1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 * Copyright (c) 2007-2008,2010
7 * Swinburne University of Technology, Melbourne, Australia.
8 * Copyright (c) 2009-2010 Lawrence Stewart <lstewart@freebsd.org>
9 * Copyright (c) 2010 The FreeBSD Foundation
10 * Copyright (c) 2010-2011 Juniper Networks, Inc.
11 * All rights reserved.
12 *
13 * Portions of this software were developed at the Centre for Advanced Internet
14 * Architectures, Swinburne University of Technology, by Lawrence Stewart,
15 * James Healy and David Hayes, made possible in part by a grant from the Cisco
16 * University Research Program Fund at Community Foundation Silicon Valley.
17 *
18 * Portions of this software were developed at the Centre for Advanced
19 * Internet Architectures, Swinburne University of Technology, Melbourne,
20 * Australia by David Hayes under sponsorship from the FreeBSD Foundation.
21 *
22 * Portions of this software were developed by Robert N. M. Watson under
23 * contract to Juniper Networks, Inc.
24 *
25 * Redistribution and use in source and binary forms, with or without
26 * modification, are permitted provided that the following conditions
27 * are met:
28 * 1. Redistributions of source code must retain the above copyright
29 * notice, this list of conditions and the following disclaimer.
30 * 2. Redistributions in binary form must reproduce the above copyright
31 * notice, this list of conditions and the following disclaimer in the
32 * documentation and/or other materials provided with the distribution.
33 * 3. Neither the name of the University nor the names of its contributors
34 * may be used to endorse or promote products derived from this software
35 * without specific prior written permission.
36 *
37 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
38 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
39 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
40 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
41 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
42 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
43 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
44 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
45 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
46 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
47 * SUCH DAMAGE.
48 */
49
50 #include <sys/cdefs.h>
51 #include "opt_inet.h"
52 #include "opt_inet6.h"
53 #include "opt_ipsec.h"
54 #include "opt_rss.h"
55
56 #include <sys/param.h>
57 #include <sys/arb.h>
58 #include <sys/kernel.h>
59 #ifdef TCP_HHOOK
60 #include <sys/hhook.h>
61 #endif
62 #include <sys/malloc.h>
63 #include <sys/mbuf.h>
64 #include <sys/proc.h> /* for proc0 declaration */
65 #include <sys/protosw.h>
66 #include <sys/qmath.h>
67 #include <sys/sdt.h>
68 #include <sys/signalvar.h>
69 #include <sys/socket.h>
70 #include <sys/socketvar.h>
71 #include <sys/sysctl.h>
72 #include <sys/syslog.h>
73 #include <sys/systm.h>
74 #include <sys/stats.h>
75
76 #include <machine/cpu.h> /* before tcp_seq.h, for tcp_random18() */
77
78 #include <vm/uma.h>
79
80 #include <net/if.h>
81 #include <net/if_var.h>
82 #include <net/route.h>
83 #include <net/rss_config.h>
84 #include <net/vnet.h>
85
86 #define TCPSTATES /* for logging */
87
88 #include <netinet/in.h>
89 #include <netinet/in_kdtrace.h>
90 #include <netinet/in_pcb.h>
91 #include <netinet/in_rss.h>
92 #include <netinet/in_systm.h>
93 #include <netinet/ip.h>
94 #include <netinet/ip_icmp.h> /* required for icmp_var.h */
95 #include <netinet/icmp_var.h> /* for ICMP_BANDLIM */
96 #include <netinet/ip_var.h>
97 #include <netinet/ip_options.h>
98 #include <netinet/ip6.h>
99 #include <netinet/icmp6.h>
100 #include <netinet6/in6_pcb.h>
101 #include <netinet6/in6_rss.h>
102 #include <netinet6/in6_var.h>
103 #include <netinet6/ip6_var.h>
104 #include <netinet6/nd6.h>
105 #include <netinet/tcp.h>
106 #include <netinet/tcp_fsm.h>
107 #include <netinet/tcp_seq.h>
108 #include <netinet/tcp_timer.h>
109 #include <netinet/tcp_var.h>
110 #include <netinet/tcp_log_buf.h>
111 #include <netinet6/tcp6_var.h>
112 #include <netinet/tcpip.h>
113 #include <netinet/cc/cc.h>
114 #include <netinet/tcp_fastopen.h>
115 #ifdef TCPPCAP
116 #include <netinet/tcp_pcap.h>
117 #endif
118 #include <netinet/tcp_syncache.h>
119 #ifdef TCP_OFFLOAD
120 #include <netinet/tcp_offload.h>
121 #endif
122 #include <netinet/tcp_ecn.h>
123 #include <netinet/udp.h>
124
125 #include <netipsec/ipsec_support.h>
126
127 #include <machine/in_cksum.h>
128
129 #include <security/mac/mac_framework.h>
130
131 const int tcprexmtthresh = 3;
132
133 VNET_DEFINE(int, tcp_log_in_vain) = 0;
134 SYSCTL_INT(_net_inet_tcp, OID_AUTO, log_in_vain, CTLFLAG_VNET | CTLFLAG_RW,
135 &VNET_NAME(tcp_log_in_vain), 0,
136 "Log all incoming TCP segments to closed ports");
137
138 VNET_DEFINE(int, blackhole) = 0;
139 #define V_blackhole VNET(blackhole)
140 SYSCTL_INT(_net_inet_tcp, OID_AUTO, blackhole, CTLFLAG_VNET | CTLFLAG_RW,
141 &VNET_NAME(blackhole), 0,
142 "Do not send RST on segments to closed ports");
143
144 VNET_DEFINE(bool, blackhole_local) = false;
145 #define V_blackhole_local VNET(blackhole_local)
146 SYSCTL_BOOL(_net_inet_tcp, OID_AUTO, blackhole_local, CTLFLAG_VNET |
147 CTLFLAG_RW, &VNET_NAME(blackhole_local), false,
148 "Enforce net.inet.tcp.blackhole for locally originated packets");
149
150 VNET_DEFINE(int, tcp_delack_enabled) = 1;
151 SYSCTL_INT(_net_inet_tcp, OID_AUTO, delayed_ack, CTLFLAG_VNET | CTLFLAG_RW,
152 &VNET_NAME(tcp_delack_enabled), 0,
153 "Delay ACK to try and piggyback it onto a data packet");
154
155 VNET_DEFINE(int, drop_synfin) = 0;
156 SYSCTL_INT(_net_inet_tcp, OID_AUTO, drop_synfin, CTLFLAG_VNET | CTLFLAG_RW,
157 &VNET_NAME(drop_synfin), 0,
158 "Drop TCP packets with SYN+FIN set");
159
160 VNET_DEFINE(int, tcp_do_prr) = 1;
161 SYSCTL_INT(_net_inet_tcp, OID_AUTO, do_prr, CTLFLAG_VNET | CTLFLAG_RW,
162 &VNET_NAME(tcp_do_prr), 1,
163 "Enable Proportional Rate Reduction per RFC 6937");
164
165 VNET_DEFINE(int, tcp_do_newcwv) = 0;
166 SYSCTL_INT(_net_inet_tcp, OID_AUTO, newcwv, CTLFLAG_VNET | CTLFLAG_RW,
167 &VNET_NAME(tcp_do_newcwv), 0,
168 "Enable New Congestion Window Validation per RFC7661");
169
170 VNET_DEFINE(int, tcp_do_rfc3042) = 1;
171 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3042, CTLFLAG_VNET | CTLFLAG_RW,
172 &VNET_NAME(tcp_do_rfc3042), 0,
173 "Enable RFC 3042 (Limited Transmit)");
174
175 VNET_DEFINE(int, tcp_do_rfc3390) = 1;
176 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3390, CTLFLAG_VNET | CTLFLAG_RW,
177 &VNET_NAME(tcp_do_rfc3390), 0,
178 "Enable RFC 3390 (Increasing TCP's Initial Congestion Window)");
179
180 VNET_DEFINE(int, tcp_initcwnd_segments) = 10;
181 SYSCTL_INT(_net_inet_tcp, OID_AUTO, initcwnd_segments,
182 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_initcwnd_segments), 0,
183 "Slow-start flight size (initial congestion window) in number of segments");
184
185 VNET_DEFINE(int, tcp_do_rfc3465) = 1;
186 SYSCTL_INT(_net_inet_tcp, OID_AUTO, rfc3465, CTLFLAG_VNET | CTLFLAG_RW,
187 &VNET_NAME(tcp_do_rfc3465), 0,
188 "Enable RFC 3465 (Appropriate Byte Counting)");
189
190 VNET_DEFINE(int, tcp_abc_l_var) = 2;
191 SYSCTL_INT(_net_inet_tcp, OID_AUTO, abc_l_var, CTLFLAG_VNET | CTLFLAG_RW,
192 &VNET_NAME(tcp_abc_l_var), 2,
193 "Cap the max cwnd increment during slow-start to this number of segments");
194
195 VNET_DEFINE(int, tcp_insecure_syn) = 0;
196 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_syn, CTLFLAG_VNET | CTLFLAG_RW,
197 &VNET_NAME(tcp_insecure_syn), 0,
198 "Follow RFC793 instead of RFC5961 criteria for accepting SYN packets");
199
200 VNET_DEFINE(int, tcp_insecure_rst) = 0;
201 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_rst, CTLFLAG_VNET | CTLFLAG_RW,
202 &VNET_NAME(tcp_insecure_rst), 0,
203 "Follow RFC793 instead of RFC5961 criteria for accepting RST packets");
204
205 VNET_DEFINE(int, tcp_insecure_ack) = 0;
206 SYSCTL_INT(_net_inet_tcp, OID_AUTO, insecure_ack, CTLFLAG_VNET | CTLFLAG_RW,
207 &VNET_NAME(tcp_insecure_ack), 0,
208 "Follow RFC793 criteria for validating SEG.ACK");
209
210 VNET_DEFINE(int, tcp_recvspace) = 1024*64;
211 #define V_tcp_recvspace VNET(tcp_recvspace)
212 SYSCTL_INT(_net_inet_tcp, TCPCTL_RECVSPACE, recvspace, CTLFLAG_VNET | CTLFLAG_RW,
213 &VNET_NAME(tcp_recvspace), 0, "Initial receive socket buffer size");
214
215 VNET_DEFINE(int, tcp_do_autorcvbuf) = 1;
216 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_auto, CTLFLAG_VNET | CTLFLAG_RW,
217 &VNET_NAME(tcp_do_autorcvbuf), 0,
218 "Enable automatic receive buffer sizing");
219
220 VNET_DEFINE(int, tcp_autorcvbuf_max) = 2*1024*1024;
221 SYSCTL_INT(_net_inet_tcp, OID_AUTO, recvbuf_max, CTLFLAG_VNET | CTLFLAG_RW,
222 &VNET_NAME(tcp_autorcvbuf_max), 0,
223 "Max size of automatic receive buffer");
224
225 VNET_DEFINE(struct inpcbinfo, tcbinfo);
226
227 /*
228 * TCP statistics are stored in an array of counter(9)s, which size matches
229 * size of struct tcpstat. TCP running connection count is a regular array.
230 */
231 VNET_PCPUSTAT_DEFINE(struct tcpstat, tcpstat);
232 SYSCTL_VNET_PCPUSTAT(_net_inet_tcp, TCPCTL_STATS, stats, struct tcpstat,
233 tcpstat, "TCP statistics (struct tcpstat, netinet/tcp_var.h)");
234 VNET_DEFINE(counter_u64_t, tcps_states[TCP_NSTATES]);
235 SYSCTL_COUNTER_U64_ARRAY(_net_inet_tcp, TCPCTL_STATES, states, CTLFLAG_RD |
236 CTLFLAG_VNET, &VNET_NAME(tcps_states)[0], TCP_NSTATES,
237 "TCP connection counts by TCP state");
238
239 /*
240 * Kernel module interface for updating tcpstat. The first argument is an index
241 * into tcpstat treated as an array.
242 */
243 void
kmod_tcpstat_add(int statnum,int val)244 kmod_tcpstat_add(int statnum, int val)
245 {
246
247 counter_u64_add(VNET(tcpstat)[statnum], val);
248 }
249
250 /*
251 * Make sure that we only start a SACK loss recovery when
252 * receiving a duplicate ACK with a SACK block, and also
253 * complete SACK loss recovery in case the other end
254 * reneges.
255 */
256 static bool inline
tcp_is_sack_recovery(struct tcpcb * tp,struct tcpopt * to)257 tcp_is_sack_recovery(struct tcpcb *tp, struct tcpopt *to)
258 {
259 return ((tp->t_flags & TF_SACK_PERMIT) &&
260 ((to->to_flags & TOF_SACK) ||
261 (!TAILQ_EMPTY(&tp->snd_holes))));
262 }
263
264 #ifdef TCP_HHOOK
265 /*
266 * Wrapper for the TCP established input helper hook.
267 */
268 void
hhook_run_tcp_est_in(struct tcpcb * tp,struct tcphdr * th,struct tcpopt * to)269 hhook_run_tcp_est_in(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to)
270 {
271 struct tcp_hhook_data hhook_data;
272
273 if (V_tcp_hhh[HHOOK_TCP_EST_IN]->hhh_nhooks > 0) {
274 hhook_data.tp = tp;
275 hhook_data.th = th;
276 hhook_data.to = to;
277
278 hhook_run_hooks(V_tcp_hhh[HHOOK_TCP_EST_IN], &hhook_data,
279 &tp->t_osd);
280 }
281 }
282 #endif
283
284 /*
285 * CC wrapper hook functions
286 */
287 void
cc_ack_received(struct tcpcb * tp,struct tcphdr * th,uint16_t nsegs,uint16_t type)288 cc_ack_received(struct tcpcb *tp, struct tcphdr *th, uint16_t nsegs,
289 uint16_t type)
290 {
291 #ifdef STATS
292 int32_t gput;
293 #endif
294
295 INP_WLOCK_ASSERT(tptoinpcb(tp));
296
297 tp->t_ccv.nsegs = nsegs;
298 tp->t_ccv.bytes_this_ack = BYTES_THIS_ACK(tp, th);
299 if ((!V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd)) ||
300 (V_tcp_do_newcwv && (tp->snd_cwnd <= tp->snd_wnd) &&
301 (tp->snd_cwnd < (tcp_compute_pipe(tp) * 2))))
302 tp->t_ccv.flags |= CCF_CWND_LIMITED;
303 else
304 tp->t_ccv.flags &= ~CCF_CWND_LIMITED;
305
306 if (type == CC_ACK) {
307 #ifdef STATS
308 stats_voi_update_abs_s32(tp->t_stats, VOI_TCP_CALCFRWINDIFF,
309 ((int32_t)tp->snd_cwnd) - tp->snd_wnd);
310 if (!IN_RECOVERY(tp->t_flags))
311 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_ACKLEN,
312 tp->t_ccv.bytes_this_ack / (tcp_maxseg(tp) * nsegs));
313 if ((tp->t_flags & TF_GPUTINPROG) &&
314 SEQ_GEQ(th->th_ack, tp->gput_ack)) {
315 /*
316 * Compute goodput in bits per millisecond.
317 */
318 gput = (((int64_t)SEQ_SUB(th->th_ack, tp->gput_seq)) << 3) /
319 max(1, tcp_ts_getticks() - tp->gput_ts);
320 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_GPUT,
321 gput);
322 /*
323 * XXXLAS: This is a temporary hack, and should be
324 * chained off VOI_TCP_GPUT when stats(9) grows an API
325 * to deal with chained VOIs.
326 */
327 if (tp->t_stats_gput_prev > 0)
328 stats_voi_update_abs_s32(tp->t_stats,
329 VOI_TCP_GPUT_ND,
330 ((gput - tp->t_stats_gput_prev) * 100) /
331 tp->t_stats_gput_prev);
332 tp->t_flags &= ~TF_GPUTINPROG;
333 tp->t_stats_gput_prev = gput;
334 }
335 #endif /* STATS */
336 if (tp->snd_cwnd > tp->snd_ssthresh) {
337 tp->t_bytes_acked += tp->t_ccv.bytes_this_ack;
338 if (tp->t_bytes_acked >= tp->snd_cwnd) {
339 tp->t_bytes_acked -= tp->snd_cwnd;
340 tp->t_ccv.flags |= CCF_ABC_SENTAWND;
341 }
342 } else {
343 tp->t_ccv.flags &= ~CCF_ABC_SENTAWND;
344 tp->t_bytes_acked = 0;
345 }
346 }
347
348 if (CC_ALGO(tp)->ack_received != NULL) {
349 /* XXXLAS: Find a way to live without this */
350 tp->t_ccv.curack = th->th_ack;
351 CC_ALGO(tp)->ack_received(&tp->t_ccv, type);
352 }
353 #ifdef STATS
354 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_LCWIN, tp->snd_cwnd);
355 #endif
356 }
357
358 void
cc_conn_init(struct tcpcb * tp)359 cc_conn_init(struct tcpcb *tp)
360 {
361 struct hc_metrics_lite metrics;
362 struct inpcb *inp = tptoinpcb(tp);
363 u_int maxseg;
364 int rtt;
365
366 INP_WLOCK_ASSERT(inp);
367
368 tcp_hc_get(&inp->inp_inc, &metrics);
369 maxseg = tcp_maxseg(tp);
370
371 if (tp->t_srtt == 0 && (rtt = metrics.rmx_rtt)) {
372 tp->t_srtt = rtt;
373 TCPSTAT_INC(tcps_usedrtt);
374 if (metrics.rmx_rttvar) {
375 tp->t_rttvar = metrics.rmx_rttvar;
376 TCPSTAT_INC(tcps_usedrttvar);
377 } else {
378 /* default variation is +- 1 rtt */
379 tp->t_rttvar =
380 tp->t_srtt * TCP_RTTVAR_SCALE / TCP_RTT_SCALE;
381 }
382 TCPT_RANGESET(tp->t_rxtcur,
383 ((tp->t_srtt >> 2) + tp->t_rttvar) >> 1,
384 tp->t_rttmin, TCPTV_REXMTMAX);
385 }
386 if (metrics.rmx_ssthresh) {
387 /*
388 * There's some sort of gateway or interface
389 * buffer limit on the path. Use this to set
390 * the slow start threshold, but set the
391 * threshold to no less than 2*mss.
392 */
393 tp->snd_ssthresh = max(2 * maxseg, metrics.rmx_ssthresh);
394 TCPSTAT_INC(tcps_usedssthresh);
395 }
396
397 /*
398 * Set the initial slow-start flight size.
399 *
400 * If a SYN or SYN/ACK was lost and retransmitted, we have to
401 * reduce the initial CWND to one segment as congestion is likely
402 * requiring us to be cautious.
403 */
404 if (tp->snd_cwnd == 1)
405 tp->snd_cwnd = maxseg; /* SYN(-ACK) lost */
406 else
407 tp->snd_cwnd = tcp_compute_initwnd(maxseg);
408
409 if (CC_ALGO(tp)->conn_init != NULL)
410 CC_ALGO(tp)->conn_init(&tp->t_ccv);
411 }
412
413 void inline
cc_cong_signal(struct tcpcb * tp,struct tcphdr * th,uint32_t type)414 cc_cong_signal(struct tcpcb *tp, struct tcphdr *th, uint32_t type)
415 {
416 INP_WLOCK_ASSERT(tptoinpcb(tp));
417
418 #ifdef STATS
419 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_CSIG, type);
420 #endif
421
422 switch(type) {
423 case CC_NDUPACK:
424 if (!IN_FASTRECOVERY(tp->t_flags)) {
425 tp->snd_recover = tp->snd_max;
426 if (tp->t_flags2 & TF2_ECN_PERMIT)
427 tp->t_flags2 |= TF2_ECN_SND_CWR;
428 }
429 break;
430 case CC_ECN:
431 if (!IN_CONGRECOVERY(tp->t_flags) ||
432 /*
433 * Allow ECN reaction on ACK to CWR, if
434 * that data segment was also CE marked.
435 */
436 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
437 EXIT_CONGRECOVERY(tp->t_flags);
438 TCPSTAT_INC(tcps_ecn_rcwnd);
439 tp->snd_recover = tp->snd_max + 1;
440 if (tp->t_flags2 & TF2_ECN_PERMIT)
441 tp->t_flags2 |= TF2_ECN_SND_CWR;
442 }
443 break;
444 case CC_RTO:
445 tp->t_dupacks = 0;
446 tp->t_bytes_acked = 0;
447 if ((tp->t_rxtshift > 1) ||
448 !((tp->t_flags & TF_SACK_PERMIT) &&
449 (!TAILQ_EMPTY(&tp->snd_holes))))
450 EXIT_RECOVERY(tp->t_flags);
451 if (tp->t_flags2 & TF2_ECN_PERMIT)
452 tp->t_flags2 |= TF2_ECN_SND_CWR;
453 break;
454 case CC_RTO_ERR:
455 TCPSTAT_INC(tcps_sndrexmitbad);
456 /* RTO was unnecessary, so reset everything. */
457 tp->snd_cwnd = tp->snd_cwnd_prev;
458 tp->snd_ssthresh = tp->snd_ssthresh_prev;
459 tp->snd_recover = tp->snd_recover_prev;
460 if (tp->t_flags & TF_WASFRECOVERY)
461 ENTER_FASTRECOVERY(tp->t_flags);
462 if (tp->t_flags & TF_WASCRECOVERY)
463 ENTER_CONGRECOVERY(tp->t_flags);
464 tp->snd_nxt = tp->snd_max;
465 tp->t_flags &= ~TF_PREVVALID;
466 tp->t_badrxtwin = 0;
467 break;
468 }
469 if (SEQ_LT(tp->snd_fack, tp->snd_una) ||
470 SEQ_GT(tp->snd_fack, tp->snd_max)) {
471 tp->snd_fack = tp->snd_una;
472 }
473
474 if (CC_ALGO(tp)->cong_signal != NULL) {
475 if (th != NULL)
476 tp->t_ccv.curack = th->th_ack;
477 CC_ALGO(tp)->cong_signal(&tp->t_ccv, type);
478 }
479 }
480
481 void inline
cc_post_recovery(struct tcpcb * tp,struct tcphdr * th)482 cc_post_recovery(struct tcpcb *tp, struct tcphdr *th)
483 {
484 INP_WLOCK_ASSERT(tptoinpcb(tp));
485
486 if (CC_ALGO(tp)->post_recovery != NULL) {
487 if (SEQ_LT(tp->snd_fack, th->th_ack) ||
488 SEQ_GT(tp->snd_fack, tp->snd_max)) {
489 tp->snd_fack = th->th_ack;
490 }
491 tp->t_ccv.curack = th->th_ack;
492 CC_ALGO(tp)->post_recovery(&tp->t_ccv);
493 }
494 EXIT_RECOVERY(tp->t_flags);
495
496 tp->t_bytes_acked = 0;
497 tp->sackhint.delivered_data = 0;
498 tp->sackhint.prr_delivered = 0;
499 tp->sackhint.prr_out = 0;
500 }
501
502 /*
503 * Indicate whether this ack should be delayed. We can delay the ack if
504 * following conditions are met:
505 * - There is no delayed ack timer in progress.
506 * - Our last ack wasn't a 0-sized window. We never want to delay
507 * the ack that opens up a 0-sized window.
508 * - LRO wasn't used for this segment. We make sure by checking that the
509 * segment size is not larger than the MSS.
510 */
511 #define DELAY_ACK(tp, tlen) \
512 ((!tcp_timer_active(tp, TT_DELACK) && \
513 (tp->t_flags & TF_RXWIN0SENT) == 0) && \
514 (tlen <= tp->t_maxseg) && \
515 (V_tcp_delack_enabled || (tp->t_flags & TF_NEEDSYN)))
516
517 void inline
cc_ecnpkt_handler_flags(struct tcpcb * tp,uint16_t flags,uint8_t iptos)518 cc_ecnpkt_handler_flags(struct tcpcb *tp, uint16_t flags, uint8_t iptos)
519 {
520 INP_WLOCK_ASSERT(tptoinpcb(tp));
521
522 if (CC_ALGO(tp)->ecnpkt_handler != NULL) {
523 switch (iptos & IPTOS_ECN_MASK) {
524 case IPTOS_ECN_CE:
525 tp->t_ccv.flags |= CCF_IPHDR_CE;
526 break;
527 case IPTOS_ECN_ECT0:
528 /* FALLTHROUGH */
529 case IPTOS_ECN_ECT1:
530 /* FALLTHROUGH */
531 case IPTOS_ECN_NOTECT:
532 tp->t_ccv.flags &= ~CCF_IPHDR_CE;
533 break;
534 }
535
536 if (flags & TH_CWR)
537 tp->t_ccv.flags |= CCF_TCPHDR_CWR;
538 else
539 tp->t_ccv.flags &= ~CCF_TCPHDR_CWR;
540
541 CC_ALGO(tp)->ecnpkt_handler(&tp->t_ccv);
542
543 if (tp->t_ccv.flags & CCF_ACKNOW) {
544 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
545 tp->t_flags |= TF_ACKNOW;
546 }
547 }
548 }
549
550 void inline
cc_ecnpkt_handler(struct tcpcb * tp,struct tcphdr * th,uint8_t iptos)551 cc_ecnpkt_handler(struct tcpcb *tp, struct tcphdr *th, uint8_t iptos)
552 {
553 cc_ecnpkt_handler_flags(tp, tcp_get_flags(th), iptos);
554 }
555
556 /*
557 * TCP input handling is split into multiple parts:
558 * tcp6_input is a thin wrapper around tcp_input for the extended
559 * ip6_protox[] call format in ip6_input
560 * tcp_input handles primary segment validation, inpcb lookup and
561 * SYN processing on listen sockets
562 * tcp_do_segment processes the ACK and text of the segment for
563 * establishing, established and closing connections
564 */
565 #ifdef INET6
566 int
tcp6_input_with_port(struct mbuf ** mp,int * offp,int proto,uint16_t port)567 tcp6_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
568 {
569 struct mbuf *m;
570 struct in6_ifaddr *ia6;
571 struct ip6_hdr *ip6;
572
573 m = *mp;
574 if (m->m_len < *offp + sizeof(struct tcphdr)) {
575 m = m_pullup(m, *offp + sizeof(struct tcphdr));
576 if (m == NULL) {
577 *mp = m;
578 TCPSTAT_INC(tcps_rcvshort);
579 return (IPPROTO_DONE);
580 }
581 }
582
583 /*
584 * draft-itojun-ipv6-tcp-to-anycast
585 * better place to put this in?
586 */
587 ip6 = mtod(m, struct ip6_hdr *);
588 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
589 if (ia6 && (ia6->ia6_flags & IN6_IFF_ANYCAST)) {
590 icmp6_error(m, ICMP6_DST_UNREACH, ICMP6_DST_UNREACH_ADDR,
591 (caddr_t)&ip6->ip6_dst - (caddr_t)ip6);
592 *mp = NULL;
593 return (IPPROTO_DONE);
594 }
595
596 *mp = m;
597 return (tcp_input_with_port(mp, offp, proto, port));
598 }
599
600 int
tcp6_input(struct mbuf ** mp,int * offp,int proto)601 tcp6_input(struct mbuf **mp, int *offp, int proto)
602 {
603
604 return(tcp6_input_with_port(mp, offp, proto, 0));
605 }
606 #endif /* INET6 */
607
608 int
tcp_input_with_port(struct mbuf ** mp,int * offp,int proto,uint16_t port)609 tcp_input_with_port(struct mbuf **mp, int *offp, int proto, uint16_t port)
610 {
611 struct mbuf *m = *mp;
612 struct tcphdr *th = NULL;
613 struct ip *ip = NULL;
614 struct inpcb *inp = NULL;
615 struct tcpcb *tp = NULL;
616 struct socket *so = NULL;
617 u_char *optp = NULL;
618 int off0;
619 int optlen = 0;
620 #ifdef INET
621 int len;
622 uint8_t ipttl;
623 #endif
624 int tlen = 0, off;
625 int drop_hdrlen;
626 int thflags;
627 int rstreason = 0; /* For badport_bandlim accounting purposes */
628 int lookupflag;
629 uint8_t iptos;
630 struct m_tag *fwd_tag = NULL;
631 #ifdef INET6
632 struct ip6_hdr *ip6 = NULL;
633 int isipv6;
634 #else
635 const void *ip6 = NULL;
636 #endif /* INET6 */
637 struct tcpopt to; /* options in this segment */
638 char *s = NULL; /* address and port logging */
639
640 NET_EPOCH_ASSERT();
641
642 #ifdef INET6
643 isipv6 = (mtod(m, struct ip *)->ip_v == 6) ? 1 : 0;
644 #endif
645
646 off0 = *offp;
647 m = *mp;
648 *mp = NULL;
649 to.to_flags = 0;
650 TCPSTAT_INC(tcps_rcvtotal);
651
652 m->m_pkthdr.tcp_tun_port = port;
653 #ifdef INET6
654 if (isipv6) {
655 ip6 = mtod(m, struct ip6_hdr *);
656 th = (struct tcphdr *)((caddr_t)ip6 + off0);
657 tlen = sizeof(*ip6) + ntohs(ip6->ip6_plen) - off0;
658 if (port)
659 goto skip6_csum;
660 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID_IPV6) {
661 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
662 th->th_sum = m->m_pkthdr.csum_data;
663 else
664 th->th_sum = in6_cksum_pseudo(ip6, tlen,
665 IPPROTO_TCP, m->m_pkthdr.csum_data);
666 th->th_sum ^= 0xffff;
667 } else
668 th->th_sum = in6_cksum(m, IPPROTO_TCP, off0, tlen);
669 if (th->th_sum) {
670 TCPSTAT_INC(tcps_rcvbadsum);
671 goto drop;
672 }
673 skip6_csum:
674 /*
675 * Be proactive about unspecified IPv6 address in source.
676 * As we use all-zero to indicate unbounded/unconnected pcb,
677 * unspecified IPv6 address can be used to confuse us.
678 *
679 * Note that packets with unspecified IPv6 destination is
680 * already dropped in ip6_input.
681 */
682 KASSERT(!IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_dst),
683 ("%s: unspecified destination v6 address", __func__));
684 if (IN6_IS_ADDR_UNSPECIFIED(&ip6->ip6_src)) {
685 IP6STAT_INC(ip6s_badscope); /* XXX */
686 goto drop;
687 }
688 iptos = IPV6_TRAFFIC_CLASS(ip6);
689 }
690 #endif
691 #if defined(INET) && defined(INET6)
692 else
693 #endif
694 #ifdef INET
695 {
696 /*
697 * Get IP and TCP header together in first mbuf.
698 * Note: IP leaves IP header in first mbuf.
699 */
700 if (off0 > sizeof (struct ip)) {
701 ip_stripoptions(m);
702 off0 = sizeof(struct ip);
703 }
704 if (m->m_len < sizeof (struct tcpiphdr)) {
705 if ((m = m_pullup(m, sizeof (struct tcpiphdr)))
706 == NULL) {
707 TCPSTAT_INC(tcps_rcvshort);
708 return (IPPROTO_DONE);
709 }
710 }
711 ip = mtod(m, struct ip *);
712 th = (struct tcphdr *)((caddr_t)ip + off0);
713 tlen = ntohs(ip->ip_len) - off0;
714
715 iptos = ip->ip_tos;
716 if (port)
717 goto skip_csum;
718 if (m->m_pkthdr.csum_flags & CSUM_DATA_VALID) {
719 if (m->m_pkthdr.csum_flags & CSUM_PSEUDO_HDR)
720 th->th_sum = m->m_pkthdr.csum_data;
721 else
722 th->th_sum = in_pseudo(ip->ip_src.s_addr,
723 ip->ip_dst.s_addr,
724 htonl(m->m_pkthdr.csum_data + tlen +
725 IPPROTO_TCP));
726 th->th_sum ^= 0xffff;
727 } else {
728 struct ipovly *ipov = (struct ipovly *)ip;
729
730 /*
731 * Checksum extended TCP header and data.
732 */
733 len = off0 + tlen;
734 ipttl = ip->ip_ttl;
735 bzero(ipov->ih_x1, sizeof(ipov->ih_x1));
736 ipov->ih_len = htons(tlen);
737 th->th_sum = in_cksum(m, len);
738 /* Reset length for SDT probes. */
739 ip->ip_len = htons(len);
740 /* Reset TOS bits */
741 ip->ip_tos = iptos;
742 /* Re-initialization for later version check */
743 ip->ip_ttl = ipttl;
744 ip->ip_v = IPVERSION;
745 ip->ip_hl = off0 >> 2;
746 }
747 skip_csum:
748 if (th->th_sum && (port == 0)) {
749 TCPSTAT_INC(tcps_rcvbadsum);
750 goto drop;
751 }
752 KASSERT(ip->ip_dst.s_addr != INADDR_ANY,
753 ("%s: unspecified destination v4 address", __func__));
754 if (__predict_false(ip->ip_src.s_addr == INADDR_ANY)) {
755 IPSTAT_INC(ips_badaddr);
756 goto drop;
757 }
758 }
759 #endif /* INET */
760
761 /*
762 * Check that TCP offset makes sense,
763 * pull out TCP options and adjust length. XXX
764 */
765 off = th->th_off << 2;
766 if (off < sizeof (struct tcphdr) || off > tlen) {
767 TCPSTAT_INC(tcps_rcvbadoff);
768 goto drop;
769 }
770 tlen -= off; /* tlen is used instead of ti->ti_len */
771 if (off > sizeof (struct tcphdr)) {
772 #ifdef INET6
773 if (isipv6) {
774 if (m->m_len < off0 + off) {
775 m = m_pullup(m, off0 + off);
776 if (m == NULL) {
777 TCPSTAT_INC(tcps_rcvshort);
778 return (IPPROTO_DONE);
779 }
780 }
781 ip6 = mtod(m, struct ip6_hdr *);
782 th = (struct tcphdr *)((caddr_t)ip6 + off0);
783 }
784 #endif
785 #if defined(INET) && defined(INET6)
786 else
787 #endif
788 #ifdef INET
789 {
790 if (m->m_len < sizeof(struct ip) + off) {
791 if ((m = m_pullup(m, sizeof (struct ip) + off))
792 == NULL) {
793 TCPSTAT_INC(tcps_rcvshort);
794 return (IPPROTO_DONE);
795 }
796 ip = mtod(m, struct ip *);
797 th = (struct tcphdr *)((caddr_t)ip + off0);
798 }
799 }
800 #endif
801 optlen = off - sizeof (struct tcphdr);
802 optp = (u_char *)(th + 1);
803 }
804 thflags = tcp_get_flags(th);
805
806 /*
807 * Convert TCP protocol specific fields to host format.
808 */
809 tcp_fields_to_host(th);
810
811 /*
812 * Delay dropping TCP, IP headers, IPv6 ext headers, and TCP options.
813 */
814 drop_hdrlen = off0 + off;
815
816 /*
817 * Grab info from PACKET_TAG_IPFORWARD tag prepended to the chain.
818 */
819 if (
820 #ifdef INET6
821 (isipv6 && (m->m_flags & M_IP6_NEXTHOP))
822 #ifdef INET
823 || (!isipv6 && (m->m_flags & M_IP_NEXTHOP))
824 #endif
825 #endif
826 #if defined(INET) && !defined(INET6)
827 (m->m_flags & M_IP_NEXTHOP)
828 #endif
829 )
830 fwd_tag = m_tag_find(m, PACKET_TAG_IPFORWARD, NULL);
831
832 /*
833 * For initial SYN packets we don't need write lock on matching
834 * PCB, be it a listening one or a synchronized one. The packet
835 * shall not modify its state.
836 */
837 lookupflag = INPLOOKUP_WILDCARD |
838 ((thflags & (TH_ACK|TH_SYN)) == TH_SYN ?
839 INPLOOKUP_RLOCKPCB : INPLOOKUP_WLOCKPCB);
840 findpcb:
841 tp = NULL;
842 #ifdef INET6
843 if (isipv6 && fwd_tag != NULL) {
844 struct sockaddr_in6 *next_hop6;
845
846 next_hop6 = (struct sockaddr_in6 *)(fwd_tag + 1);
847 /*
848 * Transparently forwarded. Pretend to be the destination.
849 * Already got one like this?
850 */
851 inp = in6_pcblookup_mbuf(&V_tcbinfo,
852 &ip6->ip6_src, th->th_sport, &ip6->ip6_dst, th->th_dport,
853 lookupflag & ~INPLOOKUP_WILDCARD, m->m_pkthdr.rcvif, m);
854 if (!inp) {
855 /*
856 * It's new. Try to find the ambushing socket.
857 * Because we've rewritten the destination address,
858 * any hardware-generated hash is ignored.
859 */
860 inp = in6_pcblookup(&V_tcbinfo, &ip6->ip6_src,
861 th->th_sport, &next_hop6->sin6_addr,
862 next_hop6->sin6_port ? ntohs(next_hop6->sin6_port) :
863 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
864 }
865 } else if (isipv6) {
866 inp = in6_pcblookup_mbuf(&V_tcbinfo, &ip6->ip6_src,
867 th->th_sport, &ip6->ip6_dst, th->th_dport, lookupflag,
868 m->m_pkthdr.rcvif, m);
869 }
870 #endif /* INET6 */
871 #if defined(INET6) && defined(INET)
872 else
873 #endif
874 #ifdef INET
875 if (fwd_tag != NULL) {
876 struct sockaddr_in *next_hop;
877
878 next_hop = (struct sockaddr_in *)(fwd_tag+1);
879 /*
880 * Transparently forwarded. Pretend to be the destination.
881 * already got one like this?
882 */
883 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src, th->th_sport,
884 ip->ip_dst, th->th_dport, lookupflag & ~INPLOOKUP_WILDCARD,
885 m->m_pkthdr.rcvif, m);
886 if (!inp) {
887 /*
888 * It's new. Try to find the ambushing socket.
889 * Because we've rewritten the destination address,
890 * any hardware-generated hash is ignored.
891 */
892 inp = in_pcblookup(&V_tcbinfo, ip->ip_src,
893 th->th_sport, next_hop->sin_addr,
894 next_hop->sin_port ? ntohs(next_hop->sin_port) :
895 th->th_dport, lookupflag, m->m_pkthdr.rcvif);
896 }
897 } else
898 inp = in_pcblookup_mbuf(&V_tcbinfo, ip->ip_src,
899 th->th_sport, ip->ip_dst, th->th_dport, lookupflag,
900 m->m_pkthdr.rcvif, m);
901 #endif /* INET */
902
903 /*
904 * If the INPCB does not exist then all data in the incoming
905 * segment is discarded and an appropriate RST is sent back.
906 * XXX MRT Send RST using which routing table?
907 */
908 if (inp == NULL) {
909 if (rstreason != 0) {
910 /* We came here after second (safety) lookup. */
911 MPASS((lookupflag & INPLOOKUP_WILDCARD) == 0);
912 goto dropwithreset;
913 }
914 /*
915 * Log communication attempts to ports that are not
916 * in use.
917 */
918 if ((V_tcp_log_in_vain == 1 && (thflags & TH_SYN)) ||
919 V_tcp_log_in_vain == 2) {
920 if ((s = tcp_log_vain(NULL, th, (void *)ip, ip6)))
921 log(LOG_INFO, "%s; %s: Connection attempt "
922 "to closed port\n", s, __func__);
923 }
924 rstreason = BANDLIM_RST_CLOSEDPORT;
925 goto dropwithreset;
926 }
927 INP_LOCK_ASSERT(inp);
928
929 if ((inp->inp_flowtype == M_HASHTYPE_NONE) &&
930 !SOLISTENING(inp->inp_socket)) {
931 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE) {
932 inp->inp_flowid = m->m_pkthdr.flowid;
933 inp->inp_flowtype = M_HASHTYPE_GET(m);
934 #ifdef RSS
935 } else {
936 /* assign flowid by software RSS hash */
937 #ifdef INET6
938 if (isipv6) {
939 rss_proto_software_hash_v6(&inp->in6p_faddr,
940 &inp->in6p_laddr,
941 inp->inp_fport,
942 inp->inp_lport,
943 IPPROTO_TCP,
944 &inp->inp_flowid,
945 &inp->inp_flowtype);
946 } else
947 #endif /* INET6 */
948 {
949 rss_proto_software_hash_v4(inp->inp_faddr,
950 inp->inp_laddr,
951 inp->inp_fport,
952 inp->inp_lport,
953 IPPROTO_TCP,
954 &inp->inp_flowid,
955 &inp->inp_flowtype);
956 }
957 #endif /* RSS */
958 }
959 }
960 #if defined(IPSEC) || defined(IPSEC_SUPPORT)
961 #ifdef INET6
962 if (isipv6 && IPSEC_ENABLED(ipv6) &&
963 IPSEC_CHECK_POLICY(ipv6, m, inp) != 0) {
964 goto dropunlock;
965 }
966 #ifdef INET
967 else
968 #endif
969 #endif /* INET6 */
970 #ifdef INET
971 if (IPSEC_ENABLED(ipv4) &&
972 IPSEC_CHECK_POLICY(ipv4, m, inp) != 0) {
973 goto dropunlock;
974 }
975 #endif /* INET */
976 #endif /* IPSEC */
977
978 /*
979 * Check the minimum TTL for socket.
980 */
981 if (inp->inp_ip_minttl != 0) {
982 #ifdef INET6
983 if (isipv6) {
984 if (inp->inp_ip_minttl > ip6->ip6_hlim)
985 goto dropunlock;
986 } else
987 #endif
988 if (inp->inp_ip_minttl > ip->ip_ttl)
989 goto dropunlock;
990 }
991
992 tp = intotcpcb(inp);
993 switch (tp->t_state) {
994 case TCPS_TIME_WAIT:
995 /*
996 * A previous connection in TIMEWAIT state is supposed to catch
997 * stray or duplicate segments arriving late. If this segment
998 * was a legitimate new connection attempt, the old INPCB gets
999 * removed and we can try again to find a listening socket.
1000 */
1001 tcp_dooptions(&to, optp, optlen,
1002 (thflags & TH_SYN) ? TO_SYN : 0);
1003 /*
1004 * tcp_twcheck unlocks the inp always, and frees the m if fails.
1005 */
1006 if (tcp_twcheck(inp, &to, th, m, tlen))
1007 goto findpcb;
1008 return (IPPROTO_DONE);
1009 case TCPS_CLOSED:
1010 /*
1011 * The TCPCB may no longer exist if the connection is winding
1012 * down or it is in the CLOSED state. Either way we drop the
1013 * segment and send an appropriate response.
1014 */
1015 rstreason = BANDLIM_RST_CLOSEDPORT;
1016 goto dropwithreset;
1017 }
1018
1019 if ((tp->t_port != port) && (tp->t_state > TCPS_LISTEN)) {
1020 rstreason = BANDLIM_RST_CLOSEDPORT;
1021 goto dropwithreset;
1022 }
1023
1024 #ifdef TCP_OFFLOAD
1025 if (tp->t_flags & TF_TOE) {
1026 tcp_offload_input(tp, m);
1027 m = NULL; /* consumed by the TOE driver */
1028 goto dropunlock;
1029 }
1030 #endif
1031
1032 #ifdef MAC
1033 if (mac_inpcb_check_deliver(inp, m))
1034 goto dropunlock;
1035 #endif
1036 so = inp->inp_socket;
1037 KASSERT(so != NULL, ("%s: so == NULL", __func__));
1038 /*
1039 * When the socket is accepting connections (the INPCB is in LISTEN
1040 * state) we look into the SYN cache if this is a new connection
1041 * attempt or the completion of a previous one.
1042 */
1043 KASSERT(tp->t_state == TCPS_LISTEN || !SOLISTENING(so),
1044 ("%s: so accepting but tp %p not listening", __func__, tp));
1045 if (tp->t_state == TCPS_LISTEN && SOLISTENING(so)) {
1046 struct in_conninfo inc;
1047
1048 bzero(&inc, sizeof(inc));
1049 #ifdef INET6
1050 if (isipv6) {
1051 inc.inc_flags |= INC_ISIPV6;
1052 if (inp->inp_inc.inc_flags & INC_IPV6MINMTU)
1053 inc.inc_flags |= INC_IPV6MINMTU;
1054 inc.inc6_faddr = ip6->ip6_src;
1055 inc.inc6_laddr = ip6->ip6_dst;
1056 } else
1057 #endif
1058 {
1059 inc.inc_faddr = ip->ip_src;
1060 inc.inc_laddr = ip->ip_dst;
1061 }
1062 inc.inc_fport = th->th_sport;
1063 inc.inc_lport = th->th_dport;
1064 inc.inc_fibnum = so->so_fibnum;
1065
1066 /*
1067 * Check for an existing connection attempt in syncache if
1068 * the flag is only ACK. A successful lookup creates a new
1069 * socket appended to the listen queue in SYN_RECEIVED state.
1070 */
1071 if ((thflags & (TH_RST|TH_ACK|TH_SYN)) == TH_ACK) {
1072 /*
1073 * Parse the TCP options here because
1074 * syncookies need access to the reflected
1075 * timestamp.
1076 */
1077 tcp_dooptions(&to, optp, optlen, 0);
1078 /*
1079 * NB: syncache_expand() doesn't unlock inp.
1080 */
1081 rstreason = syncache_expand(&inc, &to, th, &so, m, port);
1082 if (rstreason < 0) {
1083 /*
1084 * A failing TCP MD5 signature comparison
1085 * must result in the segment being dropped
1086 * and must not produce any response back
1087 * to the sender.
1088 */
1089 goto dropunlock;
1090 } else if (rstreason == 0) {
1091 /*
1092 * No syncache entry, or ACK was not for our
1093 * SYN/ACK. Do our protection against double
1094 * ACK. If peer sent us 2 ACKs, then for the
1095 * first one syncache_expand() successfully
1096 * converted syncache entry into a socket,
1097 * while we were waiting on the inpcb lock. We
1098 * don't want to sent RST for the second ACK,
1099 * so we perform second lookup without wildcard
1100 * match, hoping to find the new socket. If
1101 * the ACK is stray indeed, rstreason would
1102 * hint the above code that the lookup was a
1103 * second attempt.
1104 *
1105 * NB: syncache did its own logging
1106 * of the failure cause.
1107 */
1108 INP_WUNLOCK(inp);
1109 rstreason = BANDLIM_RST_OPENPORT;
1110 lookupflag &= ~INPLOOKUP_WILDCARD;
1111 goto findpcb;
1112 }
1113 tfo_socket_result:
1114 if (so == NULL) {
1115 /*
1116 * We completed the 3-way handshake
1117 * but could not allocate a socket
1118 * either due to memory shortage,
1119 * listen queue length limits or
1120 * global socket limits. Send RST
1121 * or wait and have the remote end
1122 * retransmit the ACK for another
1123 * try.
1124 */
1125 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1126 log(LOG_DEBUG, "%s; %s: Listen socket: "
1127 "Socket allocation failed due to "
1128 "limits or memory shortage, %s\n",
1129 s, __func__,
1130 V_tcp_sc_rst_sock_fail ?
1131 "sending RST" : "try again");
1132 if (V_tcp_sc_rst_sock_fail) {
1133 rstreason = BANDLIM_UNLIMITED;
1134 goto dropwithreset;
1135 } else
1136 goto dropunlock;
1137 }
1138 /*
1139 * Socket is created in state SYN_RECEIVED.
1140 * Unlock the listen socket, lock the newly
1141 * created socket and update the tp variable.
1142 * If we came here via jump to tfo_socket_result,
1143 * then listening socket is read-locked.
1144 */
1145 INP_UNLOCK(inp); /* listen socket */
1146 inp = sotoinpcb(so);
1147 /*
1148 * New connection inpcb is already locked by
1149 * syncache_expand().
1150 */
1151 INP_WLOCK_ASSERT(inp);
1152 tp = intotcpcb(inp);
1153 KASSERT(tp->t_state == TCPS_SYN_RECEIVED,
1154 ("%s: ", __func__));
1155 /*
1156 * Process the segment and the data it
1157 * contains. tcp_do_segment() consumes
1158 * the mbuf chain and unlocks the inpcb.
1159 */
1160 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1161 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen,
1162 tlen, iptos);
1163 return (IPPROTO_DONE);
1164 }
1165 /*
1166 * Segment flag validation for new connection attempts:
1167 *
1168 * Our (SYN|ACK) response was rejected.
1169 * Check with syncache and remove entry to prevent
1170 * retransmits.
1171 *
1172 * NB: syncache_chkrst does its own logging of failure
1173 * causes.
1174 */
1175 if (thflags & TH_RST) {
1176 syncache_chkrst(&inc, th, m, port);
1177 goto dropunlock;
1178 }
1179 /*
1180 * We can't do anything without SYN.
1181 */
1182 if ((thflags & TH_SYN) == 0) {
1183 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1184 log(LOG_DEBUG, "%s; %s: Listen socket: "
1185 "SYN is missing, segment ignored\n",
1186 s, __func__);
1187 TCPSTAT_INC(tcps_badsyn);
1188 goto dropunlock;
1189 }
1190 /*
1191 * (SYN|ACK) is bogus on a listen socket.
1192 */
1193 if (thflags & TH_ACK) {
1194 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1195 log(LOG_DEBUG, "%s; %s: Listen socket: "
1196 "SYN|ACK invalid, segment rejected\n",
1197 s, __func__);
1198 syncache_badack(&inc, port); /* XXX: Not needed! */
1199 TCPSTAT_INC(tcps_badsyn);
1200 rstreason = BANDLIM_RST_OPENPORT;
1201 goto dropwithreset;
1202 }
1203 /*
1204 * If the drop_synfin option is enabled, drop all
1205 * segments with both the SYN and FIN bits set.
1206 * This prevents e.g. nmap from identifying the
1207 * TCP/IP stack.
1208 * XXX: Poor reasoning. nmap has other methods
1209 * and is constantly refining its stack detection
1210 * strategies.
1211 * XXX: This is a violation of the TCP specification
1212 * and was used by RFC1644.
1213 */
1214 if ((thflags & TH_FIN) && V_drop_synfin) {
1215 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1216 log(LOG_DEBUG, "%s; %s: Listen socket: "
1217 "SYN|FIN segment ignored (based on "
1218 "sysctl setting)\n", s, __func__);
1219 TCPSTAT_INC(tcps_badsyn);
1220 goto dropunlock;
1221 }
1222 /*
1223 * Segment's flags are (SYN) or (SYN|FIN).
1224 *
1225 * TH_PUSH, TH_URG, TH_ECE, TH_CWR are ignored
1226 * as they do not affect the state of the TCP FSM.
1227 * The data pointed to by TH_URG and th_urp is ignored.
1228 */
1229 KASSERT((thflags & (TH_RST|TH_ACK)) == 0,
1230 ("%s: Listen socket: TH_RST or TH_ACK set", __func__));
1231 KASSERT(thflags & (TH_SYN),
1232 ("%s: Listen socket: TH_SYN not set", __func__));
1233 INP_RLOCK_ASSERT(inp);
1234 #ifdef INET6
1235 /*
1236 * If deprecated address is forbidden,
1237 * we do not accept SYN to deprecated interface
1238 * address to prevent any new inbound connection from
1239 * getting established.
1240 * When we do not accept SYN, we send a TCP RST,
1241 * with deprecated source address (instead of dropping
1242 * it). We compromise it as it is much better for peer
1243 * to send a RST, and RST will be the final packet
1244 * for the exchange.
1245 *
1246 * If we do not forbid deprecated addresses, we accept
1247 * the SYN packet. RFC2462 does not suggest dropping
1248 * SYN in this case.
1249 * If we decipher RFC2462 5.5.4, it says like this:
1250 * 1. use of deprecated addr with existing
1251 * communication is okay - "SHOULD continue to be
1252 * used"
1253 * 2. use of it with new communication:
1254 * (2a) "SHOULD NOT be used if alternate address
1255 * with sufficient scope is available"
1256 * (2b) nothing mentioned otherwise.
1257 * Here we fall into (2b) case as we have no choice in
1258 * our source address selection - we must obey the peer.
1259 *
1260 * The wording in RFC2462 is confusing, and there are
1261 * multiple description text for deprecated address
1262 * handling - worse, they are not exactly the same.
1263 * I believe 5.5.4 is the best one, so we follow 5.5.4.
1264 */
1265 if (isipv6 && !V_ip6_use_deprecated) {
1266 struct in6_ifaddr *ia6;
1267
1268 ia6 = in6ifa_ifwithaddr(&ip6->ip6_dst, 0 /* XXX */, false);
1269 if (ia6 != NULL &&
1270 (ia6->ia6_flags & IN6_IFF_DEPRECATED)) {
1271 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1272 log(LOG_DEBUG, "%s; %s: Listen socket: "
1273 "Connection attempt to deprecated "
1274 "IPv6 address rejected\n",
1275 s, __func__);
1276 rstreason = BANDLIM_RST_OPENPORT;
1277 goto dropwithreset;
1278 }
1279 }
1280 #endif /* INET6 */
1281 /*
1282 * Basic sanity checks on incoming SYN requests:
1283 * Don't respond if the destination is a link layer
1284 * broadcast according to RFC1122 4.2.3.10, p. 104.
1285 * If it is from this socket it must be forged.
1286 * Don't respond if the source or destination is a
1287 * global or subnet broad- or multicast address.
1288 * Note that it is quite possible to receive unicast
1289 * link-layer packets with a broadcast IP address. Use
1290 * in_broadcast() to find them.
1291 */
1292 if (m->m_flags & (M_BCAST|M_MCAST)) {
1293 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1294 log(LOG_DEBUG, "%s; %s: Listen socket: "
1295 "Connection attempt from broad- or multicast "
1296 "link layer address ignored\n", s, __func__);
1297 goto dropunlock;
1298 }
1299 #ifdef INET6
1300 if (isipv6) {
1301 if (th->th_dport == th->th_sport &&
1302 IN6_ARE_ADDR_EQUAL(&ip6->ip6_dst, &ip6->ip6_src)) {
1303 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1304 log(LOG_DEBUG, "%s; %s: Listen socket: "
1305 "Connection attempt to/from self "
1306 "ignored\n", s, __func__);
1307 goto dropunlock;
1308 }
1309 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
1310 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src)) {
1311 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1312 log(LOG_DEBUG, "%s; %s: Listen socket: "
1313 "Connection attempt from/to multicast "
1314 "address ignored\n", s, __func__);
1315 goto dropunlock;
1316 }
1317 }
1318 #endif
1319 #if defined(INET) && defined(INET6)
1320 else
1321 #endif
1322 #ifdef INET
1323 {
1324 if (th->th_dport == th->th_sport &&
1325 ip->ip_dst.s_addr == ip->ip_src.s_addr) {
1326 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1327 log(LOG_DEBUG, "%s; %s: Listen socket: "
1328 "Connection attempt from/to self "
1329 "ignored\n", s, __func__);
1330 goto dropunlock;
1331 }
1332 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
1333 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
1334 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
1335 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif)) {
1336 if ((s = tcp_log_addrs(&inc, th, NULL, NULL)))
1337 log(LOG_DEBUG, "%s; %s: Listen socket: "
1338 "Connection attempt from/to broad- "
1339 "or multicast address ignored\n",
1340 s, __func__);
1341 goto dropunlock;
1342 }
1343 }
1344 #endif
1345 /*
1346 * SYN appears to be valid. Create compressed TCP state
1347 * for syncache.
1348 */
1349 TCP_PROBE3(debug__input, tp, th, m);
1350 tcp_dooptions(&to, optp, optlen, TO_SYN);
1351 if ((so = syncache_add(&inc, &to, th, inp, so, m, NULL, NULL,
1352 iptos, port)) != NULL)
1353 goto tfo_socket_result;
1354
1355 /*
1356 * Entry added to syncache and mbuf consumed.
1357 * Only the listen socket is unlocked by syncache_add().
1358 */
1359 return (IPPROTO_DONE);
1360 } else if (tp->t_state == TCPS_LISTEN) {
1361 /*
1362 * When a listen socket is torn down the SO_ACCEPTCONN
1363 * flag is removed first while connections are drained
1364 * from the accept queue in a unlock/lock cycle of the
1365 * ACCEPT_LOCK, opening a race condition allowing a SYN
1366 * attempt go through unhandled.
1367 */
1368 goto dropunlock;
1369 }
1370 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1371 if (tp->t_flags & TF_SIGNATURE) {
1372 tcp_dooptions(&to, optp, optlen, thflags);
1373 if ((to.to_flags & TOF_SIGNATURE) == 0) {
1374 TCPSTAT_INC(tcps_sig_err_nosigopt);
1375 goto dropunlock;
1376 }
1377 if (!TCPMD5_ENABLED() ||
1378 TCPMD5_INPUT(m, th, to.to_signature) != 0)
1379 goto dropunlock;
1380 }
1381 #endif
1382 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1383
1384 /*
1385 * Segment belongs to a connection in SYN_SENT, ESTABLISHED or later
1386 * state. tcp_do_segment() always consumes the mbuf chain, unlocks
1387 * the inpcb, and unlocks pcbinfo.
1388 *
1389 * XXXGL: in case of a pure SYN arriving on existing connection
1390 * TCP stacks won't need to modify the PCB, they would either drop
1391 * the segment silently, or send a challenge ACK. However, we try
1392 * to upgrade the lock, because calling convention for stacks is
1393 * write-lock on PCB. If upgrade fails, drop the SYN.
1394 */
1395 if ((lookupflag & INPLOOKUP_RLOCKPCB) && INP_TRY_UPGRADE(inp) == 0)
1396 goto dropunlock;
1397
1398 tp->t_fb->tfb_tcp_do_segment(tp, m, th, drop_hdrlen, tlen, iptos);
1399 return (IPPROTO_DONE);
1400
1401 dropwithreset:
1402 /*
1403 * When blackholing do not respond with a RST but
1404 * completely ignore the segment and drop it.
1405 */
1406 if (((rstreason == BANDLIM_RST_OPENPORT && V_blackhole == 3) ||
1407 (rstreason == BANDLIM_RST_CLOSEDPORT &&
1408 ((V_blackhole == 1 && (thflags & TH_SYN)) || V_blackhole > 1))) &&
1409 (V_blackhole_local || (
1410 #ifdef INET6
1411 isipv6 ? !in6_localaddr(&ip6->ip6_src) :
1412 #endif
1413 #ifdef INET
1414 !in_localip(ip->ip_src)
1415 #else
1416 true
1417 #endif
1418 )))
1419 goto dropunlock;
1420 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1421 tcp_dropwithreset(m, th, tp, tlen, rstreason);
1422 m = NULL; /* mbuf chain got consumed. */
1423
1424 dropunlock:
1425 if (m != NULL)
1426 TCP_PROBE5(receive, NULL, tp, m, tp, th);
1427
1428 if (inp != NULL)
1429 INP_UNLOCK(inp);
1430
1431 drop:
1432 if (s != NULL)
1433 free(s, M_TCPLOG);
1434 if (m != NULL)
1435 m_freem(m);
1436 return (IPPROTO_DONE);
1437 }
1438
1439 /*
1440 * Automatic sizing of receive socket buffer. Often the send
1441 * buffer size is not optimally adjusted to the actual network
1442 * conditions at hand (delay bandwidth product). Setting the
1443 * buffer size too small limits throughput on links with high
1444 * bandwidth and high delay (eg. trans-continental/oceanic links).
1445 *
1446 * On the receive side the socket buffer memory is only rarely
1447 * used to any significant extent. This allows us to be much
1448 * more aggressive in scaling the receive socket buffer. For
1449 * the case that the buffer space is actually used to a large
1450 * extent and we run out of kernel memory we can simply drop
1451 * the new segments; TCP on the sender will just retransmit it
1452 * later. Setting the buffer size too big may only consume too
1453 * much kernel memory if the application doesn't read() from
1454 * the socket or packet loss or reordering makes use of the
1455 * reassembly queue.
1456 *
1457 * The criteria to step up the receive buffer one notch are:
1458 * 1. Application has not set receive buffer size with
1459 * SO_RCVBUF. Setting SO_RCVBUF clears SB_AUTOSIZE.
1460 * 2. the number of bytes received during 1/2 of an sRTT
1461 * is at least 3/8 of the current socket buffer size.
1462 * 3. receive buffer size has not hit maximal automatic size;
1463 *
1464 * If all of the criteria are met we increaset the socket buffer
1465 * by a 1/2 (bounded by the max). This allows us to keep ahead
1466 * of slow-start but also makes it so our peer never gets limited
1467 * by our rwnd which we then open up causing a burst.
1468 *
1469 * This algorithm does two steps per RTT at most and only if
1470 * we receive a bulk stream w/o packet losses or reorderings.
1471 * Shrinking the buffer during idle times is not necessary as
1472 * it doesn't consume any memory when idle.
1473 *
1474 * TODO: Only step up if the application is actually serving
1475 * the buffer to better manage the socket buffer resources.
1476 */
1477 int
tcp_autorcvbuf(struct mbuf * m,struct tcphdr * th,struct socket * so,struct tcpcb * tp,int tlen)1478 tcp_autorcvbuf(struct mbuf *m, struct tcphdr *th, struct socket *so,
1479 struct tcpcb *tp, int tlen)
1480 {
1481 int newsize = 0;
1482
1483 if (V_tcp_do_autorcvbuf && (so->so_rcv.sb_flags & SB_AUTOSIZE) &&
1484 tp->t_srtt != 0 && tp->rfbuf_ts != 0 &&
1485 TCP_TS_TO_TICKS(tcp_ts_getticks() - tp->rfbuf_ts) >
1486 ((tp->t_srtt >> TCP_RTT_SHIFT)/2)) {
1487 if (tp->rfbuf_cnt > ((so->so_rcv.sb_hiwat / 2)/ 4 * 3) &&
1488 so->so_rcv.sb_hiwat < V_tcp_autorcvbuf_max) {
1489 newsize = min((so->so_rcv.sb_hiwat + (so->so_rcv.sb_hiwat/2)), V_tcp_autorcvbuf_max);
1490 }
1491 TCP_PROBE6(receive__autoresize, NULL, tp, m, tp, th, newsize);
1492
1493 /* Start over with next RTT. */
1494 tp->rfbuf_ts = 0;
1495 tp->rfbuf_cnt = 0;
1496 } else {
1497 tp->rfbuf_cnt += tlen; /* add up */
1498 }
1499 return (newsize);
1500 }
1501
1502 int
tcp_input(struct mbuf ** mp,int * offp,int proto)1503 tcp_input(struct mbuf **mp, int *offp, int proto)
1504 {
1505 return(tcp_input_with_port(mp, offp, proto, 0));
1506 }
1507
1508 static void
tcp_handle_wakeup(struct tcpcb * tp)1509 tcp_handle_wakeup(struct tcpcb *tp)
1510 {
1511
1512 INP_WLOCK_ASSERT(tptoinpcb(tp));
1513
1514 if (tp->t_flags & TF_WAKESOR) {
1515 struct socket *so = tptosocket(tp);
1516
1517 tp->t_flags &= ~TF_WAKESOR;
1518 SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1519 sorwakeup_locked(so);
1520 }
1521 }
1522
1523 void
tcp_do_segment(struct tcpcb * tp,struct mbuf * m,struct tcphdr * th,int drop_hdrlen,int tlen,uint8_t iptos)1524 tcp_do_segment(struct tcpcb *tp, struct mbuf *m, struct tcphdr *th,
1525 int drop_hdrlen, int tlen, uint8_t iptos)
1526 {
1527 uint16_t thflags;
1528 int acked, ourfinisacked, needoutput = 0;
1529 sackstatus_t sack_changed;
1530 int rstreason, todrop, win, incforsyn = 0;
1531 uint32_t tiwin;
1532 uint16_t nsegs;
1533 char *s;
1534 struct inpcb *inp = tptoinpcb(tp);
1535 struct socket *so = tptosocket(tp);
1536 struct in_conninfo *inc = &inp->inp_inc;
1537 struct mbuf *mfree;
1538 struct tcpopt to;
1539 int tfo_syn;
1540 u_int maxseg = 0;
1541
1542 thflags = tcp_get_flags(th);
1543 tp->sackhint.last_sack_ack = 0;
1544 sack_changed = SACK_NOCHANGE;
1545 nsegs = max(1, m->m_pkthdr.lro_nsegs);
1546
1547 NET_EPOCH_ASSERT();
1548 INP_WLOCK_ASSERT(inp);
1549 KASSERT(tp->t_state > TCPS_LISTEN, ("%s: TCPS_LISTEN",
1550 __func__));
1551 KASSERT(tp->t_state != TCPS_TIME_WAIT, ("%s: TCPS_TIME_WAIT",
1552 __func__));
1553
1554 #ifdef TCPPCAP
1555 /* Save segment, if requested. */
1556 tcp_pcap_add(th, m, &(tp->t_inpkts));
1557 #endif
1558 TCP_LOG_EVENT(tp, th, &so->so_rcv, &so->so_snd, TCP_LOG_IN, 0,
1559 tlen, NULL, true);
1560
1561 if ((thflags & TH_SYN) && (thflags & TH_FIN) && V_drop_synfin) {
1562 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1563 log(LOG_DEBUG, "%s; %s: "
1564 "SYN|FIN segment ignored (based on "
1565 "sysctl setting)\n", s, __func__);
1566 free(s, M_TCPLOG);
1567 }
1568 goto drop;
1569 }
1570
1571 /*
1572 * If a segment with the ACK-bit set arrives in the SYN-SENT state
1573 * check SEQ.ACK first.
1574 */
1575 if ((tp->t_state == TCPS_SYN_SENT) && (thflags & TH_ACK) &&
1576 (SEQ_LEQ(th->th_ack, tp->iss) || SEQ_GT(th->th_ack, tp->snd_max))) {
1577 rstreason = BANDLIM_UNLIMITED;
1578 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1579 goto dropwithreset;
1580 }
1581
1582 /*
1583 * Segment received on connection.
1584 * Reset idle time and keep-alive timer.
1585 * XXX: This should be done after segment
1586 * validation to ignore broken/spoofed segs.
1587 */
1588 if (tp->t_idle_reduce &&
1589 (tp->snd_max == tp->snd_una) &&
1590 ((ticks - tp->t_rcvtime) >= tp->t_rxtcur))
1591 cc_after_idle(tp);
1592 tp->t_rcvtime = ticks;
1593
1594 if (thflags & TH_FIN)
1595 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_FIN);
1596 /*
1597 * Scale up the window into a 32-bit value.
1598 * For the SYN_SENT state the scale is zero.
1599 */
1600 tiwin = th->th_win << tp->snd_scale;
1601 #ifdef STATS
1602 stats_voi_update_abs_ulong(tp->t_stats, VOI_TCP_FRWIN, tiwin);
1603 #endif
1604
1605 /*
1606 * TCP ECN processing.
1607 */
1608 if (tcp_ecn_input_segment(tp, thflags, tlen,
1609 tcp_packets_this_ack(tp, th->th_ack),
1610 iptos))
1611 cc_cong_signal(tp, th, CC_ECN);
1612
1613 /*
1614 * Parse options on any incoming segment.
1615 */
1616 tcp_dooptions(&to, (u_char *)(th + 1),
1617 (th->th_off << 2) - sizeof(struct tcphdr),
1618 (thflags & TH_SYN) ? TO_SYN : 0);
1619 if (tp->t_flags2 & TF2_PROC_SACK_PROHIBIT) {
1620 /*
1621 * We don't look at sack's from the
1622 * peer because the MSS is too small which
1623 * can subject us to an attack.
1624 */
1625 to.to_flags &= ~TOF_SACK;
1626 }
1627 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
1628 if ((tp->t_flags & TF_SIGNATURE) != 0 &&
1629 (to.to_flags & TOF_SIGNATURE) == 0) {
1630 TCPSTAT_INC(tcps_sig_err_sigopt);
1631 /* XXX: should drop? */
1632 }
1633 #endif
1634 /*
1635 * If echoed timestamp is later than the current time,
1636 * fall back to non RFC1323 RTT calculation. Normalize
1637 * timestamp if syncookies were used when this connection
1638 * was established.
1639 */
1640 if ((to.to_flags & TOF_TS) && (to.to_tsecr != 0)) {
1641 to.to_tsecr -= tp->ts_offset;
1642 if (TSTMP_GT(to.to_tsecr, tcp_ts_getticks())) {
1643 to.to_tsecr = 0;
1644 } else if (tp->t_rxtshift == 1 &&
1645 tp->t_flags & TF_PREVVALID &&
1646 tp->t_badrxtwin != 0 &&
1647 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin)) {
1648 cc_cong_signal(tp, th, CC_RTO_ERR);
1649 }
1650 }
1651 /*
1652 * Process options only when we get SYN/ACK back. The SYN case
1653 * for incoming connections is handled in tcp_syncache.
1654 * According to RFC1323 the window field in a SYN (i.e., a <SYN>
1655 * or <SYN,ACK>) segment itself is never scaled.
1656 * XXX this is traditional behavior, may need to be cleaned up.
1657 */
1658 if (tp->t_state == TCPS_SYN_SENT && (thflags & TH_SYN)) {
1659 /* Handle parallel SYN for ECN */
1660 tcp_ecn_input_parallel_syn(tp, thflags, iptos);
1661 if ((to.to_flags & TOF_SCALE) &&
1662 (tp->t_flags & TF_REQ_SCALE) &&
1663 !(tp->t_flags & TF_NOOPT)) {
1664 tp->t_flags |= TF_RCVD_SCALE;
1665 tp->snd_scale = to.to_wscale;
1666 } else {
1667 tp->t_flags &= ~TF_REQ_SCALE;
1668 }
1669 /*
1670 * Initial send window. It will be updated with
1671 * the next incoming segment to the scaled value.
1672 */
1673 tp->snd_wnd = th->th_win;
1674 if ((to.to_flags & TOF_TS) &&
1675 (tp->t_flags & TF_REQ_TSTMP) &&
1676 !(tp->t_flags & TF_NOOPT)) {
1677 tp->t_flags |= TF_RCVD_TSTMP;
1678 tp->ts_recent = to.to_tsval;
1679 tp->ts_recent_age = tcp_ts_getticks();
1680 } else {
1681 tp->t_flags &= ~TF_REQ_TSTMP;
1682 }
1683 if (to.to_flags & TOF_MSS) {
1684 tcp_mss(tp, to.to_mss);
1685 }
1686 if ((tp->t_flags & TF_SACK_PERMIT) &&
1687 (!(to.to_flags & TOF_SACKPERM) ||
1688 (tp->t_flags & TF_NOOPT))) {
1689 tp->t_flags &= ~TF_SACK_PERMIT;
1690 }
1691 if (tp->t_flags & TF_FASTOPEN) {
1692 if ((to.to_flags & TOF_FASTOPEN) &&
1693 !(tp->t_flags & TF_NOOPT)) {
1694 uint16_t mss;
1695
1696 if (to.to_flags & TOF_MSS) {
1697 mss = to.to_mss;
1698 } else {
1699 if ((inp->inp_vflag & INP_IPV6) != 0) {
1700 mss = TCP6_MSS;
1701 } else {
1702 mss = TCP_MSS;
1703 }
1704 }
1705 tcp_fastopen_update_cache(tp, mss,
1706 to.to_tfo_len, to.to_tfo_cookie);
1707 } else {
1708 tcp_fastopen_disable_path(tp);
1709 }
1710 }
1711 }
1712
1713 /*
1714 * If timestamps were negotiated during SYN/ACK and a
1715 * segment without a timestamp is received, silently drop
1716 * the segment, unless it is a RST segment or missing timestamps are
1717 * tolerated.
1718 * See section 3.2 of RFC 7323.
1719 */
1720 if ((tp->t_flags & TF_RCVD_TSTMP) && !(to.to_flags & TOF_TS)) {
1721 if (((thflags & TH_RST) != 0) || V_tcp_tolerate_missing_ts) {
1722 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1723 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1724 "segment processed normally\n",
1725 s, __func__);
1726 free(s, M_TCPLOG);
1727 }
1728 } else {
1729 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1730 log(LOG_DEBUG, "%s; %s: Timestamp missing, "
1731 "segment silently dropped\n", s, __func__);
1732 free(s, M_TCPLOG);
1733 }
1734 goto drop;
1735 }
1736 }
1737 /*
1738 * If timestamps were not negotiated during SYN/ACK and a
1739 * segment with a timestamp is received, ignore the
1740 * timestamp and process the packet normally.
1741 * See section 3.2 of RFC 7323.
1742 */
1743 if (!(tp->t_flags & TF_RCVD_TSTMP) && (to.to_flags & TOF_TS)) {
1744 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
1745 log(LOG_DEBUG, "%s; %s: Timestamp not expected, "
1746 "segment processed normally\n", s, __func__);
1747 free(s, M_TCPLOG);
1748 }
1749 }
1750
1751 /*
1752 * Header prediction: check for the two common cases
1753 * of a uni-directional data xfer. If the packet has
1754 * no control flags, is in-sequence, the window didn't
1755 * change and we're not retransmitting, it's a
1756 * candidate. If the length is zero and the ack moved
1757 * forward, we're the sender side of the xfer. Just
1758 * free the data acked & wake any higher level process
1759 * that was blocked waiting for space. If the length
1760 * is non-zero and the ack didn't move, we're the
1761 * receiver side. If we're getting packets in-order
1762 * (the reassembly queue is empty), add the data to
1763 * the socket buffer and note that we need a delayed ack.
1764 * Make sure that the hidden state-flags are also off.
1765 * Since we check for TCPS_ESTABLISHED first, it can only
1766 * be TH_NEEDSYN.
1767 */
1768 if (tp->t_state == TCPS_ESTABLISHED &&
1769 th->th_seq == tp->rcv_nxt &&
1770 (thflags & (TH_SYN|TH_FIN|TH_RST|TH_URG|TH_ACK)) == TH_ACK &&
1771 tp->snd_nxt == tp->snd_max &&
1772 tiwin && tiwin == tp->snd_wnd &&
1773 ((tp->t_flags & (TF_NEEDSYN|TF_NEEDFIN)) == 0) &&
1774 SEGQ_EMPTY(tp) &&
1775 ((to.to_flags & TOF_TS) == 0 ||
1776 TSTMP_GEQ(to.to_tsval, tp->ts_recent)) ) {
1777 /*
1778 * If last ACK falls within this segment's sequence numbers,
1779 * record the timestamp.
1780 * NOTE that the test is modified according to the latest
1781 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
1782 */
1783 if ((to.to_flags & TOF_TS) != 0 &&
1784 SEQ_LEQ(th->th_seq, tp->last_ack_sent)) {
1785 tp->ts_recent_age = tcp_ts_getticks();
1786 tp->ts_recent = to.to_tsval;
1787 }
1788
1789 if (tlen == 0) {
1790 if (SEQ_GT(th->th_ack, tp->snd_una) &&
1791 SEQ_LEQ(th->th_ack, tp->snd_max) &&
1792 !IN_RECOVERY(tp->t_flags) &&
1793 (to.to_flags & TOF_SACK) == 0 &&
1794 TAILQ_EMPTY(&tp->snd_holes)) {
1795 /*
1796 * This is a pure ack for outstanding data.
1797 */
1798 TCPSTAT_INC(tcps_predack);
1799
1800 /*
1801 * "bad retransmit" recovery without timestamps.
1802 */
1803 if ((to.to_flags & TOF_TS) == 0 &&
1804 tp->t_rxtshift == 1 &&
1805 tp->t_flags & TF_PREVVALID &&
1806 tp->t_badrxtwin != 0 &&
1807 TSTMP_LT(ticks, tp->t_badrxtwin)) {
1808 cc_cong_signal(tp, th, CC_RTO_ERR);
1809 }
1810
1811 /*
1812 * Recalculate the transmit timer / rtt.
1813 *
1814 * Some boxes send broken timestamp replies
1815 * during the SYN+ACK phase, ignore
1816 * timestamps of 0 or we could calculate a
1817 * huge RTT and blow up the retransmit timer.
1818 */
1819 if ((to.to_flags & TOF_TS) != 0 &&
1820 to.to_tsecr) {
1821 uint32_t t;
1822
1823 t = tcp_ts_getticks() - to.to_tsecr;
1824 if (!tp->t_rttlow || tp->t_rttlow > t)
1825 tp->t_rttlow = t;
1826 tcp_xmit_timer(tp,
1827 TCP_TS_TO_TICKS(t) + 1);
1828 } else if (tp->t_rtttime &&
1829 SEQ_GT(th->th_ack, tp->t_rtseq)) {
1830 if (!tp->t_rttlow ||
1831 tp->t_rttlow > ticks - tp->t_rtttime)
1832 tp->t_rttlow = ticks - tp->t_rtttime;
1833 tcp_xmit_timer(tp,
1834 ticks - tp->t_rtttime);
1835 }
1836 acked = BYTES_THIS_ACK(tp, th);
1837
1838 #ifdef TCP_HHOOK
1839 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
1840 hhook_run_tcp_est_in(tp, th, &to);
1841 #endif
1842
1843 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
1844 TCPSTAT_ADD(tcps_rcvackbyte, acked);
1845 sbdrop(&so->so_snd, acked);
1846 if (SEQ_GT(tp->snd_una, tp->snd_recover) &&
1847 SEQ_LEQ(th->th_ack, tp->snd_recover))
1848 tp->snd_recover = th->th_ack - 1;
1849
1850 /*
1851 * Let the congestion control algorithm update
1852 * congestion control related information. This
1853 * typically means increasing the congestion
1854 * window.
1855 */
1856 cc_ack_received(tp, th, nsegs, CC_ACK);
1857
1858 tp->snd_una = th->th_ack;
1859 /*
1860 * Pull snd_wl2 up to prevent seq wrap relative
1861 * to th_ack.
1862 */
1863 tp->snd_wl2 = th->th_ack;
1864 tp->t_dupacks = 0;
1865 m_freem(m);
1866
1867 /*
1868 * If all outstanding data are acked, stop
1869 * retransmit timer, otherwise restart timer
1870 * using current (possibly backed-off) value.
1871 * If process is waiting for space,
1872 * wakeup/selwakeup/signal. If data
1873 * are ready to send, let tcp_output
1874 * decide between more output or persist.
1875 */
1876 TCP_PROBE3(debug__input, tp, th, m);
1877 /*
1878 * Clear t_acktime if remote side has ACKd
1879 * all data in the socket buffer.
1880 * Otherwise, update t_acktime if we received
1881 * a sufficiently large ACK.
1882 */
1883 if (sbavail(&so->so_snd) == 0)
1884 tp->t_acktime = 0;
1885 else if (acked > 1)
1886 tp->t_acktime = ticks;
1887 if (tp->snd_una == tp->snd_max)
1888 tcp_timer_activate(tp, TT_REXMT, 0);
1889 else if (!tcp_timer_active(tp, TT_PERSIST))
1890 tcp_timer_activate(tp, TT_REXMT,
1891 TP_RXTCUR(tp));
1892 sowwakeup(so);
1893 /*
1894 * Only call tcp_output when there
1895 * is new data available to be sent
1896 * or we need to send an ACK.
1897 */
1898 if ((tp->t_flags & TF_ACKNOW) ||
1899 (sbavail(&so->so_snd) >=
1900 SEQ_SUB(tp->snd_max, tp->snd_una))) {
1901 (void) tcp_output(tp);
1902 }
1903 goto check_delack;
1904 }
1905 } else if (th->th_ack == tp->snd_una &&
1906 tlen <= sbspace(&so->so_rcv)) {
1907 int newsize = 0; /* automatic sockbuf scaling */
1908
1909 /*
1910 * This is a pure, in-sequence data packet with
1911 * nothing on the reassembly queue and we have enough
1912 * buffer space to take it.
1913 */
1914 /* Clean receiver SACK report if present */
1915 if ((tp->t_flags & TF_SACK_PERMIT) && tp->rcv_numsacks)
1916 tcp_clean_sackreport(tp);
1917 TCPSTAT_INC(tcps_preddat);
1918 tp->rcv_nxt += tlen;
1919 if (tlen &&
1920 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1921 (tp->t_fbyte_in == 0)) {
1922 tp->t_fbyte_in = ticks;
1923 if (tp->t_fbyte_in == 0)
1924 tp->t_fbyte_in = 1;
1925 if (tp->t_fbyte_out && tp->t_fbyte_in)
1926 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1927 }
1928 /*
1929 * Pull snd_wl1 up to prevent seq wrap relative to
1930 * th_seq.
1931 */
1932 tp->snd_wl1 = th->th_seq;
1933 /*
1934 * Pull rcv_up up to prevent seq wrap relative to
1935 * rcv_nxt.
1936 */
1937 tp->rcv_up = tp->rcv_nxt;
1938 TCPSTAT_ADD(tcps_rcvpack, nsegs);
1939 TCPSTAT_ADD(tcps_rcvbyte, tlen);
1940 TCP_PROBE3(debug__input, tp, th, m);
1941
1942 newsize = tcp_autorcvbuf(m, th, so, tp, tlen);
1943
1944 /* Add data to socket buffer. */
1945 SOCKBUF_LOCK(&so->so_rcv);
1946 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1947 m_freem(m);
1948 } else {
1949 /*
1950 * Set new socket buffer size.
1951 * Give up when limit is reached.
1952 */
1953 if (newsize)
1954 if (!sbreserve_locked(so, SO_RCV,
1955 newsize, NULL))
1956 so->so_rcv.sb_flags &= ~SB_AUTOSIZE;
1957 m_adj(m, drop_hdrlen); /* delayed header drop */
1958 sbappendstream_locked(&so->so_rcv, m, 0);
1959 }
1960 /* NB: sorwakeup_locked() does an implicit unlock. */
1961 sorwakeup_locked(so);
1962 if (DELAY_ACK(tp, tlen)) {
1963 tp->t_flags |= TF_DELACK;
1964 } else {
1965 tp->t_flags |= TF_ACKNOW;
1966 (void) tcp_output(tp);
1967 }
1968 goto check_delack;
1969 }
1970 }
1971
1972 /*
1973 * Calculate amount of space in receive window,
1974 * and then do TCP input processing.
1975 * Receive window is amount of space in rcv queue,
1976 * but not less than advertised window.
1977 */
1978 win = sbspace(&so->so_rcv);
1979 if (win < 0)
1980 win = 0;
1981 tp->rcv_wnd = imax(win, (int)(tp->rcv_adv - tp->rcv_nxt));
1982
1983 switch (tp->t_state) {
1984 /*
1985 * If the state is SYN_RECEIVED:
1986 * if seg contains an ACK, but not for our SYN/ACK, send a RST.
1987 */
1988 case TCPS_SYN_RECEIVED:
1989 if (thflags & TH_RST) {
1990 /* Handle RST segments later. */
1991 break;
1992 }
1993 if ((thflags & TH_ACK) &&
1994 (SEQ_LEQ(th->th_ack, tp->snd_una) ||
1995 SEQ_GT(th->th_ack, tp->snd_max))) {
1996 rstreason = BANDLIM_RST_OPENPORT;
1997 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
1998 goto dropwithreset;
1999 }
2000 if (tp->t_flags & TF_FASTOPEN) {
2001 /*
2002 * When a TFO connection is in SYN_RECEIVED, the
2003 * only valid packets are the initial SYN, a
2004 * retransmit/copy of the initial SYN (possibly with
2005 * a subset of the original data), a valid ACK, a
2006 * FIN, or a RST.
2007 */
2008 if ((thflags & (TH_SYN|TH_ACK)) == (TH_SYN|TH_ACK)) {
2009 rstreason = BANDLIM_RST_OPENPORT;
2010 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2011 goto dropwithreset;
2012 } else if (thflags & TH_SYN) {
2013 /* non-initial SYN is ignored */
2014 if ((tcp_timer_active(tp, TT_DELACK) ||
2015 tcp_timer_active(tp, TT_REXMT)))
2016 goto drop;
2017 } else if (!(thflags & (TH_ACK|TH_FIN|TH_RST))) {
2018 goto drop;
2019 }
2020 }
2021 break;
2022
2023 /*
2024 * If the state is SYN_SENT:
2025 * if seg contains a RST with valid ACK (SEQ.ACK has already
2026 * been verified), then drop the connection.
2027 * if seg contains a RST without an ACK, drop the seg.
2028 * if seg does not contain SYN, then drop the seg.
2029 * Otherwise this is an acceptable SYN segment
2030 * initialize tp->rcv_nxt and tp->irs
2031 * if seg contains ack then advance tp->snd_una
2032 * if seg contains an ECE and ECN support is enabled, the stream
2033 * is ECN capable.
2034 * if SYN has been acked change to ESTABLISHED else SYN_RCVD state
2035 * arrange for segment to be acked (eventually)
2036 * continue processing rest of data/controls, beginning with URG
2037 */
2038 case TCPS_SYN_SENT:
2039 if ((thflags & (TH_ACK|TH_RST)) == (TH_ACK|TH_RST)) {
2040 TCP_PROBE5(connect__refused, NULL, tp,
2041 m, tp, th);
2042 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2043 tp = tcp_drop(tp, ECONNREFUSED);
2044 }
2045 if (thflags & TH_RST)
2046 goto drop;
2047 if (!(thflags & TH_SYN))
2048 goto drop;
2049
2050 tp->irs = th->th_seq;
2051 tcp_rcvseqinit(tp);
2052 if (thflags & TH_ACK) {
2053 int tfo_partial_ack = 0;
2054
2055 TCPSTAT_INC(tcps_connects);
2056 soisconnected(so);
2057 #ifdef MAC
2058 mac_socketpeer_set_from_mbuf(m, so);
2059 #endif
2060 /* Do window scaling on this connection? */
2061 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2062 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2063 tp->rcv_scale = tp->request_r_scale;
2064 }
2065 tp->rcv_adv += min(tp->rcv_wnd,
2066 TCP_MAXWIN << tp->rcv_scale);
2067 tp->snd_una++; /* SYN is acked */
2068 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
2069 tp->snd_nxt = tp->snd_una;
2070 /*
2071 * If not all the data that was sent in the TFO SYN
2072 * has been acked, resend the remainder right away.
2073 */
2074 if ((tp->t_flags & TF_FASTOPEN) &&
2075 (tp->snd_una != tp->snd_max)) {
2076 tp->snd_nxt = th->th_ack;
2077 tfo_partial_ack = 1;
2078 }
2079 /*
2080 * If there's data, delay ACK; if there's also a FIN
2081 * ACKNOW will be turned on later.
2082 */
2083 if (DELAY_ACK(tp, tlen) && tlen != 0 && !tfo_partial_ack)
2084 tcp_timer_activate(tp, TT_DELACK,
2085 tcp_delacktime);
2086 else
2087 tp->t_flags |= TF_ACKNOW;
2088
2089 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2090
2091 /*
2092 * Received <SYN,ACK> in SYN_SENT[*] state.
2093 * Transitions:
2094 * SYN_SENT --> ESTABLISHED
2095 * SYN_SENT* --> FIN_WAIT_1
2096 */
2097 tp->t_starttime = ticks;
2098 if (tp->t_flags & TF_NEEDFIN) {
2099 tp->t_acktime = ticks;
2100 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2101 tp->t_flags &= ~TF_NEEDFIN;
2102 thflags &= ~TH_SYN;
2103 } else {
2104 tcp_state_change(tp, TCPS_ESTABLISHED);
2105 TCP_PROBE5(connect__established, NULL, tp,
2106 m, tp, th);
2107 cc_conn_init(tp);
2108 tcp_timer_activate(tp, TT_KEEP,
2109 TP_KEEPIDLE(tp));
2110 }
2111 } else {
2112 /*
2113 * Received initial SYN in SYN-SENT[*] state =>
2114 * simultaneous open.
2115 * If it succeeds, connection is * half-synchronized.
2116 * Otherwise, do 3-way handshake:
2117 * SYN-SENT -> SYN-RECEIVED
2118 * SYN-SENT* -> SYN-RECEIVED*
2119 */
2120 tp->t_flags |= (TF_ACKNOW | TF_NEEDSYN | TF_SONOTCONN);
2121 tcp_timer_activate(tp, TT_REXMT, 0);
2122 tcp_state_change(tp, TCPS_SYN_RECEIVED);
2123 }
2124
2125 /*
2126 * Advance th->th_seq to correspond to first data byte.
2127 * If data, trim to stay within window,
2128 * dropping FIN if necessary.
2129 */
2130 th->th_seq++;
2131 if (tlen > tp->rcv_wnd) {
2132 todrop = tlen - tp->rcv_wnd;
2133 m_adj(m, -todrop);
2134 tlen = tp->rcv_wnd;
2135 thflags &= ~TH_FIN;
2136 TCPSTAT_INC(tcps_rcvpackafterwin);
2137 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2138 }
2139 tp->snd_wl1 = th->th_seq - 1;
2140 tp->rcv_up = th->th_seq;
2141 /*
2142 * Client side of transaction: already sent SYN and data.
2143 * If the remote host used T/TCP to validate the SYN,
2144 * our data will be ACK'd; if so, enter normal data segment
2145 * processing in the middle of step 5, ack processing.
2146 * Otherwise, goto step 6.
2147 */
2148 if (thflags & TH_ACK)
2149 goto process_ACK;
2150
2151 goto step6;
2152 }
2153
2154 /*
2155 * States other than LISTEN or SYN_SENT.
2156 * First check the RST flag and sequence number since reset segments
2157 * are exempt from the timestamp and connection count tests. This
2158 * fixes a bug introduced by the Stevens, vol. 2, p. 960 bugfix
2159 * below which allowed reset segments in half the sequence space
2160 * to fall though and be processed (which gives forged reset
2161 * segments with a random sequence number a 50 percent chance of
2162 * killing a connection).
2163 * Then check timestamp, if present.
2164 * Then check the connection count, if present.
2165 * Then check that at least some bytes of segment are within
2166 * receive window. If segment begins before rcv_nxt,
2167 * drop leading data (and SYN); if nothing left, just ack.
2168 */
2169 if (thflags & TH_RST) {
2170 /*
2171 * RFC5961 Section 3.2
2172 *
2173 * - RST drops connection only if SEG.SEQ == RCV.NXT.
2174 * - If RST is in window, we send challenge ACK.
2175 *
2176 * Note: to take into account delayed ACKs, we should
2177 * test against last_ack_sent instead of rcv_nxt.
2178 * Note 2: we handle special case of closed window, not
2179 * covered by the RFC.
2180 */
2181 if ((SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2182 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) ||
2183 (tp->rcv_wnd == 0 && tp->last_ack_sent == th->th_seq)) {
2184 KASSERT(tp->t_state != TCPS_SYN_SENT,
2185 ("%s: TH_RST for TCPS_SYN_SENT th %p tp %p",
2186 __func__, th, tp));
2187
2188 if (V_tcp_insecure_rst ||
2189 tp->last_ack_sent == th->th_seq) {
2190 TCPSTAT_INC(tcps_drops);
2191 /* Drop the connection. */
2192 switch (tp->t_state) {
2193 case TCPS_SYN_RECEIVED:
2194 so->so_error = ECONNREFUSED;
2195 goto close;
2196 case TCPS_ESTABLISHED:
2197 case TCPS_FIN_WAIT_1:
2198 case TCPS_FIN_WAIT_2:
2199 case TCPS_CLOSE_WAIT:
2200 case TCPS_CLOSING:
2201 case TCPS_LAST_ACK:
2202 so->so_error = ECONNRESET;
2203 close:
2204 /* FALLTHROUGH */
2205 default:
2206 tcp_log_end_status(tp, TCP_EI_STATUS_CLIENT_RST);
2207 tp = tcp_close(tp);
2208 }
2209 } else {
2210 TCPSTAT_INC(tcps_badrst);
2211 tcp_send_challenge_ack(tp, th, m);
2212 m = NULL;
2213 }
2214 }
2215 goto drop;
2216 }
2217
2218 /*
2219 * RFC5961 Section 4.2
2220 * Send challenge ACK for any SYN in synchronized state.
2221 */
2222 if ((thflags & TH_SYN) && tp->t_state != TCPS_SYN_SENT &&
2223 tp->t_state != TCPS_SYN_RECEIVED) {
2224 TCPSTAT_INC(tcps_badsyn);
2225 if (V_tcp_insecure_syn &&
2226 SEQ_GEQ(th->th_seq, tp->last_ack_sent) &&
2227 SEQ_LT(th->th_seq, tp->last_ack_sent + tp->rcv_wnd)) {
2228 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2229 tp = tcp_drop(tp, ECONNRESET);
2230 rstreason = BANDLIM_UNLIMITED;
2231 } else {
2232 tcp_ecn_input_syn_sent(tp, thflags, iptos);
2233 tcp_send_challenge_ack(tp, th, m);
2234 m = NULL;
2235 }
2236 goto drop;
2237 }
2238
2239 /*
2240 * RFC 1323 PAWS: If we have a timestamp reply on this segment
2241 * and it's less than ts_recent, drop it.
2242 */
2243 if ((to.to_flags & TOF_TS) != 0 && tp->ts_recent &&
2244 TSTMP_LT(to.to_tsval, tp->ts_recent)) {
2245 /* Check to see if ts_recent is over 24 days old. */
2246 if (tcp_ts_getticks() - tp->ts_recent_age > TCP_PAWS_IDLE) {
2247 /*
2248 * Invalidate ts_recent. If this segment updates
2249 * ts_recent, the age will be reset later and ts_recent
2250 * will get a valid value. If it does not, setting
2251 * ts_recent to zero will at least satisfy the
2252 * requirement that zero be placed in the timestamp
2253 * echo reply when ts_recent isn't valid. The
2254 * age isn't reset until we get a valid ts_recent
2255 * because we don't want out-of-order segments to be
2256 * dropped when ts_recent is old.
2257 */
2258 tp->ts_recent = 0;
2259 } else {
2260 TCPSTAT_INC(tcps_rcvduppack);
2261 TCPSTAT_ADD(tcps_rcvdupbyte, tlen);
2262 TCPSTAT_INC(tcps_pawsdrop);
2263 if (tlen)
2264 goto dropafterack;
2265 goto drop;
2266 }
2267 }
2268
2269 /*
2270 * In the SYN-RECEIVED state, validate that the packet belongs to
2271 * this connection before trimming the data to fit the receive
2272 * window. Check the sequence number versus IRS since we know
2273 * the sequence numbers haven't wrapped. This is a partial fix
2274 * for the "LAND" DoS attack.
2275 */
2276 if (tp->t_state == TCPS_SYN_RECEIVED && SEQ_LT(th->th_seq, tp->irs)) {
2277 rstreason = BANDLIM_RST_OPENPORT;
2278 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
2279 goto dropwithreset;
2280 }
2281
2282 todrop = tp->rcv_nxt - th->th_seq;
2283 if (todrop > 0) {
2284 if (thflags & TH_SYN) {
2285 thflags &= ~TH_SYN;
2286 th->th_seq++;
2287 if (th->th_urp > 1)
2288 th->th_urp--;
2289 else
2290 thflags &= ~TH_URG;
2291 todrop--;
2292 }
2293 /*
2294 * Following if statement from Stevens, vol. 2, p. 960.
2295 */
2296 if (todrop > tlen
2297 || (todrop == tlen && (thflags & TH_FIN) == 0)) {
2298 /*
2299 * Any valid FIN must be to the left of the window.
2300 * At this point the FIN must be a duplicate or out
2301 * of sequence; drop it.
2302 */
2303 thflags &= ~TH_FIN;
2304
2305 /*
2306 * Send an ACK to resynchronize and drop any data.
2307 * But keep on processing for RST or ACK.
2308 */
2309 tp->t_flags |= TF_ACKNOW;
2310 todrop = tlen;
2311 TCPSTAT_INC(tcps_rcvduppack);
2312 TCPSTAT_ADD(tcps_rcvdupbyte, todrop);
2313 } else {
2314 TCPSTAT_INC(tcps_rcvpartduppack);
2315 TCPSTAT_ADD(tcps_rcvpartdupbyte, todrop);
2316 }
2317 /*
2318 * DSACK - add SACK block for dropped range
2319 */
2320 if ((todrop > 0) && (tp->t_flags & TF_SACK_PERMIT)) {
2321 tcp_update_sack_list(tp, th->th_seq,
2322 th->th_seq + todrop);
2323 /*
2324 * ACK now, as the next in-sequence segment
2325 * will clear the DSACK block again
2326 */
2327 tp->t_flags |= TF_ACKNOW;
2328 }
2329 drop_hdrlen += todrop; /* drop from the top afterwards */
2330 th->th_seq += todrop;
2331 tlen -= todrop;
2332 if (th->th_urp > todrop)
2333 th->th_urp -= todrop;
2334 else {
2335 thflags &= ~TH_URG;
2336 th->th_urp = 0;
2337 }
2338 }
2339
2340 /*
2341 * If new data are received on a connection after the
2342 * user processes are gone, then RST the other end if
2343 * no FIN has been processed.
2344 */
2345 if ((tp->t_flags & TF_CLOSED) && tlen > 0 &&
2346 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
2347 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
2348 log(LOG_DEBUG, "%s; %s: %s: Received %d bytes of data "
2349 "after socket was closed, "
2350 "sending RST and removing tcpcb\n",
2351 s, __func__, tcpstates[tp->t_state], tlen);
2352 free(s, M_TCPLOG);
2353 }
2354 tcp_log_end_status(tp, TCP_EI_STATUS_DATA_A_CLOSE);
2355 /* tcp_close will kill the inp pre-log the Reset */
2356 tcp_log_end_status(tp, TCP_EI_STATUS_SERVER_RST);
2357 tp = tcp_close(tp);
2358 TCPSTAT_INC(tcps_rcvafterclose);
2359 rstreason = BANDLIM_UNLIMITED;
2360 goto dropwithreset;
2361 }
2362
2363 /*
2364 * If segment ends after window, drop trailing data
2365 * (and PUSH and FIN); if nothing left, just ACK.
2366 */
2367 todrop = (th->th_seq + tlen) - (tp->rcv_nxt + tp->rcv_wnd);
2368 if (todrop > 0) {
2369 TCPSTAT_INC(tcps_rcvpackafterwin);
2370 if (todrop >= tlen) {
2371 TCPSTAT_ADD(tcps_rcvbyteafterwin, tlen);
2372 /*
2373 * If window is closed can only take segments at
2374 * window edge, and have to drop data and PUSH from
2375 * incoming segments. Continue processing, but
2376 * remember to ack. Otherwise, drop segment
2377 * and ack.
2378 */
2379 if (tp->rcv_wnd == 0 && th->th_seq == tp->rcv_nxt) {
2380 tp->t_flags |= TF_ACKNOW;
2381 TCPSTAT_INC(tcps_rcvwinprobe);
2382 } else
2383 goto dropafterack;
2384 } else
2385 TCPSTAT_ADD(tcps_rcvbyteafterwin, todrop);
2386 m_adj(m, -todrop);
2387 tlen -= todrop;
2388 thflags &= ~(TH_PUSH|TH_FIN);
2389 }
2390
2391 /*
2392 * If last ACK falls within this segment's sequence numbers,
2393 * record its timestamp.
2394 * NOTE:
2395 * 1) That the test incorporates suggestions from the latest
2396 * proposal of the tcplw@cray.com list (Braden 1993/04/26).
2397 * 2) That updating only on newer timestamps interferes with
2398 * our earlier PAWS tests, so this check should be solely
2399 * predicated on the sequence space of this segment.
2400 * 3) That we modify the segment boundary check to be
2401 * Last.ACK.Sent <= SEG.SEQ + SEG.Len
2402 * instead of RFC1323's
2403 * Last.ACK.Sent < SEG.SEQ + SEG.Len,
2404 * This modified check allows us to overcome RFC1323's
2405 * limitations as described in Stevens TCP/IP Illustrated
2406 * Vol. 2 p.869. In such cases, we can still calculate the
2407 * RTT correctly when RCV.NXT == Last.ACK.Sent.
2408 */
2409 if ((to.to_flags & TOF_TS) != 0 &&
2410 SEQ_LEQ(th->th_seq, tp->last_ack_sent) &&
2411 SEQ_LEQ(tp->last_ack_sent, th->th_seq + tlen +
2412 ((thflags & (TH_SYN|TH_FIN)) != 0))) {
2413 tp->ts_recent_age = tcp_ts_getticks();
2414 tp->ts_recent = to.to_tsval;
2415 }
2416
2417 /*
2418 * If the ACK bit is off: if in SYN-RECEIVED state or SENDSYN
2419 * flag is on (half-synchronized state), then queue data for
2420 * later processing; else drop segment and return.
2421 */
2422 if ((thflags & TH_ACK) == 0) {
2423 if (tp->t_state == TCPS_SYN_RECEIVED ||
2424 (tp->t_flags & TF_NEEDSYN)) {
2425 if (tp->t_state == TCPS_SYN_RECEIVED &&
2426 (tp->t_flags & TF_FASTOPEN)) {
2427 tp->snd_wnd = tiwin;
2428 cc_conn_init(tp);
2429 }
2430 goto step6;
2431 } else if (tp->t_flags & TF_ACKNOW)
2432 goto dropafterack;
2433 else
2434 goto drop;
2435 }
2436
2437 /*
2438 * Ack processing.
2439 */
2440 if (SEQ_GEQ(tp->snd_una, tp->iss + (TCP_MAXWIN << tp->snd_scale))) {
2441 /* Checking SEG.ACK against ISS is definitely redundant. */
2442 tp->t_flags2 |= TF2_NO_ISS_CHECK;
2443 }
2444 if (!V_tcp_insecure_ack) {
2445 tcp_seq seq_min;
2446 bool ghost_ack_check;
2447
2448 if (tp->t_flags2 & TF2_NO_ISS_CHECK) {
2449 /* Check for too old ACKs (RFC 5961, Section 5.2). */
2450 seq_min = tp->snd_una - tp->max_sndwnd;
2451 ghost_ack_check = false;
2452 } else {
2453 if (SEQ_GT(tp->iss + 1, tp->snd_una - tp->max_sndwnd)) {
2454 /* Checking for ghost ACKs is stricter. */
2455 seq_min = tp->iss + 1;
2456 ghost_ack_check = true;
2457 } else {
2458 /*
2459 * Checking for too old ACKs (RFC 5961,
2460 * Section 5.2) is stricter.
2461 */
2462 seq_min = tp->snd_una - tp->max_sndwnd;
2463 ghost_ack_check = false;
2464 }
2465 }
2466 if (SEQ_LT(th->th_ack, seq_min)) {
2467 if (ghost_ack_check)
2468 TCPSTAT_INC(tcps_rcvghostack);
2469 else
2470 TCPSTAT_INC(tcps_rcvacktooold);
2471 tcp_send_challenge_ack(tp, th, m);
2472 m = NULL;
2473 goto drop;
2474 }
2475 }
2476 switch (tp->t_state) {
2477 /*
2478 * In SYN_RECEIVED state, the ack ACKs our SYN, so enter
2479 * ESTABLISHED state and continue processing.
2480 * The ACK was checked above.
2481 */
2482 case TCPS_SYN_RECEIVED:
2483
2484 TCPSTAT_INC(tcps_connects);
2485 if (tp->t_flags & TF_SONOTCONN) {
2486 /*
2487 * Usually SYN_RECEIVED had been created from a LISTEN,
2488 * and solisten_enqueue() has already marked the socket
2489 * layer as connected. If it didn't, which can happen
2490 * only with an accept_filter(9), then the tp is marked
2491 * with TF_SONOTCONN. The other reason for this mark
2492 * to be set is a simultaneous open, a SYN_RECEIVED
2493 * that had been created from SYN_SENT.
2494 */
2495 tp->t_flags &= ~TF_SONOTCONN;
2496 soisconnected(so);
2497 }
2498 /* Do window scaling? */
2499 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2500 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2501 tp->rcv_scale = tp->request_r_scale;
2502 }
2503 tp->snd_wnd = tiwin;
2504 /*
2505 * Make transitions:
2506 * SYN-RECEIVED -> ESTABLISHED
2507 * SYN-RECEIVED* -> FIN-WAIT-1
2508 */
2509 tp->t_starttime = ticks;
2510 if ((tp->t_flags & TF_FASTOPEN) && tp->t_tfo_pending) {
2511 tcp_fastopen_decrement_counter(tp->t_tfo_pending);
2512 tp->t_tfo_pending = NULL;
2513 }
2514 if (tp->t_flags & TF_NEEDFIN) {
2515 tp->t_acktime = ticks;
2516 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2517 tp->t_flags &= ~TF_NEEDFIN;
2518 } else {
2519 tcp_state_change(tp, TCPS_ESTABLISHED);
2520 TCP_PROBE5(accept__established, NULL, tp,
2521 m, tp, th);
2522 /*
2523 * TFO connections call cc_conn_init() during SYN
2524 * processing. Calling it again here for such
2525 * connections is not harmless as it would undo the
2526 * snd_cwnd reduction that occurs when a TFO SYN|ACK
2527 * is retransmitted.
2528 */
2529 if (!(tp->t_flags & TF_FASTOPEN))
2530 cc_conn_init(tp);
2531 tcp_timer_activate(tp, TT_KEEP, TP_KEEPIDLE(tp));
2532 }
2533 /*
2534 * Account for the ACK of our SYN prior to
2535 * regular ACK processing below, except for
2536 * simultaneous SYN, which is handled later.
2537 */
2538 if (SEQ_GT(th->th_ack, tp->snd_una) && !(tp->t_flags & TF_NEEDSYN))
2539 incforsyn = 1;
2540 /*
2541 * If segment contains data or ACK, will call tcp_reass()
2542 * later; if not, do so now to pass queued data to user.
2543 */
2544 if (tlen == 0 && (thflags & TH_FIN) == 0) {
2545 (void) tcp_reass(tp, (struct tcphdr *)0, NULL, 0,
2546 (struct mbuf *)0);
2547 tcp_handle_wakeup(tp);
2548 }
2549 tp->snd_wl1 = th->th_seq - 1;
2550 /* FALLTHROUGH */
2551
2552 /*
2553 * In ESTABLISHED state: drop duplicate ACKs; ACK out of range
2554 * ACKs. If the ack is in the range
2555 * tp->snd_una < th->th_ack <= tp->snd_max
2556 * then advance tp->snd_una to th->th_ack and drop
2557 * data from the retransmission queue. If this ACK reflects
2558 * more up to date window information we update our window information.
2559 */
2560 case TCPS_ESTABLISHED:
2561 case TCPS_FIN_WAIT_1:
2562 case TCPS_FIN_WAIT_2:
2563 case TCPS_CLOSE_WAIT:
2564 case TCPS_CLOSING:
2565 case TCPS_LAST_ACK:
2566 if (SEQ_GT(th->th_ack, tp->snd_max)) {
2567 TCPSTAT_INC(tcps_rcvacktoomuch);
2568 goto dropafterack;
2569 }
2570 if (tcp_is_sack_recovery(tp, &to)) {
2571 sack_changed = tcp_sack_doack(tp, &to, th->th_ack);
2572 if ((sack_changed != SACK_NOCHANGE) &&
2573 (tp->t_flags & TF_LRD)) {
2574 tcp_sack_lost_retransmission(tp, th);
2575 }
2576 } else
2577 /*
2578 * Reset the value so that previous (valid) value
2579 * from the last ack with SACK doesn't get used.
2580 */
2581 tp->sackhint.sacked_bytes = 0;
2582
2583 #ifdef TCP_HHOOK
2584 /* Run HHOOK_TCP_ESTABLISHED_IN helper hooks. */
2585 hhook_run_tcp_est_in(tp, th, &to);
2586 #endif
2587
2588 if (SEQ_LEQ(th->th_ack, tp->snd_una)) {
2589 maxseg = tcp_maxseg(tp);
2590 if (tlen == 0 &&
2591 (tiwin == tp->snd_wnd ||
2592 (tp->t_flags & TF_SACK_PERMIT))) {
2593 /*
2594 * If this is the first time we've seen a
2595 * FIN from the remote, this is not a
2596 * duplicate and it needs to be processed
2597 * normally. This happens during a
2598 * simultaneous close.
2599 */
2600 if ((thflags & TH_FIN) &&
2601 (TCPS_HAVERCVDFIN(tp->t_state) == 0)) {
2602 tp->t_dupacks = 0;
2603 break;
2604 }
2605 TCPSTAT_INC(tcps_rcvdupack);
2606 /*
2607 * If we have outstanding data (other than
2608 * a window probe), this is a completely
2609 * duplicate ack (ie, window info didn't
2610 * change and FIN isn't set),
2611 * the ack is the biggest we've
2612 * seen and we've seen exactly our rexmt
2613 * threshold of them, assume a packet
2614 * has been dropped and retransmit it.
2615 * Kludge snd_nxt & the congestion
2616 * window so we send only this one
2617 * packet.
2618 *
2619 * We know we're losing at the current
2620 * window size so do congestion avoidance
2621 * (set ssthresh to half the current window
2622 * and pull our congestion window back to
2623 * the new ssthresh).
2624 *
2625 * Dup acks mean that packets have left the
2626 * network (they're now cached at the receiver)
2627 * so bump cwnd by the amount in the receiver
2628 * to keep a constant cwnd packets in the
2629 * network.
2630 *
2631 * When using TCP ECN, notify the peer that
2632 * we reduced the cwnd.
2633 */
2634 /*
2635 * Following 2 kinds of acks should not affect
2636 * dupack counting:
2637 * 1) Old acks
2638 * 2) Acks with SACK but without any new SACK
2639 * information in them. These could result from
2640 * any anomaly in the network like a switch
2641 * duplicating packets or a possible DoS attack.
2642 */
2643 if (th->th_ack != tp->snd_una ||
2644 (tcp_is_sack_recovery(tp, &to) &&
2645 (sack_changed == SACK_NOCHANGE))) {
2646 break;
2647 } else if (!tcp_timer_active(tp, TT_REXMT)) {
2648 tp->t_dupacks = 0;
2649 } else if (++tp->t_dupacks > tcprexmtthresh ||
2650 IN_FASTRECOVERY(tp->t_flags)) {
2651 cc_ack_received(tp, th, nsegs,
2652 CC_DUPACK);
2653 if (V_tcp_do_prr &&
2654 IN_FASTRECOVERY(tp->t_flags) &&
2655 (tp->t_flags & TF_SACK_PERMIT)) {
2656 tcp_do_prr_ack(tp, th, &to,
2657 sack_changed, &maxseg);
2658 } else if (tcp_is_sack_recovery(tp, &to) &&
2659 IN_FASTRECOVERY(tp->t_flags)) {
2660 int awnd;
2661
2662 /*
2663 * Compute the amount of data in flight first.
2664 * We can inject new data into the pipe iff
2665 * we have less than 1/2 the original window's
2666 * worth of data in flight.
2667 */
2668 if (V_tcp_do_newsack) {
2669 awnd = tcp_compute_pipe(tp);
2670 } else {
2671 awnd = (tp->snd_nxt - tp->snd_fack) +
2672 tp->sackhint.sack_bytes_rexmit;
2673 }
2674 if (awnd < tp->snd_ssthresh) {
2675 tp->snd_cwnd += maxseg;
2676 if (tp->snd_cwnd > tp->snd_ssthresh)
2677 tp->snd_cwnd = tp->snd_ssthresh;
2678 }
2679 } else {
2680 tp->snd_cwnd += maxseg;
2681 }
2682 (void) tcp_output(tp);
2683 goto drop;
2684 } else if (tp->t_dupacks == tcprexmtthresh ||
2685 (tp->t_flags & TF_SACK_PERMIT &&
2686 V_tcp_do_newsack &&
2687 tp->sackhint.sacked_bytes >
2688 (tcprexmtthresh - 1) * maxseg)) {
2689 enter_recovery:
2690 /*
2691 * Above is the RFC6675 trigger condition of
2692 * more than (dupthresh-1)*maxseg sacked data.
2693 * If the count of holes in the
2694 * scoreboard is >= dupthresh, we could
2695 * also enter loss recovery, but don't
2696 * have that value readily available.
2697 */
2698 tp->t_dupacks = tcprexmtthresh;
2699 tcp_seq onxt = tp->snd_nxt;
2700
2701 /*
2702 * If we're doing sack, or prr, check
2703 * to see if we're already in sack
2704 * recovery. If we're not doing sack,
2705 * check to see if we're in newreno
2706 * recovery.
2707 */
2708 if (V_tcp_do_prr ||
2709 (tp->t_flags & TF_SACK_PERMIT)) {
2710 if (IN_FASTRECOVERY(tp->t_flags)) {
2711 tp->t_dupacks = 0;
2712 break;
2713 }
2714 } else {
2715 if (SEQ_LEQ(th->th_ack,
2716 tp->snd_recover)) {
2717 tp->t_dupacks = 0;
2718 break;
2719 }
2720 }
2721 /* Congestion signal before ack. */
2722 cc_cong_signal(tp, th, CC_NDUPACK);
2723 cc_ack_received(tp, th, nsegs,
2724 CC_DUPACK);
2725 tcp_timer_activate(tp, TT_REXMT, 0);
2726 tp->t_rtttime = 0;
2727 if (V_tcp_do_prr) {
2728 /*
2729 * snd_ssthresh is already updated by
2730 * cc_cong_signal.
2731 */
2732 if (tcp_is_sack_recovery(tp, &to)) {
2733 /*
2734 * Exclude Limited Transmit
2735 * segments here
2736 */
2737 tp->sackhint.prr_delivered =
2738 maxseg;
2739 } else {
2740 tp->sackhint.prr_delivered =
2741 imin(tp->snd_max - tp->snd_una,
2742 imin(INT_MAX / 65536,
2743 tp->t_dupacks) * maxseg);
2744 }
2745 tp->sackhint.recover_fs = max(1,
2746 tp->snd_nxt - tp->snd_una);
2747 }
2748 if (tcp_is_sack_recovery(tp, &to)) {
2749 TCPSTAT_INC(tcps_sack_recovery_episode);
2750 tp->snd_recover = tp->snd_nxt;
2751 tp->snd_cwnd = maxseg;
2752 (void) tcp_output(tp);
2753 if (SEQ_GT(th->th_ack, tp->snd_una)) {
2754 goto resume_partialack;
2755 }
2756 goto drop;
2757 }
2758 tp->snd_nxt = th->th_ack;
2759 tp->snd_cwnd = maxseg;
2760 (void) tcp_output(tp);
2761 KASSERT(tp->snd_limited <= 2,
2762 ("%s: tp->snd_limited too big",
2763 __func__));
2764 tp->snd_cwnd = tp->snd_ssthresh +
2765 maxseg *
2766 (tp->t_dupacks - tp->snd_limited);
2767 if (SEQ_GT(onxt, tp->snd_nxt))
2768 tp->snd_nxt = onxt;
2769 goto drop;
2770 } else if (V_tcp_do_rfc3042) {
2771 /*
2772 * Process first and second duplicate
2773 * ACKs. Each indicates a segment
2774 * leaving the network, creating room
2775 * for more. Make sure we can send a
2776 * packet on reception of each duplicate
2777 * ACK by increasing snd_cwnd by one
2778 * segment. Restore the original
2779 * snd_cwnd after packet transmission.
2780 */
2781 cc_ack_received(tp, th, nsegs, CC_DUPACK);
2782 uint32_t oldcwnd = tp->snd_cwnd;
2783 tcp_seq oldsndmax = tp->snd_max;
2784 u_int sent;
2785 int avail;
2786
2787 KASSERT(tp->t_dupacks == 1 ||
2788 tp->t_dupacks == 2,
2789 ("%s: dupacks not 1 or 2",
2790 __func__));
2791 if (tp->t_dupacks == 1)
2792 tp->snd_limited = 0;
2793 tp->snd_cwnd =
2794 (tp->snd_nxt - tp->snd_una) +
2795 (tp->t_dupacks - tp->snd_limited) *
2796 maxseg;
2797 /*
2798 * Only call tcp_output when there
2799 * is new data available to be sent
2800 * or we need to send an ACK.
2801 */
2802 SOCKBUF_LOCK(&so->so_snd);
2803 avail = sbavail(&so->so_snd);
2804 SOCKBUF_UNLOCK(&so->so_snd);
2805 if (tp->t_flags & TF_ACKNOW ||
2806 (avail >=
2807 SEQ_SUB(tp->snd_nxt, tp->snd_una))) {
2808 (void) tcp_output(tp);
2809 }
2810 sent = SEQ_SUB(tp->snd_max, oldsndmax);
2811 if (sent > maxseg) {
2812 KASSERT((tp->t_dupacks == 2 &&
2813 tp->snd_limited == 0) ||
2814 (sent == maxseg + 1 &&
2815 tp->t_flags & TF_SENTFIN),
2816 ("%s: sent too much",
2817 __func__));
2818 tp->snd_limited = 2;
2819 } else if (sent > 0) {
2820 ++tp->snd_limited;
2821 }
2822 tp->snd_cwnd = oldcwnd;
2823 goto drop;
2824 }
2825 }
2826 break;
2827 } else {
2828 /*
2829 * This ack is advancing the left edge, reset the
2830 * counter.
2831 */
2832 tp->t_dupacks = 0;
2833 /*
2834 * If this ack also has new SACK info, increment the
2835 * counter as per rfc6675. The variable
2836 * sack_changed tracks all changes to the SACK
2837 * scoreboard, including when partial ACKs without
2838 * SACK options are received, and clear the scoreboard
2839 * from the left side. Such partial ACKs should not be
2840 * counted as dupacks here.
2841 */
2842 if (tcp_is_sack_recovery(tp, &to) &&
2843 (sack_changed != SACK_NOCHANGE)) {
2844 tp->t_dupacks++;
2845 /* limit overhead by setting maxseg last */
2846 if (!IN_FASTRECOVERY(tp->t_flags) &&
2847 (tp->sackhint.sacked_bytes >
2848 ((tcprexmtthresh - 1) *
2849 (maxseg = tcp_maxseg(tp))))) {
2850 goto enter_recovery;
2851 }
2852 }
2853 }
2854
2855 resume_partialack:
2856 KASSERT(SEQ_GT(th->th_ack, tp->snd_una),
2857 ("%s: th_ack <= snd_una", __func__));
2858
2859 /*
2860 * If the congestion window was inflated to account
2861 * for the other side's cached packets, retract it.
2862 */
2863 if (SEQ_LT(th->th_ack, tp->snd_recover)) {
2864 if (IN_FASTRECOVERY(tp->t_flags)) {
2865 if (tp->t_flags & TF_SACK_PERMIT) {
2866 if (V_tcp_do_prr &&
2867 (to.to_flags & TOF_SACK)) {
2868 tcp_timer_activate(tp,
2869 TT_REXMT, 0);
2870 tp->t_rtttime = 0;
2871 tcp_do_prr_ack(tp, th, &to,
2872 sack_changed, &maxseg);
2873 tp->t_flags |= TF_ACKNOW;
2874 (void) tcp_output(tp);
2875 } else {
2876 tcp_sack_partialack(tp, th,
2877 &maxseg);
2878 }
2879 } else {
2880 tcp_newreno_partial_ack(tp, th);
2881 }
2882 } else if (IN_CONGRECOVERY(tp->t_flags) &&
2883 (V_tcp_do_prr)) {
2884 tp->sackhint.delivered_data =
2885 BYTES_THIS_ACK(tp, th);
2886 tp->snd_fack = th->th_ack;
2887 /*
2888 * During ECN cwnd reduction
2889 * always use PRR-SSRB
2890 */
2891 tcp_do_prr_ack(tp, th, &to, SACK_CHANGE,
2892 &maxseg);
2893 (void) tcp_output(tp);
2894 }
2895 }
2896 /*
2897 * If we reach this point, ACK is not a duplicate,
2898 * i.e., it ACKs something we sent.
2899 */
2900 if (tp->t_flags & TF_NEEDSYN) {
2901 /*
2902 * T/TCP: Connection was half-synchronized, and our
2903 * SYN has been ACK'd (so connection is now fully
2904 * synchronized). Go to non-starred state,
2905 * increment snd_una for ACK of SYN, and check if
2906 * we can do window scaling.
2907 */
2908 tp->t_flags &= ~TF_NEEDSYN;
2909 tp->snd_una++;
2910 /* Do window scaling? */
2911 if ((tp->t_flags & (TF_RCVD_SCALE|TF_REQ_SCALE)) ==
2912 (TF_RCVD_SCALE|TF_REQ_SCALE)) {
2913 tp->rcv_scale = tp->request_r_scale;
2914 /* Send window already scaled. */
2915 }
2916 }
2917
2918 process_ACK:
2919 INP_WLOCK_ASSERT(inp);
2920
2921 /*
2922 * Adjust for the SYN bit in sequence space,
2923 * but don't account for it in cwnd calculations.
2924 * This is for the SYN_RECEIVED, non-simultaneous
2925 * SYN case. SYN_SENT and simultaneous SYN are
2926 * treated elsewhere.
2927 */
2928 if (incforsyn)
2929 tp->snd_una++;
2930 acked = BYTES_THIS_ACK(tp, th);
2931 KASSERT(acked >= 0, ("%s: acked unexepectedly negative "
2932 "(tp->snd_una=%u, th->th_ack=%u, tp=%p, m=%p)", __func__,
2933 tp->snd_una, th->th_ack, tp, m));
2934 TCPSTAT_ADD(tcps_rcvackpack, nsegs);
2935 TCPSTAT_ADD(tcps_rcvackbyte, acked);
2936
2937 /*
2938 * If we just performed our first retransmit, and the ACK
2939 * arrives within our recovery window, then it was a mistake
2940 * to do the retransmit in the first place. Recover our
2941 * original cwnd and ssthresh, and proceed to transmit where
2942 * we left off.
2943 */
2944 if (tp->t_rxtshift == 1 &&
2945 tp->t_flags & TF_PREVVALID &&
2946 tp->t_badrxtwin != 0 &&
2947 to.to_flags & TOF_TS &&
2948 to.to_tsecr != 0 &&
2949 TSTMP_LT(to.to_tsecr, tp->t_badrxtwin))
2950 cc_cong_signal(tp, th, CC_RTO_ERR);
2951
2952 /*
2953 * If we have a timestamp reply, update smoothed
2954 * round trip time. If no timestamp is present but
2955 * transmit timer is running and timed sequence
2956 * number was acked, update smoothed round trip time.
2957 * Since we now have an rtt measurement, cancel the
2958 * timer backoff (cf., Phil Karn's retransmit alg.).
2959 * Recompute the initial retransmit timer.
2960 *
2961 * Some boxes send broken timestamp replies
2962 * during the SYN+ACK phase, ignore
2963 * timestamps of 0 or we could calculate a
2964 * huge RTT and blow up the retransmit timer.
2965 */
2966 if ((to.to_flags & TOF_TS) != 0 && to.to_tsecr) {
2967 uint32_t t;
2968
2969 t = tcp_ts_getticks() - to.to_tsecr;
2970 if (!tp->t_rttlow || tp->t_rttlow > t)
2971 tp->t_rttlow = t;
2972 tcp_xmit_timer(tp, TCP_TS_TO_TICKS(t) + 1);
2973 } else if (tp->t_rtttime && SEQ_GT(th->th_ack, tp->t_rtseq)) {
2974 if (!tp->t_rttlow || tp->t_rttlow > ticks - tp->t_rtttime)
2975 tp->t_rttlow = ticks - tp->t_rtttime;
2976 tcp_xmit_timer(tp, ticks - tp->t_rtttime);
2977 }
2978
2979 SOCKBUF_LOCK(&so->so_snd);
2980 /*
2981 * Clear t_acktime if remote side has ACKd all data in the
2982 * socket buffer and FIN (if applicable).
2983 * Otherwise, update t_acktime if we received a sufficiently
2984 * large ACK.
2985 */
2986 if ((tp->t_state <= TCPS_CLOSE_WAIT &&
2987 acked == sbavail(&so->so_snd)) ||
2988 acked > sbavail(&so->so_snd))
2989 tp->t_acktime = 0;
2990 else if (acked > 1)
2991 tp->t_acktime = ticks;
2992
2993 /*
2994 * If all outstanding data is acked, stop retransmit
2995 * timer and remember to restart (more output or persist).
2996 * If there is more data to be acked, restart retransmit
2997 * timer, using current (possibly backed-off) value.
2998 */
2999 if (th->th_ack == tp->snd_max) {
3000 tcp_timer_activate(tp, TT_REXMT, 0);
3001 needoutput = 1;
3002 } else if (!tcp_timer_active(tp, TT_PERSIST))
3003 tcp_timer_activate(tp, TT_REXMT, TP_RXTCUR(tp));
3004
3005 /*
3006 * If no data (only SYN) was ACK'd,
3007 * skip rest of ACK processing.
3008 */
3009 if (acked == 0) {
3010 SOCKBUF_UNLOCK(&so->so_snd);
3011 goto step6;
3012 }
3013
3014 /*
3015 * Let the congestion control algorithm update congestion
3016 * control related information. This typically means increasing
3017 * the congestion window.
3018 */
3019 cc_ack_received(tp, th, nsegs, CC_ACK);
3020
3021 if (acked > sbavail(&so->so_snd)) {
3022 if (tp->snd_wnd >= sbavail(&so->so_snd))
3023 tp->snd_wnd -= sbavail(&so->so_snd);
3024 else
3025 tp->snd_wnd = 0;
3026 mfree = sbcut_locked(&so->so_snd,
3027 (int)sbavail(&so->so_snd));
3028 ourfinisacked = 1;
3029 } else {
3030 mfree = sbcut_locked(&so->so_snd, acked);
3031 if (tp->snd_wnd >= (uint32_t) acked)
3032 tp->snd_wnd -= acked;
3033 else
3034 tp->snd_wnd = 0;
3035 ourfinisacked = 0;
3036 }
3037 /* NB: sowwakeup_locked() does an implicit unlock. */
3038 sowwakeup_locked(so);
3039 m_freem(mfree);
3040 /* Detect una wraparound. */
3041 if (!IN_RECOVERY(tp->t_flags) &&
3042 SEQ_GT(tp->snd_una, tp->snd_recover) &&
3043 SEQ_LEQ(th->th_ack, tp->snd_recover))
3044 tp->snd_recover = th->th_ack - 1;
3045 tp->snd_una = th->th_ack;
3046 if (IN_RECOVERY(tp->t_flags) &&
3047 SEQ_GEQ(th->th_ack, tp->snd_recover)) {
3048 cc_post_recovery(tp, th);
3049 }
3050 if (tp->t_flags & TF_SACK_PERMIT) {
3051 if (SEQ_GT(tp->snd_una, tp->snd_recover))
3052 tp->snd_recover = tp->snd_una;
3053 }
3054 if (SEQ_LT(tp->snd_nxt, tp->snd_una))
3055 tp->snd_nxt = tp->snd_una;
3056
3057 switch (tp->t_state) {
3058 /*
3059 * In FIN_WAIT_1 STATE in addition to the processing
3060 * for the ESTABLISHED state if our FIN is now acknowledged
3061 * then enter FIN_WAIT_2.
3062 */
3063 case TCPS_FIN_WAIT_1:
3064 if (ourfinisacked) {
3065 /*
3066 * If we can't receive any more
3067 * data, then closing user can proceed.
3068 * Starting the timer is contrary to the
3069 * specification, but if we don't get a FIN
3070 * we'll hang forever.
3071 */
3072 if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3073 tcp_free_sackholes(tp);
3074 soisdisconnected(so);
3075 tcp_timer_activate(tp, TT_2MSL,
3076 (tcp_fast_finwait2_recycle ?
3077 tcp_finwait2_timeout :
3078 TP_MAXIDLE(tp)));
3079 }
3080 tcp_state_change(tp, TCPS_FIN_WAIT_2);
3081 }
3082 break;
3083
3084 /*
3085 * In CLOSING STATE in addition to the processing for
3086 * the ESTABLISHED state if the ACK acknowledges our FIN
3087 * then enter the TIME-WAIT state, otherwise ignore
3088 * the segment.
3089 */
3090 case TCPS_CLOSING:
3091 if (ourfinisacked) {
3092 tcp_twstart(tp);
3093 m_freem(m);
3094 return;
3095 }
3096 break;
3097
3098 /*
3099 * In LAST_ACK, we may still be waiting for data to drain
3100 * and/or to be acked, as well as for the ack of our FIN.
3101 * If our FIN is now acknowledged, delete the TCB,
3102 * enter the closed state and return.
3103 */
3104 case TCPS_LAST_ACK:
3105 if (ourfinisacked) {
3106 tp = tcp_close(tp);
3107 goto drop;
3108 }
3109 break;
3110 }
3111 }
3112
3113 step6:
3114 INP_WLOCK_ASSERT(inp);
3115
3116 /*
3117 * Update window information.
3118 * Don't look at window if no ACK: TAC's send garbage on first SYN.
3119 */
3120 if ((thflags & TH_ACK) &&
3121 (SEQ_LT(tp->snd_wl1, th->th_seq) ||
3122 (tp->snd_wl1 == th->th_seq && (SEQ_LT(tp->snd_wl2, th->th_ack) ||
3123 (tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd))))) {
3124 /* keep track of pure window updates */
3125 if (tlen == 0 &&
3126 tp->snd_wl2 == th->th_ack && tiwin > tp->snd_wnd)
3127 TCPSTAT_INC(tcps_rcvwinupd);
3128 tp->snd_wnd = tiwin;
3129 tp->snd_wl1 = th->th_seq;
3130 tp->snd_wl2 = th->th_ack;
3131 if (tp->snd_wnd > tp->max_sndwnd)
3132 tp->max_sndwnd = tp->snd_wnd;
3133 needoutput = 1;
3134 }
3135
3136 /*
3137 * Process segments with URG.
3138 */
3139 if ((thflags & TH_URG) && th->th_urp &&
3140 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3141 /*
3142 * This is a kludge, but if we receive and accept
3143 * random urgent pointers, we'll crash in
3144 * soreceive. It's hard to imagine someone
3145 * actually wanting to send this much urgent data.
3146 */
3147 SOCKBUF_LOCK(&so->so_rcv);
3148 if (th->th_urp + sbavail(&so->so_rcv) > sb_max) {
3149 th->th_urp = 0; /* XXX */
3150 thflags &= ~TH_URG; /* XXX */
3151 SOCKBUF_UNLOCK(&so->so_rcv); /* XXX */
3152 goto dodata; /* XXX */
3153 }
3154 /*
3155 * If this segment advances the known urgent pointer,
3156 * then mark the data stream. This should not happen
3157 * in CLOSE_WAIT, CLOSING, LAST_ACK or TIME_WAIT STATES since
3158 * a FIN has been received from the remote side.
3159 * In these states we ignore the URG.
3160 *
3161 * According to RFC961 (Assigned Protocols),
3162 * the urgent pointer points to the last octet
3163 * of urgent data. We continue, however,
3164 * to consider it to indicate the first octet
3165 * of data past the urgent section as the original
3166 * spec states (in one of two places).
3167 */
3168 if (SEQ_GT(th->th_seq+th->th_urp, tp->rcv_up)) {
3169 tp->rcv_up = th->th_seq + th->th_urp;
3170 so->so_oobmark = sbavail(&so->so_rcv) +
3171 (tp->rcv_up - tp->rcv_nxt) - 1;
3172 if (so->so_oobmark == 0)
3173 so->so_rcv.sb_state |= SBS_RCVATMARK;
3174 sohasoutofband(so);
3175 tp->t_oobflags &= ~(TCPOOB_HAVEDATA | TCPOOB_HADDATA);
3176 }
3177 SOCKBUF_UNLOCK(&so->so_rcv);
3178 /*
3179 * Remove out of band data so doesn't get presented to user.
3180 * This can happen independent of advancing the URG pointer,
3181 * but if two URG's are pending at once, some out-of-band
3182 * data may creep in... ick.
3183 */
3184 if (th->th_urp <= (uint32_t)tlen &&
3185 !(so->so_options & SO_OOBINLINE)) {
3186 /* hdr drop is delayed */
3187 tcp_pulloutofband(so, th, m, drop_hdrlen);
3188 }
3189 } else {
3190 /*
3191 * If no out of band data is expected,
3192 * pull receive urgent pointer along
3193 * with the receive window.
3194 */
3195 if (SEQ_GT(tp->rcv_nxt, tp->rcv_up))
3196 tp->rcv_up = tp->rcv_nxt;
3197 }
3198 dodata: /* XXX */
3199 INP_WLOCK_ASSERT(inp);
3200
3201 /*
3202 * Process the segment text, merging it into the TCP sequencing queue,
3203 * and arranging for acknowledgment of receipt if necessary.
3204 * This process logically involves adjusting tp->rcv_wnd as data
3205 * is presented to the user (this happens in tcp_usrreq.c,
3206 * case PRU_RCVD). If a FIN has already been received on this
3207 * connection then we just ignore the text.
3208 */
3209 tfo_syn = ((tp->t_state == TCPS_SYN_RECEIVED) &&
3210 (tp->t_flags & TF_FASTOPEN));
3211 if ((tlen || (thflags & TH_FIN) || (tfo_syn && tlen > 0)) &&
3212 TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3213 tcp_seq save_start = th->th_seq;
3214 tcp_seq save_rnxt = tp->rcv_nxt;
3215 int save_tlen = tlen;
3216 m_adj(m, drop_hdrlen); /* delayed header drop */
3217 /*
3218 * Insert segment which includes th into TCP reassembly queue
3219 * with control block tp. Set thflags to whether reassembly now
3220 * includes a segment with FIN. This handles the common case
3221 * inline (segment is the next to be received on an established
3222 * connection, and the queue is empty), avoiding linkage into
3223 * and removal from the queue and repetition of various
3224 * conversions.
3225 * Set DELACK for segments received in order, but ack
3226 * immediately when segments are out of order (so
3227 * fast retransmit can work).
3228 */
3229 if (th->th_seq == tp->rcv_nxt &&
3230 SEGQ_EMPTY(tp) &&
3231 (TCPS_HAVEESTABLISHED(tp->t_state) ||
3232 tfo_syn)) {
3233 if (DELAY_ACK(tp, tlen) || tfo_syn)
3234 tp->t_flags |= TF_DELACK;
3235 else
3236 tp->t_flags |= TF_ACKNOW;
3237 tp->rcv_nxt += tlen;
3238 if (tlen &&
3239 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
3240 (tp->t_fbyte_in == 0)) {
3241 tp->t_fbyte_in = ticks;
3242 if (tp->t_fbyte_in == 0)
3243 tp->t_fbyte_in = 1;
3244 if (tp->t_fbyte_out && tp->t_fbyte_in)
3245 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
3246 }
3247 thflags = tcp_get_flags(th) & TH_FIN;
3248 TCPSTAT_INC(tcps_rcvpack);
3249 TCPSTAT_ADD(tcps_rcvbyte, tlen);
3250 SOCKBUF_LOCK(&so->so_rcv);
3251 if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3252 m_freem(m);
3253 else
3254 sbappendstream_locked(&so->so_rcv, m, 0);
3255 tp->t_flags |= TF_WAKESOR;
3256 } else {
3257 /*
3258 * XXX: Due to the header drop above "th" is
3259 * theoretically invalid by now. Fortunately
3260 * m_adj() doesn't actually frees any mbufs
3261 * when trimming from the head.
3262 */
3263 tcp_seq temp = save_start;
3264
3265 thflags = tcp_reass(tp, th, &temp, &tlen, m);
3266 tp->t_flags |= TF_ACKNOW;
3267 }
3268 if ((tp->t_flags & TF_SACK_PERMIT) &&
3269 (save_tlen > 0) &&
3270 TCPS_HAVEESTABLISHED(tp->t_state)) {
3271 if ((tlen == 0) && (SEQ_LT(save_start, save_rnxt))) {
3272 /*
3273 * DSACK actually handled in the fastpath
3274 * above.
3275 */
3276 tcp_update_sack_list(tp, save_start,
3277 save_start + save_tlen);
3278 } else if ((tlen > 0) && SEQ_GT(tp->rcv_nxt, save_rnxt)) {
3279 if ((tp->rcv_numsacks >= 1) &&
3280 (tp->sackblks[0].end == save_start)) {
3281 /*
3282 * Partial overlap, recorded at todrop
3283 * above.
3284 */
3285 tcp_update_sack_list(tp,
3286 tp->sackblks[0].start,
3287 tp->sackblks[0].end);
3288 } else {
3289 tcp_update_dsack_list(tp, save_start,
3290 save_start + save_tlen);
3291 }
3292 } else if (tlen >= save_tlen) {
3293 /* Update of sackblks. */
3294 tcp_update_dsack_list(tp, save_start,
3295 save_start + save_tlen);
3296 } else if (tlen > 0) {
3297 tcp_update_dsack_list(tp, save_start,
3298 save_start + tlen);
3299 }
3300 }
3301 tcp_handle_wakeup(tp);
3302 #if 0
3303 /*
3304 * Note the amount of data that peer has sent into
3305 * our window, in order to estimate the sender's
3306 * buffer size.
3307 * XXX: Unused.
3308 */
3309 if (SEQ_GT(tp->rcv_adv, tp->rcv_nxt))
3310 len = so->so_rcv.sb_hiwat - (tp->rcv_adv - tp->rcv_nxt);
3311 else
3312 len = so->so_rcv.sb_hiwat;
3313 #endif
3314 } else {
3315 if ((s = tcp_log_addrs(inc, th, NULL, NULL))) {
3316 if (tlen > 0) {
3317 if ((thflags & TH_FIN) != 0) {
3318 log(LOG_DEBUG, "%s; %s: %s: "
3319 "Received %d bytes of data and FIN "
3320 "after having received a FIN, "
3321 "just dropping both\n",
3322 s, __func__,
3323 tcpstates[tp->t_state], tlen);
3324 } else {
3325 log(LOG_DEBUG, "%s; %s: %s: "
3326 "Received %d bytes of data "
3327 "after having received a FIN, "
3328 "just dropping it\n",
3329 s, __func__,
3330 tcpstates[tp->t_state], tlen);
3331 }
3332 } else {
3333 if ((thflags & TH_FIN) != 0) {
3334 log(LOG_DEBUG, "%s; %s: %s: "
3335 "Received FIN "
3336 "after having received a FIN, "
3337 "just dropping it\n",
3338 s, __func__,
3339 tcpstates[tp->t_state]);
3340 }
3341 }
3342 free(s, M_TCPLOG);
3343 }
3344 m_freem(m);
3345 thflags &= ~TH_FIN;
3346 }
3347
3348 /*
3349 * If FIN is received ACK the FIN and let the user know
3350 * that the connection is closing.
3351 */
3352 if (thflags & TH_FIN) {
3353 if (TCPS_HAVERCVDFIN(tp->t_state) == 0) {
3354 /* The socket upcall is handled by socantrcvmore. */
3355 socantrcvmore(so);
3356 /*
3357 * If connection is half-synchronized
3358 * (ie NEEDSYN flag on) then delay ACK,
3359 * so it may be piggybacked when SYN is sent.
3360 * Otherwise, since we received a FIN then no
3361 * more input can be expected, send ACK now.
3362 */
3363 if (tp->t_flags & TF_NEEDSYN)
3364 tp->t_flags |= TF_DELACK;
3365 else
3366 tp->t_flags |= TF_ACKNOW;
3367 tp->rcv_nxt++;
3368 }
3369 switch (tp->t_state) {
3370 /*
3371 * In SYN_RECEIVED and ESTABLISHED STATES
3372 * enter the CLOSE_WAIT state.
3373 */
3374 case TCPS_SYN_RECEIVED:
3375 tp->t_starttime = ticks;
3376 /* FALLTHROUGH */
3377 case TCPS_ESTABLISHED:
3378 tcp_state_change(tp, TCPS_CLOSE_WAIT);
3379 break;
3380
3381 /*
3382 * If still in FIN_WAIT_1 STATE FIN has not been acked so
3383 * enter the CLOSING state.
3384 */
3385 case TCPS_FIN_WAIT_1:
3386 tcp_state_change(tp, TCPS_CLOSING);
3387 break;
3388
3389 /*
3390 * In FIN_WAIT_2 state enter the TIME_WAIT state,
3391 * starting the time-wait timer, turning off the other
3392 * standard timers.
3393 */
3394 case TCPS_FIN_WAIT_2:
3395 tcp_twstart(tp);
3396 return;
3397 }
3398 }
3399 TCP_PROBE3(debug__input, tp, th, m);
3400
3401 /*
3402 * Return any desired output.
3403 */
3404 if (needoutput || (tp->t_flags & TF_ACKNOW)) {
3405 (void) tcp_output(tp);
3406 }
3407 check_delack:
3408 INP_WLOCK_ASSERT(inp);
3409
3410 if (tp->t_flags & TF_DELACK) {
3411 tp->t_flags &= ~TF_DELACK;
3412 tcp_timer_activate(tp, TT_DELACK, tcp_delacktime);
3413 }
3414 INP_WUNLOCK(inp);
3415 return;
3416
3417 dropafterack:
3418 /*
3419 * Generate an ACK dropping incoming segment if it occupies
3420 * sequence space, where the ACK reflects our state.
3421 *
3422 * We can now skip the test for the RST flag since all
3423 * paths to this code happen after packets containing
3424 * RST have been dropped.
3425 *
3426 * In the SYN-RECEIVED state, don't send an ACK unless the
3427 * segment we received passes the SYN-RECEIVED ACK test.
3428 * If it fails send a RST. This breaks the loop in the
3429 * "LAND" DoS attack, and also prevents an ACK storm
3430 * between two listening ports that have been sent forged
3431 * SYN segments, each with the source address of the other.
3432 */
3433 if (tp->t_state == TCPS_SYN_RECEIVED && (thflags & TH_ACK) &&
3434 (SEQ_GT(tp->snd_una, th->th_ack) ||
3435 SEQ_GT(th->th_ack, tp->snd_max)) ) {
3436 rstreason = BANDLIM_RST_OPENPORT;
3437 tcp_log_end_status(tp, TCP_EI_STATUS_RST_IN_FRONT);
3438 goto dropwithreset;
3439 }
3440 TCP_PROBE3(debug__input, tp, th, m);
3441 tp->t_flags |= TF_ACKNOW;
3442 (void) tcp_output(tp);
3443 INP_WUNLOCK(inp);
3444 m_freem(m);
3445 return;
3446
3447 dropwithreset:
3448 if (tp != NULL) {
3449 tcp_dropwithreset(m, th, tp, tlen, rstreason);
3450 INP_WUNLOCK(inp);
3451 } else
3452 tcp_dropwithreset(m, th, NULL, tlen, rstreason);
3453 return;
3454
3455 drop:
3456 /*
3457 * Drop space held by incoming segment and return.
3458 */
3459 TCP_PROBE3(debug__input, tp, th, m);
3460 if (tp != NULL) {
3461 INP_WUNLOCK(inp);
3462 }
3463 m_freem(m);
3464 }
3465
3466 /*
3467 * Issue RST and make ACK acceptable to originator of segment.
3468 * The mbuf must still include the original packet header.
3469 * tp may be NULL.
3470 */
3471 void
tcp_dropwithreset(struct mbuf * m,struct tcphdr * th,struct tcpcb * tp,int tlen,int rstreason)3472 tcp_dropwithreset(struct mbuf *m, struct tcphdr *th, struct tcpcb *tp,
3473 int tlen, int rstreason)
3474 {
3475 #ifdef INET
3476 struct ip *ip;
3477 #endif
3478 #ifdef INET6
3479 struct ip6_hdr *ip6;
3480 #endif
3481
3482 if (tp != NULL) {
3483 INP_LOCK_ASSERT(tptoinpcb(tp));
3484 }
3485
3486 /* Don't bother if destination was broadcast/multicast. */
3487 if ((tcp_get_flags(th) & TH_RST) || m->m_flags & (M_BCAST|M_MCAST))
3488 goto drop;
3489 #ifdef INET6
3490 if (mtod(m, struct ip *)->ip_v == 6) {
3491 ip6 = mtod(m, struct ip6_hdr *);
3492 if (IN6_IS_ADDR_MULTICAST(&ip6->ip6_dst) ||
3493 IN6_IS_ADDR_MULTICAST(&ip6->ip6_src))
3494 goto drop;
3495 /* IPv6 anycast check is done at tcp6_input() */
3496 }
3497 #endif
3498 #if defined(INET) && defined(INET6)
3499 else
3500 #endif
3501 #ifdef INET
3502 {
3503 ip = mtod(m, struct ip *);
3504 if (IN_MULTICAST(ntohl(ip->ip_dst.s_addr)) ||
3505 IN_MULTICAST(ntohl(ip->ip_src.s_addr)) ||
3506 ip->ip_src.s_addr == htonl(INADDR_BROADCAST) ||
3507 in_broadcast(ip->ip_dst, m->m_pkthdr.rcvif))
3508 goto drop;
3509 }
3510 #endif
3511
3512 /* Perform bandwidth limiting. */
3513 if (badport_bandlim(rstreason) < 0)
3514 goto drop;
3515
3516 /* tcp_respond consumes the mbuf chain. */
3517 if (tcp_get_flags(th) & TH_ACK) {
3518 tcp_respond(tp, mtod(m, void *), th, m, (tcp_seq)0,
3519 th->th_ack, TH_RST);
3520 } else {
3521 if (tcp_get_flags(th) & TH_SYN)
3522 tlen++;
3523 if (tcp_get_flags(th) & TH_FIN)
3524 tlen++;
3525 tcp_respond(tp, mtod(m, void *), th, m, th->th_seq+tlen,
3526 (tcp_seq)0, TH_RST|TH_ACK);
3527 }
3528 return;
3529 drop:
3530 m_freem(m);
3531 }
3532
3533 /*
3534 * Parse TCP options and place in tcpopt.
3535 */
3536 void
tcp_dooptions(struct tcpopt * to,u_char * cp,int cnt,int flags)3537 tcp_dooptions(struct tcpopt *to, u_char *cp, int cnt, int flags)
3538 {
3539 int opt, optlen;
3540
3541 to->to_flags = 0;
3542 for (; cnt > 0; cnt -= optlen, cp += optlen) {
3543 opt = cp[0];
3544 if (opt == TCPOPT_EOL)
3545 break;
3546 if (opt == TCPOPT_NOP)
3547 optlen = 1;
3548 else {
3549 if (cnt < 2)
3550 break;
3551 optlen = cp[1];
3552 if (optlen < 2 || optlen > cnt)
3553 break;
3554 }
3555 switch (opt) {
3556 case TCPOPT_MAXSEG:
3557 if (optlen != TCPOLEN_MAXSEG)
3558 continue;
3559 if (!(flags & TO_SYN))
3560 continue;
3561 to->to_flags |= TOF_MSS;
3562 bcopy((char *)cp + 2,
3563 (char *)&to->to_mss, sizeof(to->to_mss));
3564 to->to_mss = ntohs(to->to_mss);
3565 break;
3566 case TCPOPT_WINDOW:
3567 if (optlen != TCPOLEN_WINDOW)
3568 continue;
3569 if (!(flags & TO_SYN))
3570 continue;
3571 to->to_flags |= TOF_SCALE;
3572 to->to_wscale = min(cp[2], TCP_MAX_WINSHIFT);
3573 break;
3574 case TCPOPT_TIMESTAMP:
3575 if (optlen != TCPOLEN_TIMESTAMP)
3576 continue;
3577 to->to_flags |= TOF_TS;
3578 bcopy((char *)cp + 2,
3579 (char *)&to->to_tsval, sizeof(to->to_tsval));
3580 to->to_tsval = ntohl(to->to_tsval);
3581 bcopy((char *)cp + 6,
3582 (char *)&to->to_tsecr, sizeof(to->to_tsecr));
3583 to->to_tsecr = ntohl(to->to_tsecr);
3584 break;
3585 case TCPOPT_SIGNATURE:
3586 /*
3587 * In order to reply to a host which has set the
3588 * TCP_SIGNATURE option in its initial SYN, we have
3589 * to record the fact that the option was observed
3590 * here for the syncache code to perform the correct
3591 * response.
3592 */
3593 if (optlen != TCPOLEN_SIGNATURE)
3594 continue;
3595 to->to_flags |= TOF_SIGNATURE;
3596 to->to_signature = cp + 2;
3597 break;
3598 case TCPOPT_SACK_PERMITTED:
3599 if (optlen != TCPOLEN_SACK_PERMITTED)
3600 continue;
3601 if (!(flags & TO_SYN))
3602 continue;
3603 if (!V_tcp_do_sack)
3604 continue;
3605 to->to_flags |= TOF_SACKPERM;
3606 break;
3607 case TCPOPT_SACK:
3608 if (optlen <= 2 || (optlen - 2) % TCPOLEN_SACK != 0)
3609 continue;
3610 if (flags & TO_SYN)
3611 continue;
3612 to->to_flags |= TOF_SACK;
3613 to->to_nsacks = (optlen - 2) / TCPOLEN_SACK;
3614 to->to_sacks = cp + 2;
3615 TCPSTAT_INC(tcps_sack_rcv_blocks);
3616 break;
3617 case TCPOPT_FAST_OPEN:
3618 /*
3619 * Cookie length validation is performed by the
3620 * server side cookie checking code or the client
3621 * side cookie cache update code.
3622 */
3623 if (!(flags & TO_SYN))
3624 continue;
3625 if (!V_tcp_fastopen_client_enable &&
3626 !V_tcp_fastopen_server_enable)
3627 continue;
3628 to->to_flags |= TOF_FASTOPEN;
3629 to->to_tfo_len = optlen - 2;
3630 to->to_tfo_cookie = to->to_tfo_len ? cp + 2 : NULL;
3631 break;
3632 default:
3633 continue;
3634 }
3635 }
3636 }
3637
3638 /*
3639 * Pull out of band byte out of a segment so
3640 * it doesn't appear in the user's data queue.
3641 * It is still reflected in the segment length for
3642 * sequencing purposes.
3643 */
3644 void
tcp_pulloutofband(struct socket * so,struct tcphdr * th,struct mbuf * m,int off)3645 tcp_pulloutofband(struct socket *so, struct tcphdr *th, struct mbuf *m,
3646 int off)
3647 {
3648 int cnt = off + th->th_urp - 1;
3649
3650 while (cnt >= 0) {
3651 if (m->m_len > cnt) {
3652 char *cp = mtod(m, caddr_t) + cnt;
3653 struct tcpcb *tp = sototcpcb(so);
3654
3655 INP_WLOCK_ASSERT(tptoinpcb(tp));
3656
3657 tp->t_iobc = *cp;
3658 tp->t_oobflags |= TCPOOB_HAVEDATA;
3659 bcopy(cp+1, cp, (unsigned)(m->m_len - cnt - 1));
3660 m->m_len--;
3661 if (m->m_flags & M_PKTHDR)
3662 m->m_pkthdr.len--;
3663 return;
3664 }
3665 cnt -= m->m_len;
3666 m = m->m_next;
3667 if (m == NULL)
3668 break;
3669 }
3670 panic("tcp_pulloutofband");
3671 }
3672
3673 /*
3674 * Collect new round-trip time estimate
3675 * and update averages and current timeout.
3676 */
3677 void
tcp_xmit_timer(struct tcpcb * tp,int rtt)3678 tcp_xmit_timer(struct tcpcb *tp, int rtt)
3679 {
3680 int delta;
3681
3682 INP_WLOCK_ASSERT(tptoinpcb(tp));
3683
3684 TCPSTAT_INC(tcps_rttupdated);
3685 if (tp->t_rttupdated < UCHAR_MAX)
3686 tp->t_rttupdated++;
3687 #ifdef STATS
3688 stats_voi_update_abs_u32(tp->t_stats, VOI_TCP_RTT,
3689 imax(0, rtt * 1000 / hz));
3690 #endif
3691 if ((tp->t_srtt != 0) && (tp->t_rxtshift <= TCP_RTT_INVALIDATE)) {
3692 /*
3693 * srtt is stored as fixed point with 5 bits after the
3694 * binary point (i.e., scaled by 8). The following magic
3695 * is equivalent to the smoothing algorithm in rfc793 with
3696 * an alpha of .875 (srtt = rtt/8 + srtt*7/8 in fixed
3697 * point). Adjust rtt to origin 0.
3698 */
3699 delta = ((rtt - 1) << TCP_DELTA_SHIFT)
3700 - (tp->t_srtt >> (TCP_RTT_SHIFT - TCP_DELTA_SHIFT));
3701
3702 if ((tp->t_srtt += delta) <= 0)
3703 tp->t_srtt = 1;
3704
3705 /*
3706 * We accumulate a smoothed rtt variance (actually, a
3707 * smoothed mean difference), then set the retransmit
3708 * timer to smoothed rtt + 4 times the smoothed variance.
3709 * rttvar is stored as fixed point with 4 bits after the
3710 * binary point (scaled by 16). The following is
3711 * equivalent to rfc793 smoothing with an alpha of .75
3712 * (rttvar = rttvar*3/4 + |delta| / 4). This replaces
3713 * rfc793's wired-in beta.
3714 */
3715 if (delta < 0)
3716 delta = -delta;
3717 delta -= tp->t_rttvar >> (TCP_RTTVAR_SHIFT - TCP_DELTA_SHIFT);
3718 if ((tp->t_rttvar += delta) <= 0)
3719 tp->t_rttvar = 1;
3720 } else {
3721 /*
3722 * No rtt measurement yet - use the unsmoothed rtt.
3723 * Set the variance to half the rtt (so our first
3724 * retransmit happens at 3*rtt).
3725 */
3726 tp->t_srtt = rtt << TCP_RTT_SHIFT;
3727 tp->t_rttvar = rtt << (TCP_RTTVAR_SHIFT - 1);
3728 }
3729 tp->t_rtttime = 0;
3730 tp->t_rxtshift = 0;
3731
3732 /*
3733 * the retransmit should happen at rtt + 4 * rttvar.
3734 * Because of the way we do the smoothing, srtt and rttvar
3735 * will each average +1/2 tick of bias. When we compute
3736 * the retransmit timer, we want 1/2 tick of rounding and
3737 * 1 extra tick because of +-1/2 tick uncertainty in the
3738 * firing of the timer. The bias will give us exactly the
3739 * 1.5 tick we need. But, because the bias is
3740 * statistical, we have to test that we don't drop below
3741 * the minimum feasible timer (which is 2 ticks).
3742 */
3743 TCPT_RANGESET(tp->t_rxtcur, TCP_REXMTVAL(tp),
3744 max(tp->t_rttmin, rtt + 2), TCPTV_REXMTMAX);
3745
3746 /*
3747 * We received an ack for a packet that wasn't retransmitted;
3748 * it is probably safe to discard any error indications we've
3749 * received recently. This isn't quite right, but close enough
3750 * for now (a route might have failed after we sent a segment,
3751 * and the return path might not be symmetrical).
3752 */
3753 tp->t_softerror = 0;
3754 }
3755
3756 /*
3757 * Determine a reasonable value for maxseg size.
3758 * If the route is known, check route for mtu.
3759 * If none, use an mss that can be handled on the outgoing interface
3760 * without forcing IP to fragment. If no route is found, route has no mtu,
3761 * or the destination isn't local, use a default, hopefully conservative
3762 * size (usually 512 or the default IP max size, but no more than the mtu
3763 * of the interface), as we can't discover anything about intervening
3764 * gateways or networks. We also initialize the congestion/slow start
3765 * window to be a single segment if the destination isn't local.
3766 * While looking at the routing entry, we also initialize other path-dependent
3767 * parameters from pre-set or cached values in the routing entry.
3768 *
3769 * NOTE that resulting t_maxseg doesn't include space for TCP options or
3770 * IP options, e.g. IPSEC data, since length of this data may vary, and
3771 * thus it is calculated for every segment separately in tcp_output().
3772 *
3773 * NOTE that this routine is only called when we process an incoming
3774 * segment, or an ICMP need fragmentation datagram. Outgoing SYN/ACK MSS
3775 * settings are handled in tcp_mssopt().
3776 */
3777 void
tcp_mss_update(struct tcpcb * tp,int offer,int mtuoffer,struct hc_metrics_lite * metricptr,struct tcp_ifcap * cap)3778 tcp_mss_update(struct tcpcb *tp, int offer, int mtuoffer,
3779 struct hc_metrics_lite *metricptr, struct tcp_ifcap *cap)
3780 {
3781 int mss = 0;
3782 uint32_t maxmtu = 0;
3783 struct inpcb *inp = tptoinpcb(tp);
3784 struct hc_metrics_lite metrics;
3785 #ifdef INET6
3786 int isipv6 = ((inp->inp_vflag & INP_IPV6) != 0) ? 1 : 0;
3787 size_t min_protoh = isipv6 ?
3788 sizeof (struct ip6_hdr) + sizeof (struct tcphdr) :
3789 sizeof (struct tcpiphdr);
3790 #else
3791 size_t min_protoh = sizeof(struct tcpiphdr);
3792 #endif
3793
3794 INP_WLOCK_ASSERT(inp);
3795
3796 if (tp->t_port)
3797 min_protoh += V_tcp_udp_tunneling_overhead;
3798 if (mtuoffer != -1) {
3799 KASSERT(offer == -1, ("%s: conflict", __func__));
3800 offer = mtuoffer - min_protoh;
3801 }
3802
3803 /* Initialize. */
3804 #ifdef INET6
3805 if (isipv6) {
3806 maxmtu = tcp_maxmtu6(&inp->inp_inc, cap);
3807 tp->t_maxseg = V_tcp_v6mssdflt;
3808 }
3809 #endif
3810 #if defined(INET) && defined(INET6)
3811 else
3812 #endif
3813 #ifdef INET
3814 {
3815 maxmtu = tcp_maxmtu(&inp->inp_inc, cap);
3816 tp->t_maxseg = V_tcp_mssdflt;
3817 }
3818 #endif
3819
3820 /*
3821 * No route to sender, stay with default mss and return.
3822 */
3823 if (maxmtu == 0) {
3824 /*
3825 * In case we return early we need to initialize metrics
3826 * to a defined state as tcp_hc_get() would do for us
3827 * if there was no cache hit.
3828 */
3829 if (metricptr != NULL)
3830 bzero(metricptr, sizeof(struct hc_metrics_lite));
3831 return;
3832 }
3833
3834 /* What have we got? */
3835 switch (offer) {
3836 case 0:
3837 /*
3838 * Offer == 0 means that there was no MSS on the SYN
3839 * segment, in this case we use tcp_mssdflt as
3840 * already assigned to t_maxseg above.
3841 */
3842 offer = tp->t_maxseg;
3843 break;
3844
3845 case -1:
3846 /*
3847 * Offer == -1 means that we didn't receive SYN yet.
3848 */
3849 /* FALLTHROUGH */
3850
3851 default:
3852 /*
3853 * Prevent DoS attack with too small MSS. Round up
3854 * to at least minmss.
3855 */
3856 offer = max(offer, V_tcp_minmss);
3857 }
3858
3859 /*
3860 * rmx information is now retrieved from tcp_hostcache.
3861 */
3862 tcp_hc_get(&inp->inp_inc, &metrics);
3863 if (metricptr != NULL)
3864 bcopy(&metrics, metricptr, sizeof(struct hc_metrics_lite));
3865
3866 /*
3867 * If there's a discovered mtu in tcp hostcache, use it.
3868 * Else, use the link mtu.
3869 */
3870 if (metrics.rmx_mtu)
3871 mss = min(metrics.rmx_mtu, maxmtu) - min_protoh;
3872 else {
3873 #ifdef INET6
3874 if (isipv6) {
3875 mss = maxmtu - min_protoh;
3876 if (!V_path_mtu_discovery &&
3877 !in6_localaddr(&inp->in6p_faddr))
3878 mss = min(mss, V_tcp_v6mssdflt);
3879 }
3880 #endif
3881 #if defined(INET) && defined(INET6)
3882 else
3883 #endif
3884 #ifdef INET
3885 {
3886 mss = maxmtu - min_protoh;
3887 if (!V_path_mtu_discovery &&
3888 !in_localaddr(inp->inp_faddr))
3889 mss = min(mss, V_tcp_mssdflt);
3890 }
3891 #endif
3892 /*
3893 * XXX - The above conditional (mss = maxmtu - min_protoh)
3894 * probably violates the TCP spec.
3895 * The problem is that, since we don't know the
3896 * other end's MSS, we are supposed to use a conservative
3897 * default. But, if we do that, then MTU discovery will
3898 * never actually take place, because the conservative
3899 * default is much less than the MTUs typically seen
3900 * on the Internet today. For the moment, we'll sweep
3901 * this under the carpet.
3902 *
3903 * The conservative default might not actually be a problem
3904 * if the only case this occurs is when sending an initial
3905 * SYN with options and data to a host we've never talked
3906 * to before. Then, they will reply with an MSS value which
3907 * will get recorded and the new parameters should get
3908 * recomputed. For Further Study.
3909 */
3910 }
3911 mss = min(mss, offer);
3912
3913 /*
3914 * Sanity check: make sure that maxseg will be large
3915 * enough to allow some data on segments even if the
3916 * all the option space is used (40bytes). Otherwise
3917 * funny things may happen in tcp_output.
3918 *
3919 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3920 */
3921 mss = max(mss, 64);
3922
3923 tp->t_maxseg = mss;
3924 if (tp->t_maxseg < V_tcp_mssdflt) {
3925 /*
3926 * The MSS is so small we should not process incoming
3927 * SACK's since we are subject to attack in such a
3928 * case.
3929 */
3930 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3931 } else {
3932 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3933 }
3934
3935 }
3936
3937 void
tcp_mss(struct tcpcb * tp,int offer)3938 tcp_mss(struct tcpcb *tp, int offer)
3939 {
3940 int mss;
3941 uint32_t bufsize;
3942 struct inpcb *inp = tptoinpcb(tp);
3943 struct socket *so;
3944 struct hc_metrics_lite metrics;
3945 struct tcp_ifcap cap;
3946
3947 KASSERT(tp != NULL, ("%s: tp == NULL", __func__));
3948
3949 bzero(&cap, sizeof(cap));
3950 tcp_mss_update(tp, offer, -1, &metrics, &cap);
3951
3952 mss = tp->t_maxseg;
3953
3954 /*
3955 * If there's a pipesize, change the socket buffer to that size,
3956 * don't change if sb_hiwat is different than default (then it
3957 * has been changed on purpose with setsockopt).
3958 * Make the socket buffers an integral number of mss units;
3959 * if the mss is larger than the socket buffer, decrease the mss.
3960 */
3961 so = inp->inp_socket;
3962 SOCKBUF_LOCK(&so->so_snd);
3963 if ((so->so_snd.sb_hiwat == V_tcp_sendspace) && metrics.rmx_sendpipe)
3964 bufsize = metrics.rmx_sendpipe;
3965 else
3966 bufsize = so->so_snd.sb_hiwat;
3967 if (bufsize < mss)
3968 mss = bufsize;
3969 else {
3970 bufsize = roundup(bufsize, mss);
3971 if (bufsize > sb_max)
3972 bufsize = sb_max;
3973 if (bufsize > so->so_snd.sb_hiwat)
3974 (void)sbreserve_locked(so, SO_SND, bufsize, NULL);
3975 }
3976 SOCKBUF_UNLOCK(&so->so_snd);
3977 /*
3978 * Sanity check: make sure that maxseg will be large
3979 * enough to allow some data on segments even if the
3980 * all the option space is used (40bytes). Otherwise
3981 * funny things may happen in tcp_output.
3982 *
3983 * XXXGL: shouldn't we reserve space for IP/IPv6 options?
3984 */
3985 tp->t_maxseg = max(mss, 64);
3986 if (tp->t_maxseg < V_tcp_mssdflt) {
3987 /*
3988 * The MSS is so small we should not process incoming
3989 * SACK's since we are subject to attack in such a
3990 * case.
3991 */
3992 tp->t_flags2 |= TF2_PROC_SACK_PROHIBIT;
3993 } else {
3994 tp->t_flags2 &= ~TF2_PROC_SACK_PROHIBIT;
3995 }
3996
3997 SOCKBUF_LOCK(&so->so_rcv);
3998 if ((so->so_rcv.sb_hiwat == V_tcp_recvspace) && metrics.rmx_recvpipe)
3999 bufsize = metrics.rmx_recvpipe;
4000 else
4001 bufsize = so->so_rcv.sb_hiwat;
4002 if (bufsize > mss) {
4003 bufsize = roundup(bufsize, mss);
4004 if (bufsize > sb_max)
4005 bufsize = sb_max;
4006 if (bufsize > so->so_rcv.sb_hiwat)
4007 (void)sbreserve_locked(so, SO_RCV, bufsize, NULL);
4008 }
4009 SOCKBUF_UNLOCK(&so->so_rcv);
4010
4011 /* Check the interface for TSO capabilities. */
4012 if (cap.ifcap & CSUM_TSO) {
4013 tp->t_flags |= TF_TSO;
4014 tp->t_tsomax = cap.tsomax;
4015 tp->t_tsomaxsegcount = cap.tsomaxsegcount;
4016 tp->t_tsomaxsegsize = cap.tsomaxsegsize;
4017 if (cap.ipsec_tso)
4018 tp->t_flags2 |= TF2_IPSEC_TSO;
4019 }
4020 }
4021
4022 /*
4023 * Determine the MSS option to send on an outgoing SYN.
4024 */
4025 int
tcp_mssopt(struct in_conninfo * inc)4026 tcp_mssopt(struct in_conninfo *inc)
4027 {
4028 int mss = 0;
4029 uint32_t thcmtu = 0;
4030 uint32_t maxmtu = 0;
4031 size_t min_protoh;
4032
4033 KASSERT(inc != NULL, ("tcp_mssopt with NULL in_conninfo pointer"));
4034
4035 #ifdef INET6
4036 if (inc->inc_flags & INC_ISIPV6) {
4037 mss = V_tcp_v6mssdflt;
4038 maxmtu = tcp_maxmtu6(inc, NULL);
4039 min_protoh = sizeof(struct ip6_hdr) + sizeof(struct tcphdr);
4040 }
4041 #endif
4042 #if defined(INET) && defined(INET6)
4043 else
4044 #endif
4045 #ifdef INET
4046 {
4047 mss = V_tcp_mssdflt;
4048 maxmtu = tcp_maxmtu(inc, NULL);
4049 min_protoh = sizeof(struct tcpiphdr);
4050 }
4051 #endif
4052 #if defined(INET6) || defined(INET)
4053 thcmtu = tcp_hc_getmtu(inc); /* IPv4 and IPv6 */
4054 #endif
4055
4056 if (maxmtu && thcmtu)
4057 mss = min(maxmtu, thcmtu) - min_protoh;
4058 else if (maxmtu || thcmtu)
4059 mss = max(maxmtu, thcmtu) - min_protoh;
4060
4061 return (mss);
4062 }
4063
4064 void
tcp_do_prr_ack(struct tcpcb * tp,struct tcphdr * th,struct tcpopt * to,sackstatus_t sack_changed,u_int * maxsegp)4065 tcp_do_prr_ack(struct tcpcb *tp, struct tcphdr *th, struct tcpopt *to,
4066 sackstatus_t sack_changed, u_int *maxsegp)
4067 {
4068 int snd_cnt = 0, limit = 0, del_data = 0, pipe = 0;
4069 u_int maxseg;
4070
4071 INP_WLOCK_ASSERT(tptoinpcb(tp));
4072
4073 if (*maxsegp == 0) {
4074 *maxsegp = tcp_maxseg(tp);
4075 }
4076 maxseg = *maxsegp;
4077 /*
4078 * Compute the amount of data that this ACK is indicating
4079 * (del_data) and an estimate of how many bytes are in the
4080 * network.
4081 */
4082 if (tcp_is_sack_recovery(tp, to) ||
4083 (IN_CONGRECOVERY(tp->t_flags) &&
4084 !IN_FASTRECOVERY(tp->t_flags))) {
4085 del_data = tp->sackhint.delivered_data;
4086 if (V_tcp_do_newsack)
4087 pipe = tcp_compute_pipe(tp);
4088 else
4089 pipe = (tp->snd_nxt - tp->snd_fack) +
4090 tp->sackhint.sack_bytes_rexmit;
4091 } else {
4092 if (tp->sackhint.prr_delivered < (tcprexmtthresh * maxseg +
4093 tp->snd_recover - tp->snd_una)) {
4094 del_data = maxseg;
4095 }
4096 pipe = imax(0, tp->snd_max - tp->snd_una -
4097 imin(INT_MAX / 65536, tp->t_dupacks) * maxseg);
4098 }
4099 tp->sackhint.prr_delivered += del_data;
4100 /*
4101 * Proportional Rate Reduction
4102 */
4103 if (pipe >= tp->snd_ssthresh) {
4104 if (tp->sackhint.recover_fs == 0)
4105 tp->sackhint.recover_fs =
4106 imax(1, tp->snd_nxt - tp->snd_una);
4107 snd_cnt = howmany((long)tp->sackhint.prr_delivered *
4108 tp->snd_ssthresh, tp->sackhint.recover_fs) -
4109 tp->sackhint.prr_out + maxseg - 1;
4110 } else {
4111 /*
4112 * PRR 6937bis heuristic:
4113 * - A partial ack without SACK block beneath snd_recover
4114 * indicates further loss.
4115 * - An SACK scoreboard update adding a new hole indicates
4116 * further loss, so be conservative and send at most one
4117 * segment.
4118 * - Prevent ACK splitting attacks, by being conservative
4119 * when no new data is acked.
4120 */
4121 if ((sack_changed == SACK_NEWLOSS) || (del_data == 0)) {
4122 limit = tp->sackhint.prr_delivered -
4123 tp->sackhint.prr_out;
4124 } else {
4125 limit = imax(tp->sackhint.prr_delivered -
4126 tp->sackhint.prr_out, del_data) +
4127 maxseg;
4128 }
4129 snd_cnt = imin((tp->snd_ssthresh - pipe), limit);
4130 }
4131 snd_cnt = imax(snd_cnt, 0) / maxseg;
4132 /*
4133 * Send snd_cnt new data into the network in response to this ack.
4134 * If there is going to be a SACK retransmission, adjust snd_cwnd
4135 * accordingly.
4136 */
4137 if (IN_FASTRECOVERY(tp->t_flags)) {
4138 if (tcp_is_sack_recovery(tp, to)) {
4139 tp->snd_cwnd = tp->snd_nxt - tp->snd_recover +
4140 tp->sackhint.sack_bytes_rexmit +
4141 (snd_cnt * maxseg);
4142 } else {
4143 tp->snd_cwnd = (tp->snd_max - tp->snd_una) +
4144 (snd_cnt * maxseg);
4145 }
4146 } else if (IN_CONGRECOVERY(tp->t_flags)) {
4147 tp->snd_cwnd = pipe - del_data + (snd_cnt * maxseg);
4148 }
4149 tp->snd_cwnd = imax(maxseg, tp->snd_cwnd);
4150 }
4151
4152 /*
4153 * On a partial ack arrives, force the retransmission of the
4154 * next unacknowledged segment. Do not clear tp->t_dupacks.
4155 * By setting snd_nxt to ti_ack, this forces retransmission timer to
4156 * be started again.
4157 */
4158 void
tcp_newreno_partial_ack(struct tcpcb * tp,struct tcphdr * th)4159 tcp_newreno_partial_ack(struct tcpcb *tp, struct tcphdr *th)
4160 {
4161 tcp_seq onxt = tp->snd_nxt;
4162 uint32_t ocwnd = tp->snd_cwnd;
4163 u_int maxseg = tcp_maxseg(tp);
4164
4165 INP_WLOCK_ASSERT(tptoinpcb(tp));
4166
4167 tcp_timer_activate(tp, TT_REXMT, 0);
4168 tp->t_rtttime = 0;
4169 tp->snd_nxt = th->th_ack;
4170 /*
4171 * Set snd_cwnd to one segment beyond acknowledged offset.
4172 * (tp->snd_una has not yet been updated when this function is called.)
4173 */
4174 tp->snd_cwnd = maxseg + BYTES_THIS_ACK(tp, th);
4175 tp->t_flags |= TF_ACKNOW;
4176 (void) tcp_output(tp);
4177 tp->snd_cwnd = ocwnd;
4178 if (SEQ_GT(onxt, tp->snd_nxt))
4179 tp->snd_nxt = onxt;
4180 /*
4181 * Partial window deflation. Relies on fact that tp->snd_una
4182 * not updated yet.
4183 */
4184 if (tp->snd_cwnd > BYTES_THIS_ACK(tp, th))
4185 tp->snd_cwnd -= BYTES_THIS_ACK(tp, th);
4186 else
4187 tp->snd_cwnd = 0;
4188 tp->snd_cwnd += maxseg;
4189 }
4190
4191 int
tcp_compute_pipe(struct tcpcb * tp)4192 tcp_compute_pipe(struct tcpcb *tp)
4193 {
4194 if (tp->t_fb->tfb_compute_pipe == NULL) {
4195 return (tp->snd_max - tp->snd_una +
4196 tp->sackhint.sack_bytes_rexmit -
4197 tp->sackhint.sacked_bytes -
4198 tp->sackhint.lost_bytes);
4199 } else {
4200 return((*tp->t_fb->tfb_compute_pipe)(tp));
4201 }
4202 }
4203
4204 uint32_t
tcp_compute_initwnd(uint32_t maxseg)4205 tcp_compute_initwnd(uint32_t maxseg)
4206 {
4207 /*
4208 * Calculate the Initial Window, also used as Restart Window
4209 *
4210 * RFC5681 Section 3.1 specifies the default conservative values.
4211 * RFC3390 specifies slightly more aggressive values.
4212 * RFC6928 increases it to ten segments.
4213 * Support for user specified value for initial flight size.
4214 */
4215 if (V_tcp_initcwnd_segments)
4216 return min(V_tcp_initcwnd_segments * maxseg,
4217 max(2 * maxseg, V_tcp_initcwnd_segments * 1460));
4218 else if (V_tcp_do_rfc3390)
4219 return min(4 * maxseg, max(2 * maxseg, 4380));
4220 else {
4221 /* Per RFC5681 Section 3.1 */
4222 if (maxseg > 2190)
4223 return (2 * maxseg);
4224 else if (maxseg > 1095)
4225 return (3 * maxseg);
4226 else
4227 return (4 * maxseg);
4228 }
4229 }
4230