1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1982, 1986, 1988, 1993
5 * The Regents of the University of California.
6 * Copyright (c) 2006-2007 Robert N. M. Watson
7 * Copyright (c) 2010-2011 Juniper Networks, Inc.
8 * All rights reserved.
9 *
10 * Portions of this software were developed by Robert N. M. Watson under
11 * contract to Juniper Networks, Inc.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * From: @(#)tcp_usrreq.c 8.2 (Berkeley) 1/3/94
38 */
39
40 #include <sys/cdefs.h>
41 __FBSDID("$FreeBSD$");
42
43 #include "opt_ddb.h"
44 #include "opt_inet.h"
45 #include "opt_inet6.h"
46 #include "opt_ipsec.h"
47 #include "opt_kern_tls.h"
48 #include "opt_tcpdebug.h"
49
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/arb.h>
53 #include <sys/limits.h>
54 #include <sys/malloc.h>
55 #include <sys/refcount.h>
56 #include <sys/kernel.h>
57 #include <sys/ktls.h>
58 #include <sys/qmath.h>
59 #include <sys/sysctl.h>
60 #include <sys/mbuf.h>
61 #ifdef INET6
62 #include <sys/domain.h>
63 #endif /* INET6 */
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/protosw.h>
67 #include <sys/proc.h>
68 #include <sys/jail.h>
69 #include <sys/syslog.h>
70 #include <sys/stats.h>
71
72 #ifdef DDB
73 #include <ddb/ddb.h>
74 #endif
75
76 #include <net/if.h>
77 #include <net/if_var.h>
78 #include <net/route.h>
79 #include <net/vnet.h>
80
81 #include <netinet/in.h>
82 #include <netinet/in_kdtrace.h>
83 #include <netinet/in_pcb.h>
84 #include <netinet/in_systm.h>
85 #include <netinet/in_var.h>
86 #include <netinet/ip_var.h>
87 #ifdef INET6
88 #include <netinet/ip6.h>
89 #include <netinet6/in6_pcb.h>
90 #include <netinet6/ip6_var.h>
91 #include <netinet6/scope6_var.h>
92 #endif
93 #include <netinet/tcp.h>
94 #include <netinet/tcp_fsm.h>
95 #include <netinet/tcp_seq.h>
96 #include <netinet/tcp_timer.h>
97 #include <netinet/tcp_var.h>
98 #include <netinet/tcp_log_buf.h>
99 #include <netinet/tcpip.h>
100 #include <netinet/cc/cc.h>
101 #include <netinet/tcp_fastopen.h>
102 #include <netinet/tcp_hpts.h>
103 #ifdef TCPPCAP
104 #include <netinet/tcp_pcap.h>
105 #endif
106 #ifdef TCPDEBUG
107 #include <netinet/tcp_debug.h>
108 #endif
109 #ifdef TCP_OFFLOAD
110 #include <netinet/tcp_offload.h>
111 #endif
112 #include <netipsec/ipsec_support.h>
113
114 #include <vm/vm.h>
115 #include <vm/vm_param.h>
116 #include <vm/pmap.h>
117 #include <vm/vm_extern.h>
118 #include <vm/vm_map.h>
119 #include <vm/vm_page.h>
120
121 /*
122 * TCP protocol interface to socket abstraction.
123 */
124 #ifdef INET
125 static int tcp_connect(struct tcpcb *, struct sockaddr *,
126 struct thread *td);
127 #endif /* INET */
128 #ifdef INET6
129 static int tcp6_connect(struct tcpcb *, struct sockaddr *,
130 struct thread *td);
131 #endif /* INET6 */
132 static void tcp_disconnect(struct tcpcb *);
133 static void tcp_usrclosed(struct tcpcb *);
134 static void tcp_fill_info(struct tcpcb *, struct tcp_info *);
135
136 static int tcp_pru_options_support(struct tcpcb *tp, int flags);
137
138 #ifdef TCPDEBUG
139 #define TCPDEBUG0 int ostate = 0
140 #define TCPDEBUG1() ostate = tp ? tp->t_state : 0
141 #define TCPDEBUG2(req) if (tp && (so->so_options & SO_DEBUG)) \
142 tcp_trace(TA_USER, ostate, tp, 0, 0, req)
143 #else
144 #define TCPDEBUG0
145 #define TCPDEBUG1()
146 #define TCPDEBUG2(req)
147 #endif
148
149 /*
150 * tcp_require_unique port requires a globally-unique source port for each
151 * outgoing connection. The default is to require the 4-tuple to be unique.
152 */
153 VNET_DEFINE(int, tcp_require_unique_port) = 0;
154 SYSCTL_INT(_net_inet_tcp, OID_AUTO, require_unique_port,
155 CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(tcp_require_unique_port), 0,
156 "Require globally-unique ephemeral port for outgoing connections");
157 #define V_tcp_require_unique_port VNET(tcp_require_unique_port)
158
159 /*
160 * TCP attaches to socket via pru_attach(), reserving space,
161 * and an internet control block.
162 */
163 static int
tcp_usr_attach(struct socket * so,int proto,struct thread * td)164 tcp_usr_attach(struct socket *so, int proto, struct thread *td)
165 {
166 struct inpcb *inp;
167 struct tcpcb *tp = NULL;
168 int error;
169 TCPDEBUG0;
170
171 inp = sotoinpcb(so);
172 KASSERT(inp == NULL, ("tcp_usr_attach: inp != NULL"));
173 TCPDEBUG1();
174
175 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
176 error = soreserve(so, V_tcp_sendspace, V_tcp_recvspace);
177 if (error)
178 goto out;
179 }
180
181 so->so_rcv.sb_flags |= SB_AUTOSIZE;
182 so->so_snd.sb_flags |= SB_AUTOSIZE;
183 error = in_pcballoc(so, &V_tcbinfo);
184 if (error)
185 goto out;
186 inp = sotoinpcb(so);
187 #ifdef INET6
188 if (inp->inp_vflag & INP_IPV6PROTO) {
189 inp->inp_vflag |= INP_IPV6;
190 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
191 inp->inp_vflag |= INP_IPV4;
192 inp->in6p_hops = -1; /* use kernel default */
193 }
194 else
195 #endif
196 inp->inp_vflag |= INP_IPV4;
197 tp = tcp_newtcpcb(inp);
198 if (tp == NULL) {
199 error = ENOBUFS;
200 in_pcbdetach(inp);
201 in_pcbfree(inp);
202 goto out;
203 }
204 tp->t_state = TCPS_CLOSED;
205 INP_WUNLOCK(inp);
206 TCPSTATES_INC(TCPS_CLOSED);
207 if ((so->so_options & SO_LINGER) && so->so_linger == 0)
208 so->so_linger = TCP_LINGERTIME;
209 out:
210 TCPDEBUG2(PRU_ATTACH);
211 TCP_PROBE2(debug__user, tp, PRU_ATTACH);
212 return (error);
213 }
214
215 /*
216 * tcp_usr_detach is called when the socket layer loses its final reference
217 * to the socket, be it a file descriptor reference, a reference from TCP,
218 * etc. At this point, there is only one case in which we will keep around
219 * inpcb state: time wait.
220 */
221 static void
tcp_usr_detach(struct socket * so)222 tcp_usr_detach(struct socket *so)
223 {
224 struct inpcb *inp;
225 struct tcpcb *tp;
226
227 inp = sotoinpcb(so);
228 KASSERT(inp != NULL, ("%s: inp == NULL", __func__));
229 INP_WLOCK(inp);
230 KASSERT(so->so_pcb == inp && inp->inp_socket == so,
231 ("%s: socket %p inp %p mismatch", __func__, so, inp));
232
233 tp = intotcpcb(inp);
234
235 if (inp->inp_flags & INP_TIMEWAIT) {
236 /*
237 * There are two cases to handle: one in which the time wait
238 * state is being discarded (INP_DROPPED), and one in which
239 * this connection will remain in timewait. In the former,
240 * it is time to discard all state (except tcptw, which has
241 * already been discarded by the timewait close code, which
242 * should be further up the call stack somewhere). In the
243 * latter case, we detach from the socket, but leave the pcb
244 * present until timewait ends.
245 *
246 * XXXRW: Would it be cleaner to free the tcptw here?
247 *
248 * Astute question indeed, from twtcp perspective there are
249 * four cases to consider:
250 *
251 * #1 tcp_usr_detach is called at tcptw creation time by
252 * tcp_twstart, then do not discard the newly created tcptw
253 * and leave inpcb present until timewait ends
254 * #2 tcp_usr_detach is called at tcptw creation time by
255 * tcp_twstart, but connection is local and tw will be
256 * discarded immediately
257 * #3 tcp_usr_detach is called at timewait end (or reuse) by
258 * tcp_twclose, then the tcptw has already been discarded
259 * (or reused) and inpcb is freed here
260 * #4 tcp_usr_detach is called() after timewait ends (or reuse)
261 * (e.g. by soclose), then tcptw has already been discarded
262 * (or reused) and inpcb is freed here
263 *
264 * In all three cases the tcptw should not be freed here.
265 */
266 if (inp->inp_flags & INP_DROPPED) {
267 in_pcbdetach(inp);
268 if (__predict_true(tp == NULL)) {
269 in_pcbfree(inp);
270 } else {
271 /*
272 * This case should not happen as in TIMEWAIT
273 * state the inp should not be destroyed before
274 * its tcptw. If INVARIANTS is defined, panic.
275 */
276 #ifdef INVARIANTS
277 panic("%s: Panic before an inp double-free: "
278 "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
279 , __func__);
280 #else
281 log(LOG_ERR, "%s: Avoid an inp double-free: "
282 "INP_TIMEWAIT && INP_DROPPED && tp != NULL"
283 , __func__);
284 #endif
285 INP_WUNLOCK(inp);
286 }
287 } else {
288 in_pcbdetach(inp);
289 INP_WUNLOCK(inp);
290 }
291 } else {
292 /*
293 * If the connection is not in timewait, we consider two
294 * two conditions: one in which no further processing is
295 * necessary (dropped || embryonic), and one in which TCP is
296 * not yet done, but no longer requires the socket, so the
297 * pcb will persist for the time being.
298 *
299 * XXXRW: Does the second case still occur?
300 */
301 if (inp->inp_flags & INP_DROPPED ||
302 tp->t_state < TCPS_SYN_SENT) {
303 tcp_discardcb(tp);
304 in_pcbdetach(inp);
305 in_pcbfree(inp);
306 } else {
307 in_pcbdetach(inp);
308 INP_WUNLOCK(inp);
309 }
310 }
311 }
312
313 #ifdef LVS_TCPOPT_TOA
314
315 #ifndef TCPOPT_TOA
316 #define TCPOPT_TOA 254
317 #define TCPOLEN_TOA 8
318 #endif
319
320 struct toa_data {
321 uint8_t opcode;
322 uint8_t opsize;
323 uint16_t port;
324 uint32_t ip;
325 };
326
327 static int
toa_getpeeraddr(struct socket * so,struct sockaddr ** nam)328 toa_getpeeraddr(struct socket *so, struct sockaddr **nam)
329 {
330 int ret;
331 struct toa_data *toa;
332 struct sockaddr_in *sin;
333
334 ret = in_getpeeraddr(so, nam);
335 if (ret) {
336 return ret;
337 }
338
339 toa = (struct toa_data *)so->so_toa;
340 if (toa->opcode == TCPOPT_TOA && toa->opsize == TCPOLEN_TOA) {
341 sin = (struct sockaddr_in *)(*nam);
342
343 sin->sin_addr.s_addr = toa->ip;
344 sin->sin_port = toa->port;
345 }
346
347 return 0;
348 }
349 #endif
350
351 #ifdef INET
352 /*
353 * Give the socket an address.
354 */
355 static int
tcp_usr_bind(struct socket * so,struct sockaddr * nam,struct thread * td)356 tcp_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
357 {
358 int error = 0;
359 struct inpcb *inp;
360 struct tcpcb *tp = NULL;
361 struct sockaddr_in *sinp;
362
363 sinp = (struct sockaddr_in *)nam;
364 if (nam->sa_len != sizeof (*sinp))
365 return (EINVAL);
366 /*
367 * Must check for multicast addresses and disallow binding
368 * to them.
369 */
370 if (sinp->sin_family == AF_INET &&
371 IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
372 return (EAFNOSUPPORT);
373
374 TCPDEBUG0;
375 inp = sotoinpcb(so);
376 KASSERT(inp != NULL, ("tcp_usr_bind: inp == NULL"));
377 INP_WLOCK(inp);
378 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
379 error = EINVAL;
380 goto out;
381 }
382 tp = intotcpcb(inp);
383 TCPDEBUG1();
384 INP_HASH_WLOCK(&V_tcbinfo);
385 error = in_pcbbind(inp, nam, td->td_ucred);
386 INP_HASH_WUNLOCK(&V_tcbinfo);
387 out:
388 TCPDEBUG2(PRU_BIND);
389 TCP_PROBE2(debug__user, tp, PRU_BIND);
390 INP_WUNLOCK(inp);
391
392 return (error);
393 }
394 #endif /* INET */
395
396 #ifdef INET6
397 static int
tcp6_usr_bind(struct socket * so,struct sockaddr * nam,struct thread * td)398 tcp6_usr_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
399 {
400 int error = 0;
401 struct inpcb *inp;
402 struct tcpcb *tp = NULL;
403 struct sockaddr_in6 *sin6;
404 u_char vflagsav;
405
406 sin6 = (struct sockaddr_in6 *)nam;
407 if (nam->sa_len != sizeof (*sin6))
408 return (EINVAL);
409 /*
410 * Must check for multicast addresses and disallow binding
411 * to them.
412 */
413 if (sin6->sin6_family == AF_INET6 &&
414 IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
415 return (EAFNOSUPPORT);
416
417 TCPDEBUG0;
418 inp = sotoinpcb(so);
419 KASSERT(inp != NULL, ("tcp6_usr_bind: inp == NULL"));
420 INP_WLOCK(inp);
421 vflagsav = inp->inp_vflag;
422 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
423 error = EINVAL;
424 goto out;
425 }
426 tp = intotcpcb(inp);
427 TCPDEBUG1();
428 INP_HASH_WLOCK(&V_tcbinfo);
429 inp->inp_vflag &= ~INP_IPV4;
430 inp->inp_vflag |= INP_IPV6;
431 #ifdef INET
432 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0) {
433 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
434 inp->inp_vflag |= INP_IPV4;
435 else if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
436 struct sockaddr_in sin;
437
438 in6_sin6_2_sin(&sin, sin6);
439 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
440 error = EAFNOSUPPORT;
441 INP_HASH_WUNLOCK(&V_tcbinfo);
442 goto out;
443 }
444 inp->inp_vflag |= INP_IPV4;
445 inp->inp_vflag &= ~INP_IPV6;
446 error = in_pcbbind(inp, (struct sockaddr *)&sin,
447 td->td_ucred);
448 INP_HASH_WUNLOCK(&V_tcbinfo);
449 goto out;
450 }
451 }
452 #endif
453 error = in6_pcbbind(inp, nam, td->td_ucred);
454 INP_HASH_WUNLOCK(&V_tcbinfo);
455 out:
456 if (error != 0)
457 inp->inp_vflag = vflagsav;
458 TCPDEBUG2(PRU_BIND);
459 TCP_PROBE2(debug__user, tp, PRU_BIND);
460 INP_WUNLOCK(inp);
461 return (error);
462 }
463 #endif /* INET6 */
464
465 #ifdef INET
466 /*
467 * Prepare to accept connections.
468 */
469 static int
tcp_usr_listen(struct socket * so,int backlog,struct thread * td)470 tcp_usr_listen(struct socket *so, int backlog, struct thread *td)
471 {
472 int error = 0;
473 struct inpcb *inp;
474 struct tcpcb *tp = NULL;
475
476 TCPDEBUG0;
477 inp = sotoinpcb(so);
478 KASSERT(inp != NULL, ("tcp_usr_listen: inp == NULL"));
479 INP_WLOCK(inp);
480 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
481 error = EINVAL;
482 goto out;
483 }
484 tp = intotcpcb(inp);
485 TCPDEBUG1();
486 SOCK_LOCK(so);
487 error = solisten_proto_check(so);
488 INP_HASH_WLOCK(&V_tcbinfo);
489 if (error == 0 && inp->inp_lport == 0)
490 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
491 INP_HASH_WUNLOCK(&V_tcbinfo);
492 if (error == 0) {
493 tcp_state_change(tp, TCPS_LISTEN);
494 solisten_proto(so, backlog);
495 #ifdef TCP_OFFLOAD
496 if ((so->so_options & SO_NO_OFFLOAD) == 0)
497 tcp_offload_listen_start(tp);
498 #endif
499 }
500 SOCK_UNLOCK(so);
501
502 if (IS_FASTOPEN(tp->t_flags))
503 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
504
505 out:
506 TCPDEBUG2(PRU_LISTEN);
507 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
508 INP_WUNLOCK(inp);
509 return (error);
510 }
511 #endif /* INET */
512
513 #ifdef INET6
514 static int
tcp6_usr_listen(struct socket * so,int backlog,struct thread * td)515 tcp6_usr_listen(struct socket *so, int backlog, struct thread *td)
516 {
517 int error = 0;
518 struct inpcb *inp;
519 struct tcpcb *tp = NULL;
520 u_char vflagsav;
521
522 TCPDEBUG0;
523 inp = sotoinpcb(so);
524 KASSERT(inp != NULL, ("tcp6_usr_listen: inp == NULL"));
525 INP_WLOCK(inp);
526 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
527 error = EINVAL;
528 goto out;
529 }
530 vflagsav = inp->inp_vflag;
531 tp = intotcpcb(inp);
532 TCPDEBUG1();
533 SOCK_LOCK(so);
534 error = solisten_proto_check(so);
535 INP_HASH_WLOCK(&V_tcbinfo);
536 if (error == 0 && inp->inp_lport == 0) {
537 inp->inp_vflag &= ~INP_IPV4;
538 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) == 0)
539 inp->inp_vflag |= INP_IPV4;
540 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
541 }
542 INP_HASH_WUNLOCK(&V_tcbinfo);
543 if (error == 0) {
544 tcp_state_change(tp, TCPS_LISTEN);
545 solisten_proto(so, backlog);
546 #ifdef TCP_OFFLOAD
547 if ((so->so_options & SO_NO_OFFLOAD) == 0)
548 tcp_offload_listen_start(tp);
549 #endif
550 }
551 SOCK_UNLOCK(so);
552
553 if (IS_FASTOPEN(tp->t_flags))
554 tp->t_tfo_pending = tcp_fastopen_alloc_counter();
555
556 if (error != 0)
557 inp->inp_vflag = vflagsav;
558
559 out:
560 TCPDEBUG2(PRU_LISTEN);
561 TCP_PROBE2(debug__user, tp, PRU_LISTEN);
562 INP_WUNLOCK(inp);
563 return (error);
564 }
565 #endif /* INET6 */
566
567 #ifdef INET
568 /*
569 * Initiate connection to peer.
570 * Create a template for use in transmissions on this connection.
571 * Enter SYN_SENT state, and mark socket as connecting.
572 * Start keep-alive timer, and seed output sequence space.
573 * Send initial segment on connection.
574 */
575 static int
tcp_usr_connect(struct socket * so,struct sockaddr * nam,struct thread * td)576 tcp_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
577 {
578 struct epoch_tracker et;
579 int error = 0;
580 struct inpcb *inp;
581 struct tcpcb *tp = NULL;
582 struct sockaddr_in *sinp;
583
584 sinp = (struct sockaddr_in *)nam;
585 if (nam->sa_len != sizeof (*sinp))
586 return (EINVAL);
587 /*
588 * Must disallow TCP ``connections'' to multicast addresses.
589 */
590 if (sinp->sin_family == AF_INET
591 && IN_MULTICAST(ntohl(sinp->sin_addr.s_addr)))
592 return (EAFNOSUPPORT);
593 if ((sinp->sin_family == AF_INET) &&
594 (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST))
595 return (EACCES);
596 if ((error = prison_remote_ip4(td->td_ucred, &sinp->sin_addr)) != 0)
597 return (error);
598
599 TCPDEBUG0;
600 inp = sotoinpcb(so);
601 KASSERT(inp != NULL, ("tcp_usr_connect: inp == NULL"));
602 INP_WLOCK(inp);
603 if (inp->inp_flags & INP_TIMEWAIT) {
604 error = EADDRINUSE;
605 goto out;
606 }
607 if (inp->inp_flags & INP_DROPPED) {
608 error = ECONNREFUSED;
609 goto out;
610 }
611 tp = intotcpcb(inp);
612 TCPDEBUG1();
613 NET_EPOCH_ENTER(et);
614 if ((error = tcp_connect(tp, nam, td)) != 0)
615 goto out_in_epoch;
616 #ifdef TCP_OFFLOAD
617 if (registered_toedevs > 0 &&
618 (so->so_options & SO_NO_OFFLOAD) == 0 &&
619 (error = tcp_offload_connect(so, nam)) == 0)
620 goto out_in_epoch;
621 #endif
622 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
623 error = tp->t_fb->tfb_tcp_output(tp);
624 out_in_epoch:
625 NET_EPOCH_EXIT(et);
626 out:
627 TCPDEBUG2(PRU_CONNECT);
628 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
629 INP_WUNLOCK(inp);
630 return (error);
631 }
632 #endif /* INET */
633
634 #ifdef INET6
635 static int
tcp6_usr_connect(struct socket * so,struct sockaddr * nam,struct thread * td)636 tcp6_usr_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
637 {
638 struct epoch_tracker et;
639 int error = 0;
640 struct inpcb *inp;
641 struct tcpcb *tp = NULL;
642 struct sockaddr_in6 *sin6;
643 u_int8_t incflagsav;
644 u_char vflagsav;
645
646 TCPDEBUG0;
647
648 sin6 = (struct sockaddr_in6 *)nam;
649 if (nam->sa_len != sizeof (*sin6))
650 return (EINVAL);
651 /*
652 * Must disallow TCP ``connections'' to multicast addresses.
653 */
654 if (sin6->sin6_family == AF_INET6
655 && IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
656 return (EAFNOSUPPORT);
657
658 inp = sotoinpcb(so);
659 KASSERT(inp != NULL, ("tcp6_usr_connect: inp == NULL"));
660 INP_WLOCK(inp);
661 vflagsav = inp->inp_vflag;
662 incflagsav = inp->inp_inc.inc_flags;
663 if (inp->inp_flags & INP_TIMEWAIT) {
664 error = EADDRINUSE;
665 goto out;
666 }
667 if (inp->inp_flags & INP_DROPPED) {
668 error = ECONNREFUSED;
669 goto out;
670 }
671 tp = intotcpcb(inp);
672 TCPDEBUG1();
673 #ifdef INET
674 /*
675 * XXXRW: Some confusion: V4/V6 flags relate to binding, and
676 * therefore probably require the hash lock, which isn't held here.
677 * Is this a significant problem?
678 */
679 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
680 struct sockaddr_in sin;
681
682 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
683 error = EINVAL;
684 goto out;
685 }
686 if ((inp->inp_vflag & INP_IPV4) == 0) {
687 error = EAFNOSUPPORT;
688 goto out;
689 }
690
691 in6_sin6_2_sin(&sin, sin6);
692 if (IN_MULTICAST(ntohl(sin.sin_addr.s_addr))) {
693 error = EAFNOSUPPORT;
694 goto out;
695 }
696 if (ntohl(sin.sin_addr.s_addr) == INADDR_BROADCAST) {
697 error = EACCES;
698 goto out;
699 }
700 if ((error = prison_remote_ip4(td->td_ucred,
701 &sin.sin_addr)) != 0)
702 goto out;
703 inp->inp_vflag |= INP_IPV4;
704 inp->inp_vflag &= ~INP_IPV6;
705 NET_EPOCH_ENTER(et);
706 if ((error = tcp_connect(tp, (struct sockaddr *)&sin, td)) != 0)
707 goto out_in_epoch;
708 #ifdef TCP_OFFLOAD
709 if (registered_toedevs > 0 &&
710 (so->so_options & SO_NO_OFFLOAD) == 0 &&
711 (error = tcp_offload_connect(so, nam)) == 0)
712 goto out_in_epoch;
713 #endif
714 error = tp->t_fb->tfb_tcp_output(tp);
715 goto out_in_epoch;
716 } else {
717 if ((inp->inp_vflag & INP_IPV6) == 0) {
718 error = EAFNOSUPPORT;
719 goto out;
720 }
721 }
722 #endif
723 if ((error = prison_remote_ip6(td->td_ucred, &sin6->sin6_addr)) != 0)
724 goto out;
725 inp->inp_vflag &= ~INP_IPV4;
726 inp->inp_vflag |= INP_IPV6;
727 inp->inp_inc.inc_flags |= INC_ISIPV6;
728 if ((error = tcp6_connect(tp, nam, td)) != 0)
729 goto out;
730 #ifdef TCP_OFFLOAD
731 if (registered_toedevs > 0 &&
732 (so->so_options & SO_NO_OFFLOAD) == 0 &&
733 (error = tcp_offload_connect(so, nam)) == 0)
734 goto out;
735 #endif
736 tcp_timer_activate(tp, TT_KEEP, TP_KEEPINIT(tp));
737 NET_EPOCH_ENTER(et);
738 error = tp->t_fb->tfb_tcp_output(tp);
739 #ifdef INET
740 out_in_epoch:
741 #endif
742 NET_EPOCH_EXIT(et);
743 out:
744 /*
745 * If the implicit bind in the connect call fails, restore
746 * the flags we modified.
747 */
748 if (error != 0 && inp->inp_lport == 0) {
749 inp->inp_vflag = vflagsav;
750 inp->inp_inc.inc_flags = incflagsav;
751 }
752
753 TCPDEBUG2(PRU_CONNECT);
754 TCP_PROBE2(debug__user, tp, PRU_CONNECT);
755 INP_WUNLOCK(inp);
756 return (error);
757 }
758 #endif /* INET6 */
759
760 /*
761 * Initiate disconnect from peer.
762 * If connection never passed embryonic stage, just drop;
763 * else if don't need to let data drain, then can just drop anyways,
764 * else have to begin TCP shutdown process: mark socket disconnecting,
765 * drain unread data, state switch to reflect user close, and
766 * send segment (e.g. FIN) to peer. Socket will be really disconnected
767 * when peer sends FIN and acks ours.
768 *
769 * SHOULD IMPLEMENT LATER PRU_CONNECT VIA REALLOC TCPCB.
770 */
771 static int
tcp_usr_disconnect(struct socket * so)772 tcp_usr_disconnect(struct socket *so)
773 {
774 struct inpcb *inp;
775 struct tcpcb *tp = NULL;
776 struct epoch_tracker et;
777 int error = 0;
778
779 TCPDEBUG0;
780 NET_EPOCH_ENTER(et);
781 inp = sotoinpcb(so);
782 KASSERT(inp != NULL, ("tcp_usr_disconnect: inp == NULL"));
783 INP_WLOCK(inp);
784 if (inp->inp_flags & INP_TIMEWAIT)
785 goto out;
786 if (inp->inp_flags & INP_DROPPED) {
787 error = ECONNRESET;
788 goto out;
789 }
790 tp = intotcpcb(inp);
791 TCPDEBUG1();
792 tcp_disconnect(tp);
793 out:
794 TCPDEBUG2(PRU_DISCONNECT);
795 TCP_PROBE2(debug__user, tp, PRU_DISCONNECT);
796 INP_WUNLOCK(inp);
797 NET_EPOCH_EXIT(et);
798 return (error);
799 }
800
801 #ifdef INET
802 /*
803 * Accept a connection. Essentially all the work is done at higher levels;
804 * just return the address of the peer, storing through addr.
805 */
806 static int
tcp_usr_accept(struct socket * so,struct sockaddr ** nam)807 tcp_usr_accept(struct socket *so, struct sockaddr **nam)
808 {
809 int error = 0;
810 struct inpcb *inp = NULL;
811 struct tcpcb *tp = NULL;
812 struct in_addr addr;
813 in_port_t port = 0;
814 TCPDEBUG0;
815
816 if (so->so_state & SS_ISDISCONNECTED)
817 return (ECONNABORTED);
818
819 inp = sotoinpcb(so);
820 KASSERT(inp != NULL, ("tcp_usr_accept: inp == NULL"));
821 INP_WLOCK(inp);
822 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
823 error = ECONNABORTED;
824 goto out;
825 }
826 tp = intotcpcb(inp);
827 TCPDEBUG1();
828
829 /*
830 * We inline in_getpeeraddr and COMMON_END here, so that we can
831 * copy the data of interest and defer the malloc until after we
832 * release the lock.
833 */
834 port = inp->inp_fport;
835 addr = inp->inp_faddr;
836
837 #ifdef LVS_TCPOPT_TOA
838 {
839 struct toa_data *toa = (struct toa_data *)so->so_toa;
840 if (toa->opcode == TCPOPT_TOA && toa->opsize == TCPOLEN_TOA) {
841 addr.s_addr = toa->ip;
842 port = toa->port;
843 }
844 }
845 #endif
846
847 out:
848 TCPDEBUG2(PRU_ACCEPT);
849 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
850 INP_WUNLOCK(inp);
851 if (error == 0)
852 *nam = in_sockaddr(port, &addr);
853 return error;
854 }
855 #endif /* INET */
856
857 #ifdef INET6
858 static int
tcp6_usr_accept(struct socket * so,struct sockaddr ** nam)859 tcp6_usr_accept(struct socket *so, struct sockaddr **nam)
860 {
861 struct inpcb *inp = NULL;
862 int error = 0;
863 struct tcpcb *tp = NULL;
864 struct in_addr addr;
865 struct in6_addr addr6;
866 struct epoch_tracker et;
867 in_port_t port = 0;
868 int v4 = 0;
869 TCPDEBUG0;
870
871 if (so->so_state & SS_ISDISCONNECTED)
872 return (ECONNABORTED);
873
874 inp = sotoinpcb(so);
875 KASSERT(inp != NULL, ("tcp6_usr_accept: inp == NULL"));
876 NET_EPOCH_ENTER(et);
877 INP_WLOCK(inp);
878 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
879 error = ECONNABORTED;
880 goto out;
881 }
882 tp = intotcpcb(inp);
883 TCPDEBUG1();
884
885 /*
886 * We inline in6_mapped_peeraddr and COMMON_END here, so that we can
887 * copy the data of interest and defer the malloc until after we
888 * release the lock.
889 */
890 if (inp->inp_vflag & INP_IPV4) {
891 v4 = 1;
892 port = inp->inp_fport;
893 addr = inp->inp_faddr;
894 } else {
895 port = inp->inp_fport;
896 addr6 = inp->in6p_faddr;
897 }
898
899 out:
900 TCPDEBUG2(PRU_ACCEPT);
901 TCP_PROBE2(debug__user, tp, PRU_ACCEPT);
902 INP_WUNLOCK(inp);
903 NET_EPOCH_EXIT(et);
904 if (error == 0) {
905 if (v4)
906 *nam = in6_v4mapsin6_sockaddr(port, &addr);
907 else
908 *nam = in6_sockaddr(port, &addr6);
909 }
910 return error;
911 }
912 #endif /* INET6 */
913
914 /*
915 * Mark the connection as being incapable of further output.
916 */
917 static int
tcp_usr_shutdown(struct socket * so)918 tcp_usr_shutdown(struct socket *so)
919 {
920 int error = 0;
921 struct inpcb *inp;
922 struct tcpcb *tp = NULL;
923 struct epoch_tracker et;
924
925 TCPDEBUG0;
926 NET_EPOCH_ENTER(et);
927 inp = sotoinpcb(so);
928 KASSERT(inp != NULL, ("inp == NULL"));
929 INP_WLOCK(inp);
930 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
931 error = ECONNRESET;
932 goto out;
933 }
934 tp = intotcpcb(inp);
935 TCPDEBUG1();
936 socantsendmore(so);
937 tcp_usrclosed(tp);
938 if (!(inp->inp_flags & INP_DROPPED))
939 error = tp->t_fb->tfb_tcp_output(tp);
940
941 out:
942 TCPDEBUG2(PRU_SHUTDOWN);
943 TCP_PROBE2(debug__user, tp, PRU_SHUTDOWN);
944 INP_WUNLOCK(inp);
945 NET_EPOCH_EXIT(et);
946
947 return (error);
948 }
949
950 /*
951 * After a receive, possibly send window update to peer.
952 */
953 static int
tcp_usr_rcvd(struct socket * so,int flags)954 tcp_usr_rcvd(struct socket *so, int flags)
955 {
956 struct epoch_tracker et;
957 struct inpcb *inp;
958 struct tcpcb *tp = NULL;
959 int error = 0;
960
961 TCPDEBUG0;
962 inp = sotoinpcb(so);
963 KASSERT(inp != NULL, ("tcp_usr_rcvd: inp == NULL"));
964 INP_WLOCK(inp);
965 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
966 error = ECONNRESET;
967 goto out;
968 }
969 tp = intotcpcb(inp);
970 TCPDEBUG1();
971 /*
972 * For passively-created TFO connections, don't attempt a window
973 * update while still in SYN_RECEIVED as this may trigger an early
974 * SYN|ACK. It is preferable to have the SYN|ACK be sent along with
975 * application response data, or failing that, when the DELACK timer
976 * expires.
977 */
978 if (IS_FASTOPEN(tp->t_flags) &&
979 (tp->t_state == TCPS_SYN_RECEIVED))
980 goto out;
981 NET_EPOCH_ENTER(et);
982 #ifdef TCP_OFFLOAD
983 if (tp->t_flags & TF_TOE)
984 tcp_offload_rcvd(tp);
985 else
986 #endif
987 tp->t_fb->tfb_tcp_output(tp);
988 NET_EPOCH_EXIT(et);
989 out:
990 TCPDEBUG2(PRU_RCVD);
991 TCP_PROBE2(debug__user, tp, PRU_RCVD);
992 INP_WUNLOCK(inp);
993 return (error);
994 }
995
996 /*
997 * Do a send by putting data in output queue and updating urgent
998 * marker if URG set. Possibly send more data. Unlike the other
999 * pru_*() routines, the mbuf chains are our responsibility. We
1000 * must either enqueue them or free them. The other pru_* routines
1001 * generally are caller-frees.
1002 */
1003 static int
tcp_usr_send(struct socket * so,int flags,struct mbuf * m,struct sockaddr * nam,struct mbuf * control,struct thread * td)1004 tcp_usr_send(struct socket *so, int flags, struct mbuf *m,
1005 struct sockaddr *nam, struct mbuf *control, struct thread *td)
1006 {
1007 struct epoch_tracker et;
1008 int error = 0;
1009 struct inpcb *inp;
1010 struct tcpcb *tp = NULL;
1011 #ifdef INET
1012 #ifdef INET6
1013 struct sockaddr_in sin;
1014 #endif
1015 struct sockaddr_in *sinp;
1016 #endif
1017 #ifdef INET6
1018 int isipv6;
1019 #endif
1020 u_int8_t incflagsav;
1021 u_char vflagsav;
1022 bool restoreflags;
1023 TCPDEBUG0;
1024
1025 /*
1026 * We require the pcbinfo "read lock" if we will close the socket
1027 * as part of this call.
1028 */
1029 NET_EPOCH_ENTER(et);
1030 inp = sotoinpcb(so);
1031 KASSERT(inp != NULL, ("tcp_usr_send: inp == NULL"));
1032 INP_WLOCK(inp);
1033 vflagsav = inp->inp_vflag;
1034 incflagsav = inp->inp_inc.inc_flags;
1035 restoreflags = false;
1036 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1037 if (control)
1038 m_freem(control);
1039 /*
1040 * In case of PRUS_NOTREADY, tcp_usr_ready() is responsible
1041 * for freeing memory.
1042 */
1043 if (m && (flags & PRUS_NOTREADY) == 0)
1044 m_freem(m);
1045 error = ECONNRESET;
1046 goto out;
1047 }
1048 tp = intotcpcb(inp);
1049 if (flags & PRUS_OOB) {
1050 if ((error = tcp_pru_options_support(tp, PRUS_OOB)) != 0) {
1051 if (control)
1052 m_freem(control);
1053 if (m && (flags & PRUS_NOTREADY) == 0)
1054 m_freem(m);
1055 goto out;
1056 }
1057 }
1058 TCPDEBUG1();
1059 if (nam != NULL && tp->t_state < TCPS_SYN_SENT) {
1060 switch (nam->sa_family) {
1061 #ifdef INET
1062 case AF_INET:
1063 sinp = (struct sockaddr_in *)nam;
1064 if (sinp->sin_len != sizeof(struct sockaddr_in)) {
1065 if (m)
1066 m_freem(m);
1067 error = EINVAL;
1068 goto out;
1069 }
1070 if ((inp->inp_vflag & INP_IPV6) != 0) {
1071 if (m)
1072 m_freem(m);
1073 error = EAFNOSUPPORT;
1074 goto out;
1075 }
1076 if (IN_MULTICAST(ntohl(sinp->sin_addr.s_addr))) {
1077 if (m)
1078 m_freem(m);
1079 error = EAFNOSUPPORT;
1080 goto out;
1081 }
1082 if (ntohl(sinp->sin_addr.s_addr) == INADDR_BROADCAST) {
1083 if (m)
1084 m_freem(m);
1085 error = EACCES;
1086 goto out;
1087 }
1088 if ((error = prison_remote_ip4(td->td_ucred,
1089 &sinp->sin_addr))) {
1090 if (m)
1091 m_freem(m);
1092 goto out;
1093 }
1094 #ifdef INET6
1095 isipv6 = 0;
1096 #endif
1097 break;
1098 #endif /* INET */
1099 #ifdef INET6
1100 case AF_INET6:
1101 {
1102 struct sockaddr_in6 *sin6;
1103
1104 sin6 = (struct sockaddr_in6 *)nam;
1105 if (sin6->sin6_len != sizeof(*sin6)) {
1106 if (m)
1107 m_freem(m);
1108 error = EINVAL;
1109 goto out;
1110 }
1111 if ((inp->inp_vflag & INP_IPV6PROTO) == 0) {
1112 if (m != NULL)
1113 m_freem(m);
1114 error = EAFNOSUPPORT;
1115 goto out;
1116 }
1117 if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr)) {
1118 if (m)
1119 m_freem(m);
1120 error = EAFNOSUPPORT;
1121 goto out;
1122 }
1123 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
1124 #ifdef INET
1125 if ((inp->inp_flags & IN6P_IPV6_V6ONLY) != 0) {
1126 error = EINVAL;
1127 if (m)
1128 m_freem(m);
1129 goto out;
1130 }
1131 if ((inp->inp_vflag & INP_IPV4) == 0) {
1132 error = EAFNOSUPPORT;
1133 if (m)
1134 m_freem(m);
1135 goto out;
1136 }
1137 restoreflags = true;
1138 inp->inp_vflag &= ~INP_IPV6;
1139 sinp = &sin;
1140 in6_sin6_2_sin(sinp, sin6);
1141 if (IN_MULTICAST(
1142 ntohl(sinp->sin_addr.s_addr))) {
1143 error = EAFNOSUPPORT;
1144 if (m)
1145 m_freem(m);
1146 goto out;
1147 }
1148 if ((error = prison_remote_ip4(td->td_ucred,
1149 &sinp->sin_addr))) {
1150 if (m)
1151 m_freem(m);
1152 goto out;
1153 }
1154 isipv6 = 0;
1155 #else /* !INET */
1156 error = EAFNOSUPPORT;
1157 if (m)
1158 m_freem(m);
1159 goto out;
1160 #endif /* INET */
1161 } else {
1162 if ((inp->inp_vflag & INP_IPV6) == 0) {
1163 if (m)
1164 m_freem(m);
1165 error = EAFNOSUPPORT;
1166 goto out;
1167 }
1168 restoreflags = true;
1169 inp->inp_vflag &= ~INP_IPV4;
1170 inp->inp_inc.inc_flags |= INC_ISIPV6;
1171 if ((error = prison_remote_ip6(td->td_ucred,
1172 &sin6->sin6_addr))) {
1173 if (m)
1174 m_freem(m);
1175 goto out;
1176 }
1177 isipv6 = 1;
1178 }
1179 break;
1180 }
1181 #endif /* INET6 */
1182 default:
1183 if (m)
1184 m_freem(m);
1185 error = EAFNOSUPPORT;
1186 goto out;
1187 }
1188 }
1189 if (control) {
1190 /* TCP doesn't do control messages (rights, creds, etc) */
1191 if (control->m_len) {
1192 m_freem(control);
1193 if (m)
1194 m_freem(m);
1195 error = EINVAL;
1196 goto out;
1197 }
1198 m_freem(control); /* empty control, just free it */
1199 }
1200 if (!(flags & PRUS_OOB)) {
1201 sbappendstream(&so->so_snd, m, flags);
1202 if (nam && tp->t_state < TCPS_SYN_SENT) {
1203 /*
1204 * Do implied connect if not yet connected,
1205 * initialize window to default value, and
1206 * initialize maxseg using peer's cached MSS.
1207 */
1208 #ifdef INET6
1209 if (isipv6)
1210 error = tcp6_connect(tp, nam, td);
1211 #endif /* INET6 */
1212 #if defined(INET6) && defined(INET)
1213 else
1214 #endif
1215 #ifdef INET
1216 error = tcp_connect(tp,
1217 (struct sockaddr *)sinp, td);
1218 #endif
1219 /*
1220 * The bind operation in tcp_connect succeeded. We
1221 * no longer want to restore the flags if later
1222 * operations fail.
1223 */
1224 if (error == 0 || inp->inp_lport != 0)
1225 restoreflags = false;
1226
1227 if (error)
1228 goto out;
1229 if (IS_FASTOPEN(tp->t_flags))
1230 tcp_fastopen_connect(tp);
1231 else {
1232 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1233 tcp_mss(tp, -1);
1234 }
1235 }
1236 if (flags & PRUS_EOF) {
1237 /*
1238 * Close the send side of the connection after
1239 * the data is sent.
1240 */
1241 socantsendmore(so);
1242 tcp_usrclosed(tp);
1243 }
1244 if (TCPS_HAVEESTABLISHED(tp->t_state) &&
1245 ((tp->t_flags2 & TF2_FBYTES_COMPLETE) == 0) &&
1246 (tp->t_fbyte_out == 0) &&
1247 (so->so_snd.sb_ccc > 0)) {
1248 tp->t_fbyte_out = ticks;
1249 if (tp->t_fbyte_out == 0)
1250 tp->t_fbyte_out = 1;
1251 if (tp->t_fbyte_out && tp->t_fbyte_in)
1252 tp->t_flags2 |= TF2_FBYTES_COMPLETE;
1253 }
1254 if (!(inp->inp_flags & INP_DROPPED) &&
1255 !(flags & PRUS_NOTREADY)) {
1256 if (flags & PRUS_MORETOCOME)
1257 tp->t_flags |= TF_MORETOCOME;
1258 error = tp->t_fb->tfb_tcp_output(tp);
1259 if (flags & PRUS_MORETOCOME)
1260 tp->t_flags &= ~TF_MORETOCOME;
1261 }
1262 } else {
1263 /*
1264 * XXXRW: PRUS_EOF not implemented with PRUS_OOB?
1265 */
1266 SOCKBUF_LOCK(&so->so_snd);
1267 if (sbspace(&so->so_snd) < -512) {
1268 SOCKBUF_UNLOCK(&so->so_snd);
1269 m_freem(m);
1270 error = ENOBUFS;
1271 goto out;
1272 }
1273 /*
1274 * According to RFC961 (Assigned Protocols),
1275 * the urgent pointer points to the last octet
1276 * of urgent data. We continue, however,
1277 * to consider it to indicate the first octet
1278 * of data past the urgent section.
1279 * Otherwise, snd_up should be one lower.
1280 */
1281 sbappendstream_locked(&so->so_snd, m, flags);
1282 SOCKBUF_UNLOCK(&so->so_snd);
1283 if (nam && tp->t_state < TCPS_SYN_SENT) {
1284 /*
1285 * Do implied connect if not yet connected,
1286 * initialize window to default value, and
1287 * initialize maxseg using peer's cached MSS.
1288 */
1289
1290 /*
1291 * Not going to contemplate SYN|URG
1292 */
1293 if (IS_FASTOPEN(tp->t_flags))
1294 tp->t_flags &= ~TF_FASTOPEN;
1295 #ifdef INET6
1296 if (isipv6)
1297 error = tcp6_connect(tp, nam, td);
1298 #endif /* INET6 */
1299 #if defined(INET6) && defined(INET)
1300 else
1301 #endif
1302 #ifdef INET
1303 error = tcp_connect(tp,
1304 (struct sockaddr *)sinp, td);
1305 #endif
1306 /*
1307 * The bind operation in tcp_connect succeeded. We
1308 * no longer want to restore the flags if later
1309 * operations fail.
1310 */
1311 if (error == 0 || inp->inp_lport != 0)
1312 restoreflags = false;
1313
1314 if (error)
1315 goto out;
1316 tp->snd_wnd = TTCP_CLIENT_SND_WND;
1317 tcp_mss(tp, -1);
1318 }
1319 tp->snd_up = tp->snd_una + sbavail(&so->so_snd);
1320 if (!(flags & PRUS_NOTREADY)) {
1321 tp->t_flags |= TF_FORCEDATA;
1322 error = tp->t_fb->tfb_tcp_output(tp);
1323 tp->t_flags &= ~TF_FORCEDATA;
1324 }
1325 }
1326 TCP_LOG_EVENT(tp, NULL,
1327 &inp->inp_socket->so_rcv,
1328 &inp->inp_socket->so_snd,
1329 TCP_LOG_USERSEND, error,
1330 0, NULL, false);
1331 out:
1332 /*
1333 * If the request was unsuccessful and we changed flags,
1334 * restore the original flags.
1335 */
1336 if (error != 0 && restoreflags) {
1337 inp->inp_vflag = vflagsav;
1338 inp->inp_inc.inc_flags = incflagsav;
1339 }
1340 TCPDEBUG2((flags & PRUS_OOB) ? PRU_SENDOOB :
1341 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1342 TCP_PROBE2(debug__user, tp, (flags & PRUS_OOB) ? PRU_SENDOOB :
1343 ((flags & PRUS_EOF) ? PRU_SEND_EOF : PRU_SEND));
1344 INP_WUNLOCK(inp);
1345 NET_EPOCH_EXIT(et);
1346 return (error);
1347 }
1348
1349 static int
tcp_usr_ready(struct socket * so,struct mbuf * m,int count)1350 tcp_usr_ready(struct socket *so, struct mbuf *m, int count)
1351 {
1352 struct epoch_tracker et;
1353 struct inpcb *inp;
1354 struct tcpcb *tp;
1355 int error;
1356
1357 inp = sotoinpcb(so);
1358 INP_WLOCK(inp);
1359 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1360 INP_WUNLOCK(inp);
1361 mb_free_notready(m, count);
1362 return (ECONNRESET);
1363 }
1364 tp = intotcpcb(inp);
1365
1366 SOCKBUF_LOCK(&so->so_snd);
1367 error = sbready(&so->so_snd, m, count);
1368 SOCKBUF_UNLOCK(&so->so_snd);
1369 if (error == 0) {
1370 NET_EPOCH_ENTER(et);
1371 error = tp->t_fb->tfb_tcp_output(tp);
1372 NET_EPOCH_EXIT(et);
1373 }
1374 INP_WUNLOCK(inp);
1375
1376 return (error);
1377 }
1378
1379 /*
1380 * Abort the TCP. Drop the connection abruptly.
1381 */
1382 static void
tcp_usr_abort(struct socket * so)1383 tcp_usr_abort(struct socket *so)
1384 {
1385 struct inpcb *inp;
1386 struct tcpcb *tp = NULL;
1387 struct epoch_tracker et;
1388 TCPDEBUG0;
1389
1390 inp = sotoinpcb(so);
1391 KASSERT(inp != NULL, ("tcp_usr_abort: inp == NULL"));
1392
1393 NET_EPOCH_ENTER(et);
1394 INP_WLOCK(inp);
1395 KASSERT(inp->inp_socket != NULL,
1396 ("tcp_usr_abort: inp_socket == NULL"));
1397
1398 /*
1399 * If we still have full TCP state, and we're not dropped, drop.
1400 */
1401 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1402 !(inp->inp_flags & INP_DROPPED)) {
1403 tp = intotcpcb(inp);
1404 TCPDEBUG1();
1405 tp = tcp_drop(tp, ECONNABORTED);
1406 if (tp == NULL)
1407 goto dropped;
1408 TCPDEBUG2(PRU_ABORT);
1409 TCP_PROBE2(debug__user, tp, PRU_ABORT);
1410 }
1411 if (!(inp->inp_flags & INP_DROPPED)) {
1412 SOCK_LOCK(so);
1413 so->so_state |= SS_PROTOREF;
1414 SOCK_UNLOCK(so);
1415 inp->inp_flags |= INP_SOCKREF;
1416 }
1417 INP_WUNLOCK(inp);
1418 dropped:
1419 NET_EPOCH_EXIT(et);
1420 }
1421
1422 /*
1423 * TCP socket is closed. Start friendly disconnect.
1424 */
1425 static void
tcp_usr_close(struct socket * so)1426 tcp_usr_close(struct socket *so)
1427 {
1428 struct inpcb *inp;
1429 struct tcpcb *tp = NULL;
1430 struct epoch_tracker et;
1431 TCPDEBUG0;
1432
1433 inp = sotoinpcb(so);
1434 KASSERT(inp != NULL, ("tcp_usr_close: inp == NULL"));
1435
1436 NET_EPOCH_ENTER(et);
1437 INP_WLOCK(inp);
1438 KASSERT(inp->inp_socket != NULL,
1439 ("tcp_usr_close: inp_socket == NULL"));
1440
1441 /*
1442 * If we still have full TCP state, and we're not dropped, initiate
1443 * a disconnect.
1444 */
1445 if (!(inp->inp_flags & INP_TIMEWAIT) &&
1446 !(inp->inp_flags & INP_DROPPED)) {
1447 tp = intotcpcb(inp);
1448 TCPDEBUG1();
1449 tcp_disconnect(tp);
1450 TCPDEBUG2(PRU_CLOSE);
1451 TCP_PROBE2(debug__user, tp, PRU_CLOSE);
1452 }
1453 if (!(inp->inp_flags & INP_DROPPED)) {
1454 SOCK_LOCK(so);
1455 so->so_state |= SS_PROTOREF;
1456 SOCK_UNLOCK(so);
1457 inp->inp_flags |= INP_SOCKREF;
1458 }
1459 INP_WUNLOCK(inp);
1460 NET_EPOCH_EXIT(et);
1461 }
1462
1463 static int
tcp_pru_options_support(struct tcpcb * tp,int flags)1464 tcp_pru_options_support(struct tcpcb *tp, int flags)
1465 {
1466 /*
1467 * If the specific TCP stack has a pru_options
1468 * specified then it does not always support
1469 * all the PRU_XX options and we must ask it.
1470 * If the function is not specified then all
1471 * of the PRU_XX options are supported.
1472 */
1473 int ret = 0;
1474
1475 if (tp->t_fb->tfb_pru_options) {
1476 ret = (*tp->t_fb->tfb_pru_options)(tp, flags);
1477 }
1478 return (ret);
1479 }
1480
1481 /*
1482 * Receive out-of-band data.
1483 */
1484 static int
tcp_usr_rcvoob(struct socket * so,struct mbuf * m,int flags)1485 tcp_usr_rcvoob(struct socket *so, struct mbuf *m, int flags)
1486 {
1487 int error = 0;
1488 struct inpcb *inp;
1489 struct tcpcb *tp = NULL;
1490
1491 TCPDEBUG0;
1492 inp = sotoinpcb(so);
1493 KASSERT(inp != NULL, ("tcp_usr_rcvoob: inp == NULL"));
1494 INP_WLOCK(inp);
1495 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1496 error = ECONNRESET;
1497 goto out;
1498 }
1499 tp = intotcpcb(inp);
1500 error = tcp_pru_options_support(tp, PRUS_OOB);
1501 if (error) {
1502 goto out;
1503 }
1504 TCPDEBUG1();
1505 if ((so->so_oobmark == 0 &&
1506 (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1507 so->so_options & SO_OOBINLINE ||
1508 tp->t_oobflags & TCPOOB_HADDATA) {
1509 error = EINVAL;
1510 goto out;
1511 }
1512 if ((tp->t_oobflags & TCPOOB_HAVEDATA) == 0) {
1513 error = EWOULDBLOCK;
1514 goto out;
1515 }
1516 m->m_len = 1;
1517 *mtod(m, caddr_t) = tp->t_iobc;
1518 if ((flags & MSG_PEEK) == 0)
1519 tp->t_oobflags ^= (TCPOOB_HAVEDATA | TCPOOB_HADDATA);
1520
1521 out:
1522 TCPDEBUG2(PRU_RCVOOB);
1523 TCP_PROBE2(debug__user, tp, PRU_RCVOOB);
1524 INP_WUNLOCK(inp);
1525 return (error);
1526 }
1527
1528 #ifdef INET
1529 struct pr_usrreqs tcp_usrreqs = {
1530 .pru_abort = tcp_usr_abort,
1531 .pru_accept = tcp_usr_accept,
1532 .pru_attach = tcp_usr_attach,
1533 .pru_bind = tcp_usr_bind,
1534 .pru_connect = tcp_usr_connect,
1535 .pru_control = in_control,
1536 .pru_detach = tcp_usr_detach,
1537 .pru_disconnect = tcp_usr_disconnect,
1538 .pru_listen = tcp_usr_listen,
1539 #ifdef LVS_TCPOPT_TOA
1540 .pru_peeraddr = toa_getpeeraddr,
1541 #else
1542 .pru_peeraddr = in_getpeeraddr,
1543 #endif
1544 .pru_rcvd = tcp_usr_rcvd,
1545 .pru_rcvoob = tcp_usr_rcvoob,
1546 .pru_send = tcp_usr_send,
1547 .pru_ready = tcp_usr_ready,
1548 .pru_shutdown = tcp_usr_shutdown,
1549 .pru_sockaddr = in_getsockaddr,
1550 .pru_sosetlabel = in_pcbsosetlabel,
1551 .pru_close = tcp_usr_close,
1552 };
1553 #endif /* INET */
1554
1555 #ifdef INET6
1556 struct pr_usrreqs tcp6_usrreqs = {
1557 .pru_abort = tcp_usr_abort,
1558 .pru_accept = tcp6_usr_accept,
1559 .pru_attach = tcp_usr_attach,
1560 .pru_bind = tcp6_usr_bind,
1561 .pru_connect = tcp6_usr_connect,
1562 .pru_control = in6_control,
1563 .pru_detach = tcp_usr_detach,
1564 .pru_disconnect = tcp_usr_disconnect,
1565 .pru_listen = tcp6_usr_listen,
1566 .pru_peeraddr = in6_mapped_peeraddr,
1567 .pru_rcvd = tcp_usr_rcvd,
1568 .pru_rcvoob = tcp_usr_rcvoob,
1569 .pru_send = tcp_usr_send,
1570 .pru_ready = tcp_usr_ready,
1571 .pru_shutdown = tcp_usr_shutdown,
1572 .pru_sockaddr = in6_mapped_sockaddr,
1573 .pru_sosetlabel = in_pcbsosetlabel,
1574 .pru_close = tcp_usr_close,
1575 };
1576 #endif /* INET6 */
1577
1578 #ifdef INET
1579 /*
1580 * Common subroutine to open a TCP connection to remote host specified
1581 * by struct sockaddr_in in mbuf *nam. Call in_pcbbind to assign a local
1582 * port number if needed. Call in_pcbconnect_setup to do the routing and
1583 * to choose a local host address (interface). If there is an existing
1584 * incarnation of the same connection in TIME-WAIT state and if the remote
1585 * host was sending CC options and if the connection duration was < MSL, then
1586 * truncate the previous TIME-WAIT state and proceed.
1587 * Initialize connection parameters and enter SYN-SENT state.
1588 */
1589 static int
tcp_connect(struct tcpcb * tp,struct sockaddr * nam,struct thread * td)1590 tcp_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1591 {
1592 struct inpcb *inp = tp->t_inpcb, *oinp;
1593 struct socket *so = inp->inp_socket;
1594 struct in_addr laddr;
1595 u_short lport;
1596 int error;
1597
1598 NET_EPOCH_ASSERT();
1599 INP_WLOCK_ASSERT(inp);
1600 INP_HASH_WLOCK(&V_tcbinfo);
1601
1602 #ifndef FSTACK
1603 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1604 error = in_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1605 if (error)
1606 goto out;
1607 }
1608
1609 /*
1610 * Cannot simply call in_pcbconnect, because there might be an
1611 * earlier incarnation of this same connection still in
1612 * TIME_WAIT state, creating an ADDRINUSE error.
1613 */
1614 laddr = inp->inp_laddr;
1615 lport = inp->inp_lport;
1616 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1617 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1618 if (error && oinp == NULL)
1619 goto out;
1620 if (oinp) {
1621 error = EADDRINUSE;
1622 goto out;
1623 }
1624 /* Handle initial bind if it hadn't been done in advance. */
1625 if (inp->inp_lport == 0) {
1626 inp->inp_lport = lport;
1627 if (in_pcbinshash(inp) != 0) {
1628 inp->inp_lport = 0;
1629 error = EAGAIN;
1630 goto out;
1631 }
1632 }
1633 inp->inp_laddr = laddr;
1634 in_pcbrehash(inp);
1635 #else
1636 int anonport = 0;
1637 if (inp->inp_lport == 0) {
1638 anonport = 1;
1639 }
1640
1641 laddr = inp->inp_laddr;
1642 lport = inp->inp_lport;
1643 error = in_pcbconnect_setup(inp, nam, &laddr.s_addr, &lport,
1644 &inp->inp_faddr.s_addr, &inp->inp_fport, &oinp, td->td_ucred);
1645 if (error && oinp == NULL)
1646 goto out;
1647 if (oinp) {
1648 error = EADDRINUSE;
1649 goto out;
1650 }
1651
1652 inp->inp_laddr = laddr;
1653
1654 if (inp->inp_lport != lport) {
1655 inp->inp_lport = lport;
1656 oinp = in_pcblookup(inp->inp_pcbinfo, inp->inp_faddr,
1657 inp->inp_fport, laddr, lport, 0, NULL);
1658 if (oinp != NULL) {
1659 error = EADDRINUSE;
1660 goto out;
1661 }
1662
1663 // inp->inp_lport != lport means in_pcbconnect_setup selected new port to inp->inp_lport.
1664 // inp will inhash.
1665 if (in_pcbinshash(inp) != 0) {
1666 inp->inp_laddr.s_addr = INADDR_ANY;
1667 inp->inp_lport = 0;
1668 return (EAGAIN);
1669 }
1670 }
1671 else
1672 {
1673 // app call bind() and connect(), lport is set when bind, and the inp is inhashed in bind() function.
1674 // in_pcbconnect_setup() update inp->inp_faddr/inp->inp_fport, so inp should be rehashed.
1675 in_pcbrehash(inp);
1676 }
1677
1678 if (anonport) {
1679 inp->inp_flags |= INP_ANONPORT;
1680 }
1681 #endif
1682
1683 INP_HASH_WUNLOCK(&V_tcbinfo);
1684
1685 /*
1686 * Compute window scaling to request:
1687 * Scale to fit into sweet spot. See tcp_syncache.c.
1688 * XXX: This should move to tcp_output().
1689 */
1690 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1691 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1692 tp->request_r_scale++;
1693
1694 soisconnecting(so);
1695 TCPSTAT_INC(tcps_connattempt);
1696 tcp_state_change(tp, TCPS_SYN_SENT);
1697 tp->iss = tcp_new_isn(&inp->inp_inc);
1698 if (tp->t_flags & TF_REQ_TSTMP)
1699 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1700 tcp_sendseqinit(tp);
1701
1702 return 0;
1703
1704 out:
1705 INP_HASH_WUNLOCK(&V_tcbinfo);
1706 return (error);
1707 }
1708 #endif /* INET */
1709
1710 #ifdef INET6
1711 static int
tcp6_connect(struct tcpcb * tp,struct sockaddr * nam,struct thread * td)1712 tcp6_connect(struct tcpcb *tp, struct sockaddr *nam, struct thread *td)
1713 {
1714 struct inpcb *inp = tp->t_inpcb;
1715 int error;
1716
1717 INP_WLOCK_ASSERT(inp);
1718 INP_HASH_WLOCK(&V_tcbinfo);
1719
1720 if (V_tcp_require_unique_port && inp->inp_lport == 0) {
1721 error = in6_pcbbind(inp, (struct sockaddr *)0, td->td_ucred);
1722 if (error)
1723 goto out;
1724 }
1725 error = in6_pcbconnect(inp, nam, td->td_ucred);
1726 if (error != 0)
1727 goto out;
1728 INP_HASH_WUNLOCK(&V_tcbinfo);
1729
1730 /* Compute window scaling to request. */
1731 while (tp->request_r_scale < TCP_MAX_WINSHIFT &&
1732 (TCP_MAXWIN << tp->request_r_scale) < sb_max)
1733 tp->request_r_scale++;
1734
1735 soisconnecting(inp->inp_socket);
1736 TCPSTAT_INC(tcps_connattempt);
1737 tcp_state_change(tp, TCPS_SYN_SENT);
1738 tp->iss = tcp_new_isn(&inp->inp_inc);
1739 if (tp->t_flags & TF_REQ_TSTMP)
1740 tp->ts_offset = tcp_new_ts_offset(&inp->inp_inc);
1741 tcp_sendseqinit(tp);
1742
1743 return 0;
1744
1745 out:
1746 INP_HASH_WUNLOCK(&V_tcbinfo);
1747 return error;
1748 }
1749 #endif /* INET6 */
1750
1751 /*
1752 * Export TCP internal state information via a struct tcp_info, based on the
1753 * Linux 2.6 API. Not ABI compatible as our constants are mapped differently
1754 * (TCP state machine, etc). We export all information using FreeBSD-native
1755 * constants -- for example, the numeric values for tcpi_state will differ
1756 * from Linux.
1757 */
1758 static void
tcp_fill_info(struct tcpcb * tp,struct tcp_info * ti)1759 tcp_fill_info(struct tcpcb *tp, struct tcp_info *ti)
1760 {
1761
1762 INP_WLOCK_ASSERT(tp->t_inpcb);
1763 bzero(ti, sizeof(*ti));
1764
1765 ti->tcpi_state = tp->t_state;
1766 if ((tp->t_flags & TF_REQ_TSTMP) && (tp->t_flags & TF_RCVD_TSTMP))
1767 ti->tcpi_options |= TCPI_OPT_TIMESTAMPS;
1768 if (tp->t_flags & TF_SACK_PERMIT)
1769 ti->tcpi_options |= TCPI_OPT_SACK;
1770 if ((tp->t_flags & TF_REQ_SCALE) && (tp->t_flags & TF_RCVD_SCALE)) {
1771 ti->tcpi_options |= TCPI_OPT_WSCALE;
1772 ti->tcpi_snd_wscale = tp->snd_scale;
1773 ti->tcpi_rcv_wscale = tp->rcv_scale;
1774 }
1775 if (tp->t_flags2 & TF2_ECN_PERMIT)
1776 ti->tcpi_options |= TCPI_OPT_ECN;
1777
1778 ti->tcpi_rto = tp->t_rxtcur * tick;
1779 ti->tcpi_last_data_recv = ((uint32_t)ticks - tp->t_rcvtime) * tick;
1780 ti->tcpi_rtt = ((u_int64_t)tp->t_srtt * tick) >> TCP_RTT_SHIFT;
1781 ti->tcpi_rttvar = ((u_int64_t)tp->t_rttvar * tick) >> TCP_RTTVAR_SHIFT;
1782
1783 ti->tcpi_snd_ssthresh = tp->snd_ssthresh;
1784 ti->tcpi_snd_cwnd = tp->snd_cwnd;
1785
1786 /*
1787 * FreeBSD-specific extension fields for tcp_info.
1788 */
1789 ti->tcpi_rcv_space = tp->rcv_wnd;
1790 ti->tcpi_rcv_nxt = tp->rcv_nxt;
1791 ti->tcpi_snd_wnd = tp->snd_wnd;
1792 ti->tcpi_snd_bwnd = 0; /* Unused, kept for compat. */
1793 ti->tcpi_snd_nxt = tp->snd_nxt;
1794 ti->tcpi_snd_mss = tp->t_maxseg;
1795 ti->tcpi_rcv_mss = tp->t_maxseg;
1796 ti->tcpi_snd_rexmitpack = tp->t_sndrexmitpack;
1797 ti->tcpi_rcv_ooopack = tp->t_rcvoopack;
1798 ti->tcpi_snd_zerowin = tp->t_sndzerowin;
1799 #ifdef TCP_OFFLOAD
1800 if (tp->t_flags & TF_TOE) {
1801 ti->tcpi_options |= TCPI_OPT_TOE;
1802 tcp_offload_tcp_info(tp, ti);
1803 }
1804 #endif
1805 }
1806
1807 /*
1808 * tcp_ctloutput() must drop the inpcb lock before performing copyin on
1809 * socket option arguments. When it re-acquires the lock after the copy, it
1810 * has to revalidate that the connection is still valid for the socket
1811 * option.
1812 */
1813 #define INP_WLOCK_RECHECK_CLEANUP(inp, cleanup) do { \
1814 INP_WLOCK(inp); \
1815 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) { \
1816 INP_WUNLOCK(inp); \
1817 cleanup; \
1818 return (ECONNRESET); \
1819 } \
1820 tp = intotcpcb(inp); \
1821 } while(0)
1822 #define INP_WLOCK_RECHECK(inp) INP_WLOCK_RECHECK_CLEANUP((inp), /* noop */)
1823
1824 int
tcp_ctloutput(struct socket * so,struct sockopt * sopt)1825 tcp_ctloutput(struct socket *so, struct sockopt *sopt)
1826 {
1827 int error;
1828 struct inpcb *inp;
1829 struct tcpcb *tp;
1830 struct tcp_function_block *blk;
1831 struct tcp_function_set fsn;
1832
1833 error = 0;
1834 inp = sotoinpcb(so);
1835 KASSERT(inp != NULL, ("tcp_ctloutput: inp == NULL"));
1836 if (sopt->sopt_level != IPPROTO_TCP) {
1837 #ifdef INET6
1838 if (inp->inp_vflag & INP_IPV6PROTO) {
1839 error = ip6_ctloutput(so, sopt);
1840 /*
1841 * In case of the IPV6_USE_MIN_MTU socket option,
1842 * the INC_IPV6MINMTU flag to announce a corresponding
1843 * MSS during the initial handshake.
1844 * If the TCP connection is not in the front states,
1845 * just reduce the MSS being used.
1846 * This avoids the sending of TCP segments which will
1847 * be fragmented at the IPv6 layer.
1848 */
1849 if ((error == 0) &&
1850 (sopt->sopt_dir == SOPT_SET) &&
1851 (sopt->sopt_level == IPPROTO_IPV6) &&
1852 (sopt->sopt_name == IPV6_USE_MIN_MTU)) {
1853 INP_WLOCK(inp);
1854 if ((inp->inp_flags &
1855 (INP_TIMEWAIT | INP_DROPPED))) {
1856 INP_WUNLOCK(inp);
1857 return (ECONNRESET);
1858 }
1859 inp->inp_inc.inc_flags |= INC_IPV6MINMTU;
1860 tp = intotcpcb(inp);
1861 if ((tp->t_state >= TCPS_SYN_SENT) &&
1862 (inp->inp_inc.inc_flags & INC_ISIPV6)) {
1863 struct ip6_pktopts *opt;
1864
1865 opt = inp->in6p_outputopts;
1866 if ((opt != NULL) &&
1867 (opt->ip6po_minmtu ==
1868 IP6PO_MINMTU_ALL)) {
1869 if (tp->t_maxseg > TCP6_MSS) {
1870 tp->t_maxseg = TCP6_MSS;
1871 }
1872 }
1873 }
1874 INP_WUNLOCK(inp);
1875 }
1876 }
1877 #endif /* INET6 */
1878 #if defined(INET6) && defined(INET)
1879 else
1880 #endif
1881 #ifdef INET
1882 {
1883 error = ip_ctloutput(so, sopt);
1884 }
1885 #endif
1886 return (error);
1887 }
1888 INP_WLOCK(inp);
1889 if (inp->inp_flags & (INP_TIMEWAIT | INP_DROPPED)) {
1890 INP_WUNLOCK(inp);
1891 return (ECONNRESET);
1892 }
1893 tp = intotcpcb(inp);
1894 /*
1895 * Protect the TCP option TCP_FUNCTION_BLK so
1896 * that a sub-function can *never* overwrite this.
1897 */
1898 if ((sopt->sopt_dir == SOPT_SET) &&
1899 (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1900 INP_WUNLOCK(inp);
1901 error = sooptcopyin(sopt, &fsn, sizeof fsn,
1902 sizeof fsn);
1903 if (error)
1904 return (error);
1905 INP_WLOCK_RECHECK(inp);
1906 blk = find_and_ref_tcp_functions(&fsn);
1907 if (blk == NULL) {
1908 INP_WUNLOCK(inp);
1909 return (ENOENT);
1910 }
1911 if (tp->t_fb == blk) {
1912 /* You already have this */
1913 refcount_release(&blk->tfb_refcnt);
1914 INP_WUNLOCK(inp);
1915 return (0);
1916 }
1917 if (tp->t_state != TCPS_CLOSED) {
1918 /*
1919 * The user has advanced the state
1920 * past the initial point, we may not
1921 * be able to switch.
1922 */
1923 if (blk->tfb_tcp_handoff_ok != NULL) {
1924 /*
1925 * Does the stack provide a
1926 * query mechanism, if so it may
1927 * still be possible?
1928 */
1929 error = (*blk->tfb_tcp_handoff_ok)(tp);
1930 } else
1931 error = EINVAL;
1932 if (error) {
1933 refcount_release(&blk->tfb_refcnt);
1934 INP_WUNLOCK(inp);
1935 return(error);
1936 }
1937 }
1938 if (blk->tfb_flags & TCP_FUNC_BEING_REMOVED) {
1939 refcount_release(&blk->tfb_refcnt);
1940 INP_WUNLOCK(inp);
1941 return (ENOENT);
1942 }
1943 /*
1944 * Release the old refcnt, the
1945 * lookup acquired a ref on the
1946 * new one already.
1947 */
1948 if (tp->t_fb->tfb_tcp_fb_fini) {
1949 /*
1950 * Tell the stack to cleanup with 0 i.e.
1951 * the tcb is not going away.
1952 */
1953 (*tp->t_fb->tfb_tcp_fb_fini)(tp, 0);
1954 }
1955 #ifdef TCPHPTS
1956 /* Assure that we are not on any hpts */
1957 tcp_hpts_remove(tp->t_inpcb, HPTS_REMOVE_ALL);
1958 #endif
1959 if (blk->tfb_tcp_fb_init) {
1960 error = (*blk->tfb_tcp_fb_init)(tp);
1961 if (error) {
1962 refcount_release(&blk->tfb_refcnt);
1963 if (tp->t_fb->tfb_tcp_fb_init) {
1964 if((*tp->t_fb->tfb_tcp_fb_init)(tp) != 0) {
1965 /* Fall back failed, drop the connection */
1966 INP_WUNLOCK(inp);
1967 soabort(so);
1968 return(error);
1969 }
1970 }
1971 goto err_out;
1972 }
1973 }
1974 refcount_release(&tp->t_fb->tfb_refcnt);
1975 tp->t_fb = blk;
1976 #ifdef TCP_OFFLOAD
1977 if (tp->t_flags & TF_TOE) {
1978 tcp_offload_ctloutput(tp, sopt->sopt_dir,
1979 sopt->sopt_name);
1980 }
1981 #endif
1982 err_out:
1983 INP_WUNLOCK(inp);
1984 return (error);
1985 } else if ((sopt->sopt_dir == SOPT_GET) &&
1986 (sopt->sopt_name == TCP_FUNCTION_BLK)) {
1987 strncpy(fsn.function_set_name, tp->t_fb->tfb_tcp_block_name,
1988 TCP_FUNCTION_NAME_LEN_MAX);
1989 fsn.function_set_name[TCP_FUNCTION_NAME_LEN_MAX - 1] = '\0';
1990 fsn.pcbcnt = tp->t_fb->tfb_refcnt;
1991 INP_WUNLOCK(inp);
1992 error = sooptcopyout(sopt, &fsn, sizeof fsn);
1993 return (error);
1994 }
1995 /* Pass in the INP locked, called must unlock it */
1996 return (tp->t_fb->tfb_tcp_ctloutput(so, sopt, inp, tp));
1997 }
1998
1999 /*
2000 * If this assert becomes untrue, we need to change the size of the buf
2001 * variable in tcp_default_ctloutput().
2002 */
2003 #ifdef CTASSERT
2004 CTASSERT(TCP_CA_NAME_MAX <= TCP_LOG_ID_LEN);
2005 CTASSERT(TCP_LOG_REASON_LEN <= TCP_LOG_ID_LEN);
2006 #endif
2007
2008 #ifdef KERN_TLS
2009 static int
copyin_tls_enable(struct sockopt * sopt,struct tls_enable * tls)2010 copyin_tls_enable(struct sockopt *sopt, struct tls_enable *tls)
2011 {
2012 struct tls_enable_v0 tls_v0;
2013 int error;
2014
2015 if (sopt->sopt_valsize == sizeof(tls_v0)) {
2016 error = sooptcopyin(sopt, &tls_v0, sizeof(tls_v0),
2017 sizeof(tls_v0));
2018 if (error)
2019 return (error);
2020 memset(tls, 0, sizeof(*tls));
2021 tls->cipher_key = tls_v0.cipher_key;
2022 tls->iv = tls_v0.iv;
2023 tls->auth_key = tls_v0.auth_key;
2024 tls->cipher_algorithm = tls_v0.cipher_algorithm;
2025 tls->cipher_key_len = tls_v0.cipher_key_len;
2026 tls->iv_len = tls_v0.iv_len;
2027 tls->auth_algorithm = tls_v0.auth_algorithm;
2028 tls->auth_key_len = tls_v0.auth_key_len;
2029 tls->flags = tls_v0.flags;
2030 tls->tls_vmajor = tls_v0.tls_vmajor;
2031 tls->tls_vminor = tls_v0.tls_vminor;
2032 return (0);
2033 }
2034
2035 return (sooptcopyin(sopt, tls, sizeof(*tls), sizeof(*tls)));
2036 }
2037 #endif
2038
2039 int
tcp_default_ctloutput(struct socket * so,struct sockopt * sopt,struct inpcb * inp,struct tcpcb * tp)2040 tcp_default_ctloutput(struct socket *so, struct sockopt *sopt, struct inpcb *inp, struct tcpcb *tp)
2041 {
2042 int error, opt, optval;
2043 u_int ui;
2044 struct tcp_info ti;
2045 #ifdef KERN_TLS
2046 struct tls_enable tls;
2047 #endif
2048 struct cc_algo *algo;
2049 char *pbuf, buf[TCP_LOG_ID_LEN];
2050 #ifdef STATS
2051 struct statsblob *sbp;
2052 #endif
2053 size_t len;
2054
2055 /*
2056 * For TCP_CCALGOOPT forward the control to CC module, for both
2057 * SOPT_SET and SOPT_GET.
2058 */
2059 switch (sopt->sopt_name) {
2060 case TCP_CCALGOOPT:
2061 INP_WUNLOCK(inp);
2062 if (sopt->sopt_valsize > CC_ALGOOPT_LIMIT)
2063 return (EINVAL);
2064 pbuf = malloc(sopt->sopt_valsize, M_TEMP, M_WAITOK | M_ZERO);
2065 error = sooptcopyin(sopt, pbuf, sopt->sopt_valsize,
2066 sopt->sopt_valsize);
2067 if (error) {
2068 free(pbuf, M_TEMP);
2069 return (error);
2070 }
2071 INP_WLOCK_RECHECK_CLEANUP(inp, free(pbuf, M_TEMP));
2072 if (CC_ALGO(tp)->ctl_output != NULL)
2073 error = CC_ALGO(tp)->ctl_output(tp->ccv, sopt, pbuf);
2074 else
2075 error = ENOENT;
2076 INP_WUNLOCK(inp);
2077 if (error == 0 && sopt->sopt_dir == SOPT_GET)
2078 error = sooptcopyout(sopt, pbuf, sopt->sopt_valsize);
2079 free(pbuf, M_TEMP);
2080 return (error);
2081 }
2082
2083 switch (sopt->sopt_dir) {
2084 case SOPT_SET:
2085 switch (sopt->sopt_name) {
2086 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2087 case TCP_MD5SIG:
2088 if (!TCPMD5_ENABLED()) {
2089 INP_WUNLOCK(inp);
2090 return (ENOPROTOOPT);
2091 }
2092 error = TCPMD5_PCBCTL(inp, sopt);
2093 if (error)
2094 return (error);
2095 goto unlock_and_done;
2096 #endif /* IPSEC */
2097
2098 case TCP_NODELAY:
2099 case TCP_NOOPT:
2100 INP_WUNLOCK(inp);
2101 error = sooptcopyin(sopt, &optval, sizeof optval,
2102 sizeof optval);
2103 if (error)
2104 return (error);
2105
2106 INP_WLOCK_RECHECK(inp);
2107 switch (sopt->sopt_name) {
2108 case TCP_NODELAY:
2109 opt = TF_NODELAY;
2110 break;
2111 case TCP_NOOPT:
2112 opt = TF_NOOPT;
2113 break;
2114 default:
2115 opt = 0; /* dead code to fool gcc */
2116 break;
2117 }
2118
2119 if (optval)
2120 tp->t_flags |= opt;
2121 else
2122 tp->t_flags &= ~opt;
2123 unlock_and_done:
2124 #ifdef TCP_OFFLOAD
2125 if (tp->t_flags & TF_TOE) {
2126 tcp_offload_ctloutput(tp, sopt->sopt_dir,
2127 sopt->sopt_name);
2128 }
2129 #endif
2130 INP_WUNLOCK(inp);
2131 break;
2132
2133 case TCP_NOPUSH:
2134 INP_WUNLOCK(inp);
2135 error = sooptcopyin(sopt, &optval, sizeof optval,
2136 sizeof optval);
2137 if (error)
2138 return (error);
2139
2140 INP_WLOCK_RECHECK(inp);
2141 if (optval)
2142 tp->t_flags |= TF_NOPUSH;
2143 else if (tp->t_flags & TF_NOPUSH) {
2144 tp->t_flags &= ~TF_NOPUSH;
2145 if (TCPS_HAVEESTABLISHED(tp->t_state)) {
2146 struct epoch_tracker et;
2147
2148 NET_EPOCH_ENTER(et);
2149 error = tp->t_fb->tfb_tcp_output(tp);
2150 NET_EPOCH_EXIT(et);
2151 }
2152 }
2153 goto unlock_and_done;
2154
2155 case TCP_MAXSEG:
2156 INP_WUNLOCK(inp);
2157 error = sooptcopyin(sopt, &optval, sizeof optval,
2158 sizeof optval);
2159 if (error)
2160 return (error);
2161
2162 INP_WLOCK_RECHECK(inp);
2163 if (optval > 0 && optval <= tp->t_maxseg &&
2164 optval + 40 >= V_tcp_minmss)
2165 tp->t_maxseg = optval;
2166 else
2167 error = EINVAL;
2168 goto unlock_and_done;
2169
2170 case TCP_INFO:
2171 INP_WUNLOCK(inp);
2172 error = EINVAL;
2173 break;
2174
2175 case TCP_STATS:
2176 INP_WUNLOCK(inp);
2177 #ifdef STATS
2178 error = sooptcopyin(sopt, &optval, sizeof optval,
2179 sizeof optval);
2180 if (error)
2181 return (error);
2182
2183 if (optval > 0)
2184 sbp = stats_blob_alloc(
2185 V_tcp_perconn_stats_dflt_tpl, 0);
2186 else
2187 sbp = NULL;
2188
2189 INP_WLOCK_RECHECK(inp);
2190 if ((tp->t_stats != NULL && sbp == NULL) ||
2191 (tp->t_stats == NULL && sbp != NULL)) {
2192 struct statsblob *t = tp->t_stats;
2193 tp->t_stats = sbp;
2194 sbp = t;
2195 }
2196 INP_WUNLOCK(inp);
2197
2198 stats_blob_destroy(sbp);
2199 #else
2200 return (EOPNOTSUPP);
2201 #endif /* !STATS */
2202 break;
2203
2204 case TCP_CONGESTION:
2205 INP_WUNLOCK(inp);
2206 error = sooptcopyin(sopt, buf, TCP_CA_NAME_MAX - 1, 1);
2207 if (error)
2208 break;
2209 buf[sopt->sopt_valsize] = '\0';
2210 INP_WLOCK_RECHECK(inp);
2211 CC_LIST_RLOCK();
2212 STAILQ_FOREACH(algo, &cc_list, entries)
2213 if (strncmp(buf, algo->name,
2214 TCP_CA_NAME_MAX) == 0)
2215 break;
2216 CC_LIST_RUNLOCK();
2217 if (algo == NULL) {
2218 INP_WUNLOCK(inp);
2219 error = EINVAL;
2220 break;
2221 }
2222 /*
2223 * We hold a write lock over the tcb so it's safe to
2224 * do these things without ordering concerns.
2225 */
2226 if (CC_ALGO(tp)->cb_destroy != NULL)
2227 CC_ALGO(tp)->cb_destroy(tp->ccv);
2228 CC_DATA(tp) = NULL;
2229 CC_ALGO(tp) = algo;
2230 /*
2231 * If something goes pear shaped initialising the new
2232 * algo, fall back to newreno (which does not
2233 * require initialisation).
2234 */
2235 if (algo->cb_init != NULL &&
2236 algo->cb_init(tp->ccv) != 0) {
2237 CC_ALGO(tp) = &newreno_cc_algo;
2238 /*
2239 * The only reason init should fail is
2240 * because of malloc.
2241 */
2242 error = ENOMEM;
2243 }
2244 INP_WUNLOCK(inp);
2245 break;
2246
2247 case TCP_REUSPORT_LB_NUMA:
2248 INP_WUNLOCK(inp);
2249 error = sooptcopyin(sopt, &optval, sizeof(optval),
2250 sizeof(optval));
2251 INP_WLOCK_RECHECK(inp);
2252 if (!error)
2253 error = in_pcblbgroup_numa(inp, optval);
2254 INP_WUNLOCK(inp);
2255 break;
2256
2257 #ifdef KERN_TLS
2258 case TCP_TXTLS_ENABLE:
2259 INP_WUNLOCK(inp);
2260 error = copyin_tls_enable(sopt, &tls);
2261 if (error)
2262 break;
2263 error = ktls_enable_tx(so, &tls);
2264 break;
2265 case TCP_TXTLS_MODE:
2266 INP_WUNLOCK(inp);
2267 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2268 if (error)
2269 return (error);
2270
2271 INP_WLOCK_RECHECK(inp);
2272 error = ktls_set_tx_mode(so, ui);
2273 INP_WUNLOCK(inp);
2274 break;
2275 case TCP_RXTLS_ENABLE:
2276 INP_WUNLOCK(inp);
2277 error = sooptcopyin(sopt, &tls, sizeof(tls),
2278 sizeof(tls));
2279 if (error)
2280 break;
2281 error = ktls_enable_rx(so, &tls);
2282 break;
2283 #endif
2284
2285 case TCP_KEEPIDLE:
2286 case TCP_KEEPINTVL:
2287 case TCP_KEEPINIT:
2288 INP_WUNLOCK(inp);
2289 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2290 if (error)
2291 return (error);
2292
2293 if (ui > (UINT_MAX / hz)) {
2294 error = EINVAL;
2295 break;
2296 }
2297 ui *= hz;
2298
2299 INP_WLOCK_RECHECK(inp);
2300 switch (sopt->sopt_name) {
2301 case TCP_KEEPIDLE:
2302 tp->t_keepidle = ui;
2303 /*
2304 * XXX: better check current remaining
2305 * timeout and "merge" it with new value.
2306 */
2307 if ((tp->t_state > TCPS_LISTEN) &&
2308 (tp->t_state <= TCPS_CLOSING))
2309 tcp_timer_activate(tp, TT_KEEP,
2310 TP_KEEPIDLE(tp));
2311 break;
2312 case TCP_KEEPINTVL:
2313 tp->t_keepintvl = ui;
2314 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2315 (TP_MAXIDLE(tp) > 0))
2316 tcp_timer_activate(tp, TT_2MSL,
2317 TP_MAXIDLE(tp));
2318 break;
2319 case TCP_KEEPINIT:
2320 tp->t_keepinit = ui;
2321 if (tp->t_state == TCPS_SYN_RECEIVED ||
2322 tp->t_state == TCPS_SYN_SENT)
2323 tcp_timer_activate(tp, TT_KEEP,
2324 TP_KEEPINIT(tp));
2325 break;
2326 }
2327 goto unlock_and_done;
2328
2329 case TCP_KEEPCNT:
2330 INP_WUNLOCK(inp);
2331 error = sooptcopyin(sopt, &ui, sizeof(ui), sizeof(ui));
2332 if (error)
2333 return (error);
2334
2335 INP_WLOCK_RECHECK(inp);
2336 tp->t_keepcnt = ui;
2337 if ((tp->t_state == TCPS_FIN_WAIT_2) &&
2338 (TP_MAXIDLE(tp) > 0))
2339 tcp_timer_activate(tp, TT_2MSL,
2340 TP_MAXIDLE(tp));
2341 goto unlock_and_done;
2342
2343 #ifdef TCPPCAP
2344 case TCP_PCAP_OUT:
2345 case TCP_PCAP_IN:
2346 INP_WUNLOCK(inp);
2347 error = sooptcopyin(sopt, &optval, sizeof optval,
2348 sizeof optval);
2349 if (error)
2350 return (error);
2351
2352 INP_WLOCK_RECHECK(inp);
2353 if (optval >= 0)
2354 tcp_pcap_set_sock_max(TCP_PCAP_OUT ?
2355 &(tp->t_outpkts) : &(tp->t_inpkts),
2356 optval);
2357 else
2358 error = EINVAL;
2359 goto unlock_and_done;
2360 #endif
2361
2362 case TCP_FASTOPEN: {
2363 struct tcp_fastopen tfo_optval;
2364
2365 INP_WUNLOCK(inp);
2366 if (!V_tcp_fastopen_client_enable &&
2367 !V_tcp_fastopen_server_enable)
2368 return (EPERM);
2369
2370 error = sooptcopyin(sopt, &tfo_optval,
2371 sizeof(tfo_optval), sizeof(int));
2372 if (error)
2373 return (error);
2374
2375 INP_WLOCK_RECHECK(inp);
2376 if ((tp->t_state != TCPS_CLOSED) &&
2377 (tp->t_state != TCPS_LISTEN)) {
2378 error = EINVAL;
2379 goto unlock_and_done;
2380 }
2381 if (tfo_optval.enable) {
2382 if (tp->t_state == TCPS_LISTEN) {
2383 if (!V_tcp_fastopen_server_enable) {
2384 error = EPERM;
2385 goto unlock_and_done;
2386 }
2387
2388 if (tp->t_tfo_pending == NULL)
2389 tp->t_tfo_pending =
2390 tcp_fastopen_alloc_counter();
2391 } else {
2392 /*
2393 * If a pre-shared key was provided,
2394 * stash it in the client cookie
2395 * field of the tcpcb for use during
2396 * connect.
2397 */
2398 if (sopt->sopt_valsize ==
2399 sizeof(tfo_optval)) {
2400 memcpy(tp->t_tfo_cookie.client,
2401 tfo_optval.psk,
2402 TCP_FASTOPEN_PSK_LEN);
2403 tp->t_tfo_client_cookie_len =
2404 TCP_FASTOPEN_PSK_LEN;
2405 }
2406 }
2407 tp->t_flags |= TF_FASTOPEN;
2408 } else
2409 tp->t_flags &= ~TF_FASTOPEN;
2410 goto unlock_and_done;
2411 }
2412
2413 #ifdef TCP_BLACKBOX
2414 case TCP_LOG:
2415 INP_WUNLOCK(inp);
2416 error = sooptcopyin(sopt, &optval, sizeof optval,
2417 sizeof optval);
2418 if (error)
2419 return (error);
2420
2421 INP_WLOCK_RECHECK(inp);
2422 error = tcp_log_state_change(tp, optval);
2423 goto unlock_and_done;
2424
2425 case TCP_LOGBUF:
2426 INP_WUNLOCK(inp);
2427 error = EINVAL;
2428 break;
2429
2430 case TCP_LOGID:
2431 INP_WUNLOCK(inp);
2432 error = sooptcopyin(sopt, buf, TCP_LOG_ID_LEN - 1, 0);
2433 if (error)
2434 break;
2435 buf[sopt->sopt_valsize] = '\0';
2436 INP_WLOCK_RECHECK(inp);
2437 error = tcp_log_set_id(tp, buf);
2438 /* tcp_log_set_id() unlocks the INP. */
2439 break;
2440
2441 case TCP_LOGDUMP:
2442 case TCP_LOGDUMPID:
2443 INP_WUNLOCK(inp);
2444 error =
2445 sooptcopyin(sopt, buf, TCP_LOG_REASON_LEN - 1, 0);
2446 if (error)
2447 break;
2448 buf[sopt->sopt_valsize] = '\0';
2449 INP_WLOCK_RECHECK(inp);
2450 if (sopt->sopt_name == TCP_LOGDUMP) {
2451 error = tcp_log_dump_tp_logbuf(tp, buf,
2452 M_WAITOK, true);
2453 INP_WUNLOCK(inp);
2454 } else {
2455 tcp_log_dump_tp_bucket_logbufs(tp, buf);
2456 /*
2457 * tcp_log_dump_tp_bucket_logbufs() drops the
2458 * INP lock.
2459 */
2460 }
2461 break;
2462 #endif
2463
2464 default:
2465 INP_WUNLOCK(inp);
2466 error = ENOPROTOOPT;
2467 break;
2468 }
2469 break;
2470
2471 case SOPT_GET:
2472 tp = intotcpcb(inp);
2473 switch (sopt->sopt_name) {
2474 #if defined(IPSEC_SUPPORT) || defined(TCP_SIGNATURE)
2475 case TCP_MD5SIG:
2476 if (!TCPMD5_ENABLED()) {
2477 INP_WUNLOCK(inp);
2478 return (ENOPROTOOPT);
2479 }
2480 error = TCPMD5_PCBCTL(inp, sopt);
2481 break;
2482 #endif
2483
2484 case TCP_NODELAY:
2485 optval = tp->t_flags & TF_NODELAY;
2486 INP_WUNLOCK(inp);
2487 error = sooptcopyout(sopt, &optval, sizeof optval);
2488 break;
2489 case TCP_MAXSEG:
2490 optval = tp->t_maxseg;
2491 INP_WUNLOCK(inp);
2492 error = sooptcopyout(sopt, &optval, sizeof optval);
2493 break;
2494 case TCP_NOOPT:
2495 optval = tp->t_flags & TF_NOOPT;
2496 INP_WUNLOCK(inp);
2497 error = sooptcopyout(sopt, &optval, sizeof optval);
2498 break;
2499 case TCP_NOPUSH:
2500 optval = tp->t_flags & TF_NOPUSH;
2501 INP_WUNLOCK(inp);
2502 error = sooptcopyout(sopt, &optval, sizeof optval);
2503 break;
2504 case TCP_INFO:
2505 tcp_fill_info(tp, &ti);
2506 INP_WUNLOCK(inp);
2507 error = sooptcopyout(sopt, &ti, sizeof ti);
2508 break;
2509 case TCP_STATS:
2510 {
2511 #ifdef STATS
2512 int nheld;
2513 TYPEOF_MEMBER(struct statsblob, flags) sbflags = 0;
2514
2515 error = 0;
2516 socklen_t outsbsz = sopt->sopt_valsize;
2517 if (tp->t_stats == NULL)
2518 error = ENOENT;
2519 else if (outsbsz >= tp->t_stats->cursz)
2520 outsbsz = tp->t_stats->cursz;
2521 else if (outsbsz >= sizeof(struct statsblob))
2522 outsbsz = sizeof(struct statsblob);
2523 else
2524 error = EINVAL;
2525 INP_WUNLOCK(inp);
2526 if (error)
2527 break;
2528
2529 sbp = sopt->sopt_val;
2530 nheld = atop(round_page(((vm_offset_t)sbp) +
2531 (vm_size_t)outsbsz) - trunc_page((vm_offset_t)sbp));
2532 vm_page_t ma[nheld];
2533 if (vm_fault_quick_hold_pages(
2534 &curproc->p_vmspace->vm_map, (vm_offset_t)sbp,
2535 outsbsz, VM_PROT_READ | VM_PROT_WRITE, ma,
2536 nheld) < 0) {
2537 error = EFAULT;
2538 break;
2539 }
2540
2541 if ((error = copyin_nofault(&(sbp->flags), &sbflags,
2542 SIZEOF_MEMBER(struct statsblob, flags))))
2543 goto unhold;
2544
2545 INP_WLOCK_RECHECK(inp);
2546 error = stats_blob_snapshot(&sbp, outsbsz, tp->t_stats,
2547 sbflags | SB_CLONE_USRDSTNOFAULT);
2548 INP_WUNLOCK(inp);
2549 sopt->sopt_valsize = outsbsz;
2550 unhold:
2551 vm_page_unhold_pages(ma, nheld);
2552 #else
2553 INP_WUNLOCK(inp);
2554 error = EOPNOTSUPP;
2555 #endif /* !STATS */
2556 break;
2557 }
2558 case TCP_CONGESTION:
2559 len = strlcpy(buf, CC_ALGO(tp)->name, TCP_CA_NAME_MAX);
2560 INP_WUNLOCK(inp);
2561 error = sooptcopyout(sopt, buf, len + 1);
2562 break;
2563 case TCP_KEEPIDLE:
2564 case TCP_KEEPINTVL:
2565 case TCP_KEEPINIT:
2566 case TCP_KEEPCNT:
2567 switch (sopt->sopt_name) {
2568 case TCP_KEEPIDLE:
2569 ui = TP_KEEPIDLE(tp) / hz;
2570 break;
2571 case TCP_KEEPINTVL:
2572 ui = TP_KEEPINTVL(tp) / hz;
2573 break;
2574 case TCP_KEEPINIT:
2575 ui = TP_KEEPINIT(tp) / hz;
2576 break;
2577 case TCP_KEEPCNT:
2578 ui = TP_KEEPCNT(tp);
2579 break;
2580 }
2581 INP_WUNLOCK(inp);
2582 error = sooptcopyout(sopt, &ui, sizeof(ui));
2583 break;
2584 #ifdef TCPPCAP
2585 case TCP_PCAP_OUT:
2586 case TCP_PCAP_IN:
2587 optval = tcp_pcap_get_sock_max(TCP_PCAP_OUT ?
2588 &(tp->t_outpkts) : &(tp->t_inpkts));
2589 INP_WUNLOCK(inp);
2590 error = sooptcopyout(sopt, &optval, sizeof optval);
2591 break;
2592 #endif
2593 case TCP_FASTOPEN:
2594 optval = tp->t_flags & TF_FASTOPEN;
2595 INP_WUNLOCK(inp);
2596 error = sooptcopyout(sopt, &optval, sizeof optval);
2597 break;
2598 #ifdef TCP_BLACKBOX
2599 case TCP_LOG:
2600 optval = tp->t_logstate;
2601 INP_WUNLOCK(inp);
2602 error = sooptcopyout(sopt, &optval, sizeof(optval));
2603 break;
2604 case TCP_LOGBUF:
2605 /* tcp_log_getlogbuf() does INP_WUNLOCK(inp) */
2606 error = tcp_log_getlogbuf(sopt, tp);
2607 break;
2608 case TCP_LOGID:
2609 len = tcp_log_get_id(tp, buf);
2610 INP_WUNLOCK(inp);
2611 error = sooptcopyout(sopt, buf, len + 1);
2612 break;
2613 case TCP_LOGDUMP:
2614 case TCP_LOGDUMPID:
2615 INP_WUNLOCK(inp);
2616 error = EINVAL;
2617 break;
2618 #endif
2619 #ifdef KERN_TLS
2620 case TCP_TXTLS_MODE:
2621 optval = ktls_get_tx_mode(so);
2622 INP_WUNLOCK(inp);
2623 error = sooptcopyout(sopt, &optval, sizeof(optval));
2624 break;
2625 case TCP_RXTLS_MODE:
2626 optval = ktls_get_rx_mode(so);
2627 INP_WUNLOCK(inp);
2628 error = sooptcopyout(sopt, &optval, sizeof(optval));
2629 break;
2630 #endif
2631 default:
2632 INP_WUNLOCK(inp);
2633 error = ENOPROTOOPT;
2634 break;
2635 }
2636 break;
2637 }
2638 return (error);
2639 }
2640 #undef INP_WLOCK_RECHECK
2641 #undef INP_WLOCK_RECHECK_CLEANUP
2642
2643 /*
2644 * Initiate (or continue) disconnect.
2645 * If embryonic state, just send reset (once).
2646 * If in ``let data drain'' option and linger null, just drop.
2647 * Otherwise (hard), mark socket disconnecting and drop
2648 * current input data; switch states based on user close, and
2649 * send segment to peer (with FIN).
2650 */
2651 static void
tcp_disconnect(struct tcpcb * tp)2652 tcp_disconnect(struct tcpcb *tp)
2653 {
2654 struct inpcb *inp = tp->t_inpcb;
2655 struct socket *so = inp->inp_socket;
2656
2657 NET_EPOCH_ASSERT();
2658 INP_WLOCK_ASSERT(inp);
2659
2660 /*
2661 * Neither tcp_close() nor tcp_drop() should return NULL, as the
2662 * socket is still open.
2663 */
2664 if (tp->t_state < TCPS_ESTABLISHED &&
2665 !(tp->t_state > TCPS_LISTEN && IS_FASTOPEN(tp->t_flags))) {
2666 tp = tcp_close(tp);
2667 KASSERT(tp != NULL,
2668 ("tcp_disconnect: tcp_close() returned NULL"));
2669 } else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
2670 tp = tcp_drop(tp, 0);
2671 KASSERT(tp != NULL,
2672 ("tcp_disconnect: tcp_drop() returned NULL"));
2673 } else {
2674 soisdisconnecting(so);
2675 sbflush(&so->so_rcv);
2676 tcp_usrclosed(tp);
2677 if (!(inp->inp_flags & INP_DROPPED))
2678 tp->t_fb->tfb_tcp_output(tp);
2679 }
2680 }
2681
2682 /*
2683 * User issued close, and wish to trail through shutdown states:
2684 * if never received SYN, just forget it. If got a SYN from peer,
2685 * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
2686 * If already got a FIN from peer, then almost done; go to LAST_ACK
2687 * state. In all other cases, have already sent FIN to peer (e.g.
2688 * after PRU_SHUTDOWN), and just have to play tedious game waiting
2689 * for peer to send FIN or not respond to keep-alives, etc.
2690 * We can let the user exit from the close as soon as the FIN is acked.
2691 */
2692 static void
tcp_usrclosed(struct tcpcb * tp)2693 tcp_usrclosed(struct tcpcb *tp)
2694 {
2695
2696 NET_EPOCH_ASSERT();
2697 INP_WLOCK_ASSERT(tp->t_inpcb);
2698
2699 switch (tp->t_state) {
2700 case TCPS_LISTEN:
2701 #ifdef TCP_OFFLOAD
2702 tcp_offload_listen_stop(tp);
2703 #endif
2704 tcp_state_change(tp, TCPS_CLOSED);
2705 /* FALLTHROUGH */
2706 case TCPS_CLOSED:
2707 tp = tcp_close(tp);
2708 /*
2709 * tcp_close() should never return NULL here as the socket is
2710 * still open.
2711 */
2712 KASSERT(tp != NULL,
2713 ("tcp_usrclosed: tcp_close() returned NULL"));
2714 break;
2715
2716 case TCPS_SYN_SENT:
2717 case TCPS_SYN_RECEIVED:
2718 tp->t_flags |= TF_NEEDFIN;
2719 break;
2720
2721 case TCPS_ESTABLISHED:
2722 tcp_state_change(tp, TCPS_FIN_WAIT_1);
2723 break;
2724
2725 case TCPS_CLOSE_WAIT:
2726 tcp_state_change(tp, TCPS_LAST_ACK);
2727 break;
2728 }
2729 if (tp->t_state >= TCPS_FIN_WAIT_2) {
2730 soisdisconnected(tp->t_inpcb->inp_socket);
2731 /* Prevent the connection hanging in FIN_WAIT_2 forever. */
2732 if (tp->t_state == TCPS_FIN_WAIT_2) {
2733 int timeout;
2734
2735 timeout = (tcp_fast_finwait2_recycle) ?
2736 tcp_finwait2_timeout : TP_MAXIDLE(tp);
2737 tcp_timer_activate(tp, TT_2MSL, timeout);
2738 }
2739 }
2740 }
2741
2742 #ifdef DDB
2743 static void
db_print_indent(int indent)2744 db_print_indent(int indent)
2745 {
2746 int i;
2747
2748 for (i = 0; i < indent; i++)
2749 db_printf(" ");
2750 }
2751
2752 static void
db_print_tstate(int t_state)2753 db_print_tstate(int t_state)
2754 {
2755
2756 switch (t_state) {
2757 case TCPS_CLOSED:
2758 db_printf("TCPS_CLOSED");
2759 return;
2760
2761 case TCPS_LISTEN:
2762 db_printf("TCPS_LISTEN");
2763 return;
2764
2765 case TCPS_SYN_SENT:
2766 db_printf("TCPS_SYN_SENT");
2767 return;
2768
2769 case TCPS_SYN_RECEIVED:
2770 db_printf("TCPS_SYN_RECEIVED");
2771 return;
2772
2773 case TCPS_ESTABLISHED:
2774 db_printf("TCPS_ESTABLISHED");
2775 return;
2776
2777 case TCPS_CLOSE_WAIT:
2778 db_printf("TCPS_CLOSE_WAIT");
2779 return;
2780
2781 case TCPS_FIN_WAIT_1:
2782 db_printf("TCPS_FIN_WAIT_1");
2783 return;
2784
2785 case TCPS_CLOSING:
2786 db_printf("TCPS_CLOSING");
2787 return;
2788
2789 case TCPS_LAST_ACK:
2790 db_printf("TCPS_LAST_ACK");
2791 return;
2792
2793 case TCPS_FIN_WAIT_2:
2794 db_printf("TCPS_FIN_WAIT_2");
2795 return;
2796
2797 case TCPS_TIME_WAIT:
2798 db_printf("TCPS_TIME_WAIT");
2799 return;
2800
2801 default:
2802 db_printf("unknown");
2803 return;
2804 }
2805 }
2806
2807 static void
db_print_tflags(u_int t_flags)2808 db_print_tflags(u_int t_flags)
2809 {
2810 int comma;
2811
2812 comma = 0;
2813 if (t_flags & TF_ACKNOW) {
2814 db_printf("%sTF_ACKNOW", comma ? ", " : "");
2815 comma = 1;
2816 }
2817 if (t_flags & TF_DELACK) {
2818 db_printf("%sTF_DELACK", comma ? ", " : "");
2819 comma = 1;
2820 }
2821 if (t_flags & TF_NODELAY) {
2822 db_printf("%sTF_NODELAY", comma ? ", " : "");
2823 comma = 1;
2824 }
2825 if (t_flags & TF_NOOPT) {
2826 db_printf("%sTF_NOOPT", comma ? ", " : "");
2827 comma = 1;
2828 }
2829 if (t_flags & TF_SENTFIN) {
2830 db_printf("%sTF_SENTFIN", comma ? ", " : "");
2831 comma = 1;
2832 }
2833 if (t_flags & TF_REQ_SCALE) {
2834 db_printf("%sTF_REQ_SCALE", comma ? ", " : "");
2835 comma = 1;
2836 }
2837 if (t_flags & TF_RCVD_SCALE) {
2838 db_printf("%sTF_RECVD_SCALE", comma ? ", " : "");
2839 comma = 1;
2840 }
2841 if (t_flags & TF_REQ_TSTMP) {
2842 db_printf("%sTF_REQ_TSTMP", comma ? ", " : "");
2843 comma = 1;
2844 }
2845 if (t_flags & TF_RCVD_TSTMP) {
2846 db_printf("%sTF_RCVD_TSTMP", comma ? ", " : "");
2847 comma = 1;
2848 }
2849 if (t_flags & TF_SACK_PERMIT) {
2850 db_printf("%sTF_SACK_PERMIT", comma ? ", " : "");
2851 comma = 1;
2852 }
2853 if (t_flags & TF_NEEDSYN) {
2854 db_printf("%sTF_NEEDSYN", comma ? ", " : "");
2855 comma = 1;
2856 }
2857 if (t_flags & TF_NEEDFIN) {
2858 db_printf("%sTF_NEEDFIN", comma ? ", " : "");
2859 comma = 1;
2860 }
2861 if (t_flags & TF_NOPUSH) {
2862 db_printf("%sTF_NOPUSH", comma ? ", " : "");
2863 comma = 1;
2864 }
2865 if (t_flags & TF_MORETOCOME) {
2866 db_printf("%sTF_MORETOCOME", comma ? ", " : "");
2867 comma = 1;
2868 }
2869 if (t_flags & TF_LQ_OVERFLOW) {
2870 db_printf("%sTF_LQ_OVERFLOW", comma ? ", " : "");
2871 comma = 1;
2872 }
2873 if (t_flags & TF_LASTIDLE) {
2874 db_printf("%sTF_LASTIDLE", comma ? ", " : "");
2875 comma = 1;
2876 }
2877 if (t_flags & TF_RXWIN0SENT) {
2878 db_printf("%sTF_RXWIN0SENT", comma ? ", " : "");
2879 comma = 1;
2880 }
2881 if (t_flags & TF_FASTRECOVERY) {
2882 db_printf("%sTF_FASTRECOVERY", comma ? ", " : "");
2883 comma = 1;
2884 }
2885 if (t_flags & TF_CONGRECOVERY) {
2886 db_printf("%sTF_CONGRECOVERY", comma ? ", " : "");
2887 comma = 1;
2888 }
2889 if (t_flags & TF_WASFRECOVERY) {
2890 db_printf("%sTF_WASFRECOVERY", comma ? ", " : "");
2891 comma = 1;
2892 }
2893 if (t_flags & TF_SIGNATURE) {
2894 db_printf("%sTF_SIGNATURE", comma ? ", " : "");
2895 comma = 1;
2896 }
2897 if (t_flags & TF_FORCEDATA) {
2898 db_printf("%sTF_FORCEDATA", comma ? ", " : "");
2899 comma = 1;
2900 }
2901 if (t_flags & TF_TSO) {
2902 db_printf("%sTF_TSO", comma ? ", " : "");
2903 comma = 1;
2904 }
2905 if (t_flags & TF_FASTOPEN) {
2906 db_printf("%sTF_FASTOPEN", comma ? ", " : "");
2907 comma = 1;
2908 }
2909 }
2910
2911 static void
db_print_tflags2(u_int t_flags2)2912 db_print_tflags2(u_int t_flags2)
2913 {
2914 int comma;
2915
2916 comma = 0;
2917 if (t_flags2 & TF2_ECN_PERMIT) {
2918 db_printf("%sTF2_ECN_PERMIT", comma ? ", " : "");
2919 comma = 1;
2920 }
2921 }
2922
2923 static void
db_print_toobflags(char t_oobflags)2924 db_print_toobflags(char t_oobflags)
2925 {
2926 int comma;
2927
2928 comma = 0;
2929 if (t_oobflags & TCPOOB_HAVEDATA) {
2930 db_printf("%sTCPOOB_HAVEDATA", comma ? ", " : "");
2931 comma = 1;
2932 }
2933 if (t_oobflags & TCPOOB_HADDATA) {
2934 db_printf("%sTCPOOB_HADDATA", comma ? ", " : "");
2935 comma = 1;
2936 }
2937 }
2938
2939 static void
db_print_tcpcb(struct tcpcb * tp,const char * name,int indent)2940 db_print_tcpcb(struct tcpcb *tp, const char *name, int indent)
2941 {
2942
2943 db_print_indent(indent);
2944 db_printf("%s at %p\n", name, tp);
2945
2946 indent += 2;
2947
2948 db_print_indent(indent);
2949 db_printf("t_segq first: %p t_segqlen: %d t_dupacks: %d\n",
2950 TAILQ_FIRST(&tp->t_segq), tp->t_segqlen, tp->t_dupacks);
2951
2952 db_print_indent(indent);
2953 db_printf("tt_rexmt: %p tt_persist: %p tt_keep: %p\n",
2954 &tp->t_timers->tt_rexmt, &tp->t_timers->tt_persist, &tp->t_timers->tt_keep);
2955
2956 db_print_indent(indent);
2957 db_printf("tt_2msl: %p tt_delack: %p t_inpcb: %p\n", &tp->t_timers->tt_2msl,
2958 &tp->t_timers->tt_delack, tp->t_inpcb);
2959
2960 db_print_indent(indent);
2961 db_printf("t_state: %d (", tp->t_state);
2962 db_print_tstate(tp->t_state);
2963 db_printf(")\n");
2964
2965 db_print_indent(indent);
2966 db_printf("t_flags: 0x%x (", tp->t_flags);
2967 db_print_tflags(tp->t_flags);
2968 db_printf(")\n");
2969
2970 db_print_indent(indent);
2971 db_printf("t_flags2: 0x%x (", tp->t_flags2);
2972 db_print_tflags2(tp->t_flags2);
2973 db_printf(")\n");
2974
2975 db_print_indent(indent);
2976 db_printf("snd_una: 0x%08x snd_max: 0x%08x snd_nxt: x0%08x\n",
2977 tp->snd_una, tp->snd_max, tp->snd_nxt);
2978
2979 db_print_indent(indent);
2980 db_printf("snd_up: 0x%08x snd_wl1: 0x%08x snd_wl2: 0x%08x\n",
2981 tp->snd_up, tp->snd_wl1, tp->snd_wl2);
2982
2983 db_print_indent(indent);
2984 db_printf("iss: 0x%08x irs: 0x%08x rcv_nxt: 0x%08x\n",
2985 tp->iss, tp->irs, tp->rcv_nxt);
2986
2987 db_print_indent(indent);
2988 db_printf("rcv_adv: 0x%08x rcv_wnd: %u rcv_up: 0x%08x\n",
2989 tp->rcv_adv, tp->rcv_wnd, tp->rcv_up);
2990
2991 db_print_indent(indent);
2992 db_printf("snd_wnd: %u snd_cwnd: %u\n",
2993 tp->snd_wnd, tp->snd_cwnd);
2994
2995 db_print_indent(indent);
2996 db_printf("snd_ssthresh: %u snd_recover: "
2997 "0x%08x\n", tp->snd_ssthresh, tp->snd_recover);
2998
2999 db_print_indent(indent);
3000 db_printf("t_rcvtime: %u t_startime: %u\n",
3001 tp->t_rcvtime, tp->t_starttime);
3002
3003 db_print_indent(indent);
3004 db_printf("t_rttime: %u t_rtsq: 0x%08x\n",
3005 tp->t_rtttime, tp->t_rtseq);
3006
3007 db_print_indent(indent);
3008 db_printf("t_rxtcur: %d t_maxseg: %u t_srtt: %d\n",
3009 tp->t_rxtcur, tp->t_maxseg, tp->t_srtt);
3010
3011 db_print_indent(indent);
3012 db_printf("t_rttvar: %d t_rxtshift: %d t_rttmin: %u "
3013 "t_rttbest: %u\n", tp->t_rttvar, tp->t_rxtshift, tp->t_rttmin,
3014 tp->t_rttbest);
3015
3016 db_print_indent(indent);
3017 db_printf("t_rttupdated: %lu max_sndwnd: %u t_softerror: %d\n",
3018 tp->t_rttupdated, tp->max_sndwnd, tp->t_softerror);
3019
3020 db_print_indent(indent);
3021 db_printf("t_oobflags: 0x%x (", tp->t_oobflags);
3022 db_print_toobflags(tp->t_oobflags);
3023 db_printf(") t_iobc: 0x%02x\n", tp->t_iobc);
3024
3025 db_print_indent(indent);
3026 db_printf("snd_scale: %u rcv_scale: %u request_r_scale: %u\n",
3027 tp->snd_scale, tp->rcv_scale, tp->request_r_scale);
3028
3029 db_print_indent(indent);
3030 db_printf("ts_recent: %u ts_recent_age: %u\n",
3031 tp->ts_recent, tp->ts_recent_age);
3032
3033 db_print_indent(indent);
3034 db_printf("ts_offset: %u last_ack_sent: 0x%08x snd_cwnd_prev: "
3035 "%u\n", tp->ts_offset, tp->last_ack_sent, tp->snd_cwnd_prev);
3036
3037 db_print_indent(indent);
3038 db_printf("snd_ssthresh_prev: %u snd_recover_prev: 0x%08x "
3039 "t_badrxtwin: %u\n", tp->snd_ssthresh_prev,
3040 tp->snd_recover_prev, tp->t_badrxtwin);
3041
3042 db_print_indent(indent);
3043 db_printf("snd_numholes: %d snd_holes first: %p\n",
3044 tp->snd_numholes, TAILQ_FIRST(&tp->snd_holes));
3045
3046 db_print_indent(indent);
3047 db_printf("snd_fack: 0x%08x rcv_numsacks: %d\n",
3048 tp->snd_fack, tp->rcv_numsacks);
3049
3050 /* Skip sackblks, sackhint. */
3051
3052 db_print_indent(indent);
3053 db_printf("t_rttlow: %d rfbuf_ts: %u rfbuf_cnt: %d\n",
3054 tp->t_rttlow, tp->rfbuf_ts, tp->rfbuf_cnt);
3055 }
3056
DB_SHOW_COMMAND(tcpcb,db_show_tcpcb)3057 DB_SHOW_COMMAND(tcpcb, db_show_tcpcb)
3058 {
3059 struct tcpcb *tp;
3060
3061 if (!have_addr) {
3062 db_printf("usage: show tcpcb <addr>\n");
3063 return;
3064 }
3065 tp = (struct tcpcb *)addr;
3066
3067 db_print_tcpcb(tp, "tcpcb", 0);
3068 }
3069 #endif
3070