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: releng/7.4/sys/netinet/tcp_input.c 215926 2010-11-27 03:12:39Z lstewart $");
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_mac.h"
40 #include "opt_tcpdebug.h"
41
42 #include <sys/param.h>
43 #include <sys/kernel.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/proc.h> /* for proc0 declaration */
47 #include <sys/protosw.h>
48 #include <sys/signalvar.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/systm.h>
54
55 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
56
57 #include <vm/uma.h>
58
59 #include <net/if.h>
60 #include <net/route.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
99 static const int tcprexmtthresh = 3;
100
101 struct tcpstat tcpstat;
102 SYSCTL_STRUCT(_net_inet_tcp, TCPCTL_STATS, stats, CTLFLAG_RW,
103 &tcpstat , tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
104
105 int tcp_log_in_vain = 0;
106 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_RW,
107 &tcp_log_in_vain, 0, "Log all incoming TCP segments to closed ports");
108
109 static int blackhole = 0;
110 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_RW,
111 &blackhole, 0, "Do not send RST on segments to closed ports");
112
113 int tcp_delack_enabled = 1;
114 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_RW,
115 &tcp_delack_enabled, 0,
116 "Delay ACK to try and piggyback it onto a data packet");
117
118 static int drop_synfin = 0;
119 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_RW,
120 &drop_synfin, 0, "Drop TCP packets with SYN+FIN set");
121
122 static int tcp_do_rfc3042 = 1;
123 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_RW,
124 &tcp_do_rfc3042, 0, "Enable RFC 3042 (Limited Transmit)");
125
126 int tcp_do_rfc3390 = 1;
127 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_RW,
128 &tcp_do_rfc3390, 0,
129 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
130
131 static int tcp_insecure_rst = 0;
132 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_RW,
133 &tcp_insecure_rst, 0,
134 "Follow the old (insecure) criteria for accepting RST packets");
135
136 int tcp_do_autorcvbuf = 1;
137 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_RW,
138 &tcp_do_autorcvbuf, 0, "Enable automatic receive buffer sizing");
139
140 int tcp_autorcvbuf_inc = 16*1024;
141 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_inc, CTLFLAG_RW,
142 &tcp_autorcvbuf_inc, 0,
143 "Incrementor step size of automatic receive buffer");
144
145 int tcp_autorcvbuf_max = 256*1024;
146 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_RW,
147 &tcp_autorcvbuf_max, 0, "Max size of automatic receive buffer");
148
149 struct inpcbhead tcb;
150 #define tcb6 tcb /* for KAME src sync over BSD*'s */
151 struct inpcbinfo tcbinfo;
152
153 static void tcp_dooptions(struct tcpopt *, u_char *, int, int);
154 static void tcp_do_segment(struct mbuf *, struct tcphdr *,
155 struct socket *, struct tcpcb *, int, int);
156 static void tcp_dropwithreset(struct mbuf *, struct tcphdr *,
157 struct tcpcb *, int, int);
158 static void tcp_pulloutofband(struct socket *,
159 struct tcphdr *, struct mbuf *, int);
160 static void tcp_xmit_timer(struct tcpcb *, int);
161 static void tcp_newreno_partial_ack(struct tcpcb *, struct tcphdr *);
162
163 /* Neighbor Discovery, Neighbor Unreachability Detection Upper layer hint. */
164 #ifdef INET6
165 #define ND6_HINT(tp) \
166 do { \
167 if ((tp) && (tp)->t_inpcb && \
168 ((tp)->t_inpcb->inp_vflag & INP_IPV6) != 0) \
169 nd6_nud_hint(NULL, NULL, 0); \
170 } while (0)
171 #else
172 #define ND6_HINT(tp)
173 #endif
174
175 /*
176 * Indicate whether this ack should be delayed. We can delay the ack if
177 * - there is no delayed ack timer in progress and
178 * - our last ack wasn't a 0-sized window. We never want to delay
179 * the ack that opens up a 0-sized window and
180 * - delayed acks are enabled or
181 * - this is a half-synchronized T/TCP connection.
182 */
183 #define DELAY_ACK(tp) \
184 ((!tcp_timer_active(tp, TT_DELACK) && \
185 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
186 (tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
187
188
189 /*
190 * TCP input handling is split into multiple parts:
191 * tcp6_input is a thin wrapper around tcp_input for the extended
192 * ip6_protox[] call format in ip6_input
193 * tcp_input handles primary segment validation, inpcb lookup and
194 * SYN processing on listen sockets
195 * tcp_do_segment processes the ACK and text of the segment for
196 * establishing, established and closing connections
197 */
198 #ifdef INET6
199 int
200 tcp6_input(struct mbuf **mp, int *offp, int proto)
201 {
202 struct mbuf *m = *mp;
203 struct in6_ifaddr *ia6;
204
205 IP6_EXTHDR_CHECK(m, *offp, sizeof(struct tcphdr), IPPROTO_DONE);
206
207 /*
208 * draft-itojun-ipv6-tcp-to-anycast
209 * better place to put this in?
210 */
211 ia6 = ip6_getdstifaddr(m);
212 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
213 struct ip6_hdr *ip6;
214
215 ip6 = mtod(m, struct ip6_hdr *);
216 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
217 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
218 return IPPROTO_DONE;
219 }
220
221 tcp_input(m, *offp);
222 return IPPROTO_DONE;
223 }
224 #endif
225
226 void
227 tcp_input(struct mbuf *m, int off0)
228 {
229 struct tcphdr *th;
230 struct ip *ip = NULL;
231 struct ipovly *ipov;
232 struct inpcb *inp = NULL;
233 struct tcpcb *tp = NULL;
234 struct socket *so = NULL;
235 u_char *optp = NULL;
236 int optlen = 0;
237 int len, tlen, off;
238 int drop_hdrlen;
239 int thflags;
240 int rstreason = 0; /* For badport_bandlim accounting purposes */
241 #ifdef IPFIREWALL_FORWARD
242 struct m_tag *fwd_tag;
243 #endif
244 #ifdef INET6
245 struct ip6_hdr *ip6 = NULL;
246 int isipv6;
247 #else
248 const void *ip6 = NULL;
249 const int isipv6 = 0;
250 #endif
251 struct tcpopt to; /* options in this segment */
252 char *s = NULL; /* address and port logging */
253
254 #ifdef TCPDEBUG
255 /*
256 * The size of tcp_saveipgen must be the size of the max ip header,
257 * now IPv6.
258 */
259 u_char tcp_saveipgen[IP6_HDR_LEN];
260 struct tcphdr tcp_savetcp;
261 short ostate = 0;
262 #endif
263
264 #ifdef INET6
265 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
266 #endif
267
268 to.to_flags = 0;
269 tcpstat.tcps_rcvtotal++;
270
271 if (isipv6) {
272 #ifdef INET6
273 /* IP6_EXTHDR_CHECK() is already done at tcp6_input(). */
274 ip6 = mtod(m, struct ip6_hdr *);
275 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
276 if (in6_cksum(m, IPPROTO_TCP, off0, tlen)) {
277 tcpstat.tcps_rcvbadsum++;
278 goto drop;
279 }
280 th = (struct tcphdr *)((caddr_t)ip6 + off0);
281
282 /*
283 * Be proactive about unspecified IPv6 address in source.
284 * As we use all-zero to indicate unbounded/unconnected pcb,
285 * unspecified IPv6 address can be used to confuse us.
286 *
287 * Note that packets with unspecified IPv6 destination is
288 * already dropped in ip6_input.
289 */
290 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
291 /* XXX stat */
292 goto drop;
293 }
294 #else
295 th = NULL; /* XXX: Avoid compiler warning. */
296 #endif
297 } else {
298 /*
299 * Get IP and TCP header together in first mbuf.
300 * Note: IP leaves IP header in first mbuf.
301 */
302 if (off0 > sizeof (struct ip)) {
303 ip_stripoptions(m, (struct mbuf *)0);
304 off0 = sizeof(struct ip);
305 }
306 if (m->m_len < sizeof (struct tcpiphdr)) {
307 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
308 == NULL) {
309 tcpstat.tcps_rcvshort++;
310 return;
311 }
312 }
313 ip = mtod(m, struct ip *);
314 ipov = (struct ipovly *)ip;
315 th = (struct tcphdr *)((caddr_t)ip + off0);
316 tlen = ip->ip_len;
317
318 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
319 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
320 th->th_sum = m->m_pkthdr.csum_data;
321 else
322 th->th_sum = in_pseudo(ip->ip_src.s_addr,
323 ip->ip_dst.s_addr,
324 htonl(m->m_pkthdr.csum_data +
325 ip->ip_len +
326 IPPROTO_TCP));
327 th->th_sum ^= 0xffff;
328 #ifdef TCPDEBUG
329 ipov->ih_len = (u_short)tlen;
330 ipov->ih_len = htons(ipov->ih_len);
331 #endif
332 } else {
333 /*
334 * Checksum extended TCP header and data.
335 */
336 len = sizeof (struct ip) + tlen;
337 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
338 ipov->ih_len = (u_short)tlen;
339 ipov->ih_len = htons(ipov->ih_len);
340 th->th_sum = in_cksum(m, len);
341 }
342 if (th->th_sum) {
343 tcpstat.tcps_rcvbadsum++;
344 goto drop;
345 }
346 /* Re-initialization for later version check */
347 ip->ip_v = IPVERSION;
348 }
349
350 /*
351 * Check that TCP offset makes sense,
352 * pull out TCP options and adjust length. XXX
353 */
354 off = th->th_off << 2;
355 if (off < sizeof (struct tcphdr) || off > tlen) {
356 tcpstat.tcps_rcvbadoff++;
357 goto drop;
358 }
359 tlen -= off; /* tlen is used instead of ti->ti_len */
360 if (off > sizeof (struct tcphdr)) {
361 if (isipv6) {
362 #ifdef INET6
363 IP6_EXTHDR_CHECK(m, off0, off, );
364 ip6 = mtod(m, struct ip6_hdr *);
365 th = (struct tcphdr *)((caddr_t)ip6 + off0);
366 #endif
367 } else {
368 if (m->m_len < sizeof(struct ip) + off) {
369 if ((m = m_pullup(m, sizeof (struct ip) + off))
370 == NULL) {
371 tcpstat.tcps_rcvshort++;
372 return;
373 }
374 ip = mtod(m, struct ip *);
375 ipov = (struct ipovly *)ip;
376 th = (struct tcphdr *)((caddr_t)ip + off0);
377 }
378 }
379 optlen = off - sizeof (struct tcphdr);
380 optp = (u_char *)(th + 1);
381 }
382 thflags = th->th_flags;
383
384 /*
385 * Convert TCP protocol specific fields to host format.
386 */
387 th->th_seq = ntohl(th->th_seq);
388 th->th_ack = ntohl(th->th_ack);
389 th->th_win = ntohs(th->th_win);
390 th->th_urp = ntohs(th->th_urp);
391
392 /*
393 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
394 */
395 drop_hdrlen = off0 + off;
396
397 /*
398 * Locate pcb for segment.
399 */
400 INP_INFO_WLOCK(&tcbinfo);
401 findpcb:
402 INP_INFO_WLOCK_ASSERT(&tcbinfo);
403 #ifdef IPFIREWALL_FORWARD
404 /*
405 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
406 */
407 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
408
409 if (fwd_tag != NULL && isipv6 == 0) { /* IPv6 support is not yet */
410 struct sockaddr_in *next_hop;
411
412 next_hop = (struct sockaddr_in *)(fwd_tag+1);
413 /*
414 * Transparently forwarded. Pretend to be the destination.
415 * already got one like this?
416 */
417 inp = in_pcblookup_hash(&tcbinfo,
418 ip->ip_src, th->th_sport,
419 ip->ip_dst, th->th_dport,
420 0, m->m_pkthdr.rcvif);
421 if (!inp) {
422 /* It's new. Try to find the ambushing socket. */
423 inp = in_pcblookup_hash(&tcbinfo,
424 ip->ip_src, th->th_sport,
425 next_hop->sin_addr,
426 next_hop->sin_port ?
427 ntohs(next_hop->sin_port) :
428 th->th_dport,
429 INPLOOKUP_WILDCARD,
430 m->m_pkthdr.rcvif);
431 }
432 /* Remove the tag from the packet. We don't need it anymore. */
433 m_tag_delete(m, fwd_tag);
434 } else
435 #endif /* IPFIREWALL_FORWARD */
436 {
437 if (isipv6) {
438 #ifdef INET6
439 inp = in6_pcblookup_hash(&tcbinfo,
440 &ip6->ip6_src, th->th_sport,
441 &ip6->ip6_dst, th->th_dport,
442 INPLOOKUP_WILDCARD,
443 m->m_pkthdr.rcvif);
444 #endif
445 } else
446 inp = in_pcblookup_hash(&tcbinfo,
447 ip->ip_src, th->th_sport,
448 ip->ip_dst, th->th_dport,
449 INPLOOKUP_WILDCARD,
450 m->m_pkthdr.rcvif);
451 }
452
453 /*
454 * If the INPCB does not exist then all data in the incoming
455 * segment is discarded and an appropriate RST is sent back.
456 * XXX MRT Send RST using which routing table?
457 */
458 if (inp == NULL) {
459 /*
460 * Log communication attempts to ports that are not
461 * in use.
462 */
463 if ((tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
464 tcp_log_in_vain == 2) {
465 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
466 log(LOG_INFO, "%s; %s: Connection attempt "
467 "to closed port\n", s, __func__);
468 }
469 /*
470 * When blackholing do not respond with a RST but
471 * completely ignore the segment and drop it.
472 */
473 if ((blackhole == 1 && (thflags & TH_SYN)) ||
474 blackhole == 2)
475 goto dropunlock;
476
477 rstreason = BANDLIM_RST_CLOSEDPORT;
478 goto dropwithreset;
479 }
480 INP_WLOCK(inp);
481
482 #ifdef IPSEC
483 #ifdef INET6
484 if (isipv6 && ipsec6_in_reject(m, inp)) {
485 ipsec6stat.in_polvio++;
486 goto dropunlock;
487 } else
488 #endif /* INET6 */
489 if (ipsec4_in_reject(m, inp) != 0) {
490 ipsec4stat.in_polvio++;
491 goto dropunlock;
492 }
493 #endif /* IPSEC */
494
495 /*
496 * Check the minimum TTL for socket.
497 */
498 if (inp->inp_ip_minttl != 0) {
499 #ifdef INET6
500 if (isipv6 && inp->inp_ip_minttl > ip6->ip6_hlim)
501 goto dropunlock;
502 else
503 #endif
504 if (inp->inp_ip_minttl > ip->ip_ttl)
505 goto dropunlock;
506 }
507
508 /*
509 * A previous connection in TIMEWAIT state is supposed to catch
510 * stray or duplicate segments arriving late. If this segment
511 * was a legitimate new connection attempt the old INPCB gets
512 * removed and we can try again to find a listening socket.
513 */
514 if (inp->inp_flags & INP_TIMEWAIT) {
515 if (thflags & TH_SYN)
516 tcp_dooptions(&to, optp, optlen, TO_SYN);
517 /*
518 * NB: tcp_twcheck unlocks the INP and frees the mbuf.
519 */
520 if (tcp_twcheck(inp, &to, th, m, tlen))
521 goto findpcb;
522 INP_INFO_WUNLOCK(&tcbinfo);
523 return;
524 }
525 /*
526 * The TCPCB may no longer exist if the connection is winding
527 * down or it is in the CLOSED state. Either way we drop the
528 * segment and send an appropriate response.
529 */
530 tp = intotcpcb(inp);
531 if (tp == NULL || tp->t_state == TCPS_CLOSED) {
532 rstreason = BANDLIM_RST_CLOSEDPORT;
533 goto dropwithreset;
534 }
535
536 #ifdef MAC
537 INP_WLOCK_ASSERT(inp);
538 if (mac_check_inpcb_deliver(inp, m))
539 goto dropunlock;
540 #endif
541 so = inp->inp_socket;
542 KASSERT(so != NULL, ("%s: so == NULL", __func__));
543 #ifdef TCPDEBUG
544 if (so->so_options & SO_DEBUG) {
545 ostate = tp->t_state;
546 if (isipv6) {
547 #ifdef INET6
548 bcopy((char *)ip6, (char *)tcp_saveipgen, sizeof(*ip6));
549 #endif
550 } else
551 bcopy((char *)ip, (char *)tcp_saveipgen, sizeof(*ip));
552 tcp_savetcp = *th;
553 }
554 #endif
555 /*
556 * When the socket is accepting connections (the INPCB is in LISTEN
557 * state) we look into the SYN cache if this is a new connection
558 * attempt or the completion of a previous one.
559 */
560 if (so->so_options & SO_ACCEPTCONN) {
561 struct in_conninfo inc;
562
563 KASSERT(tp->t_state == TCPS_LISTEN, ("%s: so accepting but "
564 "tp not listening", __func__));
565
566 bzero(&inc, sizeof(inc));
567 #ifdef INET6
568 if (isipv6) {
569 inc.inc_flags |= INC_ISIPV6;
570 inc.inc6_faddr = ip6->ip6_src;
571 inc.inc6_laddr = ip6->ip6_dst;
572 } else
573 #endif
574 {
575 inc.inc_faddr = ip->ip_src;
576 inc.inc_laddr = ip->ip_dst;
577 }
578 inc.inc_fport = th->th_sport;
579 inc.inc_lport = th->th_dport;
580 inc.inc_fibnum = so->so_fibnum;
581
582 /*
583 * Check for an existing connection attempt in syncache if
584 * the flag is only ACK. A successful lookup creates a new
585 * socket appended to the listen queue in SYN_RECEIVED state.
586 */
587 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
588 /*
589 * Parse the TCP options here because
590 * syncookies need access to the reflected
591 * timestamp.
592 */
593 tcp_dooptions(&to, optp, optlen, 0);
594 /*
595 * NB: syncache_expand() doesn't unlock
596 * inp and tcpinfo locks.
597 */
598 if (!syncache_expand(&inc, &to, th, &so, m)) {
599 /*
600 * No syncache entry or ACK was not
601 * for our SYN/ACK. Send a RST.
602 * NB: syncache did its own logging
603 * of the failure cause.
604 */
605 rstreason = BANDLIM_RST_OPENPORT;
606 goto dropwithreset;
607 }
608 if (so == NULL) {
609 /*
610 * We completed the 3-way handshake
611 * but could not allocate a socket
612 * either due to memory shortage,
613 * listen queue length limits or
614 * global socket limits. Send RST
615 * or wait and have the remote end
616 * retransmit the ACK for another
617 * try.
618 */
619 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
620 log(LOG_DEBUG, "%s; %s: Listen socket: "
621 "Socket allocation failed due to "
622 "limits or memory shortage, %s\n",
623 s, __func__, (tcp_sc_rst_sock_fail ?
624 "sending RST" : "try again"));
625 if (tcp_sc_rst_sock_fail) {
626 rstreason = BANDLIM_UNLIMITED;
627 goto dropwithreset;
628 } else
629 goto dropunlock;
630 }
631 /*
632 * Socket is created in state SYN_RECEIVED.
633 * Unlock the listen socket, lock the newly
634 * created socket and update the tp variable.
635 */
636 INP_WUNLOCK(inp); /* listen socket */
637 inp = sotoinpcb(so);
638 INP_WLOCK(inp); /* new connection */
639 tp = intotcpcb(inp);
640 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
641 ("%s: ", __func__));
642 /*
643 * Process the segment and the data it
644 * contains. tcp_do_segment() consumes
645 * the mbuf chain and unlocks the inpcb.
646 */
647 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen);
648 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
649 return;
650 }
651 /*
652 * Segment flag validation for new connection attempts:
653 *
654 * Our (SYN|ACK) response was rejected.
655 * Check with syncache and remove entry to prevent
656 * retransmits.
657 *
658 * NB: syncache_chkrst does its own logging of failure
659 * causes.
660 */
661 if (thflags & TH_RST) {
662 syncache_chkrst(&inc, th);
663 goto dropunlock;
664 }
665 /*
666 * We can't do anything without SYN.
667 */
668 if ((thflags & TH_SYN) == 0) {
669 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
670 log(LOG_DEBUG, "%s; %s: Listen socket: "
671 "SYN is missing, segment ignored\n",
672 s, __func__);
673 tcpstat.tcps_badsyn++;
674 goto dropunlock;
675 }
676 /*
677 * (SYN|ACK) is bogus on a listen socket.
678 */
679 if (thflags & TH_ACK) {
680 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
681 log(LOG_DEBUG, "%s; %s: Listen socket: "
682 "SYN|ACK invalid, segment rejected\n",
683 s, __func__);
684 syncache_badack(&inc); /* XXX: Not needed! */
685 tcpstat.tcps_badsyn++;
686 rstreason = BANDLIM_RST_OPENPORT;
687 goto dropwithreset;
688 }
689 /*
690 * If the drop_synfin option is enabled, drop all
691 * segments with both the SYN and FIN bits set.
692 * This prevents e.g. nmap from identifying the
693 * TCP/IP stack.
694 * XXX: Poor reasoning. nmap has other methods
695 * and is constantly refining its stack detection
696 * strategies.
697 * XXX: This is a violation of the TCP specification
698 * and was used by RFC1644.
699 */
700 if ((thflags & TH_FIN) && drop_synfin) {
701 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
702 log(LOG_DEBUG, "%s; %s: Listen socket: "
703 "SYN|FIN segment ignored (based on "
704 "sysctl setting)\n", s, __func__);
705 tcpstat.tcps_badsyn++;
706 goto dropunlock;
707 }
708 /*
709 * Segment's flags are (SYN) or (SYN|FIN).
710 *
711 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
712 * as they do not affect the state of the TCP FSM.
713 * The data pointed to by TH_URG and th_urp is ignored.
714 */
715 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
716 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
717 KASSERT(thflags & (TH_SYN),
718 ("%s: Listen socket: TH_SYN not set", __func__));
719 #ifdef INET6
720 /*
721 * If deprecated address is forbidden,
722 * we do not accept SYN to deprecated interface
723 * address to prevent any new inbound connection from
724 * getting established.
725 * When we do not accept SYN, we send a TCP RST,
726 * with deprecated source address (instead of dropping
727 * it). We compromise it as it is much better for peer
728 * to send a RST, and RST will be the final packet
729 * for the exchange.
730 *
731 * If we do not forbid deprecated addresses, we accept
732 * the SYN packet. RFC2462 does not suggest dropping
733 * SYN in this case.
734 * If we decipher RFC2462 5.5.4, it says like this:
735 * 1. use of deprecated addr with existing
736 * communication is okay - "SHOULD continue to be
737 * used"
738 * 2. use of it with new communication:
739 * (2a) "SHOULD NOT be used if alternate address
740 * with sufficient scope is available"
741 * (2b) nothing mentioned otherwise.
742 * Here we fall into (2b) case as we have no choice in
743 * our source address selection - we must obey the peer.
744 *
745 * The wording in RFC2462 is confusing, and there are
746 * multiple description text for deprecated address
747 * handling - worse, they are not exactly the same.
748 * I believe 5.5.4 is the best one, so we follow 5.5.4.
749 */
750 if (isipv6 && !ip6_use_deprecated) {
751 struct in6_ifaddr *ia6;
752
753 if ((ia6 = ip6_getdstifaddr(m)) &&
754 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
755 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
756 log(LOG_DEBUG, "%s; %s: Listen socket: "
757 "Connection attempt to deprecated "
758 "IPv6 address rejected\n",
759 s, __func__);
760 rstreason = BANDLIM_RST_OPENPORT;
761 goto dropwithreset;
762 }
763 }
764 #endif
765 /*
766 * Basic sanity checks on incoming SYN requests:
767 * Don't respond if the destination is a link layer
768 * broadcast according to RFC1122 4.2.3.10, p. 104.
769 * If it is from this socket it must be forged.
770 * Don't respond if the source or destination is a
771 * global or subnet broad- or multicast address.
772 * Note that it is quite possible to receive unicast
773 * link-layer packets with a broadcast IP address. Use
774 * in_broadcast() to find them.
775 */
776 if (m->m_flags & (M_BCAST|M_MCAST)) {
777 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
778 log(LOG_DEBUG, "%s; %s: Listen socket: "
779 "Connection attempt from broad- or multicast "
780 "link layer address ignored\n", s, __func__);
781 goto dropunlock;
782 }
783 if (isipv6) {
784 #ifdef INET6
785 if (th->th_dport == th->th_sport &&
786 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
787 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
788 log(LOG_DEBUG, "%s; %s: Listen socket: "
789 "Connection attempt to/from self "
790 "ignored\n", s, __func__);
791 goto dropunlock;
792 }
793 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
794 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
795 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
796 log(LOG_DEBUG, "%s; %s: Listen socket: "
797 "Connection attempt from/to multicast "
798 "address ignored\n", s, __func__);
799 goto dropunlock;
800 }
801 #endif
802 } else {
803 if (th->th_dport == th->th_sport &&
804 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
805 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
806 log(LOG_DEBUG, "%s; %s: Listen socket: "
807 "Connection attempt from/to self "
808 "ignored\n", s, __func__);
809 goto dropunlock;
810 }
811 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
812 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
813 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
814 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
815 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
816 log(LOG_DEBUG, "%s; %s: Listen socket: "
817 "Connection attempt from/to broad- "
818 "or multicast address ignored\n",
819 s, __func__);
820 goto dropunlock;
821 }
822 }
823 /*
824 * SYN appears to be valid. Create compressed TCP state
825 * for syncache.
826 */
827 #ifdef TCPDEBUG
828 if (so->so_options & SO_DEBUG)
829 tcp_trace(TA_INPUT, ostate, tp,
830 (void *)tcp_saveipgen, &tcp_savetcp, 0);
831 #endif
832 tcp_dooptions(&to, optp, optlen, TO_SYN);
833 syncache_add(&inc, &to, th, inp, &so, m);
834 /*
835 * Entry added to syncache and mbuf consumed.
836 * Everything already unlocked by syncache_add().
837 */
838 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
839 return;
840 }
841
842 /*
843 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
844 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
845 * the inpcb, and unlocks pcbinfo.
846 */
847 tcp_do_segment(m, th, so, tp, drop_hdrlen, tlen);
848 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
849 return;
850
851 dropwithreset:
852 INP_INFO_WLOCK_ASSERT(&tcbinfo);
853 INP_INFO_WUNLOCK(&tcbinfo);
854
855 if (inp != NULL) {
856 tcp_dropwithreset(m, th, tp, tlen, rstreason);
857 INP_WUNLOCK(inp);
858 } else
859 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
860 m = NULL; /* mbuf chain got consumed. */
861 goto drop;
862
863 dropunlock:
864 INP_INFO_WLOCK_ASSERT(&tcbinfo);
865 if (inp != NULL)
866 INP_WUNLOCK(inp);
867 INP_INFO_WUNLOCK(&tcbinfo);
868
869 drop:
870 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
871 if (s != NULL)
872 free(s, M_TCPLOG);
873 if (m != NULL)
874 m_freem(m);
875 }
876
877 static void
878 tcp_do_segment(struct mbuf *m, struct tcphdr *th, struct socket *so,
879 struct tcpcb *tp, int drop_hdrlen, int tlen)
880 {
881 int thflags, acked, ourfinisacked, needoutput = 0;
882 int headlocked = 1;
883 int rstreason, todrop, win;
884 u_long tiwin;
885 struct tcpopt to;
886
887 #ifdef TCPDEBUG
888 /*
889 * The size of tcp_saveipgen must be the size of the max ip header,
890 * now IPv6.
891 */
892 u_char tcp_saveipgen[IP6_HDR_LEN];
893 struct tcphdr tcp_savetcp;
894 short ostate = 0;
895 #endif
896 thflags = th->th_flags;
897
898 INP_INFO_WLOCK_ASSERT(&tcbinfo);
899 INP_WLOCK_ASSERT(tp->t_inpcb);
900 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
901 __func__));
902 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
903 __func__));
904
905 /*
906 * Segment received on connection.
907 * Reset idle time and keep-alive timer.
908 * XXX: This should be done after segment
909 * validation to ignore broken/spoofed segs.
910 */
911 tp->t_rcvtime = ticks;
912 if (TCPS_HAVEESTABLISHED(tp->t_state))
913 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
914
915 /*
916 * Unscale the window into a 32-bit value.
917 * For the SYN_SENT state the scale is zero.
918 */
919 tiwin = th->th_win << tp->snd_scale;
920
921 /*
922 * Parse options on any incoming segment.
923 */
924 tcp_dooptions(&to, (u_char *)(th + 1),
925 (th->th_off << 2) - sizeof(struct tcphdr),
926 (thflags & TH_SYN) ? TO_SYN : 0);
927
928 /*
929 * If echoed timestamp is later than the current time,
930 * fall back to non RFC1323 RTT calculation. Normalize
931 * timestamp if syncookies were used when this connection
932 * was established.
933 */
934 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
935 to.to_tsecr -= tp->ts_offset;
936 if (TSTMP_GT(to.to_tsecr, ticks))
937 to.to_tsecr = 0;
938 }
939
940 /*
941 * Process options only when we get SYN/ACK back. The SYN case
942 * for incoming connections is handled in tcp_syncache.
943 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
944 * or <SYN,ACK>) segment itself is never scaled.
945 * XXX this is traditional behavior, may need to be cleaned up.
946 */
947 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
948 if ((to.to_flags & TOF_SCALE) &&
949 (tp->t_flags & TF_REQ_SCALE)) {
950 tp->t_flags |= TF_RCVD_SCALE;
951 tp->snd_scale = to.to_wscale;
952 }
953 /*
954 * Initial send window. It will be updated with
955 * the next incoming segment to the scaled value.
956 */
957 tp->snd_wnd = th->th_win;
958 if (to.to_flags & TOF_TS) {
959 tp->t_flags |= TF_RCVD_TSTMP;
960 tp->ts_recent = to.to_tsval;
961 tp->ts_recent_age = ticks;
962 }
963 if (to.to_flags & TOF_MSS)
964 tcp_mss(tp, to.to_mss);
965 if ((tp->t_flags & TF_SACK_PERMIT) &&
966 (to.to_flags & TOF_SACKPERM) == 0)
967 tp->t_flags &= ~TF_SACK_PERMIT;
968 }
969
970 /*
971 * Header prediction: check for the two common cases
972 * of a uni-directional data xfer. If the packet has
973 * no control flags, is in-sequence, the window didn't
974 * change and we're not retransmitting, it's a
975 * candidate. If the length is zero and the ack moved
976 * forward, we're the sender side of the xfer. Just
977 * free the data acked & wake any higher level process
978 * that was blocked waiting for space. If the length
979 * is non-zero and the ack didn't move, we're the
980 * receiver side. If we're getting packets in-order
981 * (the reassembly queue is empty), add the data to
982 * the socket buffer and note that we need a delayed ack.
983 * Make sure that the hidden state-flags are also off.
984 * Since we check for TCPS_ESTABLISHED first, it can only
985 * be TH_NEEDSYN.
986 */
987 if (tp->t_state == TCPS_ESTABLISHED &&
988 th->th_seq == tp->rcv_nxt &&
989 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
990 tp->snd_nxt == tp->snd_max &&
991 tiwin && tiwin == tp->snd_wnd &&
992 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
993 LIST_EMPTY(&tp->t_segq) &&
994 ((to.to_flags & TOF_TS) == 0 ||
995 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
996
997 /*
998 * If last ACK falls within this segment's sequence numbers,
999 * record the timestamp.
1000 * NOTE that the test is modified according to the latest
1001 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1002 */
1003 if ((to.to_flags & TOF_TS) != 0 &&
1004 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1005 tp->ts_recent_age = ticks;
1006 tp->ts_recent = to.to_tsval;
1007 }
1008
1009 if (tlen == 0) {
1010 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1011 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1012 tp->snd_cwnd >= tp->snd_wnd &&
1013 ((!tcp_do_newreno &&
1014 !(tp->t_flags & TF_SACK_PERMIT) &&
1015 tp->t_dupacks < tcprexmtthresh) ||
1016 ((tcp_do_newreno ||
1017 (tp->t_flags & TF_SACK_PERMIT)) &&
1018 !IN_FASTRECOVERY(tp) &&
1019 (to.to_flags & TOF_SACK) == 0 &&
1020 TAILQ_EMPTY(&tp->snd_holes)))) {
1021 KASSERT(headlocked,
1022 ("%s: headlocked", __func__));
1023 INP_INFO_WUNLOCK(&tcbinfo);
1024 headlocked = 0;
1025 /*
1026 * This is a pure ack for outstanding data.
1027 */
1028 ++tcpstat.tcps_predack;
1029 /*
1030 * "bad retransmit" recovery.
1031 */
1032 if (tp->t_rxtshift == 1 &&
1033 (int)(ticks - tp->t_badrxtwin) < 0) {
1034 ++tcpstat.tcps_sndrexmitbad;
1035 tp->snd_cwnd = tp->snd_cwnd_prev;
1036 tp->snd_ssthresh =
1037 tp->snd_ssthresh_prev;
1038 tp->snd_recover = tp->snd_recover_prev;
1039 if (tp->t_flags & TF_WASFRECOVERY)
1040 ENTER_FASTRECOVERY(tp);
1041 tp->snd_nxt = tp->snd_max;
1042 tp->t_badrxtwin = 0;
1043 }
1044
1045 /*
1046 * Recalculate the transmit timer / rtt.
1047 *
1048 * Some boxes send broken timestamp replies
1049 * during the SYN+ACK phase, ignore
1050 * timestamps of 0 or we could calculate a
1051 * huge RTT and blow up the retransmit timer.
1052 */
1053 if ((to.to_flags & TOF_TS) != 0 &&
1054 to.to_tsecr) {
1055 if (!tp->t_rttlow ||
1056 tp->t_rttlow > ticks - to.to_tsecr)
1057 tp->t_rttlow = ticks - to.to_tsecr;
1058 tcp_xmit_timer(tp,
1059 ticks - to.to_tsecr + 1);
1060 } else if (tp->t_rtttime &&
1061 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1062 if (!tp->t_rttlow ||
1063 tp->t_rttlow > ticks - tp->t_rtttime)
1064 tp->t_rttlow = ticks - tp->t_rtttime;
1065 tcp_xmit_timer(tp,
1066 ticks - tp->t_rtttime);
1067 }
1068 tcp_xmit_bandwidth_limit(tp, th->th_ack);
1069 acked = th->th_ack - tp->snd_una;
1070 tcpstat.tcps_rcvackpack++;
1071 tcpstat.tcps_rcvackbyte += acked;
1072 sbdrop(&so->so_snd, acked);
1073 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1074 SEQ_LEQ(th->th_ack, tp->snd_recover))
1075 tp->snd_recover = th->th_ack - 1;
1076 tp->snd_una = th->th_ack;
1077 /*
1078 * Pull snd_wl2 up to prevent seq wrap relative
1079 * to th_ack.
1080 */
1081 tp->snd_wl2 = th->th_ack;
1082 tp->t_dupacks = 0;
1083 m_freem(m);
1084 ND6_HINT(tp); /* Some progress has been made. */
1085
1086 /*
1087 * If all outstanding data are acked, stop
1088 * retransmit timer, otherwise restart timer
1089 * using current (possibly backed-off) value.
1090 * If process is waiting for space,
1091 * wakeup/selwakeup/signal. If data
1092 * are ready to send, let tcp_output
1093 * decide between more output or persist.
1094 */
1095 #ifdef TCPDEBUG
1096 if (so->so_options & SO_DEBUG)
1097 tcp_trace(TA_INPUT, ostate, tp,
1098 (void *)tcp_saveipgen,
1099 &tcp_savetcp, 0);
1100 #endif
1101 if (tp->snd_una == tp->snd_max)
1102 tcp_timer_activate(tp, TT_REXMT, 0);
1103 else if (!tcp_timer_active(tp, TT_PERSIST))
1104 tcp_timer_activate(tp, TT_REXMT,
1105 tp->t_rxtcur);
1106 sowwakeup(so);
1107 if (so->so_snd.sb_cc)
1108 (void) tcp_output(tp);
1109 goto check_delack;
1110 }
1111 } else if (th->th_ack == tp->snd_una &&
1112 tlen <= sbspace(&so->so_rcv)) {
1113 int newsize = 0; /* automatic sockbuf scaling */
1114
1115 KASSERT(headlocked, ("%s: headlocked", __func__));
1116 INP_INFO_WUNLOCK(&tcbinfo);
1117 headlocked = 0;
1118 /*
1119 * This is a pure, in-sequence data packet
1120 * with nothing on the reassembly queue and
1121 * we have enough buffer space to take it.
1122 */
1123 /* Clean receiver SACK report if present */
1124 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1125 tcp_clean_sackreport(tp);
1126 ++tcpstat.tcps_preddat;
1127 tp->rcv_nxt += tlen;
1128 /*
1129 * Pull snd_wl1 up to prevent seq wrap relative to
1130 * th_seq.
1131 */
1132 tp->snd_wl1 = th->th_seq;
1133 /*
1134 * Pull rcv_up up to prevent seq wrap relative to
1135 * rcv_nxt.
1136 */
1137 tp->rcv_up = tp->rcv_nxt;
1138 tcpstat.tcps_rcvpack++;
1139 tcpstat.tcps_rcvbyte += tlen;
1140 ND6_HINT(tp); /* Some progress has been made */
1141 #ifdef TCPDEBUG
1142 if (so->so_options & SO_DEBUG)
1143 tcp_trace(TA_INPUT, ostate, tp,
1144 (void *)tcp_saveipgen, &tcp_savetcp, 0);
1145 #endif
1146 /*
1147 * Automatic sizing of receive socket buffer. Often the send
1148 * buffer size is not optimally adjusted to the actual network
1149 * conditions at hand (delay bandwidth product). Setting the
1150 * buffer size too small limits throughput on links with high
1151 * bandwidth and high delay (eg. trans-continental/oceanic links).
1152 *
1153 * On the receive side the socket buffer memory is only rarely
1154 * used to any significant extent. This allows us to be much
1155 * more aggressive in scaling the receive socket buffer. For
1156 * the case that the buffer space is actually used to a large
1157 * extent and we run out of kernel memory we can simply drop
1158 * the new segments; TCP on the sender will just retransmit it
1159 * later. Setting the buffer size too big may only consume too
1160 * much kernel memory if the application doesn't read() from
1161 * the socket or packet loss or reordering makes use of the
1162 * reassembly queue.
1163 *
1164 * The criteria to step up the receive buffer one notch are:
1165 * 1. the number of bytes received during the time it takes
1166 * one timestamp to be reflected back to us (the RTT);
1167 * 2. received bytes per RTT is within seven eighth of the
1168 * current socket buffer size;
1169 * 3. receive buffer size has not hit maximal automatic size;
1170 *
1171 * This algorithm does one step per RTT at most and only if
1172 * we receive a bulk stream w/o packet losses or reorderings.
1173 * Shrinking the buffer during idle times is not necessary as
1174 * it doesn't consume any memory when idle.
1175 *
1176 * TODO: Only step up if the application is actually serving
1177 * the buffer to better manage the socket buffer resources.
1178 */
1179 if (tcp_do_autorcvbuf &&
1180 to.to_tsecr &&
1181 (so->so_rcv.sb_flags & SB_AUTOSIZE)) {
1182 if (TSTMP_GT(to.to_tsecr, tp->rfbuf_ts) &&
1183 to.to_tsecr - tp->rfbuf_ts < hz) {
1184 if (tp->rfbuf_cnt >
1185 (so->so_rcv.sb_hiwat / 8 * 7) &&
1186 so->so_rcv.sb_hiwat <
1187 tcp_autorcvbuf_max) {
1188 newsize =
1189 min(so->so_rcv.sb_hiwat +
1190 tcp_autorcvbuf_inc,
1191 tcp_autorcvbuf_max);
1192 }
1193 /* Start over with next RTT. */
1194 tp->rfbuf_ts = 0;
1195 tp->rfbuf_cnt = 0;
1196 } else
1197 tp->rfbuf_cnt += tlen; /* add up */
1198 }
1199
1200 /* Add data to socket buffer. */
1201 SOCKBUF_LOCK(&so->so_rcv);
1202 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1203 m_freem(m);
1204 } else {
1205 /*
1206 * Set new socket buffer size.
1207 * Give up when limit is reached.
1208 */
1209 if (newsize)
1210 if (!sbreserve_locked(&so->so_rcv,
1211 newsize, so, NULL))
1212 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1213 m_adj(m, drop_hdrlen); /* delayed header drop */
1214 sbappendstream_locked(&so->so_rcv, m);
1215 }
1216 /* NB: sorwakeup_locked() does an implicit unlock. */
1217 sorwakeup_locked(so);
1218 if (DELAY_ACK(tp)) {
1219 tp->t_flags |= TF_DELACK;
1220 } else {
1221 tp->t_flags |= TF_ACKNOW;
1222 tcp_output(tp);
1223 }
1224 goto check_delack;
1225 }
1226 }
1227
1228 /*
1229 * Calculate amount of space in receive window,
1230 * and then do TCP input processing.
1231 * Receive window is amount of space in rcv queue,
1232 * but not less than advertised window.
1233 */
1234 win = sbspace(&so->so_rcv);
1235 if (win < 0)
1236 win = 0;
1237 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1238
1239 /* Reset receive buffer auto scaling when not in bulk receive mode. */
1240 tp->rfbuf_ts = 0;
1241 tp->rfbuf_cnt = 0;
1242
1243 switch (tp->t_state) {
1244
1245 /*
1246 * If the state is SYN_RECEIVED:
1247 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1248 */
1249 case TCPS_SYN_RECEIVED:
1250 if ((thflags & TH_ACK) &&
1251 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1252 SEQ_GT(th->th_ack, tp->snd_max))) {
1253 rstreason = BANDLIM_RST_OPENPORT;
1254 goto dropwithreset;
1255 }
1256 break;
1257
1258 /*
1259 * If the state is SYN_SENT:
1260 * if seg contains an ACK, but not for our SYN, drop the input.
1261 * if seg contains a RST, then drop the connection.
1262 * if seg does not contain SYN, then drop it.
1263 * Otherwise this is an acceptable SYN segment
1264 * initialize tp->rcv_nxt and tp->irs
1265 * if seg contains ack then advance tp->snd_una
1266 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
1267 * arrange for segment to be acked (eventually)
1268 * continue processing rest of data/controls, beginning with URG
1269 */
1270 case TCPS_SYN_SENT:
1271 if ((thflags & TH_ACK) &&
1272 (SEQ_LEQ(th->th_ack, tp->iss) ||
1273 SEQ_GT(th->th_ack, tp->snd_max))) {
1274 rstreason = BANDLIM_UNLIMITED;
1275 goto dropwithreset;
1276 }
1277 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST))
1278 tp = tcp_drop(tp, ECONNREFUSED);
1279 if (thflags & TH_RST)
1280 goto drop;
1281 if (!(thflags & TH_SYN))
1282 goto drop;
1283
1284 tp->irs = th->th_seq;
1285 tcp_rcvseqinit(tp);
1286 if (thflags & TH_ACK) {
1287 tcpstat.tcps_connects++;
1288 soisconnected(so);
1289 #ifdef MAC
1290 SOCK_LOCK(so);
1291 mac_set_socket_peer_from_mbuf(m, so);
1292 SOCK_UNLOCK(so);
1293 #endif
1294 /* Do window scaling on this connection? */
1295 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1296 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1297 tp->rcv_scale = tp->request_r_scale;
1298 }
1299 tp->rcv_adv += tp->rcv_wnd;
1300 tp->snd_una++; /* SYN is acked */
1301 /*
1302 * If there's data, delay ACK; if there's also a FIN
1303 * ACKNOW will be turned on later.
1304 */
1305 if (DELAY_ACK(tp) && tlen != 0)
1306 tcp_timer_activate(tp, TT_DELACK,
1307 tcp_delacktime);
1308 else
1309 tp->t_flags |= TF_ACKNOW;
1310 /*
1311 * Received <SYN,ACK> in SYN_SENT[*] state.
1312 * Transitions:
1313 * SYN_SENT --> ESTABLISHED
1314 * SYN_SENT* --> FIN_WAIT_1
1315 */
1316 tp->t_starttime = ticks;
1317 if (tp->t_flags & TF_NEEDFIN) {
1318 tp->t_state = TCPS_FIN_WAIT_1;
1319 tp->t_flags &= ~TF_NEEDFIN;
1320 thflags &= ~TH_SYN;
1321 } else {
1322 tp->t_state = TCPS_ESTABLISHED;
1323 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1324 }
1325 } else {
1326 /*
1327 * Received initial SYN in SYN-SENT[*] state =>
1328 * simultaneous open. If segment contains CC option
1329 * and there is a cached CC, apply TAO test.
1330 * If it succeeds, connection is * half-synchronized.
1331 * Otherwise, do 3-way handshake:
1332 * SYN-SENT -> SYN-RECEIVED
1333 * SYN-SENT* -> SYN-RECEIVED*
1334 * If there was no CC option, clear cached CC value.
1335 */
1336 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN);
1337 tcp_timer_activate(tp, TT_REXMT, 0);
1338 tp->t_state = TCPS_SYN_RECEIVED;
1339 }
1340
1341 KASSERT(headlocked, ("%s: trimthenstep6: head not locked",
1342 __func__));
1343 INP_WLOCK_ASSERT(tp->t_inpcb);
1344
1345 /*
1346 * Advance th->th_seq to correspond to first data byte.
1347 * If data, trim to stay within window,
1348 * dropping FIN if necessary.
1349 */
1350 th->th_seq++;
1351 if (tlen > tp->rcv_wnd) {
1352 todrop = tlen - tp->rcv_wnd;
1353 m_adj(m, -todrop);
1354 tlen = tp->rcv_wnd;
1355 thflags &= ~TH_FIN;
1356 tcpstat.tcps_rcvpackafterwin++;
1357 tcpstat.tcps_rcvbyteafterwin += todrop;
1358 }
1359 tp->snd_wl1 = th->th_seq - 1;
1360 tp->rcv_up = th->th_seq;
1361 /*
1362 * Client side of transaction: already sent SYN and data.
1363 * If the remote host used T/TCP to validate the SYN,
1364 * our data will be ACK'd; if so, enter normal data segment
1365 * processing in the middle of step 5, ack processing.
1366 * Otherwise, goto step 6.
1367 */
1368 if (thflags & TH_ACK)
1369 goto process_ACK;
1370
1371 goto step6;
1372
1373 /*
1374 * If the state is LAST_ACK or CLOSING or TIME_WAIT:
1375 * do normal processing.
1376 *
1377 * NB: Leftover from RFC1644 T/TCP. Cases to be reused later.
1378 */
1379 case TCPS_LAST_ACK:
1380 case TCPS_CLOSING:
1381 break; /* continue normal processing */
1382 }
1383
1384 /*
1385 * States other than LISTEN or SYN_SENT.
1386 * First check the RST flag and sequence number since reset segments
1387 * are exempt from the timestamp and connection count tests. This
1388 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
1389 * below which allowed reset segments in half the sequence space
1390 * to fall though and be processed (which gives forged reset
1391 * segments with a random sequence number a 50 percent chance of
1392 * killing a connection).
1393 * Then check timestamp, if present.
1394 * Then check the connection count, if present.
1395 * Then check that at least some bytes of segment are within
1396 * receive window. If segment begins before rcv_nxt,
1397 * drop leading data (and SYN); if nothing left, just ack.
1398 *
1399 *
1400 * If the RST bit is set, check the sequence number to see
1401 * if this is a valid reset segment.
1402 * RFC 793 page 37:
1403 * In all states except SYN-SENT, all reset (RST) segments
1404 * are validated by checking their SEQ-fields. A reset is
1405 * valid if its sequence number is in the window.
1406 * Note: this does not take into account delayed ACKs, so
1407 * we should test against last_ack_sent instead of rcv_nxt.
1408 * The sequence number in the reset segment is normally an
1409 * echo of our outgoing acknowlegement numbers, but some hosts
1410 * send a reset with the sequence number at the rightmost edge
1411 * of our receive window, and we have to handle this case.
1412 * Note 2: Paul Watson's paper "Slipping in the Window" has shown
1413 * that brute force RST attacks are possible. To combat this,
1414 * we use a much stricter check while in the ESTABLISHED state,
1415 * only accepting RSTs where the sequence number is equal to
1416 * last_ack_sent. In all other states (the states in which a
1417 * RST is more likely), the more permissive check is used.
1418 * If we have multiple segments in flight, the initial reset
1419 * segment sequence numbers will be to the left of last_ack_sent,
1420 * but they will eventually catch up.
1421 * In any case, it never made sense to trim reset segments to
1422 * fit the receive window since RFC 1122 says:
1423 * 4.2.2.12 RST Segment: RFC-793 Section 3.4
1424 *
1425 * A TCP SHOULD allow a received RST segment to include data.
1426 *
1427 * DISCUSSION
1428 * It has been suggested that a RST segment could contain
1429 * ASCII text that encoded and explained the cause of the
1430 * RST. No standard has yet been established for such
1431 * data.
1432 *
1433 * If the reset segment passes the sequence number test examine
1434 * the state:
1435 * SYN_RECEIVED STATE:
1436 * If passive open, return to LISTEN state.
1437 * If active open, inform user that connection was refused.
1438 * ESTABLISHED, FIN_WAIT_1, FIN_WAIT_2, CLOSE_WAIT STATES:
1439 * Inform user that connection was reset, and close tcb.
1440 * CLOSING, LAST_ACK STATES:
1441 * Close the tcb.
1442 * TIME_WAIT STATE:
1443 * Drop the segment - see Stevens, vol. 2, p. 964 and
1444 * RFC 1337.
1445 */
1446 if (thflags & TH_RST) {
1447 if (SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1448 SEQ_LEQ(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
1449 switch (tp->t_state) {
1450
1451 case TCPS_SYN_RECEIVED:
1452 so->so_error = ECONNREFUSED;
1453 goto close;
1454
1455 case TCPS_ESTABLISHED:
1456 if (tcp_insecure_rst == 0 &&
1457 !(SEQ_GEQ(th->th_seq, tp->rcv_nxt - 1) &&
1458 SEQ_LEQ(th->th_seq, tp->rcv_nxt + 1)) &&
1459 !(SEQ_GEQ(th->th_seq, tp->last_ack_sent - 1) &&
1460 SEQ_LEQ(th->th_seq, tp->last_ack_sent + 1))) {
1461 tcpstat.tcps_badrst++;
1462 goto drop;
1463 }
1464 /* FALLTHROUGH */
1465 case TCPS_FIN_WAIT_1:
1466 case TCPS_FIN_WAIT_2:
1467 case TCPS_CLOSE_WAIT:
1468 so->so_error = ECONNRESET;
1469 close:
1470 tp->t_state = TCPS_CLOSED;
1471 tcpstat.tcps_drops++;
1472 KASSERT(headlocked, ("%s: trimthenstep6: "
1473 "tcp_close: head not locked", __func__));
1474 tp = tcp_close(tp);
1475 break;
1476
1477 case TCPS_CLOSING:
1478 case TCPS_LAST_ACK:
1479 KASSERT(headlocked, ("%s: trimthenstep6: "
1480 "tcp_close.2: head not locked", __func__));
1481 tp = tcp_close(tp);
1482 break;
1483 }
1484 }
1485 goto drop;
1486 }
1487
1488 /*
1489 * RFC 1323 PAWS: If we have a timestamp reply on this segment
1490 * and it's less than ts_recent, drop it.
1491 */
1492 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
1493 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
1494
1495 /* Check to see if ts_recent is over 24 days old. */
1496 if ((int)(ticks - tp->ts_recent_age) > TCP_PAWS_IDLE) {
1497 /*
1498 * Invalidate ts_recent. If this segment updates
1499 * ts_recent, the age will be reset later and ts_recent
1500 * will get a valid value. If it does not, setting
1501 * ts_recent to zero will at least satisfy the
1502 * requirement that zero be placed in the timestamp
1503 * echo reply when ts_recent isn't valid. The
1504 * age isn't reset until we get a valid ts_recent
1505 * because we don't want out-of-order segments to be
1506 * dropped when ts_recent is old.
1507 */
1508 tp->ts_recent = 0;
1509 } else {
1510 tcpstat.tcps_rcvduppack++;
1511 tcpstat.tcps_rcvdupbyte += tlen;
1512 tcpstat.tcps_pawsdrop++;
1513 if (tlen)
1514 goto dropafterack;
1515 goto drop;
1516 }
1517 }
1518
1519 /*
1520 * In the SYN-RECEIVED state, validate that the packet belongs to
1521 * this connection before trimming the data to fit the receive
1522 * window. Check the sequence number versus IRS since we know
1523 * the sequence numbers haven't wrapped. This is a partial fix
1524 * for the "LAND" DoS attack.
1525 */
1526 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
1527 rstreason = BANDLIM_RST_OPENPORT;
1528 goto dropwithreset;
1529 }
1530
1531 todrop = tp->rcv_nxt - th->th_seq;
1532 if (todrop > 0) {
1533 if (thflags & TH_SYN) {
1534 thflags &= ~TH_SYN;
1535 th->th_seq++;
1536 if (th->th_urp > 1)
1537 th->th_urp--;
1538 else
1539 thflags &= ~TH_URG;
1540 todrop--;
1541 }
1542 /*
1543 * Following if statement from Stevens, vol. 2, p. 960.
1544 */
1545 if (todrop > tlen
1546 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
1547 /*
1548 * Any valid FIN must be to the left of the window.
1549 * At this point the FIN must be a duplicate or out
1550 * of sequence; drop it.
1551 */
1552 thflags &= ~TH_FIN;
1553
1554 /*
1555 * Send an ACK to resynchronize and drop any data.
1556 * But keep on processing for RST or ACK.
1557 */
1558 tp->t_flags |= TF_ACKNOW;
1559 todrop = tlen;
1560 tcpstat.tcps_rcvduppack++;
1561 tcpstat.tcps_rcvdupbyte += todrop;
1562 } else {
1563 tcpstat.tcps_rcvpartduppack++;
1564 tcpstat.tcps_rcvpartdupbyte += todrop;
1565 }
1566 drop_hdrlen += todrop; /* drop from the top afterwards */
1567 th->th_seq += todrop;
1568 tlen -= todrop;
1569 if (th->th_urp > todrop)
1570 th->th_urp -= todrop;
1571 else {
1572 thflags &= ~TH_URG;
1573 th->th_urp = 0;
1574 }
1575 }
1576
1577 /*
1578 * If new data are received on a connection after the
1579 * user processes are gone, then RST the other end.
1580 */
1581 if ((so->so_state & SS_NOFDREF) &&
1582 tp->t_state > TCPS_CLOSE_WAIT && tlen) {
1583 char *s;
1584
1585 KASSERT(headlocked, ("%s: trimthenstep6: tcp_close.3: head "
1586 "not locked", __func__));
1587 if ((s = tcp_log_addrs(&tp->t_inpcb->inp_inc, th, NULL, NULL))) {
1588 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data after socket "
1589 "was closed, sending RST and removing tcpcb\n",
1590 s, __func__, tcpstates[tp->t_state], tlen);
1591 free(s, M_TCPLOG);
1592 }
1593 tp = tcp_close(tp);
1594 tcpstat.tcps_rcvafterclose++;
1595 rstreason = BANDLIM_UNLIMITED;
1596 goto dropwithreset;
1597 }
1598
1599 /*
1600 * If segment ends after window, drop trailing data
1601 * (and PUSH and FIN); if nothing left, just ACK.
1602 */
1603 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
1604 if (todrop > 0) {
1605 tcpstat.tcps_rcvpackafterwin++;
1606 if (todrop >= tlen) {
1607 tcpstat.tcps_rcvbyteafterwin += tlen;
1608 /*
1609 * If window is closed can only take segments at
1610 * window edge, and have to drop data and PUSH from
1611 * incoming segments. Continue processing, but
1612 * remember to ack. Otherwise, drop segment
1613 * and ack.
1614 */
1615 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
1616 tp->t_flags |= TF_ACKNOW;
1617 tcpstat.tcps_rcvwinprobe++;
1618 } else
1619 goto dropafterack;
1620 } else
1621 tcpstat.tcps_rcvbyteafterwin += todrop;
1622 m_adj(m, -todrop);
1623 tlen -= todrop;
1624 thflags &= ~(TH_PUSH|TH_FIN);
1625 }
1626
1627 /*
1628 * If last ACK falls within this segment's sequence numbers,
1629 * record its timestamp.
1630 * NOTE:
1631 * 1) That the test incorporates suggestions from the latest
1632 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1633 * 2) That updating only on newer timestamps interferes with
1634 * our earlier PAWS tests, so this check should be solely
1635 * predicated on the sequence space of this segment.
1636 * 3) That we modify the segment boundary check to be
1637 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
1638 * instead of RFC1323's
1639 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
1640 * This modified check allows us to overcome RFC1323's
1641 * limitations as described in Stevens TCP/IP Illustrated
1642 * Vol. 2 p.869. In such cases, we can still calculate the
1643 * RTT correctly when RCV.NXT == Last.ACK.Sent.
1644 */
1645 if ((to.to_flags & TOF_TS) != 0 &&
1646 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
1647 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
1648 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
1649 tp->ts_recent_age = ticks;
1650 tp->ts_recent = to.to_tsval;
1651 }
1652
1653 /*
1654 * If a SYN is in the window, then this is an
1655 * error and we send an RST and drop the connection.
1656 */
1657 if (thflags & TH_SYN) {
1658 KASSERT(headlocked, ("%s: tcp_drop: trimthenstep6: "
1659 "head not locked", __func__));
1660 tp = tcp_drop(tp, ECONNRESET);
1661 rstreason = BANDLIM_UNLIMITED;
1662 goto drop;
1663 }
1664
1665 /*
1666 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
1667 * flag is on (half-synchronized state), then queue data for
1668 * later processing; else drop segment and return.
1669 */
1670 if ((thflags & TH_ACK) == 0) {
1671 if (tp->t_state == TCPS_SYN_RECEIVED ||
1672 (tp->t_flags & TF_NEEDSYN))
1673 goto step6;
1674 else if (tp->t_flags & TF_ACKNOW)
1675 goto dropafterack;
1676 else
1677 goto drop;
1678 }
1679
1680 /*
1681 * Ack processing.
1682 */
1683 switch (tp->t_state) {
1684
1685 /*
1686 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
1687 * ESTABLISHED state and continue processing.
1688 * The ACK was checked above.
1689 */
1690 case TCPS_SYN_RECEIVED:
1691
1692 tcpstat.tcps_connects++;
1693 soisconnected(so);
1694 /* Do window scaling? */
1695 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1696 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1697 tp->rcv_scale = tp->request_r_scale;
1698 tp->snd_wnd = tiwin;
1699 }
1700 /*
1701 * Make transitions:
1702 * SYN-RECEIVED -> ESTABLISHED
1703 * SYN-RECEIVED* -> FIN-WAIT-1
1704 */
1705 tp->t_starttime = ticks;
1706 if (tp->t_flags & TF_NEEDFIN) {
1707 tp->t_state = TCPS_FIN_WAIT_1;
1708 tp->t_flags &= ~TF_NEEDFIN;
1709 } else {
1710 tp->t_state = TCPS_ESTABLISHED;
1711 tcp_timer_activate(tp, TT_KEEP, tcp_keepidle);
1712 }
1713 /*
1714 * If segment contains data or ACK, will call tcp_reass()
1715 * later; if not, do so now to pass queued data to user.
1716 */
1717 if (tlen == 0 && (thflags & TH_FIN) == 0)
1718 (void) tcp_reass(tp, (struct tcphdr *)0, 0,
1719 (struct mbuf *)0);
1720 tp->snd_wl1 = th->th_seq - 1;
1721 /* FALLTHROUGH */
1722
1723 /*
1724 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
1725 * ACKs. If the ack is in the range
1726 * tp->snd_una < th->th_ack <= tp->snd_max
1727 * then advance tp->snd_una to th->th_ack and drop
1728 * data from the retransmission queue. If this ACK reflects
1729 * more up to date window information we update our window information.
1730 */
1731 case TCPS_ESTABLISHED:
1732 case TCPS_FIN_WAIT_1:
1733 case TCPS_FIN_WAIT_2:
1734 case TCPS_CLOSE_WAIT:
1735 case TCPS_CLOSING:
1736 case TCPS_LAST_ACK:
1737 if (SEQ_GT(th->th_ack, tp->snd_max)) {
1738 tcpstat.tcps_rcvacktoomuch++;
1739 goto dropafterack;
1740 }
1741 if ((tp->t_flags & TF_SACK_PERMIT) &&
1742 ((to.to_flags & TOF_SACK) ||
1743 !TAILQ_EMPTY(&tp->snd_holes)))
1744 tcp_sack_doack(tp, &to, th->th_ack);
1745 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
1746 if (tlen == 0 && tiwin == tp->snd_wnd) {
1747 tcpstat.tcps_rcvdupack++;
1748 /*
1749 * If we have outstanding data (other than
1750 * a window probe), this is a completely
1751 * duplicate ack (ie, window info didn't
1752 * change), the ack is the biggest we've
1753 * seen and we've seen exactly our rexmt
1754 * threshhold of them, assume a packet
1755 * has been dropped and retransmit it.
1756 * Kludge snd_nxt & the congestion
1757 * window so we send only this one
1758 * packet.
1759 *
1760 * We know we're losing at the current
1761 * window size so do congestion avoidance
1762 * (set ssthresh to half the current window
1763 * and pull our congestion window back to
1764 * the new ssthresh).
1765 *
1766 * Dup acks mean that packets have left the
1767 * network (they're now cached at the receiver)
1768 * so bump cwnd by the amount in the receiver
1769 * to keep a constant cwnd packets in the
1770 * network.
1771 */
1772 if (!tcp_timer_active(tp, TT_REXMT) ||
1773 th->th_ack != tp->snd_una)
1774 tp->t_dupacks = 0;
1775 else if (++tp->t_dupacks > tcprexmtthresh ||
1776 ((tcp_do_newreno ||
1777 (tp->t_flags & TF_SACK_PERMIT)) &&
1778 IN_FASTRECOVERY(tp))) {
1779 if ((tp->t_flags & TF_SACK_PERMIT) &&
1780 IN_FASTRECOVERY(tp)) {
1781 int awnd;
1782
1783 /*
1784 * Compute the amount of data in flight first.
1785 * We can inject new data into the pipe iff
1786 * we have less than 1/2 the original window's
1787 * worth of data in flight.
1788 */
1789 awnd = (tp->snd_nxt - tp->snd_fack) +
1790 tp->sackhint.sack_bytes_rexmit;
1791 if (awnd < tp->snd_ssthresh) {
1792 tp->snd_cwnd += tp->t_maxseg;
1793 if (tp->snd_cwnd > tp->snd_ssthresh)
1794 tp->snd_cwnd = tp->snd_ssthresh;
1795 }
1796 } else
1797 tp->snd_cwnd += tp->t_maxseg;
1798 (void) tcp_output(tp);
1799 goto drop;
1800 } else if (tp->t_dupacks == tcprexmtthresh) {
1801 tcp_seq onxt = tp->snd_nxt;
1802 u_int win;
1803
1804 /*
1805 * If we're doing sack, check to
1806 * see if we're already in sack
1807 * recovery. If we're not doing sack,
1808 * check to see if we're in newreno
1809 * recovery.
1810 */
1811 if (tp->t_flags & TF_SACK_PERMIT) {
1812 if (IN_FASTRECOVERY(tp)) {
1813 tp->t_dupacks = 0;
1814 break;
1815 }
1816 } else if (tcp_do_newreno) {
1817 if (SEQ_LEQ(th->th_ack,
1818 tp->snd_recover)) {
1819 tp->t_dupacks = 0;
1820 break;
1821 }
1822 }
1823 win = min(tp->snd_wnd, tp->snd_cwnd) /
1824 2 / tp->t_maxseg;
1825 if (win < 2)
1826 win = 2;
1827 tp->snd_ssthresh = win * tp->t_maxseg;
1828 ENTER_FASTRECOVERY(tp);
1829 tp->snd_recover = tp->snd_max;
1830 tcp_timer_activate(tp, TT_REXMT, 0);
1831 tp->t_rtttime = 0;
1832 if (tp->t_flags & TF_SACK_PERMIT) {
1833 tcpstat.tcps_sack_recovery_episode++;
1834 tp->sack_newdata = tp->snd_nxt;
1835 tp->snd_cwnd = tp->t_maxseg;
1836 (void) tcp_output(tp);
1837 goto drop;
1838 }
1839 tp->snd_nxt = th->th_ack;
1840 tp->snd_cwnd = tp->t_maxseg;
1841 (void) tcp_output(tp);
1842 KASSERT(tp->snd_limited <= 2,
1843 ("%s: tp->snd_limited too big",
1844 __func__));
1845 tp->snd_cwnd = tp->snd_ssthresh +
1846 tp->t_maxseg *
1847 (tp->t_dupacks - tp->snd_limited);
1848 if (SEQ_GT(onxt, tp->snd_nxt))
1849 tp->snd_nxt = onxt;
1850 goto drop;
1851 } else if (tcp_do_rfc3042) {
1852 u_long oldcwnd = tp->snd_cwnd;
1853 tcp_seq oldsndmax = tp->snd_max;
1854 u_int sent;
1855
1856 KASSERT(tp->t_dupacks == 1 ||
1857 tp->t_dupacks == 2,
1858 ("%s: dupacks not 1 or 2",
1859 __func__));
1860 if (tp->t_dupacks == 1)
1861 tp->snd_limited = 0;
1862 tp->snd_cwnd =
1863 (tp->snd_nxt - tp->snd_una) +
1864 (tp->t_dupacks - tp->snd_limited) *
1865 tp->t_maxseg;
1866 (void) tcp_output(tp);
1867 sent = tp->snd_max - oldsndmax;
1868 if (sent > tp->t_maxseg) {
1869 KASSERT((tp->t_dupacks == 2 &&
1870 tp->snd_limited == 0) ||
1871 (sent == tp->t_maxseg + 1 &&
1872 tp->t_flags & TF_SENTFIN),
1873 ("%s: sent too much",
1874 __func__));
1875 tp->snd_limited = 2;
1876 } else if (sent > 0)
1877 ++tp->snd_limited;
1878 tp->snd_cwnd = oldcwnd;
1879 goto drop;
1880 }
1881 } else
1882 tp->t_dupacks = 0;
1883 break;
1884 }
1885
1886 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
1887 ("%s: th_ack <= snd_una", __func__));
1888
1889 /*
1890 * If the congestion window was inflated to account
1891 * for the other side's cached packets, retract it.
1892 */
1893 if (tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) {
1894 if (IN_FASTRECOVERY(tp)) {
1895 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
1896 if (tp->t_flags & TF_SACK_PERMIT)
1897 tcp_sack_partialack(tp, th);
1898 else
1899 tcp_newreno_partial_ack(tp, th);
1900 } else {
1901 /*
1902 * Out of fast recovery.
1903 * Window inflation should have left us
1904 * with approximately snd_ssthresh
1905 * outstanding data.
1906 * But in case we would be inclined to
1907 * send a burst, better to do it via
1908 * the slow start mechanism.
1909 */
1910 if (SEQ_GT(th->th_ack +
1911 tp->snd_ssthresh,
1912 tp->snd_max))
1913 tp->snd_cwnd = tp->snd_max -
1914 th->th_ack +
1915 tp->t_maxseg;
1916 else
1917 tp->snd_cwnd = tp->snd_ssthresh;
1918 }
1919 }
1920 } else {
1921 if (tp->t_dupacks >= tcprexmtthresh &&
1922 tp->snd_cwnd > tp->snd_ssthresh)
1923 tp->snd_cwnd = tp->snd_ssthresh;
1924 }
1925 tp->t_dupacks = 0;
1926 /*
1927 * If we reach this point, ACK is not a duplicate,
1928 * i.e., it ACKs something we sent.
1929 */
1930 if (tp->t_flags & TF_NEEDSYN) {
1931 /*
1932 * T/TCP: Connection was half-synchronized, and our
1933 * SYN has been ACK'd (so connection is now fully
1934 * synchronized). Go to non-starred state,
1935 * increment snd_una for ACK of SYN, and check if
1936 * we can do window scaling.
1937 */
1938 tp->t_flags &= ~TF_NEEDSYN;
1939 tp->snd_una++;
1940 /* Do window scaling? */
1941 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
1942 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
1943 tp->rcv_scale = tp->request_r_scale;
1944 /* Send window already scaled. */
1945 }
1946 }
1947
1948 process_ACK:
1949 KASSERT(headlocked, ("%s: process_ACK: head not locked",
1950 __func__));
1951 INP_WLOCK_ASSERT(tp->t_inpcb);
1952
1953 acked = th->th_ack - tp->snd_una;
1954 tcpstat.tcps_rcvackpack++;
1955 tcpstat.tcps_rcvackbyte += acked;
1956
1957 /*
1958 * If we just performed our first retransmit, and the ACK
1959 * arrives within our recovery window, then it was a mistake
1960 * to do the retransmit in the first place. Recover our
1961 * original cwnd and ssthresh, and proceed to transmit where
1962 * we left off.
1963 */
1964 if (tp->t_rxtshift == 1 && (int)(ticks - tp->t_badrxtwin) < 0) {
1965 ++tcpstat.tcps_sndrexmitbad;
1966 tp->snd_cwnd = tp->snd_cwnd_prev;
1967 tp->snd_ssthresh = tp->snd_ssthresh_prev;
1968 tp->snd_recover = tp->snd_recover_prev;
1969 if (tp->t_flags & TF_WASFRECOVERY)
1970 ENTER_FASTRECOVERY(tp);
1971 tp->snd_nxt = tp->snd_max;
1972 tp->t_badrxtwin = 0; /* XXX probably not required */
1973 }
1974
1975 /*
1976 * If we have a timestamp reply, update smoothed
1977 * round trip time. If no timestamp is present but
1978 * transmit timer is running and timed sequence
1979 * number was acked, update smoothed round trip time.
1980 * Since we now have an rtt measurement, cancel the
1981 * timer backoff (cf., Phil Karn's retransmit alg.).
1982 * Recompute the initial retransmit timer.
1983 *
1984 * Some boxes send broken timestamp replies
1985 * during the SYN+ACK phase, ignore
1986 * timestamps of 0 or we could calculate a
1987 * huge RTT and blow up the retransmit timer.
1988 */
1989 if ((to.to_flags & TOF_TS) != 0 &&
1990 to.to_tsecr) {
1991 if (!tp->t_rttlow || tp->t_rttlow > ticks - to.to_tsecr)
1992 tp->t_rttlow = ticks - to.to_tsecr;
1993 tcp_xmit_timer(tp, ticks - to.to_tsecr + 1);
1994 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
1995 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
1996 tp->t_rttlow = ticks - tp->t_rtttime;
1997 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
1998 }
1999 tcp_xmit_bandwidth_limit(tp, th->th_ack);
2000
2001 /*
2002 * If all outstanding data is acked, stop retransmit
2003 * timer and remember to restart (more output or persist).
2004 * If there is more data to be acked, restart retransmit
2005 * timer, using current (possibly backed-off) value.
2006 */
2007 if (th->th_ack == tp->snd_max) {
2008 tcp_timer_activate(tp, TT_REXMT, 0);
2009 needoutput = 1;
2010 } else if (!tcp_timer_active(tp, TT_PERSIST))
2011 tcp_timer_activate(tp, TT_REXMT, tp->t_rxtcur);
2012
2013 /*
2014 * If no data (only SYN) was ACK'd,
2015 * skip rest of ACK processing.
2016 */
2017 if (acked == 0)
2018 goto step6;
2019
2020 /*
2021 * When new data is acked, open the congestion window.
2022 * If the window gives us less than ssthresh packets
2023 * in flight, open exponentially (maxseg per packet).
2024 * Otherwise open linearly: maxseg per window
2025 * (maxseg^2 / cwnd per packet).
2026 * If cwnd > maxseg^2, fix the cwnd increment at 1 byte
2027 * to avoid capping cwnd (as suggested in RFC2581).
2028 */
2029 if ((!tcp_do_newreno && !(tp->t_flags & TF_SACK_PERMIT)) ||
2030 !IN_FASTRECOVERY(tp)) {
2031 u_int cw = tp->snd_cwnd;
2032 u_int incr = tp->t_maxseg;
2033 if (cw > tp->snd_ssthresh)
2034 incr = max((incr * incr / cw), 1);
2035 tp->snd_cwnd = min(cw+incr, TCP_MAXWIN<<tp->snd_scale);
2036 }
2037 SOCKBUF_LOCK(&so->so_snd);
2038 if (acked > so->so_snd.sb_cc) {
2039 tp->snd_wnd -= so->so_snd.sb_cc;
2040 sbdrop_locked(&so->so_snd, (int)so->so_snd.sb_cc);
2041 ourfinisacked = 1;
2042 } else {
2043 sbdrop_locked(&so->so_snd, acked);
2044 tp->snd_wnd -= acked;
2045 ourfinisacked = 0;
2046 }
2047 /* NB: sowwakeup_locked() does an implicit unlock. */
2048 sowwakeup_locked(so);
2049 /* Detect una wraparound. */
2050 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2051 !IN_FASTRECOVERY(tp) &&
2052 SEQ_GT(tp->snd_una, tp->snd_recover) &&
2053 SEQ_LEQ(th->th_ack, tp->snd_recover))
2054 tp->snd_recover = th->th_ack - 1;
2055 if ((tcp_do_newreno || (tp->t_flags & TF_SACK_PERMIT)) &&
2056 IN_FASTRECOVERY(tp) &&
2057 SEQ_GEQ(th->th_ack, tp->snd_recover))
2058 EXIT_FASTRECOVERY(tp);
2059 tp->snd_una = th->th_ack;
2060 if (tp->t_flags & TF_SACK_PERMIT) {
2061 if (SEQ_GT(tp->snd_una, tp->snd_recover))
2062 tp->snd_recover = tp->snd_una;
2063 }
2064 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2065 tp->snd_nxt = tp->snd_una;
2066
2067 switch (tp->t_state) {
2068
2069 /*
2070 * In FIN_WAIT_1 STATE in addition to the processing
2071 * for the ESTABLISHED state if our FIN is now acknowledged
2072 * then enter FIN_WAIT_2.
2073 */
2074 case TCPS_FIN_WAIT_1:
2075 if (ourfinisacked) {
2076 /*
2077 * If we can't receive any more
2078 * data, then closing user can proceed.
2079 * Starting the timer is contrary to the
2080 * specification, but if we don't get a FIN
2081 * we'll hang forever.
2082 *
2083 * XXXjl:
2084 * we should release the tp also, and use a
2085 * compressed state.
2086 */
2087 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2088 int timeout;
2089
2090 soisdisconnected(so);
2091 timeout = (tcp_fast_finwait2_recycle) ?
2092 tcp_finwait2_timeout : tcp_maxidle;
2093 tcp_timer_activate(tp, TT_2MSL, timeout);
2094 }
2095 tp->t_state = TCPS_FIN_WAIT_2;
2096 }
2097 break;
2098
2099 /*
2100 * In CLOSING STATE in addition to the processing for
2101 * the ESTABLISHED state if the ACK acknowledges our FIN
2102 * then enter the TIME-WAIT state, otherwise ignore
2103 * the segment.
2104 */
2105 case TCPS_CLOSING:
2106 if (ourfinisacked) {
2107 KASSERT(headlocked, ("%s: process_ACK: "
2108 "head not locked", __func__));
2109 tcp_twstart(tp);
2110 INP_INFO_WUNLOCK(&tcbinfo);
2111 headlocked = 0;
2112 m_freem(m);
2113 return;
2114 }
2115 break;
2116
2117 /*
2118 * In LAST_ACK, we may still be waiting for data to drain
2119 * and/or to be acked, as well as for the ack of our FIN.
2120 * If our FIN is now acknowledged, delete the TCB,
2121 * enter the closed state and return.
2122 */
2123 case TCPS_LAST_ACK:
2124 if (ourfinisacked) {
2125 KASSERT(headlocked, ("%s: process_ACK: "
2126 "tcp_close: head not locked", __func__));
2127 tp = tcp_close(tp);
2128 goto drop;
2129 }
2130 break;
2131 }
2132 }
2133
2134 step6:
2135 KASSERT(headlocked, ("%s: step6: head not locked", __func__));
2136 INP_WLOCK_ASSERT(tp->t_inpcb);
2137
2138 /*
2139 * Update window information.
2140 * Don't look at window if no ACK: TAC's send garbage on first SYN.
2141 */
2142 if ((thflags & TH_ACK) &&
2143 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
2144 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
2145 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
2146 /* keep track of pure window updates */
2147 if (tlen == 0 &&
2148 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
2149 tcpstat.tcps_rcvwinupd++;
2150 tp->snd_wnd = tiwin;
2151 tp->snd_wl1 = th->th_seq;
2152 tp->snd_wl2 = th->th_ack;
2153 if (tp->snd_wnd > tp->max_sndwnd)
2154 tp->max_sndwnd = tp->snd_wnd;
2155 needoutput = 1;
2156 }
2157
2158 /*
2159 * Process segments with URG.
2160 */
2161 if ((thflags & TH_URG) && th->th_urp &&
2162 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2163 /*
2164 * This is a kludge, but if we receive and accept
2165 * random urgent pointers, we'll crash in
2166 * soreceive. It's hard to imagine someone
2167 * actually wanting to send this much urgent data.
2168 */
2169 SOCKBUF_LOCK(&so->so_rcv);
2170 if (th->th_urp + so->so_rcv.sb_cc > sb_max) {
2171 th->th_urp = 0; /* XXX */
2172 thflags &= ~TH_URG; /* XXX */
2173 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
2174 goto dodata; /* XXX */
2175 }
2176 /*
2177 * If this segment advances the known urgent pointer,
2178 * then mark the data stream. This should not happen
2179 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
2180 * a FIN has been received from the remote side.
2181 * In these states we ignore the URG.
2182 *
2183 * According to RFC961 (Assigned Protocols),
2184 * the urgent pointer points to the last octet
2185 * of urgent data. We continue, however,
2186 * to consider it to indicate the first octet
2187 * of data past the urgent section as the original
2188 * spec states (in one of two places).
2189 */
2190 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
2191 tp->rcv_up = th->th_seq + th->th_urp;
2192 so->so_oobmark = so->so_rcv.sb_cc +
2193 (tp->rcv_up - tp->rcv_nxt) - 1;
2194 if (so->so_oobmark == 0)
2195 so->so_rcv.sb_state |= SBS_RCVATMARK;
2196 sohasoutofband(so);
2197 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
2198 }
2199 SOCKBUF_UNLOCK(&so->so_rcv);
2200 /*
2201 * Remove out of band data so doesn't get presented to user.
2202 * This can happen independent of advancing the URG pointer,
2203 * but if two URG's are pending at once, some out-of-band
2204 * data may creep in... ick.
2205 */
2206 if (th->th_urp <= (u_long)tlen &&
2207 !(so->so_options & SO_OOBINLINE)) {
2208 /* hdr drop is delayed */
2209 tcp_pulloutofband(so, th, m, drop_hdrlen);
2210 }
2211 } else {
2212 /*
2213 * If no out of band data is expected,
2214 * pull receive urgent pointer along
2215 * with the receive window.
2216 */
2217 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
2218 tp->rcv_up = tp->rcv_nxt;
2219 }
2220 dodata: /* XXX */
2221 KASSERT(headlocked, ("%s: dodata: head not locked", __func__));
2222 INP_WLOCK_ASSERT(tp->t_inpcb);
2223
2224 /*
2225 * Process the segment text, merging it into the TCP sequencing queue,
2226 * and arranging for acknowledgment of receipt if necessary.
2227 * This process logically involves adjusting tp->rcv_wnd as data
2228 * is presented to the user (this happens in tcp_usrreq.c,
2229 * case PRU_RCVD). If a FIN has already been received on this
2230 * connection then we just ignore the text.
2231 */
2232 if ((tlen || (thflags & TH_FIN)) &&
2233 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2234 tcp_seq save_start = th->th_seq;
2235 m_adj(m, drop_hdrlen); /* delayed header drop */
2236 /*
2237 * Insert segment which includes th into TCP reassembly queue
2238 * with control block tp. Set thflags to whether reassembly now
2239 * includes a segment with FIN. This handles the common case
2240 * inline (segment is the next to be received on an established
2241 * connection, and the queue is empty), avoiding linkage into
2242 * and removal from the queue and repetition of various
2243 * conversions.
2244 * Set DELACK for segments received in order, but ack
2245 * immediately when segments are out of order (so
2246 * fast retransmit can work).
2247 */
2248 if (th->th_seq == tp->rcv_nxt &&
2249 LIST_EMPTY(&tp->t_segq) &&
2250 TCPS_HAVEESTABLISHED(tp->t_state)) {
2251 if (DELAY_ACK(tp))
2252 tp->t_flags |= TF_DELACK;
2253 else
2254 tp->t_flags |= TF_ACKNOW;
2255 tp->rcv_nxt += tlen;
2256 thflags = th->th_flags & TH_FIN;
2257 tcpstat.tcps_rcvpack++;
2258 tcpstat.tcps_rcvbyte += tlen;
2259 ND6_HINT(tp);
2260 SOCKBUF_LOCK(&so->so_rcv);
2261 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
2262 m_freem(m);
2263 else
2264 sbappendstream_locked(&so->so_rcv, m);
2265 /* NB: sorwakeup_locked() does an implicit unlock. */
2266 sorwakeup_locked(so);
2267 } else {
2268 /*
2269 * XXX: Due to the header drop above "th" is
2270 * theoretically invalid by now. Fortunately
2271 * m_adj() doesn't actually frees any mbufs
2272 * when trimming from the head.
2273 */
2274 thflags = tcp_reass(tp, th, &tlen, m);
2275 tp->t_flags |= TF_ACKNOW;
2276 }
2277 if (tlen > 0 && (tp->t_flags & TF_SACK_PERMIT))
2278 tcp_update_sack_list(tp, save_start, save_start + tlen);
2279 #if 0
2280 /*
2281 * Note the amount of data that peer has sent into
2282 * our window, in order to estimate the sender's
2283 * buffer size.
2284 * XXX: Unused.
2285 */
2286 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
2287 #endif
2288 } else {
2289 m_freem(m);
2290 thflags &= ~TH_FIN;
2291 }
2292
2293 /*
2294 * If FIN is received ACK the FIN and let the user know
2295 * that the connection is closing.
2296 */
2297 if (thflags & TH_FIN) {
2298 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2299 socantrcvmore(so);
2300 /*
2301 * If connection is half-synchronized
2302 * (ie NEEDSYN flag on) then delay ACK,
2303 * so it may be piggybacked when SYN is sent.
2304 * Otherwise, since we received a FIN then no
2305 * more input can be expected, send ACK now.
2306 */
2307 if (tp->t_flags & TF_NEEDSYN)
2308 tp->t_flags |= TF_DELACK;
2309 else
2310 tp->t_flags |= TF_ACKNOW;
2311 tp->rcv_nxt++;
2312 }
2313 switch (tp->t_state) {
2314
2315 /*
2316 * In SYN_RECEIVED and ESTABLISHED STATES
2317 * enter the CLOSE_WAIT state.
2318 */
2319 case TCPS_SYN_RECEIVED:
2320 tp->t_starttime = ticks;
2321 /* FALLTHROUGH */
2322 case TCPS_ESTABLISHED:
2323 tp->t_state = TCPS_CLOSE_WAIT;
2324 break;
2325
2326 /*
2327 * If still in FIN_WAIT_1 STATE FIN has not been acked so
2328 * enter the CLOSING state.
2329 */
2330 case TCPS_FIN_WAIT_1:
2331 tp->t_state = TCPS_CLOSING;
2332 break;
2333
2334 /*
2335 * In FIN_WAIT_2 state enter the TIME_WAIT state,
2336 * starting the time-wait timer, turning off the other
2337 * standard timers.
2338 */
2339 case TCPS_FIN_WAIT_2:
2340 KASSERT(headlocked == 1, ("%s: dodata: "
2341 "TCP_FIN_WAIT_2: head not locked", __func__));
2342 tcp_twstart(tp);
2343 INP_INFO_WUNLOCK(&tcbinfo);
2344 return;
2345 }
2346 }
2347 INP_INFO_WUNLOCK(&tcbinfo);
2348 headlocked = 0;
2349 #ifdef TCPDEBUG
2350 if (so->so_options & SO_DEBUG)
2351 tcp_trace(TA_INPUT, ostate, tp, (void *)tcp_saveipgen,
2352 &tcp_savetcp, 0);
2353 #endif
2354
2355 /*
2356 * Return any desired output.
2357 */
2358 if (needoutput || (tp->t_flags & TF_ACKNOW))
2359 (void) tcp_output(tp);
2360
2361 check_delack:
2362 KASSERT(headlocked == 0, ("%s: check_delack: head locked",
2363 __func__));
2364 INP_INFO_UNLOCK_ASSERT(&tcbinfo);
2365 INP_WLOCK_ASSERT(tp->t_inpcb);
2366 if (tp->t_flags & TF_DELACK) {
2367 tp->t_flags &= ~TF_DELACK;
2368 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
2369 }
2370 INP_WUNLOCK(tp->t_inpcb);
2371 return;
2372
2373 dropafterack:
2374 KASSERT(headlocked, ("%s: dropafterack: head not locked", __func__));
2375 /*
2376 * Generate an ACK dropping incoming segment if it occupies
2377 * sequence space, where the ACK reflects our state.
2378 *
2379 * We can now skip the test for the RST flag since all
2380 * paths to this code happen after packets containing
2381 * RST have been dropped.
2382 *
2383 * In the SYN-RECEIVED state, don't send an ACK unless the
2384 * segment we received passes the SYN-RECEIVED ACK test.
2385 * If it fails send a RST. This breaks the loop in the
2386 * "LAND" DoS attack, and also prevents an ACK storm
2387 * between two listening ports that have been sent forged
2388 * SYN segments, each with the source address of the other.
2389 */
2390 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
2391 (SEQ_GT(tp->snd_una, th->th_ack) ||
2392 SEQ_GT(th->th_ack, tp->snd_max)) ) {
2393 rstreason = BANDLIM_RST_OPENPORT;
2394 goto dropwithreset;
2395 }
2396 #ifdef TCPDEBUG
2397 if (so->so_options & SO_DEBUG)
2398 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2399 &tcp_savetcp, 0);
2400 #endif
2401 KASSERT(headlocked, ("%s: headlocked should be 1", __func__));
2402 INP_INFO_WUNLOCK(&tcbinfo);
2403 tp->t_flags |= TF_ACKNOW;
2404 (void) tcp_output(tp);
2405 INP_WUNLOCK(tp->t_inpcb);
2406 m_freem(m);
2407 return;
2408
2409 dropwithreset:
2410 KASSERT(headlocked, ("%s: dropwithreset: head not locked", __func__));
2411 INP_INFO_WUNLOCK(&tcbinfo);
2412
2413 if (tp != NULL) {
2414 tcp_dropwithreset(m, th, tp, tlen, rstreason);
2415 INP_WUNLOCK(tp->t_inpcb);
2416 } else
2417 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
2418 return;
2419
2420 drop:
2421 /*
2422 * Drop space held by incoming segment and return.
2423 */
2424 #ifdef TCPDEBUG
2425 if (tp == NULL || (tp->t_inpcb->inp_socket->so_options & SO_DEBUG))
2426 tcp_trace(TA_DROP, ostate, tp, (void *)tcp_saveipgen,
2427 &tcp_savetcp, 0);
2428 #endif
2429 if (tp != NULL)
2430 INP_WUNLOCK(tp->t_inpcb);
2431 if (headlocked)
2432 INP_INFO_WUNLOCK(&tcbinfo);
2433 m_freem(m);
2434 }
2435
2436 /*
2437 * Issue RST and make ACK acceptable to originator of segment.
2438 * The mbuf must still include the original packet header.
2439 * tp may be NULL.
2440 */
2441 static void
2442 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
2443 int tlen, int rstreason)
2444 {
2445 struct ip *ip;
2446 #ifdef INET6
2447 struct ip6_hdr *ip6;
2448 #endif
2449
2450 if (tp != NULL) {
2451 INP_WLOCK_ASSERT(tp->t_inpcb);
2452 }
2453
2454 /* Don't bother if destination was broadcast/multicast. */
2455 if ((th->th_flags & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
2456 goto drop;
2457 #ifdef INET6
2458 if (mtod(m, struct ip *)->ip_v == 6) {
2459 ip6 = mtod(m, struct ip6_hdr *);
2460 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
2461 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
2462 goto drop;
2463 /* IPv6 anycast check is done at tcp6_input() */
2464 } else
2465 #endif
2466 {
2467 ip = mtod(m, struct ip *);
2468 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
2469 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
2470 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
2471 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
2472 goto drop;
2473 }
2474
2475 /* Perform bandwidth limiting. */
2476 if (badport_bandlim(rstreason) < 0)
2477 goto drop;
2478
2479 /* tcp_respond consumes the mbuf chain. */
2480 if (th->th_flags & TH_ACK) {
2481 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
2482 th->th_ack, TH_RST);
2483 } else {
2484 if (th->th_flags & TH_SYN)
2485 tlen++;
2486 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
2487 (tcp_seq)0, TH_RST|TH_ACK);
2488 }
2489 return;
2490 drop:
2491 m_freem(m);
2492 }
2493
2494 /*
2495 * Parse TCP options and place in tcpopt.
2496 */
2497 static void
2498 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
2499 {
2500 int opt, optlen;
2501
2502 to->to_flags = 0;
2503 for (; cnt > 0; cnt -= optlen, cp += optlen) {
2504 opt = cp[0];
2505 if (opt == TCPOPT_EOL)
2506 break;
2507 if (opt == TCPOPT_NOP)
2508 optlen = 1;
2509 else {
2510 if (cnt < 2)
2511 break;
2512 optlen = cp[1];
2513 if (optlen < 2 || optlen > cnt)
2514 break;
2515 }
2516 switch (opt) {
2517 case TCPOPT_MAXSEG:
2518 if (optlen != TCPOLEN_MAXSEG)
2519 continue;
2520 if (!(flags & TO_SYN))
2521 continue;
2522 to->to_flags |= TOF_MSS;
2523 bcopy((char *)cp + 2,
2524 (char *)&to->to_mss, sizeof(to->to_mss));
2525 to->to_mss = ntohs(to->to_mss);
2526 break;
2527 case TCPOPT_WINDOW:
2528 if (optlen != TCPOLEN_WINDOW)
2529 continue;
2530 if (!(flags & TO_SYN))
2531 continue;
2532 to->to_flags |= TOF_SCALE;
2533 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
2534 break;
2535 case TCPOPT_TIMESTAMP:
2536 if (optlen != TCPOLEN_TIMESTAMP)
2537 continue;
2538 to->to_flags |= TOF_TS;
2539 bcopy((char *)cp + 2,
2540 (char *)&to->to_tsval, sizeof(to->to_tsval));
2541 to->to_tsval = ntohl(to->to_tsval);
2542 bcopy((char *)cp + 6,
2543 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
2544 to->to_tsecr = ntohl(to->to_tsecr);
2545 break;
2546 #ifdef TCP_SIGNATURE
2547 /*
2548 * XXX In order to reply to a host which has set the
2549 * TCP_SIGNATURE option in its initial SYN, we have to
2550 * record the fact that the option was observed here
2551 * for the syncache code to perform the correct response.
2552 */
2553 case TCPOPT_SIGNATURE:
2554 if (optlen != TCPOLEN_SIGNATURE)
2555 continue;
2556 to->to_flags |= TOF_SIGNATURE;
2557 to->to_signature = cp + 2;
2558 break;
2559 #endif
2560 case TCPOPT_SACK_PERMITTED:
2561 if (optlen != TCPOLEN_SACK_PERMITTED)
2562 continue;
2563 if (!(flags & TO_SYN))
2564 continue;
2565 if (!tcp_do_sack)
2566 continue;
2567 to->to_flags |= TOF_SACKPERM;
2568 break;
2569 case TCPOPT_SACK:
2570 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
2571 continue;
2572 if (flags & TO_SYN)
2573 continue;
2574 to->to_flags |= TOF_SACK;
2575 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
2576 to->to_sacks = cp + 2;
2577 tcpstat.tcps_sack_rcv_blocks++;
2578 break;
2579 default:
2580 continue;
2581 }
2582 }
2583 }
2584
2585 /*
2586 * Pull out of band byte out of a segment so
2587 * it doesn't appear in the user's data queue.
2588 * It is still reflected in the segment length for
2589 * sequencing purposes.
2590 */
2591 static void
2592 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
2593 int off)
2594 {
2595 int cnt = off + th->th_urp - 1;
2596
2597 while (cnt >= 0) {
2598 if (m->m_len > cnt) {
2599 char *cp = mtod(m, caddr_t) + cnt;
2600 struct tcpcb *tp = sototcpcb(so);
2601
2602 INP_WLOCK_ASSERT(tp->t_inpcb);
2603
2604 tp->t_iobc = *cp;
2605 tp->t_oobflags |= TCPOOB_HAVEDATA;
2606 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
2607 m->m_len--;
2608 if (m->m_flags & M_PKTHDR)
2609 m->m_pkthdr.len--;
2610 return;
2611 }
2612 cnt -= m->m_len;
2613 m = m->m_next;
2614 if (m == NULL)
2615 break;
2616 }
2617 panic("tcp_pulloutofband");
2618 }
2619
2620 /*
2621 * Collect new round-trip time estimate
2622 * and update averages and current timeout.
2623 */
2624 static void
2625 tcp_xmit_timer(struct tcpcb *tp, int rtt)
2626 {
2627 int delta;
2628
2629 INP_WLOCK_ASSERT(tp->t_inpcb);
2630
2631 tcpstat.tcps_rttupdated++;
2632 tp->t_rttupdated++;
2633 if (tp->t_srtt != 0) {
2634 /*
2635 * srtt is stored as fixed point with 5 bits after the
2636 * binary point (i.e., scaled by 8). The following magic
2637 * is equivalent to the smoothing algorithm in rfc793 with
2638 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
2639 * point). Adjust rtt to origin 0.
2640 */
2641 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
2642 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
2643
2644 if ((tp->t_srtt += delta) <= 0)
2645 tp->t_srtt = 1;
2646
2647 /*
2648 * We accumulate a smoothed rtt variance (actually, a
2649 * smoothed mean difference), then set the retransmit
2650 * timer to smoothed rtt + 4 times the smoothed variance.
2651 * rttvar is stored as fixed point with 4 bits after the
2652 * binary point (scaled by 16). The following is
2653 * equivalent to rfc793 smoothing with an alpha of .75
2654 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
2655 * rfc793's wired-in beta.
2656 */
2657 if (delta < 0)
2658 delta = -delta;
2659 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
2660 if ((tp->t_rttvar += delta) <= 0)
2661 tp->t_rttvar = 1;
2662 if (tp->t_rttbest > tp->t_srtt + tp->t_rttvar)
2663 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2664 } else {
2665 /*
2666 * No rtt measurement yet - use the unsmoothed rtt.
2667 * Set the variance to half the rtt (so our first
2668 * retransmit happens at 3*rtt).
2669 */
2670 tp->t_srtt = rtt << TCP_RTT_SHIFT;
2671 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
2672 tp->t_rttbest = tp->t_srtt + tp->t_rttvar;
2673 }
2674 tp->t_rtttime = 0;
2675 tp->t_rxtshift = 0;
2676
2677 /*
2678 * the retransmit should happen at rtt + 4 * rttvar.
2679 * Because of the way we do the smoothing, srtt and rttvar
2680 * will each average +1/2 tick of bias. When we compute
2681 * the retransmit timer, we want 1/2 tick of rounding and
2682 * 1 extra tick because of +-1/2 tick uncertainty in the
2683 * firing of the timer. The bias will give us exactly the
2684 * 1.5 tick we need. But, because the bias is
2685 * statistical, we have to test that we don't drop below
2686 * the minimum feasible timer (which is 2 ticks).
2687 */
2688 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
2689 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
2690
2691 /*
2692 * We received an ack for a packet that wasn't retransmitted;
2693 * it is probably safe to discard any error indications we've
2694 * received recently. This isn't quite right, but close enough
2695 * for now (a route might have failed after we sent a segment,
2696 * and the return path might not be symmetrical).
2697 */
2698 tp->t_softerror = 0;
2699 }
2700
2701 /*
2702 * Determine a reasonable value for maxseg size.
2703 * If the route is known, check route for mtu.
2704 * If none, use an mss that can be handled on the outgoing
2705 * interface without forcing IP to fragment; if bigger than
2706 * an mbuf cluster (MCLBYTES), round down to nearest multiple of MCLBYTES
2707 * to utilize large mbufs. If no route is found, route has no mtu,
2708 * or the destination isn't local, use a default, hopefully conservative
2709 * size (usually 512 or the default IP max size, but no more than the mtu
2710 * of the interface), as we can't discover anything about intervening
2711 * gateways or networks. We also initialize the congestion/slow start
2712 * window to be a single segment if the destination isn't local.
2713 * While looking at the routing entry, we also initialize other path-dependent
2714 * parameters from pre-set or cached values in the routing entry.
2715 *
2716 * Also take into account the space needed for options that we
2717 * send regularly. Make maxseg shorter by that amount to assure
2718 * that we can send maxseg amount of data even when the options
2719 * are present. Store the upper limit of the length of options plus
2720 * data in maxopd.
2721 *
2722 * In case of T/TCP, we call this routine during implicit connection
2723 * setup as well (offer = -1), to initialize maxseg from the cached
2724 * MSS of our peer.
2725 *
2726 * NOTE that this routine is only called when we process an incoming
2727 * segment. Outgoing SYN/ACK MSS settings are handled in tcp_mssopt().
2728 */
2729 void
2730 tcp_mss_update(struct tcpcb *tp, int offer,
2731 struct hc_metrics_lite *metricptr, int *mtuflags)
2732 {
2733 int mss;
2734 u_long maxmtu;
2735 struct inpcb *inp = tp->t_inpcb;
2736 struct hc_metrics_lite metrics;
2737 int origoffer = offer;
2738 #ifdef INET6
2739 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2740 size_t min_protoh = isipv6 ?
2741 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
2742 sizeof (struct tcpiphdr);
2743 #else
2744 const size_t min_protoh = sizeof(struct tcpiphdr);
2745 #endif
2746
2747 INP_WLOCK_ASSERT(tp->t_inpcb);
2748
2749 /* Initialize. */
2750 #ifdef INET6
2751 if (isipv6) {
2752 maxmtu = tcp_maxmtu6(&inp->inp_inc, mtuflags);
2753 tp->t_maxopd = tp->t_maxseg = tcp_v6mssdflt;
2754 } else
2755 #endif
2756 {
2757 maxmtu = tcp_maxmtu(&inp->inp_inc, mtuflags);
2758 tp->t_maxopd = tp->t_maxseg = tcp_mssdflt;
2759 }
2760
2761 /*
2762 * No route to sender, stay with default mss and return.
2763 */
2764 if (maxmtu == 0) {
2765 /*
2766 * In case we return early we need to initialize metrics
2767 * to a defined state as tcp_hc_get() would do for us
2768 * if there was no cache hit.
2769 */
2770 if (metricptr != NULL)
2771 bzero(metricptr, sizeof(struct hc_metrics_lite));
2772 return;
2773 }
2774
2775 /* What have we got? */
2776 switch (offer) {
2777 case 0:
2778 /*
2779 * Offer == 0 means that there was no MSS on the SYN
2780 * segment, in this case we use tcp_mssdflt as
2781 * already assigned to t_maxopd above.
2782 */
2783 offer = tp->t_maxopd;
2784 break;
2785
2786 case -1:
2787 /*
2788 * Offer == -1 means that we didn't receive SYN yet.
2789 */
2790 /* FALLTHROUGH */
2791
2792 default:
2793 /*
2794 * Prevent DoS attack with too small MSS. Round up
2795 * to at least minmss.
2796 */
2797 offer = max(offer, tcp_minmss);
2798 }
2799
2800 /*
2801 * rmx information is now retrieved from tcp_hostcache.
2802 */
2803 tcp_hc_get(&inp->inp_inc, &metrics);
2804 if (metricptr != NULL)
2805 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
2806
2807 /*
2808 * If there's a discovered mtu int tcp hostcache, use it
2809 * else, use the link mtu.
2810 */
2811 if (metrics.rmx_mtu)
2812 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
2813 else {
2814 #ifdef INET6
2815 if (isipv6) {
2816 mss = maxmtu - min_protoh;
2817 if (!path_mtu_discovery &&
2818 !in6_localaddr(&inp->in6p_faddr))
2819 mss = min(mss, tcp_v6mssdflt);
2820 } else
2821 #endif
2822 {
2823 mss = maxmtu - min_protoh;
2824 if (!path_mtu_discovery &&
2825 !in_localaddr(inp->inp_faddr))
2826 mss = min(mss, tcp_mssdflt);
2827 }
2828 /*
2829 * XXX - The above conditional (mss = maxmtu - min_protoh)
2830 * probably violates the TCP spec.
2831 * The problem is that, since we don't know the
2832 * other end's MSS, we are supposed to use a conservative
2833 * default. But, if we do that, then MTU discovery will
2834 * never actually take place, because the conservative
2835 * default is much less than the MTUs typically seen
2836 * on the Internet today. For the moment, we'll sweep
2837 * this under the carpet.
2838 *
2839 * The conservative default might not actually be a problem
2840 * if the only case this occurs is when sending an initial
2841 * SYN with options and data to a host we've never talked
2842 * to before. Then, they will reply with an MSS value which
2843 * will get recorded and the new parameters should get
2844 * recomputed. For Further Study.
2845 */
2846 }
2847 mss = min(mss, offer);
2848
2849 /*
2850 * Sanity check: make sure that maxopd will be large
2851 * enough to allow some data on segments even if the
2852 * all the option space is used (40bytes). Otherwise
2853 * funny things may happen in tcp_output.
2854 */
2855 mss = max(mss, 64);
2856
2857 /*
2858 * maxopd stores the maximum length of data AND options
2859 * in a segment; maxseg is the amount of data in a normal
2860 * segment. We need to store this value (maxopd) apart
2861 * from maxseg, because now every segment carries options
2862 * and thus we normally have somewhat less data in segments.
2863 */
2864 tp->t_maxopd = mss;
2865
2866 /*
2867 * origoffer==-1 indicates that no segments were received yet.
2868 * In this case we just guess.
2869 */
2870 if ((tp->t_flags & (TF_REQ_TSTMP|TF_NOOPT)) == TF_REQ_TSTMP &&
2871 (origoffer == -1 ||
2872 (tp->t_flags & TF_RCVD_TSTMP) == TF_RCVD_TSTMP))
2873 mss -= TCPOLEN_TSTAMP_APPA;
2874
2875 #if (MCLBYTES & (MCLBYTES - 1)) == 0
2876 if (mss > MCLBYTES)
2877 mss &= ~(MCLBYTES-1);
2878 #else
2879 if (mss > MCLBYTES)
2880 mss = mss / MCLBYTES * MCLBYTES;
2881 #endif
2882 tp->t_maxseg = mss;
2883 }
2884
2885 void
2886 tcp_mss(struct tcpcb *tp, int offer)
2887 {
2888 int rtt, mss;
2889 u_long bufsize;
2890 struct inpcb *inp;
2891 struct socket *so;
2892 struct hc_metrics_lite metrics;
2893 int mtuflags = 0;
2894 #ifdef INET6
2895 int isipv6;
2896 #endif
2897 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
2898
2899 tcp_mss_update(tp, offer, &metrics, &mtuflags);
2900
2901 mss = tp->t_maxseg;
2902 inp = tp->t_inpcb;
2903 #ifdef INET6
2904 isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
2905 #endif
2906
2907 /*
2908 * If there's a pipesize, change the socket buffer to that size,
2909 * don't change if sb_hiwat is different than default (then it
2910 * has been changed on purpose with setsockopt).
2911 * Make the socket buffers an integral number of mss units;
2912 * if the mss is larger than the socket buffer, decrease the mss.
2913 */
2914 so = inp->inp_socket;
2915 SOCKBUF_LOCK(&so->so_snd);
2916 if ((so->so_snd.sb_hiwat == tcp_sendspace) && metrics.rmx_sendpipe)
2917 bufsize = metrics.rmx_sendpipe;
2918 else
2919 bufsize = so->so_snd.sb_hiwat;
2920 if (bufsize < mss)
2921 mss = bufsize;
2922 else {
2923 bufsize = roundup(bufsize, mss);
2924 if (bufsize > sb_max)
2925 bufsize = sb_max;
2926 if (bufsize > so->so_snd.sb_hiwat)
2927 (void)sbreserve_locked(&so->so_snd, bufsize, so, NULL);
2928 }
2929 SOCKBUF_UNLOCK(&so->so_snd);
2930 tp->t_maxseg = mss;
2931
2932 SOCKBUF_LOCK(&so->so_rcv);
2933 if ((so->so_rcv.sb_hiwat == tcp_recvspace) && metrics.rmx_recvpipe)
2934 bufsize = metrics.rmx_recvpipe;
2935 else
2936 bufsize = so->so_rcv.sb_hiwat;
2937 if (bufsize > mss) {
2938 bufsize = roundup(bufsize, mss);
2939 if (bufsize > sb_max)
2940 bufsize = sb_max;
2941 if (bufsize > so->so_rcv.sb_hiwat)
2942 (void)sbreserve_locked(&so->so_rcv, bufsize, so, NULL);
2943 }
2944 SOCKBUF_UNLOCK(&so->so_rcv);
2945 /*
2946 * While we're here, check the others too.
2947 */
2948 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
2949 tp->t_srtt = rtt;
2950 tp->t_rttbest = tp->t_srtt + TCP_RTT_SCALE;
2951 tcpstat.tcps_usedrtt++;
2952 if (metrics.rmx_rttvar) {
2953 tp->t_rttvar = metrics.rmx_rttvar;
2954 tcpstat.tcps_usedrttvar++;
2955 } else {
2956 /* default variation is +- 1 rtt */
2957 tp->t_rttvar =
2958 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
2959 }
2960 TCPT_RANGESET(tp->t_rxtcur,
2961 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
2962 tp->t_rttmin, TCPTV_REXMTMAX);
2963 }
2964 if (metrics.rmx_ssthresh) {
2965 /*
2966 * There's some sort of gateway or interface
2967 * buffer limit on the path. Use this to set
2968 * the slow start threshhold, but set the
2969 * threshold to no less than 2*mss.
2970 */
2971 tp->snd_ssthresh = max(2 * mss, metrics.rmx_ssthresh);
2972 tcpstat.tcps_usedssthresh++;
2973 }
2974 if (metrics.rmx_bandwidth)
2975 tp->snd_bandwidth = metrics.rmx_bandwidth;
2976
2977 /*
2978 * Set the slow-start flight size depending on whether this
2979 * is a local network or not.
2980 *
2981 * Extend this so we cache the cwnd too and retrieve it here.
2982 * Make cwnd even bigger than RFC3390 suggests but only if we
2983 * have previous experience with the remote host. Be careful
2984 * not make cwnd bigger than remote receive window or our own
2985 * send socket buffer. Maybe put some additional upper bound
2986 * on the retrieved cwnd. Should do incremental updates to
2987 * hostcache when cwnd collapses so next connection doesn't
2988 * overloads the path again.
2989 *
2990 * XXXAO: Initializing the CWND from the hostcache is broken
2991 * and in its current form not RFC conformant. It is disabled
2992 * until fixed or removed entirely.
2993 *
2994 * RFC3390 says only do this if SYN or SYN/ACK didn't got lost.
2995 * We currently check only in syncache_socket for that.
2996 */
2997 /* #define TCP_METRICS_CWND */
2998 #ifdef TCP_METRICS_CWND
2999 if (metrics.rmx_cwnd)
3000 tp->snd_cwnd = max(mss,
3001 min(metrics.rmx_cwnd / 2,
3002 min(tp->snd_wnd, so->so_snd.sb_hiwat)));
3003 else
3004 #endif
3005 if (tcp_do_rfc3390)
3006 tp->snd_cwnd = min(4 * mss, max(2 * mss, 4380));
3007 #ifdef INET6
3008 else if ((isipv6 && in6_localaddr(&inp->in6p_faddr)) ||
3009 (!isipv6 && in_localaddr(inp->inp_faddr)))
3010 #else
3011 else if (in_localaddr(inp->inp_faddr))
3012 #endif
3013 tp->snd_cwnd = mss * ss_fltsz_local;
3014 else
3015 tp->snd_cwnd = mss * ss_fltsz;
3016
3017 /* Check the interface for TSO capabilities. */
3018 if (mtuflags & CSUM_TSO)
3019 tp->t_flags |= TF_TSO;
3020 }
3021
3022 /*
3023 * Determine the MSS option to send on an outgoing SYN.
3024 */
3025 int
3026 tcp_mssopt(struct in_conninfo *inc)
3027 {
3028 int mss = 0;
3029 u_long maxmtu = 0;
3030 u_long thcmtu = 0;
3031 size_t min_protoh;
3032
3033 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
3034
3035 #ifdef INET6
3036 if (inc->inc_flags & INC_ISIPV6) {
3037 mss = tcp_v6mssdflt;
3038 maxmtu = tcp_maxmtu6(inc, NULL);
3039 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3040 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
3041 } else
3042 #endif
3043 {
3044 mss = tcp_mssdflt;
3045 maxmtu = tcp_maxmtu(inc, NULL);
3046 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
3047 min_protoh = sizeof(struct tcpiphdr);
3048 }
3049 if (maxmtu && thcmtu)
3050 mss = min(maxmtu, thcmtu) - min_protoh;
3051 else if (maxmtu || thcmtu)
3052 mss = max(maxmtu, thcmtu) - min_protoh;
3053
3054 return (mss);
3055 }
3056
3057
3058 /*
3059 * On a partial ack arrives, force the retransmission of the
3060 * next unacknowledged segment. Do not clear tp->t_dupacks.
3061 * By setting snd_nxt to ti_ack, this forces retransmission timer to
3062 * be started again.
3063 */
3064 static void
3065 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
3066 {
3067 tcp_seq onxt = tp->snd_nxt;
3068 u_long ocwnd = tp->snd_cwnd;
3069
3070 INP_WLOCK_ASSERT(tp->t_inpcb);
3071
3072 tcp_timer_activate(tp, TT_REXMT, 0);
3073 tp->t_rtttime = 0;
3074 tp->snd_nxt = th->th_ack;
3075 /*
3076 * Set snd_cwnd to one segment beyond acknowledged offset.
3077 * (tp->snd_una has not yet been updated when this function is called.)
3078 */
3079 tp->snd_cwnd = tp->t_maxseg + (th->th_ack - tp->snd_una);
3080 tp->t_flags |= TF_ACKNOW;
3081 (void) tcp_output(tp);
3082 tp->snd_cwnd = ocwnd;
3083 if (SEQ_GT(onxt, tp->snd_nxt))
3084 tp->snd_nxt = onxt;
3085 /*
3086 * Partial window deflation. Relies on fact that tp->snd_una
3087 * not updated yet.
3088 */
3089 if (tp->snd_cwnd > th->th_ack - tp->snd_una)
3090 tp->snd_cwnd -= th->th_ack - tp->snd_una;
3091 else
3092 tp->snd_cwnd = 0;
3093 tp->snd_cwnd += tp->t_maxseg;
3094 }
Cache object: 43d65bb88d06f53ad25543f99d666b31
|