xref: /freebsd/sys/ofed/drivers/infiniband/ulp/sdp/sdp_main.c (revision 296a4cb5c5b18f82da7a365d9f209cb9fc09003b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993, 1995
5  *      The Regents of the University of California.  All rights reserved.
6  * Copyright (c) 2004 The FreeBSD Foundation.  All rights reserved.
7  * Copyright (c) 2004-2008 Robert N. M. Watson.  All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  * Excerpts taken from tcp_subr.c, tcp_usrreq.c, uipc_socket.c
34  */
35 
36 /*
37  *
38  * Copyright (c) 2010 Isilon Systems, Inc.
39  * Copyright (c) 2010 iX Systems, Inc.
40  * Copyright (c) 2010 Panasas, Inc.
41  * All rights reserved.
42  *
43  * Redistribution and use in source and binary forms, with or without
44  * modification, are permitted provided that the following conditions
45  * are met:
46  * 1. Redistributions of source code must retain the above copyright
47  *    notice unmodified, this list of conditions, and the following
48  *    disclaimer.
49  * 2. Redistributions in binary form must reproduce the above copyright
50  *    notice, this list of conditions and the following disclaimer in the
51  *    documentation and/or other materials provided with the distribution.
52  *
53  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
54  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
55  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
56  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
57  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
58  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
59  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
60  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
61  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
62  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
63  *
64  */
65 
66 #include <sys/param.h>
67 #include <sys/eventhandler.h>
68 #include <sys/kernel.h>
69 #include <sys/malloc.h>
70 
71 #include "sdp.h"
72 
73 #include <net/if.h>
74 #include <net/route.h>
75 #include <net/vnet.h>
76 #include <sys/sysctl.h>
77 
78 uma_zone_t	sdp_zone;
79 struct rwlock	sdp_lock;
80 LIST_HEAD(, sdp_sock) sdp_list;
81 
82 struct workqueue_struct *rx_comp_wq;
83 
84 RW_SYSINIT(sdplockinit, &sdp_lock, "SDP lock");
85 #define	SDP_LIST_WLOCK()	rw_wlock(&sdp_lock)
86 #define	SDP_LIST_RLOCK()	rw_rlock(&sdp_lock)
87 #define	SDP_LIST_WUNLOCK()	rw_wunlock(&sdp_lock)
88 #define	SDP_LIST_RUNLOCK()	rw_runlock(&sdp_lock)
89 #define	SDP_LIST_WLOCK_ASSERT()	rw_assert(&sdp_lock, RW_WLOCKED)
90 #define	SDP_LIST_RLOCK_ASSERT()	rw_assert(&sdp_lock, RW_RLOCKED)
91 #define	SDP_LIST_LOCK_ASSERT()	rw_assert(&sdp_lock, RW_LOCKED)
92 
93 MALLOC_DEFINE(M_SDP, "sdp", "Sockets Direct Protocol");
94 
95 static void sdp_stop_keepalive_timer(struct socket *so);
96 
97 /*
98  * SDP protocol interface to socket abstraction.
99  */
100 /*
101  * sdp_sendspace and sdp_recvspace are the default send and receive window
102  * sizes, respectively.
103  */
104 u_long	sdp_sendspace = 1024*32;
105 u_long	sdp_recvspace = 1024*64;
106 
107 static int sdp_count;
108 
109 /*
110  * Disable async. CMA events for sockets which are being torn down.
111  */
112 static void
sdp_destroy_cma(struct sdp_sock * ssk)113 sdp_destroy_cma(struct sdp_sock *ssk)
114 {
115 
116 	if (ssk->id == NULL)
117 		return;
118 	rdma_destroy_id(ssk->id);
119 	ssk->id = NULL;
120 }
121 
122 static int
sdp_pcbbind(struct sdp_sock * ssk,struct sockaddr * nam,struct ucred * cred)123 sdp_pcbbind(struct sdp_sock *ssk, struct sockaddr *nam, struct ucred *cred)
124 {
125 	struct sockaddr_in *sin;
126 	struct sockaddr_in null;
127 	int error;
128 
129 	SDP_WLOCK_ASSERT(ssk);
130 
131 	if (ssk->lport != 0 || ssk->laddr != INADDR_ANY)
132 		return (EINVAL);
133 	/* rdma_bind_addr handles bind races.  */
134 	SDP_WUNLOCK(ssk);
135 	if (ssk->id == NULL)
136 		ssk->id = rdma_create_id(&init_net, sdp_cma_handler, ssk, RDMA_PS_SDP, IB_QPT_RC);
137 	if (ssk->id == NULL) {
138 		SDP_WLOCK(ssk);
139 		return (ENOMEM);
140 	}
141 	if (nam == NULL) {
142 		null.sin_family = AF_INET;
143 		null.sin_len = sizeof(null);
144 		null.sin_addr.s_addr = INADDR_ANY;
145 		null.sin_port = 0;
146 		bzero(&null.sin_zero, sizeof(null.sin_zero));
147 		nam = (struct sockaddr *)&null;
148 	}
149 	error = -rdma_bind_addr(ssk->id, nam);
150 	SDP_WLOCK(ssk);
151 	if (error == 0) {
152 		sin = (struct sockaddr_in *)&ssk->id->route.addr.src_addr;
153 		ssk->laddr = sin->sin_addr.s_addr;
154 		ssk->lport = sin->sin_port;
155 	} else
156 		sdp_destroy_cma(ssk);
157 	return (error);
158 }
159 
160 static void
sdp_pcbfree(struct sdp_sock * ssk)161 sdp_pcbfree(struct sdp_sock *ssk)
162 {
163 
164 	KASSERT(ssk->socket == NULL, ("ssk %p socket still attached", ssk));
165 	KASSERT((ssk->flags & SDP_DESTROY) == 0,
166 	    ("ssk %p already destroyed", ssk));
167 
168 	sdp_dbg(ssk->socket, "Freeing pcb");
169 	SDP_WLOCK_ASSERT(ssk);
170 	ssk->flags |= SDP_DESTROY;
171 	SDP_WUNLOCK(ssk);
172 	SDP_LIST_WLOCK();
173 	sdp_count--;
174 	LIST_REMOVE(ssk, list);
175 	SDP_LIST_WUNLOCK();
176 	crfree(ssk->cred);
177 	ssk->qp_active = 0;
178 	if (ssk->qp) {
179 		ib_destroy_qp(ssk->qp);
180 		ssk->qp = NULL;
181 	}
182 	sdp_tx_ring_destroy(ssk);
183 	sdp_rx_ring_destroy(ssk);
184 	sdp_destroy_cma(ssk);
185 	rw_destroy(&ssk->rx_ring.destroyed_lock);
186 	rw_destroy(&ssk->lock);
187 	uma_zfree(sdp_zone, ssk);
188 }
189 
190 static int
sdp_getsockaddr(struct socket * so,struct sockaddr * sa)191 sdp_getsockaddr(struct socket *so, struct sockaddr *sa)
192 {
193 	struct sdp_sock *ssk = sdp_sk(so);
194 
195 	SDP_RLOCK(ssk);
196 	*(struct sockaddr_in *)sa = (struct sockaddr_in ){
197 		.sin_family = AF_INET,
198 		.sin_len = sizeof(struct sockaddr_in),
199 		.sin_addr.s_addr = ssk->laddr,
200 		.sin_port = ssk->lport,
201 	};
202 	SDP_RUNLOCK(ssk);
203 
204 	return (0);
205 }
206 
207 static int
sdp_getpeeraddr(struct socket * so,struct sockaddr * sa)208 sdp_getpeeraddr(struct socket *so, struct sockaddr *sa)
209 {
210 	struct sdp_sock *ssk = sdp_sk(so);
211 
212 	SDP_RLOCK(ssk);
213 	*(struct sockaddr_in *)sa = (struct sockaddr_in ){
214 		.sin_family = AF_INET,
215 		.sin_len = sizeof(struct sockaddr_in),
216 		.sin_addr.s_addr = ssk->faddr,
217 		.sin_port = ssk->fport,
218 	};
219 	SDP_RUNLOCK(ssk);
220 
221 	return (0);
222 }
223 
224 #if 0
225 static void
226 sdp_apply_all(void (*func)(struct sdp_sock *, void *), void *arg)
227 {
228 	struct sdp_sock *ssk;
229 
230 	SDP_LIST_RLOCK();
231 	LIST_FOREACH(ssk, &sdp_list, list) {
232 		SDP_WLOCK(ssk);
233 		func(ssk, arg);
234 		SDP_WUNLOCK(ssk);
235 	}
236 	SDP_LIST_RUNLOCK();
237 }
238 #endif
239 
240 static void
sdp_output_reset(struct sdp_sock * ssk)241 sdp_output_reset(struct sdp_sock *ssk)
242 {
243 	struct rdma_cm_id *id;
244 
245 	SDP_WLOCK_ASSERT(ssk);
246 	if (ssk->id) {
247 		id = ssk->id;
248 		ssk->qp_active = 0;
249 		SDP_WUNLOCK(ssk);
250 		rdma_disconnect(id);
251 		SDP_WLOCK(ssk);
252 	}
253 	ssk->state = TCPS_CLOSED;
254 }
255 
256 /*
257  * Attempt to close a SDP socket, marking it as dropped, and freeing
258  * the socket if we hold the only reference.
259  */
260 static struct sdp_sock *
sdp_closed(struct sdp_sock * ssk)261 sdp_closed(struct sdp_sock *ssk)
262 {
263 	struct socket *so;
264 
265 	SDP_WLOCK_ASSERT(ssk);
266 
267 	ssk->flags |= SDP_DROPPED;
268 	so = ssk->socket;
269 	soisdisconnected(so);
270 	if (ssk->flags & SDP_SOCKREF) {
271 		ssk->flags &= ~SDP_SOCKREF;
272 		SDP_WUNLOCK(ssk);
273 		sorele(so);
274 		return (NULL);
275 	}
276 	return (ssk);
277 }
278 
279 /*
280  * Perform timer based shutdowns which can not operate in
281  * callout context.
282  */
283 static void
sdp_shutdown_task(void * data,int pending)284 sdp_shutdown_task(void *data, int pending)
285 {
286 	struct sdp_sock *ssk;
287 
288 	ssk = data;
289 	SDP_WLOCK(ssk);
290 	/*
291 	 * I don't think this can race with another call to pcbfree()
292 	 * because SDP_TIMEWAIT protects it.  SDP_DESTROY may be redundant.
293 	 */
294 	if (ssk->flags & SDP_DESTROY)
295 		panic("sdp_shutdown_task: Racing with pcbfree for ssk %p",
296 		    ssk);
297 	if (ssk->flags & SDP_DISCON)
298 		sdp_output_reset(ssk);
299 	/* We have to clear this so sdp_detach() will call pcbfree(). */
300 	ssk->flags &= ~(SDP_TIMEWAIT | SDP_DREQWAIT);
301 	if ((ssk->flags & SDP_DROPPED) == 0 &&
302 	    sdp_closed(ssk) == NULL)
303 		return;
304 	if (ssk->socket == NULL) {
305 		sdp_pcbfree(ssk);
306 		return;
307 	}
308 	SDP_WUNLOCK(ssk);
309 }
310 
311 /*
312  * 2msl has expired, schedule the shutdown task.
313  */
314 static void
sdp_2msl_timeout(void * data)315 sdp_2msl_timeout(void *data)
316 {
317 	struct sdp_sock *ssk;
318 
319 	ssk = data;
320 	/* Callout canceled. */
321         if (!callout_active(&ssk->keep2msl))
322 		goto out;
323         callout_deactivate(&ssk->keep2msl);
324 	/* Should be impossible, defensive programming. */
325 	if ((ssk->flags & SDP_TIMEWAIT) == 0)
326 		goto out;
327 	taskqueue_enqueue(taskqueue_thread, &ssk->shutdown_task);
328 out:
329 	SDP_WUNLOCK(ssk);
330 	return;
331 }
332 
333 /*
334  * Schedule the 2msl wait timer.
335  */
336 static void
sdp_2msl_wait(struct sdp_sock * ssk)337 sdp_2msl_wait(struct sdp_sock *ssk)
338 {
339 
340 	SDP_WLOCK_ASSERT(ssk);
341 	ssk->flags |= SDP_TIMEWAIT;
342 	ssk->state = TCPS_TIME_WAIT;
343 	soisdisconnected(ssk->socket);
344 	callout_reset(&ssk->keep2msl, TCPTV_MSL, sdp_2msl_timeout, ssk);
345 }
346 
347 /*
348  * Timed out waiting for the final fin/ack from rdma_disconnect().
349  */
350 static void
sdp_dreq_timeout(void * data)351 sdp_dreq_timeout(void *data)
352 {
353 	struct sdp_sock *ssk;
354 
355 	ssk = data;
356 	/* Callout canceled. */
357         if (!callout_active(&ssk->keep2msl))
358 		goto out;
359 	/* Callout rescheduled, probably as a different timer. */
360 	if (callout_pending(&ssk->keep2msl))
361 		goto out;
362         callout_deactivate(&ssk->keep2msl);
363 	if (ssk->state != TCPS_FIN_WAIT_1 && ssk->state != TCPS_LAST_ACK)
364 		goto out;
365 	if ((ssk->flags & SDP_DREQWAIT) == 0)
366 		goto out;
367 	ssk->flags &= ~SDP_DREQWAIT;
368 	ssk->flags |= SDP_DISCON;
369 	sdp_2msl_wait(ssk);
370 	ssk->qp_active = 0;
371 out:
372 	SDP_WUNLOCK(ssk);
373 }
374 
375 /*
376  * Received the final fin/ack.  Cancel the 2msl.
377  */
378 void
sdp_cancel_dreq_wait_timeout(struct sdp_sock * ssk)379 sdp_cancel_dreq_wait_timeout(struct sdp_sock *ssk)
380 {
381 	sdp_dbg(ssk->socket, "cancelling dreq wait timeout\n");
382 	ssk->flags &= ~SDP_DREQWAIT;
383 	sdp_2msl_wait(ssk);
384 }
385 
386 static int
sdp_init_sock(struct socket * sk)387 sdp_init_sock(struct socket *sk)
388 {
389 	struct sdp_sock *ssk = sdp_sk(sk);
390 
391 	sdp_dbg(sk, "%s\n", __func__);
392 
393 	callout_init_rw(&ssk->keep2msl, &ssk->lock, CALLOUT_RETURNUNLOCKED);
394 	TASK_INIT(&ssk->shutdown_task, 0, sdp_shutdown_task, ssk);
395 #ifdef SDP_ZCOPY
396 	INIT_DELAYED_WORK(&ssk->srcavail_cancel_work, srcavail_cancel_timeout);
397 	ssk->zcopy_thresh = -1; /* use global sdp_zcopy_thresh */
398 	ssk->tx_ring.rdma_inflight = NULL;
399 #endif
400 	atomic_set(&ssk->mseq_ack, 0);
401 	sdp_rx_ring_init(ssk);
402 	ssk->tx_ring.buffer = NULL;
403 
404 	return 0;
405 }
406 
407 /*
408  * Allocate an sdp_sock for the socket and reserve socket buffer space.
409  */
410 static int
sdp_attach(struct socket * so,int proto,struct thread * td)411 sdp_attach(struct socket *so, int proto, struct thread *td)
412 {
413 	struct sdp_sock *ssk;
414 	int error;
415 
416 	ssk = sdp_sk(so);
417 	KASSERT(ssk == NULL, ("sdp_attach: ssk already set on so %p", so));
418 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
419 		error = soreserve(so, sdp_sendspace, sdp_recvspace);
420 		if (error)
421 			return (error);
422 	}
423 	so->so_rcv.sb_flags |= SB_AUTOSIZE;
424 	so->so_snd.sb_flags |= SB_AUTOSIZE;
425 	ssk = uma_zalloc(sdp_zone, M_NOWAIT | M_ZERO);
426 	if (ssk == NULL)
427 		return (ENOBUFS);
428 	rw_init(&ssk->lock, "sdpsock");
429 	ssk->socket = so;
430 	ssk->cred = crhold(so->so_cred);
431 	so->so_pcb = (caddr_t)ssk;
432 	sdp_init_sock(so);
433 	ssk->flags = 0;
434 	ssk->qp_active = 0;
435 	ssk->state = TCPS_CLOSED;
436 	mbufq_init(&ssk->rxctlq, INT_MAX);
437 	SDP_LIST_WLOCK();
438 	LIST_INSERT_HEAD(&sdp_list, ssk, list);
439 	sdp_count++;
440 	SDP_LIST_WUNLOCK();
441 
442 	return (0);
443 }
444 
445 /*
446  * Detach SDP from the socket, potentially leaving it around for the
447  * timewait to expire.
448  */
449 static void
sdp_detach(struct socket * so)450 sdp_detach(struct socket *so)
451 {
452 	struct sdp_sock *ssk;
453 
454 	ssk = sdp_sk(so);
455 	SDP_WLOCK(ssk);
456 	KASSERT(ssk->socket != NULL, ("sdp_detach: socket is NULL"));
457 	ssk->socket->so_pcb = NULL;
458 	ssk->socket = NULL;
459 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DREQWAIT))
460 		SDP_WUNLOCK(ssk);
461 	else if (ssk->flags & SDP_DROPPED || ssk->state < TCPS_SYN_SENT)
462 		sdp_pcbfree(ssk);
463 	else
464 		panic("sdp_detach: Unexpected state, ssk %p.\n", ssk);
465 }
466 
467 /*
468  * Allocate a local address for the socket.
469  */
470 static int
sdp_bind(struct socket * so,struct sockaddr * nam,struct thread * td)471 sdp_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
472 {
473 	int error = 0;
474 	struct sdp_sock *ssk;
475 	struct sockaddr_in *sin;
476 
477 	sin = (struct sockaddr_in *)nam;
478 	if (sin->sin_family != AF_INET)
479 		return (EAFNOSUPPORT);
480 	if (nam->sa_len != sizeof(*sin))
481 		return (EINVAL);
482 	if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
483 		return (EAFNOSUPPORT);
484 
485 	ssk = sdp_sk(so);
486 	SDP_WLOCK(ssk);
487 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
488 		error = EINVAL;
489 		goto out;
490 	}
491 	error = sdp_pcbbind(ssk, nam, td->td_ucred);
492 out:
493 	SDP_WUNLOCK(ssk);
494 
495 	return (error);
496 }
497 
498 /*
499  * Prepare to accept connections.
500  */
501 static int
sdp_listen(struct socket * so,int backlog,struct thread * td)502 sdp_listen(struct socket *so, int backlog, struct thread *td)
503 {
504 	int error = 0;
505 	struct sdp_sock *ssk;
506 
507 	ssk = sdp_sk(so);
508 	SDP_WLOCK(ssk);
509 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
510 		error = EINVAL;
511 		goto out;
512 	}
513 	if (error == 0 && ssk->lport == 0)
514 		error = sdp_pcbbind(ssk, (struct sockaddr *)0, td->td_ucred);
515 	SOCK_LOCK(so);
516 	if (error == 0)
517 		error = solisten_proto_check(so);
518 	if (error == 0) {
519 		solisten_proto(so, backlog);
520 		ssk->state = TCPS_LISTEN;
521 	}
522 	SOCK_UNLOCK(so);
523 
524 out:
525 	SDP_WUNLOCK(ssk);
526 	if (error == 0)
527 		error = -rdma_listen(ssk->id, backlog);
528 	return (error);
529 }
530 
531 /*
532  * Initiate a SDP connection to nam.
533  */
534 static int
sdp_start_connect(struct sdp_sock * ssk,struct sockaddr * nam,struct thread * td)535 sdp_start_connect(struct sdp_sock *ssk, struct sockaddr *nam, struct thread *td)
536 {
537 	struct sockaddr_in src;
538 	struct socket *so;
539 	int error;
540 
541 	so = ssk->socket;
542 
543 	SDP_WLOCK_ASSERT(ssk);
544 	if (ssk->lport == 0) {
545 		error = sdp_pcbbind(ssk, (struct sockaddr *)0, td->td_ucred);
546 		if (error)
547 			return error;
548 	}
549 	src.sin_family = AF_INET;
550 	src.sin_len = sizeof(src);
551 	bzero(&src.sin_zero, sizeof(src.sin_zero));
552 	src.sin_port = ssk->lport;
553 	src.sin_addr.s_addr = ssk->laddr;
554 	soisconnecting(so);
555 	SDP_WUNLOCK(ssk);
556 	error = -rdma_resolve_addr(ssk->id, (struct sockaddr *)&src, nam,
557 	    SDP_RESOLVE_TIMEOUT);
558 	SDP_WLOCK(ssk);
559 	if (error == 0)
560 		ssk->state = TCPS_SYN_SENT;
561 
562 	return 0;
563 }
564 
565 /*
566  * Initiate SDP connection.
567  */
568 static int
sdp_connect(struct socket * so,struct sockaddr * nam,struct thread * td)569 sdp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
570 {
571 	int error = 0;
572 	struct sdp_sock *ssk;
573 	struct sockaddr_in *sin;
574 
575 	sin = (struct sockaddr_in *)nam;
576 	if (nam->sa_len != sizeof(*sin))
577 		return (EINVAL);
578 	if (sin->sin_family != AF_INET)
579 		return (EAFNOSUPPORT);
580 	if (IN_MULTICAST(ntohl(sin->sin_addr.s_addr)))
581 		return (EAFNOSUPPORT);
582 	if ((error = prison_remote_ip4(td->td_ucred, &sin->sin_addr)) != 0)
583 		return (error);
584 	ssk = sdp_sk(so);
585 	SDP_WLOCK(ssk);
586 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED))
587 		error = EINVAL;
588 	else
589 		error = sdp_start_connect(ssk, nam, td);
590 	SDP_WUNLOCK(ssk);
591 	return (error);
592 }
593 
594 /*
595  * Drop a SDP socket, reporting
596  * the specified error.  If connection is synchronized,
597  * then send a RST to peer.
598  */
599 static struct sdp_sock *
sdp_drop(struct sdp_sock * ssk,int errno)600 sdp_drop(struct sdp_sock *ssk, int errno)
601 {
602 	struct socket *so;
603 
604 	SDP_WLOCK_ASSERT(ssk);
605 	so = ssk->socket;
606 	if (TCPS_HAVERCVDSYN(ssk->state))
607 		sdp_output_reset(ssk);
608 	if (errno == ETIMEDOUT && ssk->softerror)
609 		errno = ssk->softerror;
610 	so->so_error = errno;
611 	return (sdp_closed(ssk));
612 }
613 
614 /*
615  * User issued close, and wish to trail through shutdown states:
616  * if never received SYN, just forget it.  If got a SYN from peer,
617  * but haven't sent FIN, then go to FIN_WAIT_1 state to send peer a FIN.
618  * If already got a FIN from peer, then almost done; go to LAST_ACK
619  * state.  In all other cases, have already sent FIN to peer (e.g.
620  * after PRU_SHUTDOWN), and just have to play tedious game waiting
621  * for peer to send FIN or not respond to keep-alives, etc.
622  * We can let the user exit from the close as soon as the FIN is acked.
623  */
624 static void
sdp_usrclosed(struct sdp_sock * ssk)625 sdp_usrclosed(struct sdp_sock *ssk)
626 {
627 
628 	SDP_WLOCK_ASSERT(ssk);
629 
630 	switch (ssk->state) {
631 	case TCPS_LISTEN:
632 		ssk->state = TCPS_CLOSED;
633 		SDP_WUNLOCK(ssk);
634 		sdp_destroy_cma(ssk);
635 		SDP_WLOCK(ssk);
636 		/* FALLTHROUGH */
637 	case TCPS_CLOSED:
638 		ssk = sdp_closed(ssk);
639 		/*
640 		 * sdp_closed() should never return NULL here as the socket is
641 		 * still open.
642 		 */
643 		KASSERT(ssk != NULL,
644 		    ("sdp_usrclosed: sdp_closed() returned NULL"));
645 		break;
646 
647 	case TCPS_SYN_SENT:
648 		/* FALLTHROUGH */
649 	case TCPS_SYN_RECEIVED:
650 		ssk->flags |= SDP_NEEDFIN;
651 		break;
652 
653 	case TCPS_ESTABLISHED:
654 		ssk->flags |= SDP_NEEDFIN;
655 		ssk->state = TCPS_FIN_WAIT_1;
656 		break;
657 
658 	case TCPS_CLOSE_WAIT:
659 		ssk->state = TCPS_LAST_ACK;
660 		break;
661 	}
662 	if (ssk->state >= TCPS_FIN_WAIT_2) {
663 		/* Prevent the connection hanging in FIN_WAIT_2 forever. */
664 		if (ssk->state == TCPS_FIN_WAIT_2)
665 			sdp_2msl_wait(ssk);
666 		else
667 			soisdisconnected(ssk->socket);
668 	}
669 }
670 
671 static void
sdp_output_disconnect(struct sdp_sock * ssk)672 sdp_output_disconnect(struct sdp_sock *ssk)
673 {
674 
675 	SDP_WLOCK_ASSERT(ssk);
676 	callout_reset(&ssk->keep2msl, SDP_FIN_WAIT_TIMEOUT,
677 	    sdp_dreq_timeout, ssk);
678 	ssk->flags |= SDP_NEEDFIN | SDP_DREQWAIT;
679 	sdp_post_sends(ssk, M_NOWAIT);
680 }
681 
682 /*
683  * Initiate or continue a disconnect.
684  * If embryonic state, just send reset (once).
685  * If in ``let data drain'' option and linger null, just drop.
686  * Otherwise (hard), mark socket disconnecting and drop
687  * current input data; switch states based on user close, and
688  * send segment to peer (with FIN).
689  */
690 static void
sdp_start_disconnect(struct sdp_sock * ssk)691 sdp_start_disconnect(struct sdp_sock *ssk)
692 {
693 	struct socket *so;
694 	int unread;
695 
696 	so = ssk->socket;
697 	SDP_WLOCK_ASSERT(ssk);
698 	sdp_stop_keepalive_timer(so);
699 	/*
700 	 * Neither sdp_closed() nor sdp_drop() should return NULL, as the
701 	 * socket is still open.
702 	 */
703 	if (ssk->state < TCPS_ESTABLISHED) {
704 		ssk = sdp_closed(ssk);
705 		KASSERT(ssk != NULL,
706 		    ("sdp_start_disconnect: sdp_close() returned NULL"));
707 	} else if ((so->so_options & SO_LINGER) && so->so_linger == 0) {
708 		ssk = sdp_drop(ssk, 0);
709 		KASSERT(ssk != NULL,
710 		    ("sdp_start_disconnect: sdp_drop() returned NULL"));
711 	} else {
712 		soisdisconnecting(so);
713 		unread = sbused(&so->so_rcv);
714 		sbflush(&so->so_rcv);
715 		sdp_usrclosed(ssk);
716 		if (!(ssk->flags & SDP_DROPPED)) {
717 			if (unread)
718 				sdp_output_reset(ssk);
719 			else
720 				sdp_output_disconnect(ssk);
721 		}
722 	}
723 }
724 
725 /*
726  * User initiated disconnect.
727  */
728 static int
sdp_disconnect(struct socket * so)729 sdp_disconnect(struct socket *so)
730 {
731 	struct sdp_sock *ssk;
732 	int error = 0;
733 
734 	ssk = sdp_sk(so);
735 	SDP_WLOCK(ssk);
736 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
737 		error = ECONNRESET;
738 		goto out;
739 	}
740 	sdp_start_disconnect(ssk);
741 out:
742 	SDP_WUNLOCK(ssk);
743 	return (error);
744 }
745 
746 /*
747  * Accept a connection.  Essentially all the work is done at higher levels;
748  * just return the address of the peer, storing through addr.
749  *
750  *
751  * XXX This is broken XXX
752  *
753  * The rationale for acquiring the sdp lock here is somewhat complicated,
754  * and is described in detail in the commit log entry for r175612.  Acquiring
755  * it delays an accept(2) racing with sonewconn(), which inserts the socket
756  * before the address/port fields are initialized.  A better fix would
757  * prevent the socket from being placed in the listen queue until all fields
758  * are fully initialized.
759  */
760 static int
sdp_accept(struct socket * so,struct sockaddr * sa)761 sdp_accept(struct socket *so, struct sockaddr *sa)
762 {
763 	struct sdp_sock *ssk = NULL;
764 	int error;
765 
766 	if (so->so_state & SS_ISDISCONNECTED)
767 		return (ECONNABORTED);
768 
769 	error = 0;
770 	ssk = sdp_sk(so);
771 	SDP_WLOCK(ssk);
772 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED))
773 		error = ECONNABORTED;
774 	else
775 		*(struct sockaddr_in *)sa = (struct sockaddr_in ){
776 			.sin_family = AF_INET,
777 			.sin_len = sizeof(struct sockaddr_in),
778 			.sin_addr.s_addr = ssk->faddr,
779 			.sin_port = ssk->fport,
780 		};
781 	SDP_WUNLOCK(ssk);
782 
783 	return (error);
784 }
785 
786 /*
787  * Mark the connection as being incapable of further output.
788  */
789 static int
sdp_shutdown(struct socket * so,enum shutdown_how how)790 sdp_shutdown(struct socket *so, enum shutdown_how how)
791 {
792 	struct sdp_sock *ssk = sdp_sk(so);
793 	int error = 0;
794 
795 	SOCK_LOCK(so);
796 	if ((so->so_state &
797 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
798 		SOCK_UNLOCK(so);
799 		return (ENOTCONN);
800 	}
801 	if (SOLISTENING(so)) {
802 		if (how != SHUT_WR) {
803 			so->so_error = ECONNABORTED;
804 			solisten_wakeup(so);	/* unlocks so */
805 		} else
806 			SOCK_UNLOCK(so);
807 		return (0);
808 	}
809 	SOCK_UNLOCK(so);
810 
811 	switch (how) {
812 	case SHUT_RD:
813 		socantrcvmore(so);
814 		sbrelease(so, SO_RCV);
815 		break;
816 	case SHUT_RDWR:
817 		socantrcvmore(so);
818 		sbrelease(so, SO_RCV);
819 		/* FALLTHROUGH */
820 	case SHUT_WR:
821 		SDP_WLOCK(ssk);
822 		if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
823 			SDP_WUNLOCK(ssk);
824 			error = ECONNRESET;
825 			break;
826 		}
827 		socantsendmore(so);
828 		sdp_usrclosed(ssk);
829 		if (!(ssk->flags & SDP_DROPPED))
830 			sdp_output_disconnect(ssk);
831 		SDP_WUNLOCK(ssk);
832 	}
833 	wakeup(&so->so_timeo);
834 
835 	return (error);
836 }
837 
838 static void
sdp_append(struct sdp_sock * ssk,struct sockbuf * sb,struct mbuf * mb,int cnt)839 sdp_append(struct sdp_sock *ssk, struct sockbuf *sb, struct mbuf *mb, int cnt)
840 {
841 	struct mbuf *n;
842 	int ncnt;
843 
844 	SOCKBUF_LOCK_ASSERT(sb);
845 	SBLASTRECORDCHK(sb);
846 	KASSERT(mb->m_flags & M_PKTHDR,
847 		("sdp_append: %p Missing packet header.\n", mb));
848 	n = sb->sb_lastrecord;
849 	/*
850 	 * If the queue is empty just set all pointers and proceed.
851 	 */
852 	if (n == NULL) {
853 		sb->sb_lastrecord = sb->sb_mb = sb->sb_sndptr = mb;
854 		for (; mb; mb = mb->m_next) {
855 	                sb->sb_mbtail = mb;
856 			sballoc(sb, mb);
857 		}
858 		return;
859 	}
860 	/*
861 	 * Count the number of mbufs in the current tail.
862 	 */
863 	for (ncnt = 0; n->m_next; n = n->m_next)
864 		ncnt++;
865 	n = sb->sb_lastrecord;
866 	/*
867 	 * If the two chains can fit in a single sdp packet and
868 	 * the last record has not been sent yet (WRITABLE) coalesce
869 	 * them.  The lastrecord remains the same but we must strip the
870 	 * packet header and then let sbcompress do the hard part.
871 	 */
872 	if (M_WRITABLE(n) && ncnt + cnt < SDP_MAX_SEND_SGES &&
873 	    n->m_pkthdr.len + mb->m_pkthdr.len - SDP_HEAD_SIZE <
874 	    ssk->xmit_size_goal) {
875 		m_adj(mb, SDP_HEAD_SIZE);
876 		n->m_pkthdr.len += mb->m_pkthdr.len;
877 		n->m_flags |= mb->m_flags & (M_PUSH | M_URG);
878 		m_demote(mb, 1, 0);
879 		sbcompress(sb, mb, sb->sb_mbtail);
880 		return;
881 	}
882 	/*
883 	 * Not compressible, just append to the end and adjust counters.
884 	 */
885 	sb->sb_lastrecord->m_flags |= M_PUSH;
886 	sb->sb_lastrecord->m_nextpkt = mb;
887 	sb->sb_lastrecord = mb;
888 	if (sb->sb_sndptr == NULL)
889 		sb->sb_sndptr = mb;
890 	for (; mb; mb = mb->m_next) {
891 		sb->sb_mbtail = mb;
892 		sballoc(sb, mb);
893 	}
894 }
895 
896 /*
897  * Do a send by putting data in output queue and updating urgent
898  * marker if URG set.  Possibly send more data.  Unlike the other
899  * pru_*() routines, the mbuf chains are our responsibility.  We
900  * must either enqueue them or free them.  The other pru_* routines
901  * generally are caller-frees.
902  *
903  * This comes from sendfile, normal sends will come from sdp_sosend().
904  */
905 static int
sdp_send(struct socket * so,int flags,struct mbuf * m,struct sockaddr * nam,struct mbuf * control,struct thread * td)906 sdp_send(struct socket *so, int flags, struct mbuf *m,
907     struct sockaddr *nam, struct mbuf *control, struct thread *td)
908 {
909 	struct sdp_sock *ssk;
910 	struct mbuf *n;
911 	int error;
912 	int cnt;
913 
914 	if (nam != NULL) {
915 		if (nam->sa_family != AF_INET) {
916 			if (control)
917 				m_freem(control);
918 			m_freem(m);
919 			return (EAFNOSUPPORT);
920 		}
921 		if (nam->sa_len != sizeof(struct sockaddr_in)) {
922 			if (control)
923 				m_freem(control);
924 			m_freem(m);
925 			return (EINVAL);
926 		}
927 	}
928 
929 	error = 0;
930 	ssk = sdp_sk(so);
931 	KASSERT(m->m_flags & M_PKTHDR,
932 	    ("sdp_send: %p no packet header", m));
933 	M_PREPEND(m, SDP_HEAD_SIZE, M_WAITOK);
934 	mtod(m, struct sdp_bsdh *)->mid = SDP_MID_DATA;
935 	for (n = m, cnt = 0; n->m_next; n = n->m_next)
936 		cnt++;
937 	if (cnt > SDP_MAX_SEND_SGES) {
938 		n = m_collapse(m, M_WAITOK, SDP_MAX_SEND_SGES);
939 		if (n == NULL) {
940 			m_freem(m);
941 			return (EMSGSIZE);
942 		}
943 		m = n;
944 		for (cnt = 0; n->m_next; n = n->m_next)
945 			cnt++;
946 	}
947 	SDP_WLOCK(ssk);
948 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
949 		if (control)
950 			m_freem(control);
951 		if (m)
952 			m_freem(m);
953 		error = ECONNRESET;
954 		goto out;
955 	}
956 	if (control) {
957 		/* SDP doesn't support control messages. */
958 		if (control->m_len) {
959 			m_freem(control);
960 			if (m)
961 				m_freem(m);
962 			error = EINVAL;
963 			goto out;
964 		}
965 		m_freem(control);	/* empty control, just free it */
966 	}
967 	if (!(flags & PRUS_OOB)) {
968 		SOCKBUF_LOCK(&so->so_snd);
969 		sdp_append(ssk, &so->so_snd, m, cnt);
970 		SOCKBUF_UNLOCK(&so->so_snd);
971 		if (nam && ssk->state < TCPS_SYN_SENT) {
972 			/*
973 			 * Do implied connect if not yet connected.
974 			 */
975 			error = sdp_start_connect(ssk, nam, td);
976 			if (error)
977 				goto out;
978 		}
979 		if (flags & PRUS_EOF) {
980 			/*
981 			 * Close the send side of the connection after
982 			 * the data is sent.
983 			 */
984 			socantsendmore(so);
985 			sdp_usrclosed(ssk);
986 			if (!(ssk->flags & SDP_DROPPED))
987 				sdp_output_disconnect(ssk);
988 		} else if (!(ssk->flags & SDP_DROPPED) &&
989 		    !(flags & PRUS_MORETOCOME))
990 			sdp_post_sends(ssk, M_NOWAIT);
991 		SDP_WUNLOCK(ssk);
992 		return (0);
993 	} else {
994 		SOCKBUF_LOCK(&so->so_snd);
995 		if (sbspace(&so->so_snd) < -512) {
996 			SOCKBUF_UNLOCK(&so->so_snd);
997 			m_freem(m);
998 			error = ENOBUFS;
999 			goto out;
1000 		}
1001 		/*
1002 		 * According to RFC961 (Assigned Protocols),
1003 		 * the urgent pointer points to the last octet
1004 		 * of urgent data.  We continue, however,
1005 		 * to consider it to indicate the first octet
1006 		 * of data past the urgent section.
1007 		 * Otherwise, snd_up should be one lower.
1008 		 */
1009 		m->m_flags |= M_URG | M_PUSH;
1010 		sdp_append(ssk, &so->so_snd, m, cnt);
1011 		SOCKBUF_UNLOCK(&so->so_snd);
1012 		if (nam && ssk->state < TCPS_SYN_SENT) {
1013 			/*
1014 			 * Do implied connect if not yet connected.
1015 			 */
1016 			error = sdp_start_connect(ssk, nam, td);
1017 			if (error)
1018 				goto out;
1019 		}
1020 		sdp_post_sends(ssk, M_NOWAIT);
1021 		SDP_WUNLOCK(ssk);
1022 		return (0);
1023 	}
1024 out:
1025 	SDP_WUNLOCK(ssk);
1026 	return (error);
1027 }
1028 
1029 /*
1030  * Send on a socket.  If send must go all at once and message is larger than
1031  * send buffering, then hard error.  Lock against other senders.  If must go
1032  * all at once and not enough room now, then inform user that this would
1033  * block and do nothing.  Otherwise, if nonblocking, send as much as
1034  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1035  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1036  * in mbuf chain must be small enough to send all at once.
1037  *
1038  * Returns nonzero on error, timeout or signal; callers must check for short
1039  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1040  * on return.
1041  */
1042 static int
sdp_sosend(struct socket * so,struct sockaddr * addr,struct uio * uio,struct mbuf * top,struct mbuf * control,int flags,struct thread * td)1043 sdp_sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1044     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1045 {
1046 	struct sdp_sock *ssk;
1047 	long space, resid;
1048 	int atomic;
1049 	int error;
1050 	int copy;
1051 
1052 	if (uio != NULL)
1053 		resid = uio->uio_resid;
1054 	else
1055 		resid = top->m_pkthdr.len;
1056 	atomic = top != NULL;
1057 	if (control != NULL) {
1058 		if (control->m_len) {
1059 			m_freem(control);
1060 			if (top)
1061 				m_freem(top);
1062 			return (EINVAL);
1063 		}
1064 		m_freem(control);
1065 		control = NULL;
1066 	}
1067 	/*
1068 	 * In theory resid should be unsigned.  However, space must be
1069 	 * signed, as it might be less than 0 if we over-committed, and we
1070 	 * must use a signed comparison of space and resid.  On the other
1071 	 * hand, a negative resid causes us to loop sending 0-length
1072 	 * segments to the protocol.
1073 	 *
1074 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1075 	 * type sockets since that's an error.
1076 	 */
1077 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1078 		error = EINVAL;
1079 		goto out;
1080 	}
1081 	if (td != NULL)
1082 		td->td_ru.ru_msgsnd++;
1083 
1084 	ssk = sdp_sk(so);
1085 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1086 	if (error)
1087 		goto out;
1088 
1089 restart:
1090 	do {
1091 		SOCKBUF_LOCK(&so->so_snd);
1092 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1093 			SOCKBUF_UNLOCK(&so->so_snd);
1094 			error = EPIPE;
1095 			goto release;
1096 		}
1097 		if (so->so_error) {
1098 			error = so->so_error;
1099 			so->so_error = 0;
1100 			SOCKBUF_UNLOCK(&so->so_snd);
1101 			goto release;
1102 		}
1103 		if ((so->so_state & SS_ISCONNECTED) == 0 && addr == NULL) {
1104 			SOCKBUF_UNLOCK(&so->so_snd);
1105 			error = ENOTCONN;
1106 			goto release;
1107 		}
1108 		space = sbspace(&so->so_snd);
1109 		if (flags & MSG_OOB)
1110 			space += 1024;
1111 		if (atomic && resid > ssk->xmit_size_goal - SDP_HEAD_SIZE) {
1112 			SOCKBUF_UNLOCK(&so->so_snd);
1113 			error = EMSGSIZE;
1114 			goto release;
1115 		}
1116 		if (space < resid &&
1117 		    (atomic || space < so->so_snd.sb_lowat)) {
1118 			if ((so->so_state & SS_NBIO) ||
1119 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1120 				SOCKBUF_UNLOCK(&so->so_snd);
1121 				error = EWOULDBLOCK;
1122 				goto release;
1123 			}
1124 			error = sbwait(so, SO_SND);
1125 			SOCKBUF_UNLOCK(&so->so_snd);
1126 			if (error)
1127 				goto release;
1128 			goto restart;
1129 		}
1130 		SOCKBUF_UNLOCK(&so->so_snd);
1131 		do {
1132 			if (uio == NULL) {
1133 				resid = 0;
1134 				if (flags & MSG_EOR)
1135 					top->m_flags |= M_EOR;
1136 			} else {
1137 				/*
1138 				 * Copy the data from userland into a mbuf
1139 				 * chain.  If no data is to be copied in,
1140 				 * a single empty mbuf is returned.
1141 				 */
1142 				copy = min(space,
1143 				    ssk->xmit_size_goal - SDP_HEAD_SIZE);
1144 				top = m_uiotombuf(uio, M_WAITOK, copy,
1145 				    0, M_PKTHDR |
1146 				    ((flags & MSG_EOR) ? M_EOR : 0));
1147 				if (top == NULL) {
1148 					/* only possible error */
1149 					error = EFAULT;
1150 					goto release;
1151 				}
1152 				space -= resid - uio->uio_resid;
1153 				resid = uio->uio_resid;
1154 			}
1155 			/*
1156 			 * XXX all the SBS_CANTSENDMORE checks previously
1157 			 * done could be out of date after dropping the
1158 			 * socket lock.
1159 			 */
1160 			error = sdp_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1161 			/*
1162 			 * Set EOF on the last send if the user specified
1163 			 * MSG_EOF.
1164 			 */
1165 			    ((flags & MSG_EOF) && (resid <= 0)) ? PRUS_EOF :
1166 			/* If there is more to send set PRUS_MORETOCOME. */
1167 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1168 			    top, addr, NULL, td);
1169 			top = NULL;
1170 			if (error)
1171 				goto release;
1172 		} while (resid && space > 0);
1173 	} while (resid);
1174 
1175 release:
1176 	SOCK_IO_SEND_UNLOCK(so);
1177 out:
1178 	if (top != NULL)
1179 		m_freem(top);
1180 	return (error);
1181 }
1182 
1183 /*
1184  * The part of soreceive() that implements reading non-inline out-of-band
1185  * data from a socket.  For more complete comments, see soreceive(), from
1186  * which this code originated.
1187  *
1188  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1189  * unable to return an mbuf chain to the caller.
1190  */
1191 static int
soreceive_rcvoob(struct socket * so,struct uio * uio,int flags)1192 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1193 {
1194 	struct protosw *pr = so->so_proto;
1195 	struct mbuf *m;
1196 	int error;
1197 
1198 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1199 
1200 	m = m_get(M_WAITOK, MT_DATA);
1201 	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
1202 	if (error)
1203 		goto bad;
1204 	do {
1205 		error = uiomove(mtod(m, void *),
1206 		    (int) min(uio->uio_resid, m->m_len), uio);
1207 		m = m_free(m);
1208 	} while (uio->uio_resid && error == 0 && m);
1209 bad:
1210 	if (m != NULL)
1211 		m_freem(m);
1212 	return (error);
1213 }
1214 
1215 /*
1216  * Optimized version of soreceive() for stream (TCP) sockets.
1217  */
1218 static int
sdp_sorecv(struct socket * so,struct sockaddr ** psa,struct uio * uio,struct mbuf ** mp0,struct mbuf ** controlp,int * flagsp)1219 sdp_sorecv(struct socket *so, struct sockaddr **psa, struct uio *uio,
1220     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1221 {
1222 	int len = 0, error = 0, flags, oresid;
1223 	struct sockbuf *sb;
1224 	struct mbuf *m, *n = NULL;
1225 	struct sdp_sock *ssk;
1226 
1227 	/* We only do stream sockets. */
1228 	if (so->so_type != SOCK_STREAM)
1229 		return (EINVAL);
1230 	if (psa != NULL)
1231 		*psa = NULL;
1232 	if (controlp != NULL)
1233 		return (EINVAL);
1234 	if (flagsp != NULL)
1235 		flags = *flagsp &~ MSG_EOR;
1236 	else
1237 		flags = 0;
1238 	if (flags & MSG_OOB)
1239 		return (soreceive_rcvoob(so, uio, flags));
1240 	if (mp0 != NULL)
1241 		*mp0 = NULL;
1242 
1243 	sb = &so->so_rcv;
1244 	ssk = sdp_sk(so);
1245 
1246 	/* Prevent other readers from entering the socket. */
1247 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1248 	if (error)
1249 		return (error);
1250 	SOCKBUF_LOCK(sb);
1251 
1252 	/* Easy one, no space to copyout anything. */
1253 	if (uio->uio_resid == 0) {
1254 		error = EINVAL;
1255 		goto out;
1256 	}
1257 	oresid = uio->uio_resid;
1258 
1259 	/* We will never ever get anything unless we are connected. */
1260 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1261 		/* When disconnecting there may be still some data left. */
1262 		if (sbavail(sb))
1263 			goto deliver;
1264 		if (!(so->so_state & SS_ISDISCONNECTED))
1265 			error = ENOTCONN;
1266 		goto out;
1267 	}
1268 
1269 	/* Socket buffer is empty and we shall not block. */
1270 	if (sbavail(sb) == 0 &&
1271 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1272 		error = EAGAIN;
1273 		goto out;
1274 	}
1275 
1276 restart:
1277 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1278 
1279 	/* Abort if socket has reported problems. */
1280 	if (so->so_error) {
1281 		if (sbavail(sb))
1282 			goto deliver;
1283 		if (oresid > uio->uio_resid)
1284 			goto out;
1285 		error = so->so_error;
1286 		if (!(flags & MSG_PEEK))
1287 			so->so_error = 0;
1288 		goto out;
1289 	}
1290 
1291 	/* Door is closed.  Deliver what is left, if any. */
1292 	if (sb->sb_state & SBS_CANTRCVMORE) {
1293 		if (sbavail(sb))
1294 			goto deliver;
1295 		else
1296 			goto out;
1297 	}
1298 
1299 	/* Socket buffer got some data that we shall deliver now. */
1300 	if (sbavail(sb) && !(flags & MSG_WAITALL) &&
1301 	    ((so->so_state & SS_NBIO) ||
1302 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1303 	     sbavail(sb) >= sb->sb_lowat ||
1304 	     sbavail(sb) >= uio->uio_resid ||
1305 	     sbavail(sb) >= sb->sb_hiwat) ) {
1306 		goto deliver;
1307 	}
1308 
1309 	/* On MSG_WAITALL we must wait until all data or error arrives. */
1310 	if ((flags & MSG_WAITALL) &&
1311 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_lowat))
1312 		goto deliver;
1313 
1314 	/*
1315 	 * Wait and block until (more) data comes in.
1316 	 * NB: Drops the sockbuf lock during wait.
1317 	 */
1318 	error = sbwait(so, SO_RCV);
1319 	if (error)
1320 		goto out;
1321 	goto restart;
1322 
1323 deliver:
1324 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1325 	KASSERT(sbavail(sb), ("%s: sockbuf empty", __func__));
1326 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1327 
1328 	/* Statistics. */
1329 	if (uio->uio_td)
1330 		uio->uio_td->td_ru.ru_msgrcv++;
1331 
1332 	/* Fill uio until full or current end of socket buffer is reached. */
1333 	len = min(uio->uio_resid, sbavail(sb));
1334 	if (mp0 != NULL) {
1335 		/* Dequeue as many mbufs as possible. */
1336 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1337 			for (*mp0 = m = sb->sb_mb;
1338 			     m != NULL && m->m_len <= len;
1339 			     m = m->m_next) {
1340 				len -= m->m_len;
1341 				uio->uio_resid -= m->m_len;
1342 				sbfree(sb, m);
1343 				n = m;
1344 			}
1345 			sb->sb_mb = m;
1346 			if (sb->sb_mb == NULL)
1347 				SB_EMPTY_FIXUP(sb);
1348 			n->m_next = NULL;
1349 		}
1350 		/* Copy the remainder. */
1351 		if (len > 0) {
1352 			KASSERT(sb->sb_mb != NULL,
1353 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
1354 
1355 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
1356 			if (m == NULL)
1357 				len = 0;	/* Don't flush data from sockbuf. */
1358 			else
1359 				uio->uio_resid -= m->m_len;
1360 			if (*mp0 != NULL)
1361 				n->m_next = m;
1362 			else
1363 				*mp0 = m;
1364 			if (*mp0 == NULL) {
1365 				error = ENOBUFS;
1366 				goto out;
1367 			}
1368 		}
1369 	} else {
1370 		/* NB: Must unlock socket buffer as uiomove may sleep. */
1371 		SOCKBUF_UNLOCK(sb);
1372 		error = m_mbuftouio(uio, sb->sb_mb, len);
1373 		SOCKBUF_LOCK(sb);
1374 		if (error)
1375 			goto out;
1376 	}
1377 	SBLASTRECORDCHK(sb);
1378 	SBLASTMBUFCHK(sb);
1379 
1380 	/*
1381 	 * Remove the delivered data from the socket buffer unless we
1382 	 * were only peeking.
1383 	 */
1384 	if (!(flags & MSG_PEEK)) {
1385 		if (len > 0)
1386 			sbdrop_locked(sb, len);
1387 
1388 		/* Notify protocol that we drained some data. */
1389 		SOCKBUF_UNLOCK(sb);
1390 		SDP_WLOCK(ssk);
1391 		sdp_do_posts(ssk);
1392 		SDP_WUNLOCK(ssk);
1393 		SOCKBUF_LOCK(sb);
1394 	}
1395 
1396 	/*
1397 	 * For MSG_WAITALL we may have to loop again and wait for
1398 	 * more data to come in.
1399 	 */
1400 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
1401 		goto restart;
1402 out:
1403 	SBLASTRECORDCHK(sb);
1404 	SBLASTMBUFCHK(sb);
1405 	SOCKBUF_UNLOCK(sb);
1406 	SOCK_IO_RECV_UNLOCK(so);
1407 	return (error);
1408 }
1409 
1410 /*
1411  * Abort is used to teardown a connection typically while sitting in
1412  * the accept queue.
1413  */
1414 void
sdp_abort(struct socket * so)1415 sdp_abort(struct socket *so)
1416 {
1417 	struct sdp_sock *ssk;
1418 
1419 	ssk = sdp_sk(so);
1420 	SDP_WLOCK(ssk);
1421 	/*
1422 	 * If we have not yet dropped, do it now.
1423 	 */
1424 	if (!(ssk->flags & SDP_TIMEWAIT) &&
1425 	    !(ssk->flags & SDP_DROPPED))
1426 		sdp_drop(ssk, ECONNABORTED);
1427 	KASSERT(ssk->flags & SDP_DROPPED, ("sdp_abort: %p not dropped 0x%X",
1428 	    ssk, ssk->flags));
1429 	SDP_WUNLOCK(ssk);
1430 }
1431 
1432 /*
1433  * Close a SDP socket and initiate a friendly disconnect.
1434  */
1435 static void
sdp_close(struct socket * so)1436 sdp_close(struct socket *so)
1437 {
1438 	struct sdp_sock *ssk;
1439 
1440 	ssk = sdp_sk(so);
1441 	SDP_WLOCK(ssk);
1442 	/*
1443 	 * If we have not yet dropped, do it now.
1444 	 */
1445 	if (!(ssk->flags & SDP_TIMEWAIT) &&
1446 	    !(ssk->flags & SDP_DROPPED))
1447 		sdp_start_disconnect(ssk);
1448 
1449 	/*
1450 	 * If we've still not dropped let the socket layer know we're
1451 	 * holding on to the socket and pcb for a while.
1452 	 */
1453 	if (!(ssk->flags & SDP_DROPPED)) {
1454 		ssk->flags |= SDP_SOCKREF;
1455 		soref(so);
1456 	}
1457 	SDP_WUNLOCK(ssk);
1458 }
1459 
1460 /*
1461  * User requests out-of-band data.
1462  */
1463 static int
sdp_rcvoob(struct socket * so,struct mbuf * m,int flags)1464 sdp_rcvoob(struct socket *so, struct mbuf *m, int flags)
1465 {
1466 	int error = 0;
1467 	struct sdp_sock *ssk;
1468 
1469 	ssk = sdp_sk(so);
1470 	SDP_WLOCK(ssk);
1471 	if (!rx_ring_trylock(&ssk->rx_ring)) {
1472 		SDP_WUNLOCK(ssk);
1473 		return (ECONNRESET);
1474 	}
1475 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
1476 		error = ECONNRESET;
1477 		goto out;
1478 	}
1479 	if ((so->so_oobmark == 0 &&
1480 	     (so->so_rcv.sb_state & SBS_RCVATMARK) == 0) ||
1481 	    so->so_options & SO_OOBINLINE ||
1482 	    ssk->oobflags & SDP_HADOOB) {
1483 		error = EINVAL;
1484 		goto out;
1485 	}
1486 	if ((ssk->oobflags & SDP_HAVEOOB) == 0) {
1487 		error = EWOULDBLOCK;
1488 		goto out;
1489 	}
1490 	m->m_len = 1;
1491 	*mtod(m, caddr_t) = ssk->iobc;
1492 	if ((flags & MSG_PEEK) == 0)
1493 		ssk->oobflags ^= (SDP_HAVEOOB | SDP_HADOOB);
1494 out:
1495 	rx_ring_unlock(&ssk->rx_ring);
1496 	SDP_WUNLOCK(ssk);
1497 	return (error);
1498 }
1499 
1500 void
sdp_urg(struct sdp_sock * ssk,struct mbuf * mb)1501 sdp_urg(struct sdp_sock *ssk, struct mbuf *mb)
1502 {
1503 	struct mbuf *m;
1504 	struct socket *so;
1505 
1506 	so = ssk->socket;
1507 	if (so == NULL)
1508 		return;
1509 
1510 	so->so_oobmark = sbused(&so->so_rcv) + mb->m_pkthdr.len - 1;
1511 	sohasoutofband(so);
1512 	ssk->oobflags &= ~(SDP_HAVEOOB | SDP_HADOOB);
1513 	if (!(so->so_options & SO_OOBINLINE)) {
1514 		for (m = mb; m->m_next != NULL; m = m->m_next);
1515 		ssk->iobc = *(mtod(m, char *) + m->m_len - 1);
1516 		ssk->oobflags |= SDP_HAVEOOB;
1517 		m->m_len--;
1518 		mb->m_pkthdr.len--;
1519 	}
1520 }
1521 
1522 /*
1523  * Notify a sdp socket of an asynchronous error.
1524  *
1525  * Do not wake up user since there currently is no mechanism for
1526  * reporting soft errors (yet - a kqueue filter may be added).
1527  */
1528 struct sdp_sock *
sdp_notify(struct sdp_sock * ssk,int error)1529 sdp_notify(struct sdp_sock *ssk, int error)
1530 {
1531 
1532 	SDP_WLOCK_ASSERT(ssk);
1533 
1534 	if ((ssk->flags & SDP_TIMEWAIT) ||
1535 	    (ssk->flags & SDP_DROPPED))
1536 		return (ssk);
1537 
1538 	/*
1539 	 * Ignore some errors if we are hooked up.
1540 	 */
1541 	if (ssk->state == TCPS_ESTABLISHED &&
1542 	    (error == EHOSTUNREACH || error == ENETUNREACH ||
1543 	     error == EHOSTDOWN))
1544 		return (ssk);
1545 	ssk->softerror = error;
1546 	return sdp_drop(ssk, error);
1547 }
1548 
1549 static void
sdp_keepalive_timeout(void * data)1550 sdp_keepalive_timeout(void *data)
1551 {
1552 	struct sdp_sock *ssk;
1553 
1554 	ssk = data;
1555 	/* Callout canceled. */
1556         if (!callout_active(&ssk->keep2msl))
1557                 return;
1558 	/* Callout rescheduled as a different kind of timer. */
1559 	if (callout_pending(&ssk->keep2msl))
1560 		goto out;
1561         callout_deactivate(&ssk->keep2msl);
1562 	if (ssk->flags & SDP_DROPPED ||
1563 	    (ssk->socket->so_options & SO_KEEPALIVE) == 0)
1564 		goto out;
1565 	sdp_post_keepalive(ssk);
1566 	callout_reset(&ssk->keep2msl, SDP_KEEPALIVE_TIME,
1567 	    sdp_keepalive_timeout, ssk);
1568 out:
1569 	SDP_WUNLOCK(ssk);
1570 }
1571 
1572 
1573 void
sdp_start_keepalive_timer(struct socket * so)1574 sdp_start_keepalive_timer(struct socket *so)
1575 {
1576 	struct sdp_sock *ssk;
1577 
1578 	ssk = sdp_sk(so);
1579 	if (!callout_pending(&ssk->keep2msl))
1580                 callout_reset(&ssk->keep2msl, SDP_KEEPALIVE_TIME,
1581                     sdp_keepalive_timeout, ssk);
1582 }
1583 
1584 static void
sdp_stop_keepalive_timer(struct socket * so)1585 sdp_stop_keepalive_timer(struct socket *so)
1586 {
1587 	struct sdp_sock *ssk;
1588 
1589 	ssk = sdp_sk(so);
1590 	callout_stop(&ssk->keep2msl);
1591 }
1592 
1593 /*
1594  * sdp_ctloutput() must drop the inpcb lock before performing copyin on
1595  * socket option arguments.  When it re-acquires the lock after the copy, it
1596  * has to revalidate that the connection is still valid for the socket
1597  * option.
1598  */
1599 #define SDP_WLOCK_RECHECK(inp) do {					\
1600 	SDP_WLOCK(ssk);							\
1601 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {		\
1602 		SDP_WUNLOCK(ssk);					\
1603 		return (ECONNRESET);					\
1604 	}								\
1605 } while(0)
1606 
1607 static int
sdp_ctloutput(struct socket * so,struct sockopt * sopt)1608 sdp_ctloutput(struct socket *so, struct sockopt *sopt)
1609 {
1610 	int	error, opt, optval;
1611 	struct sdp_sock *ssk;
1612 
1613 	error = 0;
1614 	ssk = sdp_sk(so);
1615 	if (sopt->sopt_level == SOL_SOCKET && sopt->sopt_name == SO_KEEPALIVE) {
1616 		SDP_WLOCK(ssk);
1617 		if (so->so_options & SO_KEEPALIVE)
1618 			sdp_start_keepalive_timer(so);
1619 		else
1620 			sdp_stop_keepalive_timer(so);
1621 		SDP_WUNLOCK(ssk);
1622 	}
1623 	if (sopt->sopt_level != IPPROTO_TCP)
1624 		return (error);
1625 
1626 	SDP_WLOCK(ssk);
1627 	if (ssk->flags & (SDP_TIMEWAIT | SDP_DROPPED)) {
1628 		SDP_WUNLOCK(ssk);
1629 		return (ECONNRESET);
1630 	}
1631 
1632 	switch (sopt->sopt_dir) {
1633 	case SOPT_SET:
1634 		switch (sopt->sopt_name) {
1635 		case TCP_NODELAY:
1636 			SDP_WUNLOCK(ssk);
1637 			error = sooptcopyin(sopt, &optval, sizeof optval,
1638 			    sizeof optval);
1639 			if (error)
1640 				return (error);
1641 
1642 			SDP_WLOCK_RECHECK(ssk);
1643 			opt = SDP_NODELAY;
1644 			if (optval)
1645 				ssk->flags |= opt;
1646 			else
1647 				ssk->flags &= ~opt;
1648 			sdp_do_posts(ssk);
1649 			SDP_WUNLOCK(ssk);
1650 			break;
1651 
1652 		default:
1653 			SDP_WUNLOCK(ssk);
1654 			error = ENOPROTOOPT;
1655 			break;
1656 		}
1657 		break;
1658 
1659 	case SOPT_GET:
1660 		switch (sopt->sopt_name) {
1661 		case TCP_NODELAY:
1662 			optval = ssk->flags & SDP_NODELAY;
1663 			SDP_WUNLOCK(ssk);
1664 			error = sooptcopyout(sopt, &optval, sizeof optval);
1665 			break;
1666 		default:
1667 			SDP_WUNLOCK(ssk);
1668 			error = ENOPROTOOPT;
1669 			break;
1670 		}
1671 		break;
1672 	}
1673 	return (error);
1674 }
1675 #undef SDP_WLOCK_RECHECK
1676 
1677 int sdp_mod_count = 0;
1678 int sdp_mod_usec = 0;
1679 
1680 void
sdp_set_default_moderation(struct sdp_sock * ssk)1681 sdp_set_default_moderation(struct sdp_sock *ssk)
1682 {
1683 	if (sdp_mod_count <= 0 || sdp_mod_usec <= 0)
1684 		return;
1685 	ib_modify_cq(ssk->rx_ring.cq, sdp_mod_count, sdp_mod_usec);
1686 }
1687 
1688 static void
sdp_dev_add(struct ib_device * device)1689 sdp_dev_add(struct ib_device *device)
1690 {
1691 	struct ib_fmr_pool_param param;
1692 	struct sdp_device *sdp_dev;
1693 
1694 	sdp_dev = malloc(sizeof(*sdp_dev), M_SDP, M_WAITOK | M_ZERO);
1695 	sdp_dev->pd = ib_alloc_pd(device, 0);
1696 	if (IS_ERR(sdp_dev->pd))
1697 		goto out_pd;
1698 	memset(&param, 0, sizeof param);
1699 	param.max_pages_per_fmr = SDP_FMR_SIZE;
1700 	param.page_shift = PAGE_SHIFT;
1701 	param.access = (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_READ);
1702 	param.pool_size = SDP_FMR_POOL_SIZE;
1703 	param.dirty_watermark = SDP_FMR_DIRTY_SIZE;
1704 	param.cache = 1;
1705 	sdp_dev->fmr_pool = ib_create_fmr_pool(sdp_dev->pd, &param);
1706 	if (IS_ERR(sdp_dev->fmr_pool))
1707 		goto out_fmr;
1708 	ib_set_client_data(device, &sdp_client, sdp_dev);
1709 	return;
1710 
1711 out_fmr:
1712 	ib_dealloc_pd(sdp_dev->pd);
1713 out_pd:
1714 	free(sdp_dev, M_SDP);
1715 }
1716 
1717 static void
sdp_dev_rem(struct ib_device * device,void * client_data)1718 sdp_dev_rem(struct ib_device *device, void *client_data)
1719 {
1720 	struct sdp_device *sdp_dev;
1721 	struct sdp_sock *ssk;
1722 
1723 	SDP_LIST_WLOCK();
1724 	LIST_FOREACH(ssk, &sdp_list, list) {
1725 		if (ssk->ib_device != device)
1726 			continue;
1727 		SDP_WLOCK(ssk);
1728 		if ((ssk->flags & SDP_DESTROY) == 0)
1729 			ssk = sdp_notify(ssk, ECONNRESET);
1730 		if (ssk)
1731 			SDP_WUNLOCK(ssk);
1732 	}
1733 	SDP_LIST_WUNLOCK();
1734 	/*
1735 	 * XXX Do I need to wait between these two?
1736 	 */
1737 	sdp_dev = ib_get_client_data(device, &sdp_client);
1738 	if (!sdp_dev)
1739 		return;
1740 	ib_flush_fmr_pool(sdp_dev->fmr_pool);
1741 	ib_destroy_fmr_pool(sdp_dev->fmr_pool);
1742 	ib_dealloc_pd(sdp_dev->pd);
1743 	free(sdp_dev, M_SDP);
1744 }
1745 
1746 struct ib_client sdp_client =
1747     { .name = "sdp", .add = sdp_dev_add, .remove = sdp_dev_rem };
1748 
1749 
1750 static int
sdp_pcblist(SYSCTL_HANDLER_ARGS)1751 sdp_pcblist(SYSCTL_HANDLER_ARGS)
1752 {
1753 	int error, n, i;
1754 	struct sdp_sock *ssk;
1755 	struct xinpgen xig;
1756 
1757 	/*
1758 	 * The process of preparing the TCB list is too time-consuming and
1759 	 * resource-intensive to repeat twice on every request.
1760 	 */
1761 	if (req->oldptr == NULL) {
1762 		n = sdp_count;
1763 		n += imax(n / 8, 10);
1764 		req->oldidx = 2 * (sizeof xig) + n * sizeof(struct xtcpcb);
1765 		return (0);
1766 	}
1767 
1768 	if (req->newptr != NULL)
1769 		return (EPERM);
1770 
1771 	/*
1772 	 * OK, now we're committed to doing something.
1773 	 */
1774 	SDP_LIST_RLOCK();
1775 	n = sdp_count;
1776 	SDP_LIST_RUNLOCK();
1777 
1778 	error = sysctl_wire_old_buffer(req, 2 * (sizeof xig)
1779 		+ n * sizeof(struct xtcpcb));
1780 	if (error != 0)
1781 		return (error);
1782 
1783 	bzero(&xig, sizeof(xig));
1784 	xig.xig_len = sizeof xig;
1785 	xig.xig_count = n;
1786 	xig.xig_gen = 0;
1787 	xig.xig_sogen = so_gencnt;
1788 	error = SYSCTL_OUT(req, &xig, sizeof xig);
1789 	if (error)
1790 		return (error);
1791 
1792 	SDP_LIST_RLOCK();
1793 	for (ssk = LIST_FIRST(&sdp_list), i = 0;
1794 	    ssk != NULL && i < n; ssk = LIST_NEXT(ssk, list)) {
1795 		struct xtcpcb xt;
1796 
1797 		SDP_RLOCK(ssk);
1798 		if (ssk->flags & SDP_TIMEWAIT) {
1799 			if (ssk->cred != NULL)
1800 				error = cr_cansee(req->td->td_ucred,
1801 				    ssk->cred);
1802 			else
1803 				error = EINVAL;	/* Skip this inp. */
1804 		} else if (ssk->socket)
1805 			error = cr_canseesocket(req->td->td_ucred,
1806 			    ssk->socket);
1807 		else
1808 			error = EINVAL;
1809 		if (error) {
1810 			error = 0;
1811 			goto next;
1812 		}
1813 
1814 		bzero(&xt, sizeof(xt));
1815 		xt.xt_len = sizeof xt;
1816 		xt.xt_inp.inp_gencnt = 0;
1817 		xt.xt_inp.inp_vflag = INP_IPV4;
1818 		memcpy(&xt.xt_inp.inp_laddr, &ssk->laddr, sizeof(ssk->laddr));
1819 		xt.xt_inp.inp_lport = ssk->lport;
1820 		memcpy(&xt.xt_inp.inp_faddr, &ssk->faddr, sizeof(ssk->faddr));
1821 		xt.xt_inp.inp_fport = ssk->fport;
1822 		xt.t_state = ssk->state;
1823 		if (ssk->socket != NULL)
1824 			sotoxsocket(ssk->socket, &xt.xt_inp.xi_socket);
1825 		xt.xt_inp.xi_socket.xso_protocol = IPPROTO_TCP;
1826 		SDP_RUNLOCK(ssk);
1827 		error = SYSCTL_OUT(req, &xt, sizeof xt);
1828 		if (error)
1829 			break;
1830 		i++;
1831 		continue;
1832 next:
1833 		SDP_RUNLOCK(ssk);
1834 	}
1835 	if (!error) {
1836 		/*
1837 		 * Give the user an updated idea of our state.
1838 		 * If the generation differs from what we told
1839 		 * her before, she knows that something happened
1840 		 * while we were processing this request, and it
1841 		 * might be necessary to retry.
1842 		 */
1843 		xig.xig_gen = 0;
1844 		xig.xig_sogen = so_gencnt;
1845 		xig.xig_count = sdp_count;
1846 		error = SYSCTL_OUT(req, &xig, sizeof xig);
1847 	}
1848 	SDP_LIST_RUNLOCK();
1849 	return (error);
1850 }
1851 
1852 SYSCTL_NODE(_net_inet, -1, sdp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1853     "SDP");
1854 
1855 SYSCTL_PROC(_net_inet_sdp, TCPCTL_PCBLIST, pcblist,
1856     CTLFLAG_RD | CTLTYPE_STRUCT | CTLFLAG_MPSAFE,
1857     0, 0, sdp_pcblist, "S,xtcpcb",
1858     "List of active SDP connections");
1859 
1860 static void
sdp_zone_change(void * tag)1861 sdp_zone_change(void *tag)
1862 {
1863 
1864 	uma_zone_set_max(sdp_zone, maxsockets);
1865 }
1866 
1867 static void
sdp_init(void * arg __unused)1868 sdp_init(void *arg __unused)
1869 {
1870 
1871 	LIST_INIT(&sdp_list);
1872 	sdp_zone = uma_zcreate("sdp_sock", sizeof(struct sdp_sock),
1873 	    NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1874 	uma_zone_set_max(sdp_zone, maxsockets);
1875 	EVENTHANDLER_REGISTER(maxsockets_change, sdp_zone_change, NULL,
1876 		EVENTHANDLER_PRI_ANY);
1877 	rx_comp_wq = create_singlethread_workqueue("rx_comp_wq");
1878 	ib_register_client(&sdp_client);
1879 }
1880 SYSINIT(sdp_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_SECOND, sdp_init, NULL);
1881 
1882 #define	SDP_PROTOSW							\
1883 	.pr_type =		SOCK_STREAM,				\
1884 	.pr_flags =		PR_CONNREQUIRED|PR_IMPLOPCL|PR_WANTRCVD,\
1885 	.pr_ctloutput =		sdp_ctloutput,				\
1886 	.pr_abort =		sdp_abort,				\
1887 	.pr_accept =		sdp_accept,				\
1888 	.pr_attach =		sdp_attach,				\
1889 	.pr_bind =		sdp_bind,				\
1890 	.pr_connect =		sdp_connect,				\
1891 	.pr_detach =		sdp_detach,				\
1892 	.pr_disconnect =	sdp_disconnect,				\
1893 	.pr_listen =		sdp_listen,				\
1894 	.pr_peeraddr =		sdp_getpeeraddr,			\
1895 	.pr_rcvoob =		sdp_rcvoob,				\
1896 	.pr_send =		sdp_send,				\
1897 	.pr_sosend =		sdp_sosend,				\
1898 	.pr_soreceive =		sdp_sorecv,				\
1899 	.pr_shutdown =		sdp_shutdown,				\
1900 	.pr_sockaddr =		sdp_getsockaddr,			\
1901 	.pr_close =		sdp_close
1902 
1903 
1904 static struct protosw sdp_ip_protosw = {
1905 	.pr_protocol =		IPPROTO_IP,
1906 	SDP_PROTOSW
1907 };
1908 static struct protosw sdp_tcp_protosw = {
1909 	.pr_protocol =		IPPROTO_TCP,
1910 	SDP_PROTOSW
1911 };
1912 
1913 static struct domain sdpdomain = {
1914 	.dom_family =		AF_INET_SDP,
1915 	.dom_name =		"SDP",
1916 	.dom_nprotosw =		2,
1917 	.dom_protosw = {
1918 		&sdp_ip_protosw,
1919 		&sdp_tcp_protosw,
1920 	},
1921 };
1922 
1923 DOMAIN_SET(sdp);
1924 
1925 int sdp_debug_level = 1;
1926 int sdp_data_debug_level = 0;
1927