnd = tp->snd_ssthresh +
2133 tp->t_maxseg *
2134 (tp->t_dupacks - tp->snd_limited);
2135 if (SEQ_GT(onxt, tp->snd_nxt))
2136 tp->snd_nxt = onxt;
2137 goto drop;
2138 } else if (V_tcp_do_rfc3042) {
2139 u_long oldcwnd = tp->snd_cwnd;
2140 tcp_seq oldsndmax = tp->snd_max;
2141 u_int sent;
2142
2143 KASSERT(tp->t_dupacks == 1 ||
2144 tp->t_dupacks == 2,
2145 ("%s: dupacks not 1 or 2",
2146 __func__));
2147 if (tp->t_dupacks == 1)
2148 tp->snd_limited = 0;
2149 tp->snd_cwnd =
2150 (tp->snd_nxt - tp->snd_una) +
2151 (tp->t_dupacks - tp->snd_limited) *
2152 tp->t_maxseg;
2153 (void) tcp_output(tp);
2154 sent = tp->snd_max - oldsndmax;
2155 if (sent > tp->t_maxseg) {
2156 KASSERT((tp->t_dupacks == 2 &&
2157 tp->snd_limited == 0) ||
2158 (sent == tp->t_maxseg + 1 &&
2159 tp->t_flags & TF_SENTFIN),
2160 ("%s: sent too much",
2161 __func__));
2162 tp->snd_limited = 2;
2163 } else if (sent > 0)
2164 ++tp->snd_limited;
2165 tp->snd_cwnd = oldcwnd;
2166 goto drop;
2167 }
2168 } else
2169 tp->t_dupacks = 0;
2170 break;
2171 }
2172
2173 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2174 ("%s: th_ack <= snd_una", __func__));
2175
2176 /*
2177 * If the congestion window was inflated to account
2178 * for the other side's cached packets, retract it.
2179 */
2180 if (V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) {
2181 if (IN_FASTRECOVERY(tp)) {
2182 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2183 if (tp->t_flags & TF_SACK_PERMIT)
2184 tcp_sack_partialack(tp, th);
2185 else
2186 tcp_newreno_partial_ack(tp, th);
2187 } else {
2188 /*
2189 * Out of fast recovery.
2190 * Window inflation should have left us
2191 * with approximately snd_ssthresh
2192 * outstanding data.
2193 * But in case we would be inclined to
2194 * send a burst, better to do it via
2195 * the slow start mechanism.
2196 */
2197 if (SEQ_GT(th->th_ack +
2198 tp->snd_ssthresh,
2199 tp->snd_max))
2200 tp->snd_cwnd = tp->snd_max -
2201 th->th_ack +
2202 tp->t_maxseg;
2203 else
2204 tp->snd_cwnd = tp->snd_ssthresh;
2205 }
2206 }
2207 } else {
2208 if (tp->t_dupacks >= tcprexmtthresh &&
2209 tp->snd_cwnd > tp->snd_ssthresh)
2210 tp->snd_cwnd = tp->snd_ssthresh;
2211 }
2212 tp->t_dupacks = 0;
2213 /*
2214 * If we reach this point, ACK is not a duplicate,
2215 * i.e., it ACKs something we sent.
2216 */
2217 if (tp->t_flags & TF_NEEDSYN) {
2218 /*
2219 * T/TCP: Connection was half-synchronized, and our
2220 * SYN has been ACK'd (so connection is now fully
2221 * synchronized). Go to non-starred state,
2222 * increment snd_una for ACK of SYN, and check if
2223 * we can do window scaling.
2224 */
2225 tp->t_flags &= ~TF_NEEDSYN;
2226 tp->snd_una++;
2227 /* Do window scaling? */
2228 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2229 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2230 tp->rcv_scale = tp->request_r_scale;
2231 /* Send window already scaled. */
2232 }
2233 }
2234
2235process_ACK:
2236 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2237 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2238 ("tcp_input: process_ACK ti_locked %d", ti_locked));
2239 INP_WLOCK_ASSERT(tp->t_inpcb);
2240
2241 acked = th->th_ack - tp->snd_una;
2242 TCPSTAT_INC(tcps_rcvackpack);
2243 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2244
2245 /*
2246 * If we just performed our first retransmit, and the ACK
2247 * arrives within our recovery window, then it was a mistake
2248 * to do the retransmit in the first place. Recover our
2249 * original cwnd and ssthresh, and proceed to transmit where
2250 * we left off.
2251 */
2252 if (tp->t_rxtshift == 1 && (int)(ticks - tp->t_badrxtwin) < 0) {
2253 TCPSTAT_INC(tcps_sndrexmitbad);
2254 tp->snd_cwnd = tp->snd_cwnd_prev;
2255 tp->snd_ssthresh = tp->snd_ssthresh_prev;
2256 tp->snd_recover = tp->snd_recover_prev;
2257 if (tp->t_flags & TF_WASFRECOVERY)
2258 ENTER_FASTRECOVERY(tp);
2259 tp->snd_nxt = tp->snd_max;
2260 tp->t_badrxtwin = 0; /* XXX probably not required */
2261 }
2262
2263 /*
2264 * If we have a timestamp reply, update smoothed
2265 * round trip time. If no timestamp is present but
2266 * transmit timer is running and timed sequence
2267 * number was acked, update smoothed round trip time.
2268 * Since we now have an rtt measurement, cancel the
2269 * timer backoff (cf., Phil Karn's retransmit alg.).
2270 * Recompute the initial retransmit timer.
2271 *
2272 * Some boxes send broken timestamp replies
2273 * during the SYN+ACK phase, ignore
2274 * timestamps of 0 or we could calculate a
2275 * huge RTT and blow up the retransmit timer.
2276 */
2277 if ((to.to_flags & TOF_TS) != 0 &&
2278 to.to_tsecr) {
2279 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
2280 tp->t_rttlow = ticks - to.to_tsecr;
2281 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
2282 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2283 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2284 tp->t_rttlow = ticks - tp->t_rtttime;
2285 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2286 }
2287 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2288
2289 /*
2290 * If all outstanding data is acked, stop retransmit
2291 * timer and remember to restart (more output or persist).
2292 * If there is more data to be acked, restart retransmit
2293 * timer, using current (possibly backed-off) value.
2294 */
2295 if (th->th_ack == tp->snd_max) {
2296 tcp_timer_activate(tp, TT_REXMT, 0);
2297 needoutput = 1;
2298 } else if (!tcp_timer_active(tp, TT_PERSIST))
2299 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2300
2301 /*
2302 * If no data (only SYN) was ACK'd,
2303 * skip rest of ACK processing.
2304 */
2305 if (acked == 0)
2306 goto step6;
2307
2308 /*
2309 * When new data is acked, open the congestion window.
2310 * Method depends on which congestion control state we're
2311 * in (slow start or cong avoid) and if ABC (RFC 3465) is
2312 * enabled.
2313 *
2314 * slow start: cwnd <= ssthresh
2315 * cong avoid: cwnd > ssthresh
2316 *
2317 * slow start and ABC (RFC 3465):
2318 * Grow cwnd exponentially by the amount of data
2319 * ACKed capping the max increment per ACK to
2320 * (abc_l_var * maxseg) bytes.
2321 *
2322 * slow start without ABC (RFC 2581):
2323 * Grow cwnd exponentially by maxseg per ACK.
2324 *
2325 * cong avoid and ABC (RFC 3465):
2326 * Grow cwnd linearly by maxseg per RTT for each
2327 * cwnd worth of ACKed data.
2328 *
2329 * cong avoid without ABC (RFC 2581):
2330 * Grow cwnd linearly by approximately maxseg per RTT using
2331 * maxseg^2 / cwnd per ACK as the increment.
2332 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte to
2333 * avoid capping cwnd.
2334 */
2335 if ((!V_tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) ||
2336 !IN_FASTRECOVERY(tp)) {
2337 u_int cw = tp->snd_cwnd;
2338 u_int incr = tp->t_maxseg;
2339 /* In congestion avoidance? */
2340 if (cw > tp->snd_ssthresh) {
2341 if (V_tcp_do_rfc3465) {
2342 tp->t_bytes_acked += acked;
2343 if (tp->t_bytes_acked >= tp->snd_cwnd)
2344 tp->t_bytes_acked -= cw;
2345 else
2346 incr = 0;
2347 }
2348 else
2349 incr = max((incr * incr / cw), 1);
2350 /*
2351 * In slow-start with ABC enabled and no RTO in sight?
2352 * (Must not use abc_l_var > 1 if slow starting after an
2353 * RTO. On RTO, snd_nxt = snd_una, so the snd_nxt ==
2354 * snd_max check is sufficient to handle this).
2355 */
2356 } else if (V_tcp_do_rfc3465 &&
2357 tp->snd_nxt == tp->snd_max)
2358 incr = min(acked,
2359 V_tcp_abc_l_var * tp->t_maxseg);
2360 /* ABC is on by default, so (incr == 0) frequently. */
2361 if (incr > 0)
2362 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2363 }
2364 SOCKBUF_LOCK(&so->so_snd);
2365 if (acked > so->so_snd.sb_cc) {
2366 tp->snd_wnd -= so->so_snd.sb_cc;
2367 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2368 ourfinisacked = 1;
2369 } else {
2370 sbdrop_locked(&so->so_snd, acked);
2371 tp->snd_wnd -= acked;
2372 ourfinisacked = 0;
2373 }
2374 /* NB: sowwakeup_locked() does an implicit unlock. */
2375 sowwakeup_locked(so);
2376 /* Detect una wraparound. */
2377 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2378 !IN_FASTRECOVERY(tp) &&
2379 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2380 SEQ_LEQ(th->th_ack, tp->snd_recover))
2381 tp->snd_recover = th->th_ack - 1;
2382 if ((V_tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2383 IN_FASTRECOVERY(tp) &&
2384 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
2385 EXIT_FASTRECOVERY(tp);
2386 tp->t_bytes_acked = 0;
2387 }
2388 tp->snd_una = th->th_ack;
2389 if (tp->t_flags & TF_SACK_PERMIT) {
2390 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2391 tp->snd_recover = tp->snd_una;
2392 }
2393 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2394 tp->snd_nxt = tp->snd_una;
2395
2396 switch (tp->t_state) {
2397
2398 /*
2399 * In FIN_WAIT_1 STATE in addition to the processing
2400 * for the ESTABLISHED state if our FIN is now acknowledged
2401 * then enter FIN_WAIT_2.
2402 */
2403 case TCPS_FIN_WAIT_1:
2404 if (ourfinisacked) {
2405 /*
2406 * If we can't receive any more
2407 * data, then closing user can proceed.
2408 * Starting the timer is contrary to the
2409 * specification, but if we don't get a FIN
2410 * we'll hang forever.
2411 *
2412 * XXXjl:
2413 * we should release the tp also, and use a
2414 * compressed state.
2415 */
2416 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2417 int timeout;
2418
2419 soisdisconnected(so);
2420 timeout = (tcp_fast_finwait2_recycle) ?
2421 tcp_finwait2_timeout : tcp_maxidle;
2422 tcp_timer_activate(tp, TT_2MSL, timeout);
2423 }
2424 tp->t_state = TCPS_FIN_WAIT_2;
2425 }
2426 break;
2427
2428 /*
2429 * In CLOSING STATE in addition to the processing for
2430 * the ESTABLISHED state if the ACK acknowledges our FIN
2431 * then enter the TIME-WAIT state, otherwise ignore
2432 * the segment.
2433 */
2434 case TCPS_CLOSING:
2435 if (ourfinisacked) {
2436 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2437 tcp_twstart(tp);
2438 INP_INFO_WUNLOCK(&V_tcbinfo);
2439 m_freem(m);
2440 return;
2441 }
2442 break;
2443
2444 /*
2445 * In LAST_ACK, we may still be waiting for data to drain
2446 * and/or to be acked, as well as for the ack of our FIN.
2447 * If our FIN is now acknowledged, delete the TCB,
2448 * enter the closed state and return.
2449 */
2450 case TCPS_LAST_ACK:
2451 if (ourfinisacked) {
2452 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2453 tp = tcp_close(tp);
2454 goto drop;
2455 }
2456 break;
2457 }
2458 }
2459
2460step6:
2461 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2462 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2463 ("tcp_do_segment: step6 ti_locked %d", ti_locked));
2464 INP_WLOCK_ASSERT(tp->t_inpcb);
2465
2466 /*
2467 * Update window information.
2468 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2469 */
2470 if ((thflags & TH_ACK) &&
2471 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2472 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2473 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2474 /* keep track of pure window updates */
2475 if (tlen == 0 &&
2476 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2477 TCPSTAT_INC(tcps_rcvwinupd);
2478 tp->snd_wnd = tiwin;
2479 tp->snd_wl1 = th->th_seq;
2480 tp->snd_wl2 = th->th_ack;
2481 if (tp->snd_wnd > tp->max_sndwnd)
2482 tp->max_sndwnd = tp->snd_wnd;
2483 needoutput = 1;
2484 }
2485
2486 /*
2487 * Process segments with URG.
2488 */
2489 if ((thflags & TH_URG) && th->th_urp &&
2490 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2491 /*
2492 * This is a kludge, but if we receive and accept
2493 * random urgent pointers, we'll crash in
2494 * soreceive. It's hard to imagine someone
2495 * actually wanting to send this much urgent data.
2496 */
2497 SOCKBUF_LOCK(&so->so_rcv);
2498 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2499 th->th_urp = 0; /* XXX */
2500 thflags &= ~TH_URG; /* XXX */
2501 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2502 goto dodata; /* XXX */
2503 }
2504 /*
2505 * If this segment advances the known urgent pointer,
2506 * then mark the data stream. This should not happen
2507 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2508 * a FIN has been received from the remote side.
2509 * In these states we ignore the URG.
2510 *
2511 * According to RFC961 (Assigned Protocols),
2512 * the urgent pointer points to the last octet
2513 * of urgent data. We continue, however,
2514 * to consider it to indicate the first octet
2515 * of data past the urgent section as the original
2516 * spec states (in one of two places).
2517 */
2518 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2519 tp->rcv_up = th->th_seq + th->th_urp;
2520 so->so_oobmark = so->so_rcv.sb_cc +
2521 (tp->rcv_up - tp->rcv_nxt) - 1;
2522 if (so->so_oobmark == 0)
2523 so->so_rcv.sb_state |= SBS_RCVATMARK;
2524 sohasoutofband(so);
2525 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2526 }
2527 SOCKBUF_UNLOCK(&so->so_rcv);
2528 /*
2529 * Remove out of band data so doesn't get presented to user.
2530 * This can happen independent of advancing the URG pointer,
2531 * but if two URG's are pending at once, some out-of-band
2532 * data may creep in... ick.
2533 */
2534 if (th->th_urp <= (u_long)tlen &&
2535 !(so->so_options & SO_OOBINLINE)) {
2536 /* hdr drop is delayed */
2537 tcp_pulloutofband(so, th, m, drop_hdrlen);
2538 }
2539 } else {
2540 /*
2541 * If no out of band data is expected,
2542 * pull receive urgent pointer along
2543 * with the receive window.
2544 */
2545 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2546 tp->rcv_up = tp->rcv_nxt;
2547 }
2548dodata: /* XXX */
2549 INP_INFO_LOCK_ASSERT(&V_tcbinfo);
2550 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2551 ("tcp_do_segment: dodata ti_locked %d", ti_locked));
2552 INP_WLOCK_ASSERT(tp->t_inpcb);
2553
2554 /*
2555 * Process the segment text, merging it into the TCP sequencing queue,
2556 * and arranging for acknowledgment of receipt if necessary.
2557 * This process logically involves adjusting tp->rcv_wnd as data
2558 * is presented to the user (this happens in tcp_usrreq.c,
2559 * case PRU_RCVD). If a FIN has already been received on this
2560 * connection then we just ignore the text.
2561 */
2562 if ((tlen || (thflags & TH_FIN)) &&
2563 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2564 tcp_seq save_start = th->th_seq;
2565 m_adj(m, drop_hdrlen); /* delayed header drop */
2566 /*
2567 * Insert segment which includes th into TCP reassembly queue
2568 * with control block tp. Set thflags to whether reassembly now
2569 * includes a segment with FIN. This handles the common case
2570 * inline (segment is the next to be received on an established
2571 * connection, and the queue is empty), avoiding linkage into
2572 * and removal from the queue and repetition of various
2573 * conversions.
2574 * Set DELACK for segments received in order, but ack
2575 * immediately when segments are out of order (so
2576 * fast retransmit can work).
2577 */
2578 if (th->th_seq == tp->rcv_nxt &&
2579 LIST_EMPTY(&tp->t_segq) &&
2580 TCPS_HAVEESTABLISHED(tp->t_state)) {
2581 if (DELAY_ACK(tp))
2582 tp->t_flags |= TF_DELACK;
2583 else
2584 tp->t_flags |= TF_ACKNOW;
2585 tp->rcv_nxt += tlen;
2586 thflags = th->th_flags & TH_FIN;
2587 TCPSTAT_INC(tcps_rcvpack);
2588 TCPSTAT_ADD(tcps_rcvbyte, tlen);
2589 ND6_HINT(tp);
2590 SOCKBUF_LOCK(&so->so_rcv);
2591 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2592 m_freem(m);
2593 else
2594 sbappendstream_locked(&so->so_rcv, m);
2595 /* NB: sorwakeup_locked() does an implicit unlock. */
2596 sorwakeup_locked(so);
2597 } else {
2598 /*
2599 * XXX: Due to the header drop above "th" is
2600 * theoretically invalid by now. Fortunately
2601 * m_adj() doesn't actually frees any mbufs
2602 * when trimming from the head.
2603 */
2604 thflags = tcp_reass(tp, th, &tlen, m);
2605 tp->t_flags |= TF_ACKNOW;
2606 }
2607 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2608 tcp_update_sack_list(tp, save_start, save_start + tlen);
2609#if 0
2610 /*
2611 * Note the amount of data that peer has sent into
2612 * our window, in order to estimate the sender's
2613 * buffer size.
2614 * XXX: Unused.
2615 */
2616 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2617#endif
2618 } else {
2619 m_freem(m);
2620 thflags &= ~TH_FIN;
2621 }
2622
2623 /*
2624 * If FIN is received ACK the FIN and let the user know
2625 * that the connection is closing.
2626 */
2627 if (thflags & TH_FIN) {
2628 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2629 socantrcvmore(so);
2630 /*
2631 * If connection is half-synchronized
2632 * (ie NEEDSYN flag on) then delay ACK,
2633 * so it may be piggybacked when SYN is sent.
2634 * Otherwise, since we received a FIN then no
2635 * more input can be expected, send ACK now.
2636 */
2637 if (tp->t_flags & TF_NEEDSYN)
2638 tp->t_flags |= TF_DELACK;
2639 else
2640 tp->t_flags |= TF_ACKNOW;
2641 tp->rcv_nxt++;
2642 }
2643 switch (tp->t_state) {
2644
2645 /*
2646 * In SYN_RECEIVED and ESTABLISHED STATES
2647 * enter the CLOSE_WAIT state.
2648 */
2649 case TCPS_SYN_RECEIVED:
2650 tp->t_starttime = ticks;
2651 /* FALLTHROUGH */
2652 case TCPS_ESTABLISHED:
2653 tp->t_state = TCPS_CLOSE_WAIT;
2654 break;
2655
2656 /*
2657 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2658 * enter the CLOSING state.
2659 */
2660 case TCPS_FIN_WAIT_1:
2661 tp->t_state = TCPS_CLOSING;
2662 break;
2663
2664 /*
2665 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2666 * starting the time-wait timer, turning off the other
2667 * standard timers.
2668 */
2669 case TCPS_FIN_WAIT_2:
2670 INP_INFO_WLOCK_ASSERT(&V_tcbinfo);
2671 KASSERT(ti_locked == TI_WLOCKED, ("%s: dodata "
2672 "TCP_FIN_WAIT_2 ti_locked: %d", __func__,
2673 ti_locked));
2674
2675 tcp_twstart(tp);
2676 INP_INFO_WUNLOCK(&V_tcbinfo);
2677 return;
2678 }
2679 }
2680 if (ti_locked == TI_RLOCKED)
2681 INP_INFO_RUNLOCK(&V_tcbinfo);
2682 else if (ti_locked == TI_WLOCKED)
2683 INP_INFO_WUNLOCK(&V_tcbinfo);
2684 else
2685 panic("%s: dodata epilogue ti_locked %d", __func__,
2686 ti_locked);
2687 ti_locked = TI_UNLOCKED;
2688
2689#ifdef TCPDEBUG
2690 if (so->so_options & SO_DEBUG)
2691 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2692 &tcp_savetcp, 0);
2693#endif
2694
2695 /*
2696 * Return any desired output.
2697 */
2698 if (needoutput || (tp->t_flags & TF_ACKNOW))
2699 (void) tcp_output(tp);
2700
2701check_delack:
2702 KASSERT(ti_locked == TI_UNLOCKED, ("%s: check_delack ti_locked %d",
2703 __func__, ti_locked));
2704 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2705 INP_WLOCK_ASSERT(tp->t_inpcb);
2706
2707 if (tp->t_flags & TF_DELACK) {
2708 tp->t_flags &= ~TF_DELACK;
2709 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2710 }
2711 INP_WUNLOCK(tp->t_inpcb);
2712 return;
2713
2714dropafterack:
2715 KASSERT(ti_locked == TI_RLOCKED || ti_locked == TI_WLOCKED,
2716 ("tcp_do_segment: dropafterack ti_locked %d", ti_locked));
2717
2718 /*
2719 * Generate an ACK dropping incoming segment if it occupies
2720 * sequence space, where the ACK reflects our state.
2721 *
2722 * We can now skip the test for the RST flag since all
2723 * paths to this code happen after packets containing
2724 * RST have been dropped.
2725 *
2726 * In the SYN-RECEIVED state, don't send an ACK unless the
2727 * segment we received passes the SYN-RECEIVED ACK test.
2728 * If it fails send a RST. This breaks the loop in the
2729 * "LAND" DoS attack, and also prevents an ACK storm
2730 * between two listening ports that have been sent forged
2731 * SYN segments, each with the source address of the other.
2732 */
2733 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2734 (SEQ_GT(tp->snd_una, th->th_ack) ||
2735 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2736 rstreason = BANDLIM_RST_OPENPORT;
2737 goto dropwithreset;
2738 }
2739#ifdef TCPDEBUG
2740 if (so->so_options & SO_DEBUG)
2741 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2742 &tcp_savetcp, 0);
2743#endif
2744 if (ti_locked == TI_RLOCKED)
2745 INP_INFO_RUNLOCK(&V_tcbinfo);
2746 else if (ti_locked == TI_WLOCKED)
2747 INP_INFO_WUNLOCK(&V_tcbinfo);
2748 else
2749 panic("%s: dropafterack epilogue ti_locked %d", __func__,
2750 ti_locked);
2751 ti_locked = TI_UNLOCKED;
2752
2753 tp->t_flags |= TF_ACKNOW;
2754 (void) tcp_output(tp);
2755 INP_WUNLOCK(tp->t_inpcb);
2756 m_freem(m);
2757 return;
2758
2759dropwithreset:
2760 if (ti_locked == TI_RLOCKED)
2761 INP_INFO_RUNLOCK(&V_tcbinfo);
2762 else if (ti_locked == TI_WLOCKED)
2763 INP_INFO_WUNLOCK(&V_tcbinfo);
2764 else
2765 panic("%s: dropwithreset ti_locked %d", __func__, ti_locked);
2766 ti_locked = TI_UNLOCKED;
2767
2768 if (tp != NULL) {
2769 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2770 INP_WUNLOCK(tp->t_inpcb);
2771 } else
2772 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
2773 return;
2774
2775drop:
2776 if (ti_locked == TI_RLOCKED)
2777 INP_INFO_RUNLOCK(&V_tcbinfo);
2778 else if (ti_locked == TI_WLOCKED)
2779 INP_INFO_WUNLOCK(&V_tcbinfo);
2780#ifdef INVARIANTS
2781 else
2782 INP_INFO_UNLOCK_ASSERT(&V_tcbinfo);
2783#endif
2784 ti_locked = TI_UNLOCKED;
2785
2786 /*
2787 * Drop space held by incoming segment and return.
2788 */
2789#ifdef TCPDEBUG
2790 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2791 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2792 &tcp_savetcp, 0);
2793#endif
2794 if (tp != NULL)
2795 INP_WUNLOCK(tp->t_inpcb);
2796 m_freem(m);
2797}
2798
2799/*
2800 * Issue RST and make ACK acceptable to originator of segment.
2801 * The mbuf must still include the original packet header.
2802 * tp may be NULL.
2803 */
2804static void
2805tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
2806 int tlen, int rstreason)
2807{
2808 struct ip *ip;
2809#ifdef INET6
2810 struct ip6_hdr *ip6;
2811#endif
2812
2813 if (tp != NULL) {
2814 INP_WLOCK_ASSERT(tp->t_inpcb);
2815 }
2816
2817 /* Don't bother if destination was broadcast/multicast. */
2818 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2819 goto drop;
2820#ifdef INET6
2821 if (mtod(m, struct ip *)->ip_v == 6) {
2822 ip6 = mtod(m, struct ip6_hdr *);
2823 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2824 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2825 goto drop;
2826 /* IPv6 anycast check is done at tcp6_input() */
2827 } else
2828#endif
2829 {
2830 ip = mtod(m, struct ip *);
2831 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2832 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2833 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2834 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2835 goto drop;
2836 }
2837
2838 /* Perform bandwidth limiting. */
2839 if (badport_bandlim(rstreason) < 0)
2840 goto drop;
2841
2842 /* tcp_respond consumes the mbuf chain. */
2843 if (th->th_flags & TH_ACK) {
2844 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
2845 th->th_ack, TH_RST);
2846 } else {
2847 if (th->th_flags & TH_SYN)
2848 tlen++;
2849 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2850 (tcp_seq)0, TH_RST|TH_ACK);
2851 }
2852 return;
2853drop:
2854 m_freem(m);
2855}
2856
2857/*
2858 * Parse TCP options and place in tcpopt.
2859 */
2860static void
2861tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
2862{
2863 int opt, optlen;
2864
2865 to->to_flags = 0;
2866 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2867 opt = cp[0];
2868 if (opt == TCPOPT_EOL)
2869 break;
2870 if (opt == TCPOPT_NOP)
2871 optlen = 1;
2872 else {
2873 if (cnt < 2)
2874 break;
2875 optlen = cp[1];
2876 if (optlen < 2 || optlen > cnt)
2877 break;
2878 }
2879 switch (opt) {
2880 case TCPOPT_MAXSEG:
2881 if (optlen != TCPOLEN_MAXSEG)
2882 continue;
2883 if (!(flags & TO_SYN))
2884 continue;
2885 to->to_flags |= TOF_MSS;
2886 bcopy((char *)cp + 2,
2887 (char *)&to->to_mss, sizeof(to->to_mss));
2888 to->to_mss = ntohs(to->to_mss);
2889 break;
2890 case TCPOPT_WINDOW:
2891 if (optlen != TCPOLEN_WINDOW)
2892 continue;
2893 if (!(flags & TO_SYN))
2894 continue;
2895 to->to_flags |= TOF_SCALE;
2896 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
2897 break;
2898 case TCPOPT_TIMESTAMP:
2899 if (optlen != TCPOLEN_TIMESTAMP)
2900 continue;
2901 to->to_flags |= TOF_TS;
2902 bcopy((char *)cp + 2,
2903 (char *)&to->to_tsval, sizeof(to->to_tsval));
2904 to->to_tsval = ntohl(to->to_tsval);
2905 bcopy((char *)cp + 6,
2906 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2907 to->to_tsecr = ntohl(to->to_tsecr);
2908 break;
2909#ifdef TCP_SIGNATURE
2910 /*
2911 * XXX In order to reply to a host which has set the
2912 * TCP_SIGNATURE option in its initial SYN, we have to
2913 * record the fact that the option was observed here
2914 * for the syncache code to perform the correct response.
2915 */
2916 case TCPOPT_SIGNATURE:
2917 if (optlen != TCPOLEN_SIGNATURE)
2918 continue;
2919 to->to_flags |= TOF_SIGNATURE;
2920 to->to_signature = cp + 2;
2921 break;
2922#endif
2923 case TCPOPT_SACK_PERMITTED:
2924 if (optlen != TCPOLEN_SACK_PERMITTED)
2925 continue;
2926 if (!(flags & TO_SYN))
2927 continue;
2928 if (!V_tcp_do_sack)
2929 continue;
2930 to->to_flags |= TOF_SACKPERM;
2931 break;
2932 case TCPOPT_SACK:
2933 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2934 continue;
2935 if (flags & TO_SYN)
2936 continue;
2937 to->to_flags |= TOF_SACK;
2938 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2939 to->to_sacks = cp + 2;
2940 TCPSTAT_INC(tcps_sack_rcv_blocks);
2941 break;
2942 default:
2943 continue;
2944 }
2945 }
2946}
2947
2948/*
2949 * Pull out of band byte out of a segment so
2950 * it doesn't appear in the user's data queue.
2951 * It is still reflected in the segment length for
2952 * sequencing purposes.
2953 */
2954static void
2955tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
2956 int off)
2957{
2958 int cnt = off + th->th_urp - 1;
2959
2960 while (cnt >= 0) {
2961 if (m->m_len > cnt) {
2962 char *cp = mtod(m, caddr_t) + cnt;
2963 struct tcpcb *tp = sototcpcb(so);
2964
2965 INP_WLOCK_ASSERT(tp->t_inpcb);
2966
2967 tp->t_iobc = *cp;
2968 tp->t_oobflags |= TCPOOB_HAVEDATA;
2969 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2970 m->m_len--;
2971 if (m->m_flags & M_PKTHDR)
2972 m->m_pkthdr.len--;
2973 return;
2974 }
2975 cnt -= m->m_len;
2976 m = m->m_next;
2977 if (m == NULL)
2978 break;
2979 }
2980 panic("tcp_pulloutofband");
2981}
2982
2983/*
2984 * Collect new round-trip time estimate
2985 * and update averages and current timeout.
2986 */
2987static void
2988tcp_xmit_timer(struct tcpcb *tp, int rtt)
2989{
2990 int delta;
2991
2992 INP_WLOCK_ASSERT(tp->t_inpcb);
2993
2994 TCPSTAT_INC(tcps_rttupdated);
2995 tp->t_rttupdated++;
2996 if (tp->t_srtt != 0) {
2997 /*
2998 * srtt is stored as fixed point with 5 bits after the
2999 * binary point (i.e., scaled by 8). The following magic
3000 * is equivalent to the smoothing algorithm in rfc793 with
3001 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3002 * point). Adjust rtt to origin 0.
3003 */
3004 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3005 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3006
3007 if ((tp->t_srtt += delta) <= 0)
3008 tp->t_srtt = 1;
3009
3010 /*
3011 * We accumulate a smoothed rtt variance (actually, a
3012 * smoothed mean difference), then set the retransmit
3013 * timer to smoothed rtt + 4 times the smoothed variance.
3014 * rttvar is stored as fixed point with 4 bits after the
3015 * binary point (scaled by 16). The following is
3016 * equivalent to rfc793 smoothing with an alpha of .75
3017 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3018 * rfc793's wired-in beta.
3019 */
3020 if (delta < 0)
3021 delta = -delta;
3022 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3023 if ((tp->t_rttvar += delta) <= 0)
3024 tp->t_rttvar = 1;
3025 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
3026 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3027 } else {
3028 /*
3029 * No rtt measurement yet - use the unsmoothed rtt.
3030 * Set the variance to half the rtt (so our first
3031 * retransmit happens at 3*rtt).
3032 */
3033 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3034 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3035 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
3036 }
3037 tp->t_rtttime = 0;
3038 tp->t_rxtshift = 0;
3039
3040 /*
3041 * the retransmit should happen at rtt + 4 * rttvar.
3042 * Because of the way we do the smoothing, srtt and rttvar
3043 * will each average +1/2 tick of bias. When we compute
3044 * the retransmit timer, we want 1/2 tick of rounding and
3045 * 1 extra tick because of +-1/2 tick uncertainty in the
3046 * firing of the timer. The bias will give us exactly the
3047 * 1.5 tick we need. But, because the bias is
3048 * statistical, we have to test that we don't drop below
3049 * the minimum feasible timer (which is 2 ticks).
3050 */
3051 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3052 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3053
3054 /*
3055 * We received an ack for a packet that wasn't retransmitted;
3056 * it is probably safe to discard any error indications we've
3057 * received recently. This isn't quite right, but close enough
3058 * for now (a route might have failed after we sent a segment,
3059 * and the return path might not be symmetrical).
3060 */
3061 tp->t_softerror = 0;
3062}
3063
3064/*
3065 * Determine a reasonable value for maxseg size.
3066 * If the route is known, check route for mtu.
3067 * If none, use an mss that can be handled on the outgoing
3068 * interface without forcing IP to fragment; if bigger than
3069 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
3070 * to utilize large mbufs. If no route is found, route has no mtu,
3071 * or the destination isn't local, use a default, hopefully conservative
3072 * size (usually 512 or the default IP max size, but no more than the mtu
3073 * of the interface), as we can't discover anything about intervening
3074 * gateways or networks. We also initialize the congestion/slow start
3075 * window to be a single segment if the destination isn't local.
3076 * While looking at the routing entry, we also initialize other path-dependent
3077 * parameters from pre-set or cached values in the routing entry.
3078 *
3079 * Also take into account the space needed for options that we
3080 * send regularly. Make maxseg shorter by that amount to assure
3081 * that we can send maxseg amount of data even when the options
3082 * are present. Store the upper limit of the length of options plus
3083 * data in maxopd.
3084 *
3085 * In case of T/TCP, we call this routine during implicit connection
3086 * setup as well (offer = -1), to initialize maxseg from the cached
3087 * MSS of our peer.
3088 *
3089 * NOTE that this routine is only called when we process an incoming
3090 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
3091 */
3092void
3093tcp_mss_update(struct tcpcb *tp, int offer,
3094 struct hc_metrics_lite *metricptr, int *mtuflags)
3095{
3096 int mss;
3097 u_long maxmtu;
3098 struct inpcb *inp = tp->t_inpcb;
3099 struct hc_metrics_lite metrics;
3100 int origoffer = offer;
3101#ifdef INET6
3102 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3103 size_t min_protoh = isipv6 ?
3104 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3105 sizeof (struct tcpiphdr);
3106#else
3107 const size_t min_protoh = sizeof(struct tcpiphdr);
3108#endif
3109
3110 INP_WLOCK_ASSERT(tp->t_inpcb);
3111
3112 /* Initialize. */
3113#ifdef INET6
3114 if (isipv6) {
3115 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
3116 tp->t_maxopd = tp->t_maxseg = V_tcp_v6mssdflt;
3117 } else
3118#endif
3119 {
3120 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
3121 tp->t_maxopd = tp->t_maxseg = V_tcp_mssdflt;
3122 }
3123
3124 /*
3125 * No route to sender, stay with default mss and return.
3126 */
3127 if (maxmtu == 0) {
3128 /*
3129 * In case we return early we need to initialize metrics
3130 * to a defined state as tcp_hc_get() would do for us
3131 * if there was no cache hit.
3132 */
3133 if (metricptr != NULL)
3134 bzero(metricptr, sizeof(struct hc_metrics_lite));
3135 return;
3136 }
3137
3138 /* What have we got? */
3139 switch (offer) {
3140 case 0:
3141 /*
3142 * Offer == 0 means that there was no MSS on the SYN
3143 * segment, in this case we use tcp_mssdflt as
3144 * already assigned to t_maxopd above.
3145 */
3146 offer = tp->t_maxopd;
3147 break;
3148
3149 case -1:
3150 /*
3151 * Offer == -1 means that we didn't receive SYN yet.
3152 */
3153 /* FALLTHROUGH */
3154
3155 default:
3156 /*
3157 * Prevent DoS attack with too small MSS. Round up
3158 * to at least minmss.
3159 */
3160 offer = max(offer, V_tcp_minmss);
3161 }
3162
3163 /*
3164 * rmx information is now retrieved from tcp_hostcache.
3165 */
3166 tcp_hc_get(&inp->inp_inc, &metrics);
3167 if (metricptr != NULL)
3168 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3169
3170 /*
3171 * If there's a discovered mtu int tcp hostcache, use it
3172 * else, use the link mtu.
3173 */
3174 if (metrics.rmx_mtu)
3175 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3176 else {
3177#ifdef INET6
3178 if (isipv6) {
3179 mss = maxmtu - min_protoh;
3180 if (!V_path_mtu_discovery &&
3181 !in6_localaddr(&inp->in6p_faddr))
3182 mss = min(mss, V_tcp_v6mssdflt);
3183 } else
3184#endif
3185 {
3186 mss = maxmtu - min_protoh;
3187 if (!V_path_mtu_discovery &&
3188 !in_localaddr(inp->inp_faddr))
3189 mss = min(mss, V_tcp_mssdflt);
3190 }
3191 /*
3192 * XXX - The above conditional (mss = maxmtu - min_protoh)
3193 * probably violates the TCP spec.
3194 * The problem is that, since we don't know the
3195 * other end's MSS, we are supposed to use a conservative
3196 * default. But, if we do that, then MTU discovery will
3197 * never actually take place, because the conservative
3198 * default is much less than the MTUs typically seen
3199 * on the Internet today. For the moment, we'll sweep
3200 * this under the carpet.
3201 *
3202 * The conservative default might not actually be a problem
3203 * if the only case this occurs is when sending an initial
3204 * SYN with options and data to a host we've never talked
3205 * to before. Then, they will reply with an MSS value which
3206 * will get recorded and the new parameters should get
3207 * recomputed. For Further Study.
3208 */
3209 }
3210 mss = min(mss, offer);
3211
3212 /*
3213 * Sanity check: make sure that maxopd will be large
3214 * enough to allow some data on segments even if the
3215 * all the option space is used (40bytes). Otherwise
3216 * funny things may happen in tcp_output.
3217 */
3218 mss = max(mss, 64);
3219
3220 /*
3221 * maxopd stores the maximum length of data AND options
3222 * in a segment; maxseg is the amount of data in a normal
3223 * segment. We need to store this value (maxopd) apart
3224 * from maxseg, because now every segment carries options
3225 * and thus we normally have somewhat less data in segments.
3226 */
3227 tp->t_maxopd = mss;
3228
3229 /*
3230 * origoffer==-1 indicates that no segments were received yet.
3231 * In this case we just guess.
3232 */
3233 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
3234 (origoffer == -1 ||
3235 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
3236 mss -= TCPOLEN_TSTAMP_APPA;
3237
3238#if (MCLBYTES & (MCLBYTES - 1)) == 0
3239 if (mss > MCLBYTES)
3240 mss &= ~(MCLBYTES-1);
3241#else
3242 if (mss > MCLBYTES)
3243 mss = mss / MCLBYTES * MCLBYTES;
3244#endif
3245 tp->t_maxseg = mss;
3246}
3247
3248void
3249tcp_mss(struct tcpcb *tp, int offer)
3250{
3251 int rtt, mss;
3252 u_long bufsize;
3253 struct inpcb *inp;
3254 struct socket *so;
3255 struct hc_metrics_lite metrics;
3256 int mtuflags = 0;
3257#ifdef INET6
3258 int isipv6;
3259#endif
3260 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3261
3262 tcp_mss_update(tp, offer, &metrics, &mtuflags);
3263
3264 mss = tp->t_maxseg;
3265 inp = tp->t_inpcb;
3266#ifdef INET6
3267 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3268#endif
3269
3270 /*
3271 * If there's a pipesize, change the socket buffer to that size,
3272 * don't change if sb_hiwat is different than default (then it
3273 * has been changed on purpose with setsockopt).
3274 * Make the socket buffers an integral number of mss units;
3275 * if the mss is larger than the socket buffer, decrease the mss.
3276 */
3277 so = inp->inp_socket;
3278 SOCKBUF_LOCK(&so->so_snd);
3279 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
3280 bufsize = metrics.rmx_sendpipe;
3281 else
3282 bufsize = so->so_snd.sb_hiwat;
3283 if (bufsize < mss)
3284 mss = bufsize;
3285 else {
3286 bufsize = roundup(bufsize, mss);
3287 if (bufsize > sb_max)
3288 bufsize = sb_max;
3289 if (bufsize > so->so_snd.sb_hiwat)
3290 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
3291 }
3292 SOCKBUF_UNLOCK(&so->so_snd);
3293 tp->t_maxseg = mss;
3294
3295 SOCKBUF_LOCK(&so->so_rcv);
3296 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
3297 bufsize = metrics.rmx_recvpipe;
3298 else
3299 bufsize = so->so_rcv.sb_hiwat;
3300 if (bufsize > mss) {
3301 bufsize = roundup(bufsize, mss);
3302 if (bufsize > sb_max)
3303 bufsize = sb_max;
3304 if (bufsize > so->so_rcv.sb_hiwat)
3305 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
3306 }
3307 SOCKBUF_UNLOCK(&so->so_rcv);
3308 /*
3309 * While we're here, check the others too.
3310 */
3311 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
3312 tp->t_srtt = rtt;
3313 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
3314 TCPSTAT_INC(tcps_usedrtt);
3315 if (metrics.rmx_rttvar) {
3316 tp->t_rttvar = metrics.rmx_rttvar;
3317 TCPSTAT_INC(tcps_usedrttvar);
3318 } else {
3319 /* default variation is +- 1 rtt */
3320 tp->t_rttvar =
3321 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
3322 }
3323 TCPT_RANGESET(tp->t_rxtcur,
3324 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
3325 tp->t_rttmin, TCPTV_REXMTMAX);
3326 }
3327 if (metrics.rmx_ssthresh) {
3328 /*
3329 * There's some sort of gateway or interface
3330 * buffer limit on the path. Use this to set
3331 * the slow start threshhold, but set the
3332 * threshold to no less than 2*mss.
3333 */
3334 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
3335 TCPSTAT_INC(tcps_usedssthresh);
3336 }
3337 if (metrics.rmx_bandwidth)
3338 tp->snd_bandwidth = metrics.rmx_bandwidth;
3339
3340 /*
3341 * Set the slow-start flight size depending on whether this
3342 * is a local network or not.
3343 *
3344 * Extend this so we cache the cwnd too and retrieve it here.
3345 * Make cwnd even bigger than RFC3390 suggests but only if we
3346 * have previous experience with the remote host. Be careful
3347 * not make cwnd bigger than remote receive window or our own
3348 * send socket buffer. Maybe put some additional upper bound
3349 * on the retrieved cwnd. Should do incremental updates to
3350 * hostcache when cwnd collapses so next connection doesn't
3351 * overloads the path again.
3352 *
3353 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
3354 * We currently check only in syncache_socket for that.
3355 */
3356#define TCP_METRICS_CWND
3357#ifdef TCP_METRICS_CWND
3358 if (metrics.rmx_cwnd)
3359 tp->snd_cwnd = max(mss,
3360 min(metrics.rmx_cwnd / 2,
3361 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3362 else
3363#endif
3364 if (V_tcp_do_rfc3390)
3365 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3366#ifdef INET6
3367 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3368 (!isipv6 && in_localaddr(inp->inp_faddr)))
3369#else
3370 else if (in_localaddr(inp->inp_faddr))
3371#endif
3372 tp->snd_cwnd = mss * V_ss_fltsz_local;
3373 else
3374 tp->snd_cwnd = mss * V_ss_fltsz;
3375
3376 /* Check the interface for TSO capabilities. */
3377 if (mtuflags & CSUM_TSO)
3378 tp->t_flags |= TF_TSO;
3379}
3380
3381/*
3382 * Determine the MSS option to send on an outgoing SYN.
3383 */
3384int
3385tcp_mssopt(struct in_conninfo *inc)
3386{
3387 int mss = 0;
3388 u_long maxmtu = 0;
3389 u_long thcmtu = 0;
3390 size_t min_protoh;
3391
3392 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3393
3394#ifdef INET6
3395 if (inc->inc_flags & INC_ISIPV6) {
3396 mss = V_tcp_v6mssdflt;
3397 maxmtu = tcp_maxmtu6(inc, NULL);
3398 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3399 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3400 } else
3401#endif
3402 {
3403 mss = V_tcp_mssdflt;
3404 maxmtu = tcp_maxmtu(inc, NULL);
3405 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3406 min_protoh = sizeof(struct tcpiphdr);
3407 }
3408 if (maxmtu && thcmtu)
3409 mss = min(maxmtu, thcmtu) - min_protoh;
3410 else if (maxmtu || thcmtu)
3411 mss = max(maxmtu, thcmtu) - min_protoh;
3412
3413 return (mss);
3414}
3415
3416
3417/*
3418 * On a partial ack arrives, force the retransmission of the
3419 * next unacknowledged segment. Do not clear tp->t_dupacks.
3420 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3421 * be started again.
3422 */
3423static void
3424tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3425{
3426 tcp_seq onxt = tp->snd_nxt;
3427 u_long ocwnd = tp->snd_cwnd;
3428
3429 INP_WLOCK_ASSERT(tp->t_inpcb);
3430
3431 tcp_timer_activate(tp, TT_REXMT, 0);
3432 tp->t_rtttime = 0;
3433 tp->snd_nxt = th->th_ack;
3434 /*
3435 * Set snd_cwnd to one segment beyond acknowledged offset.
3436 * (tp->snd_una has not yet been updated when this function is called.)
3437 */
3438 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3439 tp->t_flags |= TF_ACKNOW;
3440 (void) tcp_output(tp);
3441 tp->snd_cwnd = ocwnd;
3442 if (SEQ_GT(onxt, tp->snd_nxt))
3443 tp->snd_nxt = onxt;
3444 /*
3445 * Partial window deflation. Relies on fact that tp->snd_una
3446 * not updated yet.
3447 */
3448 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3449 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3450 else
3451 tp->snd_cwnd = 0;
3452 tp->snd_cwnd += tp->t_maxseg;
3453}