xref: /dragonfly/sys/kern/uipc_socket.c (revision 2702099d)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 4. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
63  * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
64  */
65 
66 #include "opt_inet.h"
67 #include "opt_sctp.h"
68 
69 #include <sys/param.h>
70 #include <sys/systm.h>
71 #include <sys/fcntl.h>
72 #include <sys/malloc.h>
73 #include <sys/mbuf.h>
74 #include <sys/domain.h>
75 #include <sys/file.h>			/* for struct knote */
76 #include <sys/kernel.h>
77 #include <sys/event.h>
78 #include <sys/proc.h>
79 #include <sys/protosw.h>
80 #include <sys/socket.h>
81 #include <sys/socketvar.h>
82 #include <sys/socketops.h>
83 #include <sys/resourcevar.h>
84 #include <sys/signalvar.h>
85 #include <sys/sysctl.h>
86 #include <sys/uio.h>
87 #include <sys/jail.h>
88 #include <vm/vm_zone.h>
89 #include <vm/pmap.h>
90 #include <net/netmsg2.h>
91 #include <net/netisr2.h>
92 
93 #include <sys/thread2.h>
94 #include <sys/socketvar2.h>
95 #include <sys/spinlock2.h>
96 
97 #include <machine/limits.h>
98 
99 #ifdef INET
100 extern int tcp_sosend_agglim;
101 extern int tcp_sosend_async;
102 extern int udp_sosend_async;
103 extern int udp_sosend_prepend;
104 
105 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
106 #endif /* INET */
107 
108 static void 	filt_sordetach(struct knote *kn);
109 static int 	filt_soread(struct knote *kn, long hint);
110 static void 	filt_sowdetach(struct knote *kn);
111 static int	filt_sowrite(struct knote *kn, long hint);
112 static int	filt_solisten(struct knote *kn, long hint);
113 
114 static void	sodiscard(struct socket *so);
115 static int	soclose_sync(struct socket *so, int fflag);
116 static void	soclose_fast(struct socket *so);
117 
118 static struct filterops solisten_filtops =
119 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
120 static struct filterops soread_filtops =
121 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
122 static struct filterops sowrite_filtops =
123 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
124 static struct filterops soexcept_filtops =
125 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
126 
127 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
128 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
129 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
130 
131 
132 static int somaxconn = SOMAXCONN;
133 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
134     &somaxconn, 0, "Maximum pending socket connection queue size");
135 
136 static int use_soclose_fast = 1;
137 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
138     &use_soclose_fast, 0, "Fast socket close");
139 
140 int use_soaccept_pred_fast = 1;
141 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
142     &use_soaccept_pred_fast, 0, "Fast socket accept predication");
143 
144 int use_sendfile_async = 1;
145 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
146     &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
147 
148 /*
149  * Socket operation routines.
150  * These routines are called by the routines in
151  * sys_socket.c or from a system process, and
152  * implement the semantics of socket operations by
153  * switching out to the protocol specific routines.
154  */
155 
156 /*
157  * Get a socket structure, and initialize it.
158  * Note that it would probably be better to allocate socket
159  * and PCB at the same time, but I'm not convinced that all
160  * the protocols can be easily modified to do this.
161  */
162 struct socket *
163 soalloc(int waitok, struct protosw *pr)
164 {
165 	struct socket *so;
166 	unsigned waitmask;
167 
168 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
169 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
170 	if (so) {
171 		/* XXX race condition for reentrant kernel */
172 		so->so_proto = pr;
173 		TAILQ_INIT(&so->so_aiojobq);
174 		TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
175 		TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
176 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
177 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
178 		spin_init(&so->so_rcvd_spin);
179 		netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
180 		    MSGF_DROPABLE, so->so_proto->pr_usrreqs->pru_rcvd);
181 		so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
182 		so->so_state = SS_NOFDREF;
183 		so->so_refs = 1;
184 	}
185 	return so;
186 }
187 
188 int
189 socreate(int dom, struct socket **aso, int type,
190 	int proto, struct thread *td)
191 {
192 	struct proc *p = td->td_proc;
193 	struct protosw *prp;
194 	struct socket *so;
195 	struct pru_attach_info ai;
196 	int error;
197 
198 	if (proto)
199 		prp = pffindproto(dom, proto, type);
200 	else
201 		prp = pffindtype(dom, type);
202 
203 	if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
204 		return (EPROTONOSUPPORT);
205 
206 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
207 	    prp->pr_domain->dom_family != PF_LOCAL &&
208 	    prp->pr_domain->dom_family != PF_INET &&
209 	    prp->pr_domain->dom_family != PF_INET6 &&
210 	    prp->pr_domain->dom_family != PF_ROUTE) {
211 		return (EPROTONOSUPPORT);
212 	}
213 
214 	if (prp->pr_type != type)
215 		return (EPROTOTYPE);
216 	so = soalloc(p != NULL, prp);
217 	if (so == NULL)
218 		return (ENOBUFS);
219 
220 	/*
221 	 * Callers of socreate() presumably will connect up a descriptor
222 	 * and call soclose() if they cannot.  This represents our so_refs
223 	 * (which should be 1) from soalloc().
224 	 */
225 	soclrstate(so, SS_NOFDREF);
226 
227 	/*
228 	 * Set a default port for protocol processing.  No action will occur
229 	 * on the socket on this port until an inpcb is attached to it and
230 	 * is able to match incoming packets, or until the socket becomes
231 	 * available to userland.
232 	 *
233 	 * We normally default the socket to the protocol thread on cpu 0.
234 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
235 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
236 	 */
237 	if (prp->pr_flags & PR_SYNC_PORT)
238 		so->so_port = &netisr_sync_port;
239 	else
240 		so->so_port = netisr_cpuport(0);
241 
242 	TAILQ_INIT(&so->so_incomp);
243 	TAILQ_INIT(&so->so_comp);
244 	so->so_type = type;
245 	so->so_cred = crhold(p->p_ucred);
246 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
247 	ai.p_ucred = p->p_ucred;
248 	ai.fd_rdir = p->p_fd->fd_rdir;
249 
250 	/*
251 	 * Auto-sizing of socket buffers is managed by the protocols and
252 	 * the appropriate flags must be set in the pru_attach function.
253 	 */
254 	error = so_pru_attach(so, proto, &ai);
255 	if (error) {
256 		sosetstate(so, SS_NOFDREF);
257 		sofree(so);	/* from soalloc */
258 		return error;
259 	}
260 
261 	/*
262 	 * NOTE: Returns referenced socket.
263 	 */
264 	*aso = so;
265 	return (0);
266 }
267 
268 int
269 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
270 {
271 	int error;
272 
273 	error = so_pru_bind(so, nam, td);
274 	return (error);
275 }
276 
277 static void
278 sodealloc(struct socket *so)
279 {
280 	if (so->so_rcv.ssb_hiwat)
281 		(void)chgsbsize(so->so_cred->cr_uidinfo,
282 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
283 	if (so->so_snd.ssb_hiwat)
284 		(void)chgsbsize(so->so_cred->cr_uidinfo,
285 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
286 #ifdef INET
287 	/* remove accept filter if present */
288 	if (so->so_accf != NULL)
289 		do_setopt_accept_filter(so, NULL);
290 #endif /* INET */
291 	crfree(so->so_cred);
292 	if (so->so_faddr != NULL)
293 		kfree(so->so_faddr, M_SONAME);
294 	kfree(so, M_SOCKET);
295 }
296 
297 int
298 solisten(struct socket *so, int backlog, struct thread *td)
299 {
300 	int error;
301 #ifdef SCTP
302 	short oldopt, oldqlimit;
303 #endif /* SCTP */
304 
305 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
306 		return (EINVAL);
307 
308 #ifdef SCTP
309 	oldopt = so->so_options;
310 	oldqlimit = so->so_qlimit;
311 #endif /* SCTP */
312 
313 	lwkt_gettoken(&so->so_rcv.ssb_token);
314 	if (TAILQ_EMPTY(&so->so_comp))
315 		so->so_options |= SO_ACCEPTCONN;
316 	lwkt_reltoken(&so->so_rcv.ssb_token);
317 	if (backlog < 0 || backlog > somaxconn)
318 		backlog = somaxconn;
319 	so->so_qlimit = backlog;
320 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
321 	 * the so_options (UDP model both connect's and gets inbound
322 	 * connections .. implicitly).
323 	 */
324 	error = so_pru_listen(so, td);
325 	if (error) {
326 #ifdef SCTP
327 		/* Restore the params */
328 		so->so_options = oldopt;
329 		so->so_qlimit = oldqlimit;
330 #endif /* SCTP */
331 		return (error);
332 	}
333 	return (0);
334 }
335 
336 /*
337  * Destroy a disconnected socket.  This routine is a NOP if entities
338  * still have a reference on the socket:
339  *
340  *	so_pcb -	The protocol stack still has a reference
341  *	SS_NOFDREF -	There is no longer a file pointer reference
342  */
343 void
344 sofree(struct socket *so)
345 {
346 	struct socket *head;
347 
348 	/*
349 	 * This is a bit hackish at the moment.  We need to interlock
350 	 * any accept queue we are on before we potentially lose the
351 	 * last reference to avoid races against a re-reference from
352 	 * someone operating on the queue.
353 	 */
354 	while ((head = so->so_head) != NULL) {
355 		lwkt_getpooltoken(head);
356 		if (so->so_head == head)
357 			break;
358 		lwkt_relpooltoken(head);
359 	}
360 
361 	/*
362 	 * Arbitrage the last free.
363 	 */
364 	KKASSERT(so->so_refs > 0);
365 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
366 		if (head)
367 			lwkt_relpooltoken(head);
368 		return;
369 	}
370 
371 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
372 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
373 
374 	/*
375 	 * We're done, remove ourselves from the accept queue we are
376 	 * on, if we are on one.
377 	 */
378 	if (head != NULL) {
379 		if (so->so_state & SS_INCOMP) {
380 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
381 			head->so_incqlen--;
382 		} else if (so->so_state & SS_COMP) {
383 			/*
384 			 * We must not decommission a socket that's
385 			 * on the accept(2) queue.  If we do, then
386 			 * accept(2) may hang after select(2) indicated
387 			 * that the listening socket was ready.
388 			 */
389 			lwkt_relpooltoken(head);
390 			return;
391 		} else {
392 			panic("sofree: not queued");
393 		}
394 		soclrstate(so, SS_INCOMP);
395 		so->so_head = NULL;
396 		lwkt_relpooltoken(head);
397 	}
398 	ssb_release(&so->so_snd, so);
399 	sorflush(so);
400 	sodealloc(so);
401 }
402 
403 /*
404  * Close a socket on last file table reference removal.
405  * Initiate disconnect if connected.
406  * Free socket when disconnect complete.
407  */
408 int
409 soclose(struct socket *so, int fflag)
410 {
411 	int error;
412 
413 	funsetown(&so->so_sigio);
414 	if (!use_soclose_fast ||
415 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
416 	    (so->so_options & SO_LINGER)) {
417 		error = soclose_sync(so, fflag);
418 	} else {
419 		soclose_fast(so);
420 		error = 0;
421 	}
422 	return error;
423 }
424 
425 static void
426 sodiscard(struct socket *so)
427 {
428 	lwkt_getpooltoken(so);
429 	if (so->so_options & SO_ACCEPTCONN) {
430 		struct socket *sp;
431 
432 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
433 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
434 			soclrstate(sp, SS_INCOMP);
435 			sp->so_head = NULL;
436 			so->so_incqlen--;
437 			soaborta(sp);
438 		}
439 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
440 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
441 			soclrstate(sp, SS_COMP);
442 			sp->so_head = NULL;
443 			so->so_qlen--;
444 			soaborta(sp);
445 		}
446 	}
447 	lwkt_relpooltoken(so);
448 
449 	if (so->so_state & SS_NOFDREF)
450 		panic("soclose: NOFDREF");
451 	sosetstate(so, SS_NOFDREF);	/* take ref */
452 }
453 
454 static int
455 soclose_sync(struct socket *so, int fflag)
456 {
457 	int error = 0;
458 
459 	if (so->so_pcb == NULL)
460 		goto discard;
461 	if (so->so_state & SS_ISCONNECTED) {
462 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
463 			error = sodisconnect(so);
464 			if (error)
465 				goto drop;
466 		}
467 		if (so->so_options & SO_LINGER) {
468 			if ((so->so_state & SS_ISDISCONNECTING) &&
469 			    (fflag & FNONBLOCK))
470 				goto drop;
471 			while (so->so_state & SS_ISCONNECTED) {
472 				error = tsleep(&so->so_timeo, PCATCH,
473 					       "soclos", so->so_linger * hz);
474 				if (error)
475 					break;
476 			}
477 		}
478 	}
479 drop:
480 	if (so->so_pcb) {
481 		int error2;
482 
483 		error2 = so_pru_detach(so);
484 		if (error == 0)
485 			error = error2;
486 	}
487 discard:
488 	sodiscard(so);
489 	so_pru_sync(so);	/* unpend async sending */
490 	sofree(so);		/* dispose of ref */
491 
492 	return (error);
493 }
494 
495 static void
496 soclose_sofree_async_handler(netmsg_t msg)
497 {
498 	sofree(msg->base.nm_so);
499 }
500 
501 static void
502 soclose_sofree_async(struct socket *so)
503 {
504 	struct netmsg_base *base = &so->so_clomsg;
505 
506 	netmsg_init(base, so, &netisr_apanic_rport, 0,
507 	    soclose_sofree_async_handler);
508 	lwkt_sendmsg(so->so_port, &base->lmsg);
509 }
510 
511 static void
512 soclose_disconn_async_handler(netmsg_t msg)
513 {
514 	struct socket *so = msg->base.nm_so;
515 
516 	if ((so->so_state & SS_ISCONNECTED) &&
517 	    (so->so_state & SS_ISDISCONNECTING) == 0)
518 		so_pru_disconnect_direct(so);
519 
520 	if (so->so_pcb)
521 		so_pru_detach_direct(so);
522 
523 	sodiscard(so);
524 	sofree(so);
525 }
526 
527 static void
528 soclose_disconn_async(struct socket *so)
529 {
530 	struct netmsg_base *base = &so->so_clomsg;
531 
532 	netmsg_init(base, so, &netisr_apanic_rport, 0,
533 	    soclose_disconn_async_handler);
534 	lwkt_sendmsg(so->so_port, &base->lmsg);
535 }
536 
537 static void
538 soclose_detach_async_handler(netmsg_t msg)
539 {
540 	struct socket *so = msg->base.nm_so;
541 
542 	if (so->so_pcb)
543 		so_pru_detach_direct(so);
544 
545 	sodiscard(so);
546 	sofree(so);
547 }
548 
549 static void
550 soclose_detach_async(struct socket *so)
551 {
552 	struct netmsg_base *base = &so->so_clomsg;
553 
554 	netmsg_init(base, so, &netisr_apanic_rport, 0,
555 	    soclose_detach_async_handler);
556 	lwkt_sendmsg(so->so_port, &base->lmsg);
557 }
558 
559 static void
560 soclose_fast(struct socket *so)
561 {
562 	if (so->so_pcb == NULL)
563 		goto discard;
564 
565 	if ((so->so_state & SS_ISCONNECTED) &&
566 	    (so->so_state & SS_ISDISCONNECTING) == 0) {
567 		soclose_disconn_async(so);
568 		return;
569 	}
570 
571 	if (so->so_pcb) {
572 		soclose_detach_async(so);
573 		return;
574 	}
575 
576 discard:
577 	sodiscard(so);
578 	soclose_sofree_async(so);
579 }
580 
581 /*
582  * Abort and destroy a socket.  Only one abort can be in progress
583  * at any given moment.
584  */
585 void
586 soabort(struct socket *so)
587 {
588 	soreference(so);
589 	so_pru_abort(so);
590 }
591 
592 void
593 soaborta(struct socket *so)
594 {
595 	soreference(so);
596 	so_pru_aborta(so);
597 }
598 
599 void
600 soabort_oncpu(struct socket *so)
601 {
602 	soreference(so);
603 	so_pru_abort_oncpu(so);
604 }
605 
606 /*
607  * so is passed in ref'd, which becomes owned by
608  * the cleared SS_NOFDREF flag.
609  */
610 void
611 soaccept_generic(struct socket *so)
612 {
613 	if ((so->so_state & SS_NOFDREF) == 0)
614 		panic("soaccept: !NOFDREF");
615 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
616 }
617 
618 int
619 soaccept(struct socket *so, struct sockaddr **nam)
620 {
621 	int error;
622 
623 	soaccept_generic(so);
624 	error = so_pru_accept(so, nam);
625 	return (error);
626 }
627 
628 int
629 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
630 {
631 	int error;
632 
633 	if (so->so_options & SO_ACCEPTCONN)
634 		return (EOPNOTSUPP);
635 	/*
636 	 * If protocol is connection-based, can only connect once.
637 	 * Otherwise, if connected, try to disconnect first.
638 	 * This allows user to disconnect by connecting to, e.g.,
639 	 * a null address.
640 	 */
641 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
642 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
643 	    (error = sodisconnect(so)))) {
644 		error = EISCONN;
645 	} else {
646 		/*
647 		 * Prevent accumulated error from previous connection
648 		 * from biting us.
649 		 */
650 		so->so_error = 0;
651 		error = so_pru_connect(so, nam, td);
652 	}
653 	return (error);
654 }
655 
656 int
657 soconnect2(struct socket *so1, struct socket *so2)
658 {
659 	int error;
660 
661 	error = so_pru_connect2(so1, so2);
662 	return (error);
663 }
664 
665 int
666 sodisconnect(struct socket *so)
667 {
668 	int error;
669 
670 	if ((so->so_state & SS_ISCONNECTED) == 0) {
671 		error = ENOTCONN;
672 		goto bad;
673 	}
674 	if (so->so_state & SS_ISDISCONNECTING) {
675 		error = EALREADY;
676 		goto bad;
677 	}
678 	error = so_pru_disconnect(so);
679 bad:
680 	return (error);
681 }
682 
683 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
684 /*
685  * Send on a socket.
686  * If send must go all at once and message is larger than
687  * send buffering, then hard error.
688  * Lock against other senders.
689  * If must go all at once and not enough room now, then
690  * inform user that this would block and do nothing.
691  * Otherwise, if nonblocking, send as much as possible.
692  * The data to be sent is described by "uio" if nonzero,
693  * otherwise by the mbuf chain "top" (which must be null
694  * if uio is not).  Data provided in mbuf chain must be small
695  * enough to send all at once.
696  *
697  * Returns nonzero on error, timeout or signal; callers
698  * must check for short counts if EINTR/ERESTART are returned.
699  * Data and control buffers are freed on return.
700  */
701 int
702 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
703 	struct mbuf *top, struct mbuf *control, int flags,
704 	struct thread *td)
705 {
706 	struct mbuf **mp;
707 	struct mbuf *m;
708 	size_t resid;
709 	int space, len;
710 	int clen = 0, error, dontroute, mlen;
711 	int atomic = sosendallatonce(so) || top;
712 	int pru_flags;
713 
714 	if (uio) {
715 		resid = uio->uio_resid;
716 	} else {
717 		resid = (size_t)top->m_pkthdr.len;
718 #ifdef INVARIANTS
719 		len = 0;
720 		for (m = top; m; m = m->m_next)
721 			len += m->m_len;
722 		KKASSERT(top->m_pkthdr.len == len);
723 #endif
724 	}
725 
726 	/*
727 	 * WARNING!  resid is unsigned, space and len are signed.  space
728 	 * 	     can wind up negative if the sockbuf is overcommitted.
729 	 *
730 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
731 	 * type sockets since that's an error.
732 	 */
733 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
734 		error = EINVAL;
735 		goto out;
736 	}
737 
738 	dontroute =
739 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
740 	    (so->so_proto->pr_flags & PR_ATOMIC);
741 	if (td->td_lwp != NULL)
742 		td->td_lwp->lwp_ru.ru_msgsnd++;
743 	if (control)
744 		clen = control->m_len;
745 #define	gotoerr(errcode)	{ error = errcode; goto release; }
746 
747 restart:
748 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
749 	if (error)
750 		goto out;
751 
752 	do {
753 		if (so->so_state & SS_CANTSENDMORE)
754 			gotoerr(EPIPE);
755 		if (so->so_error) {
756 			error = so->so_error;
757 			so->so_error = 0;
758 			goto release;
759 		}
760 		if ((so->so_state & SS_ISCONNECTED) == 0) {
761 			/*
762 			 * `sendto' and `sendmsg' is allowed on a connection-
763 			 * based socket if it supports implied connect.
764 			 * Return ENOTCONN if not connected and no address is
765 			 * supplied.
766 			 */
767 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
768 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
769 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
770 				    !(resid == 0 && clen != 0))
771 					gotoerr(ENOTCONN);
772 			} else if (addr == NULL)
773 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
774 				   ENOTCONN : EDESTADDRREQ);
775 		}
776 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
777 		    clen > so->so_snd.ssb_hiwat) {
778 			gotoerr(EMSGSIZE);
779 		}
780 		space = ssb_space(&so->so_snd);
781 		if (flags & MSG_OOB)
782 			space += 1024;
783 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
784 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
785 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
786 				gotoerr(EWOULDBLOCK);
787 			ssb_unlock(&so->so_snd);
788 			error = ssb_wait(&so->so_snd);
789 			if (error)
790 				goto out;
791 			goto restart;
792 		}
793 		mp = &top;
794 		space -= clen;
795 		do {
796 		    if (uio == NULL) {
797 			/*
798 			 * Data is prepackaged in "top".
799 			 */
800 			resid = 0;
801 			if (flags & MSG_EOR)
802 				top->m_flags |= M_EOR;
803 		    } else do {
804 			if (resid > INT_MAX)
805 				resid = INT_MAX;
806 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
807 				   top == NULL ? M_PKTHDR : 0, &mlen);
808 			if (top == NULL) {
809 				m->m_pkthdr.len = 0;
810 				m->m_pkthdr.rcvif = NULL;
811 			}
812 			len = imin((int)szmin(mlen, resid), space);
813 			if (resid < MINCLSIZE) {
814 				/*
815 				 * For datagram protocols, leave room
816 				 * for protocol headers in first mbuf.
817 				 */
818 				if (atomic && top == NULL && len < mlen)
819 					MH_ALIGN(m, len);
820 			}
821 			space -= len;
822 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
823 			resid = uio->uio_resid;
824 			m->m_len = len;
825 			*mp = m;
826 			top->m_pkthdr.len += len;
827 			if (error)
828 				goto release;
829 			mp = &m->m_next;
830 			if (resid == 0) {
831 				if (flags & MSG_EOR)
832 					top->m_flags |= M_EOR;
833 				break;
834 			}
835 		    } while (space > 0 && atomic);
836 		    if (dontroute)
837 			    so->so_options |= SO_DONTROUTE;
838 		    if (flags & MSG_OOB) {
839 		    	    pru_flags = PRUS_OOB;
840 		    } else if ((flags & MSG_EOF) &&
841 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
842 			       (resid == 0)) {
843 			    /*
844 			     * If the user set MSG_EOF, the protocol
845 			     * understands this flag and nothing left to
846 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
847 			     */
848 		    	    pru_flags = PRUS_EOF;
849 		    } else if (resid > 0 && space > 0) {
850 			    /* If there is more to send, set PRUS_MORETOCOME */
851 		    	    pru_flags = PRUS_MORETOCOME;
852 		    } else {
853 		    	    pru_flags = 0;
854 		    }
855 		    /*
856 		     * XXX all the SS_CANTSENDMORE checks previously
857 		     * done could be out of date.  We could have recieved
858 		     * a reset packet in an interrupt or maybe we slept
859 		     * while doing page faults in uiomove() etc. We could
860 		     * probably recheck again inside the splnet() protection
861 		     * here, but there are probably other places that this
862 		     * also happens.  We must rethink this.
863 		     */
864 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
865 		    if (dontroute)
866 			    so->so_options &= ~SO_DONTROUTE;
867 		    clen = 0;
868 		    control = NULL;
869 		    top = NULL;
870 		    mp = &top;
871 		    if (error)
872 			    goto release;
873 		} while (resid && space > 0);
874 	} while (resid);
875 
876 release:
877 	ssb_unlock(&so->so_snd);
878 out:
879 	if (top)
880 		m_freem(top);
881 	if (control)
882 		m_freem(control);
883 	return (error);
884 }
885 
886 #ifdef INET
887 /*
888  * A specialization of sosend() for UDP based on protocol-specific knowledge:
889  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
890  *	sosendallatonce() returns true,
891  *	the "atomic" variable is true,
892  *	and sosendudp() blocks until space is available for the entire send.
893  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
894  *	PR_IMPLOPCL flags set.
895  *   UDP has no out-of-band data.
896  *   UDP has no control data.
897  *   UDP does not support MSG_EOR.
898  */
899 int
900 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
901 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
902 {
903 	size_t resid;
904 	int error, pru_flags = 0;
905 	int space;
906 
907 	if (td->td_lwp != NULL)
908 		td->td_lwp->lwp_ru.ru_msgsnd++;
909 	if (control)
910 		m_freem(control);
911 
912 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
913 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
914 
915 restart:
916 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
917 	if (error)
918 		goto out;
919 
920 	if (so->so_state & SS_CANTSENDMORE)
921 		gotoerr(EPIPE);
922 	if (so->so_error) {
923 		error = so->so_error;
924 		so->so_error = 0;
925 		goto release;
926 	}
927 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
928 		gotoerr(EDESTADDRREQ);
929 	if (resid > so->so_snd.ssb_hiwat)
930 		gotoerr(EMSGSIZE);
931 	space = ssb_space(&so->so_snd);
932 	if (uio && (space < 0 || (size_t)space < resid)) {
933 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
934 			gotoerr(EWOULDBLOCK);
935 		ssb_unlock(&so->so_snd);
936 		error = ssb_wait(&so->so_snd);
937 		if (error)
938 			goto out;
939 		goto restart;
940 	}
941 
942 	if (uio) {
943 		int hdrlen = max_hdr;
944 
945 		/*
946 		 * We try to optimize out the additional mbuf
947 		 * allocations in M_PREPEND() on output path, e.g.
948 		 * - udp_output(), when it tries to prepend protocol
949 		 *   headers.
950 		 * - Link layer output function, when it tries to
951 		 *   prepend link layer header.
952 		 *
953 		 * This probably will not benefit any data that will
954 		 * be fragmented, so this optimization is only performed
955 		 * when the size of data and max size of protocol+link
956 		 * headers fit into one mbuf cluster.
957 		 */
958 		if (uio->uio_resid > MCLBYTES - hdrlen ||
959 		    !udp_sosend_prepend) {
960 			top = m_uiomove(uio);
961 			if (top == NULL)
962 				goto release;
963 		} else {
964 			int nsize;
965 
966 			top = m_getl(uio->uio_resid + hdrlen, MB_WAIT,
967 			    MT_DATA, M_PKTHDR, &nsize);
968 			KASSERT(nsize >= uio->uio_resid + hdrlen,
969 			    ("sosendudp invalid nsize %d, "
970 			     "resid %zu, hdrlen %d",
971 			     nsize, uio->uio_resid, hdrlen));
972 
973 			top->m_len = uio->uio_resid;
974 			top->m_pkthdr.len = uio->uio_resid;
975 			top->m_data += hdrlen;
976 
977 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
978 			if (error)
979 				goto out;
980 		}
981 	}
982 
983 	if (flags & MSG_DONTROUTE)
984 		pru_flags |= PRUS_DONTROUTE;
985 
986 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
987 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
988 		error = 0;
989 	} else {
990 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
991 	}
992 	top = NULL;		/* sent or freed in lower layer */
993 
994 release:
995 	ssb_unlock(&so->so_snd);
996 out:
997 	if (top)
998 		m_freem(top);
999 	return (error);
1000 }
1001 
1002 int
1003 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1004 	struct mbuf *top, struct mbuf *control, int flags,
1005 	struct thread *td)
1006 {
1007 	struct mbuf **mp;
1008 	struct mbuf *m;
1009 	size_t resid;
1010 	int space, len;
1011 	int error, mlen;
1012 	int allatonce;
1013 	int pru_flags;
1014 
1015 	if (uio) {
1016 		KKASSERT(top == NULL);
1017 		allatonce = 0;
1018 		resid = uio->uio_resid;
1019 	} else {
1020 		allatonce = 1;
1021 		resid = (size_t)top->m_pkthdr.len;
1022 #ifdef INVARIANTS
1023 		len = 0;
1024 		for (m = top; m; m = m->m_next)
1025 			len += m->m_len;
1026 		KKASSERT(top->m_pkthdr.len == len);
1027 #endif
1028 	}
1029 
1030 	/*
1031 	 * WARNING!  resid is unsigned, space and len are signed.  space
1032 	 * 	     can wind up negative if the sockbuf is overcommitted.
1033 	 *
1034 	 * Also check to make sure that MSG_EOR isn't used on TCP
1035 	 */
1036 	if (flags & MSG_EOR) {
1037 		error = EINVAL;
1038 		goto out;
1039 	}
1040 
1041 	if (control) {
1042 		/* TCP doesn't do control messages (rights, creds, etc) */
1043 		if (control->m_len) {
1044 			error = EINVAL;
1045 			goto out;
1046 		}
1047 		m_freem(control);	/* empty control, just free it */
1048 		control = NULL;
1049 	}
1050 
1051 	if (td->td_lwp != NULL)
1052 		td->td_lwp->lwp_ru.ru_msgsnd++;
1053 
1054 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1055 
1056 restart:
1057 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1058 	if (error)
1059 		goto out;
1060 
1061 	do {
1062 		if (so->so_state & SS_CANTSENDMORE)
1063 			gotoerr(EPIPE);
1064 		if (so->so_error) {
1065 			error = so->so_error;
1066 			so->so_error = 0;
1067 			goto release;
1068 		}
1069 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1070 		    (so->so_state & SS_ISCONFIRMING) == 0)
1071 			gotoerr(ENOTCONN);
1072 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1073 			gotoerr(EMSGSIZE);
1074 
1075 		space = ssb_space_prealloc(&so->so_snd);
1076 		if (flags & MSG_OOB)
1077 			space += 1024;
1078 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1079 		    space < so->so_snd.ssb_lowat) {
1080 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1081 				gotoerr(EWOULDBLOCK);
1082 			ssb_unlock(&so->so_snd);
1083 			error = ssb_wait(&so->so_snd);
1084 			if (error)
1085 				goto out;
1086 			goto restart;
1087 		}
1088 		mp = &top;
1089 		do {
1090 		    int cnt = 0, async = 0;
1091 
1092 		    if (uio == NULL) {
1093 			/*
1094 			 * Data is prepackaged in "top".
1095 			 */
1096 			resid = 0;
1097 		    } else do {
1098 			if (resid > INT_MAX)
1099 				resid = INT_MAX;
1100 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
1101 				   top == NULL ? M_PKTHDR : 0, &mlen);
1102 			if (top == NULL) {
1103 				m->m_pkthdr.len = 0;
1104 				m->m_pkthdr.rcvif = NULL;
1105 			}
1106 			len = imin((int)szmin(mlen, resid), space);
1107 			space -= len;
1108 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1109 			resid = uio->uio_resid;
1110 			m->m_len = len;
1111 			*mp = m;
1112 			top->m_pkthdr.len += len;
1113 			if (error)
1114 				goto release;
1115 			mp = &m->m_next;
1116 			if (resid == 0)
1117 				break;
1118 			++cnt;
1119 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1120 
1121 		    if (tcp_sosend_async)
1122 			    async = 1;
1123 
1124 		    if (flags & MSG_OOB) {
1125 		    	    pru_flags = PRUS_OOB;
1126 			    async = 0;
1127 		    } else if ((flags & MSG_EOF) && resid == 0) {
1128 			    pru_flags = PRUS_EOF;
1129 		    } else if (resid > 0 && space > 0) {
1130 			    /* If there is more to send, set PRUS_MORETOCOME */
1131 		    	    pru_flags = PRUS_MORETOCOME;
1132 			    async = 1;
1133 		    } else {
1134 		    	    pru_flags = 0;
1135 		    }
1136 
1137 		    if (flags & MSG_SYNC)
1138 			    async = 0;
1139 
1140 		    /*
1141 		     * XXX all the SS_CANTSENDMORE checks previously
1142 		     * done could be out of date.  We could have recieved
1143 		     * a reset packet in an interrupt or maybe we slept
1144 		     * while doing page faults in uiomove() etc. We could
1145 		     * probably recheck again inside the splnet() protection
1146 		     * here, but there are probably other places that this
1147 		     * also happens.  We must rethink this.
1148 		     */
1149 		    for (m = top; m; m = m->m_next)
1150 			    ssb_preallocstream(&so->so_snd, m);
1151 		    if (!async) {
1152 			    error = so_pru_send(so, pru_flags, top,
1153 			        NULL, NULL, td);
1154 		    } else {
1155 			    so_pru_send_async(so, pru_flags, top,
1156 			        NULL, NULL, td);
1157 			    error = 0;
1158 		    }
1159 
1160 		    top = NULL;
1161 		    mp = &top;
1162 		    if (error)
1163 			    goto release;
1164 		} while (resid && space > 0);
1165 	} while (resid);
1166 
1167 release:
1168 	ssb_unlock(&so->so_snd);
1169 out:
1170 	if (top)
1171 		m_freem(top);
1172 	if (control)
1173 		m_freem(control);
1174 	return (error);
1175 }
1176 #endif
1177 
1178 /*
1179  * Implement receive operations on a socket.
1180  *
1181  * We depend on the way that records are added to the signalsockbuf
1182  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1183  * must begin with an address if the protocol so specifies,
1184  * followed by an optional mbuf or mbufs containing ancillary data,
1185  * and then zero or more mbufs of data.
1186  *
1187  * Although the signalsockbuf is locked, new data may still be appended.
1188  * A token inside the ssb_lock deals with MP issues and still allows
1189  * the network to access the socket if we block in a uio.
1190  *
1191  * The caller may receive the data as a single mbuf chain by supplying
1192  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1193  * only for the count in uio_resid.
1194  */
1195 int
1196 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1197 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1198 {
1199 	struct mbuf *m, *n;
1200 	struct mbuf *free_chain = NULL;
1201 	int flags, len, error, offset;
1202 	struct protosw *pr = so->so_proto;
1203 	int moff, type = 0;
1204 	size_t resid, orig_resid;
1205 
1206 	if (uio)
1207 		resid = uio->uio_resid;
1208 	else
1209 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1210 	orig_resid = resid;
1211 
1212 	if (psa)
1213 		*psa = NULL;
1214 	if (controlp)
1215 		*controlp = NULL;
1216 	if (flagsp)
1217 		flags = *flagsp &~ MSG_EOR;
1218 	else
1219 		flags = 0;
1220 	if (flags & MSG_OOB) {
1221 		m = m_get(MB_WAIT, MT_DATA);
1222 		if (m == NULL)
1223 			return (ENOBUFS);
1224 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1225 		if (error)
1226 			goto bad;
1227 		if (sio) {
1228 			do {
1229 				sbappend(sio, m);
1230 				KKASSERT(resid >= (size_t)m->m_len);
1231 				resid -= (size_t)m->m_len;
1232 			} while (resid > 0 && m);
1233 		} else {
1234 			do {
1235 				uio->uio_resid = resid;
1236 				error = uiomove(mtod(m, caddr_t),
1237 						(int)szmin(resid, m->m_len),
1238 						uio);
1239 				resid = uio->uio_resid;
1240 				m = m_free(m);
1241 			} while (uio->uio_resid && error == 0 && m);
1242 		}
1243 bad:
1244 		if (m)
1245 			m_freem(m);
1246 		return (error);
1247 	}
1248 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1249 		so_pru_rcvd(so, 0);
1250 
1251 	/*
1252 	 * The token interlocks against the protocol thread while
1253 	 * ssb_lock is a blocking lock against other userland entities.
1254 	 */
1255 	lwkt_gettoken(&so->so_rcv.ssb_token);
1256 restart:
1257 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1258 	if (error)
1259 		goto done;
1260 
1261 	m = so->so_rcv.ssb_mb;
1262 	/*
1263 	 * If we have less data than requested, block awaiting more
1264 	 * (subject to any timeout) if:
1265 	 *   1. the current count is less than the low water mark, or
1266 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1267 	 *	receive operation at once if we block (resid <= hiwat).
1268 	 *   3. MSG_DONTWAIT is not set
1269 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1270 	 * we have to do the receive in sections, and thus risk returning
1271 	 * a short count if a timeout or signal occurs after we start.
1272 	 */
1273 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1274 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1275 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1276 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1277 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1278 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1279 		if (so->so_error) {
1280 			if (m)
1281 				goto dontblock;
1282 			error = so->so_error;
1283 			if ((flags & MSG_PEEK) == 0)
1284 				so->so_error = 0;
1285 			goto release;
1286 		}
1287 		if (so->so_state & SS_CANTRCVMORE) {
1288 			if (m)
1289 				goto dontblock;
1290 			else
1291 				goto release;
1292 		}
1293 		for (; m; m = m->m_next) {
1294 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1295 				m = so->so_rcv.ssb_mb;
1296 				goto dontblock;
1297 			}
1298 		}
1299 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1300 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1301 			error = ENOTCONN;
1302 			goto release;
1303 		}
1304 		if (resid == 0)
1305 			goto release;
1306 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1307 			error = EWOULDBLOCK;
1308 			goto release;
1309 		}
1310 		ssb_unlock(&so->so_rcv);
1311 		error = ssb_wait(&so->so_rcv);
1312 		if (error)
1313 			goto done;
1314 		goto restart;
1315 	}
1316 dontblock:
1317 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1318 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1319 
1320 	/*
1321 	 * note: m should be == sb_mb here.  Cache the next record while
1322 	 * cleaning up.  Note that calling m_free*() will break out critical
1323 	 * section.
1324 	 */
1325 	KKASSERT(m == so->so_rcv.ssb_mb);
1326 
1327 	/*
1328 	 * Skip any address mbufs prepending the record.
1329 	 */
1330 	if (pr->pr_flags & PR_ADDR) {
1331 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1332 		orig_resid = 0;
1333 		if (psa)
1334 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1335 		if (flags & MSG_PEEK)
1336 			m = m->m_next;
1337 		else
1338 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1339 	}
1340 
1341 	/*
1342 	 * Skip any control mbufs prepending the record.
1343 	 */
1344 #ifdef SCTP
1345 	if (pr->pr_flags & PR_ADDR_OPT) {
1346 		/*
1347 		 * For SCTP we may be getting a
1348 		 * whole message OR a partial delivery.
1349 		 */
1350 		if (m && m->m_type == MT_SONAME) {
1351 			orig_resid = 0;
1352 			if (psa)
1353 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1354 			if (flags & MSG_PEEK)
1355 				m = m->m_next;
1356 			else
1357 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1358 		}
1359 	}
1360 #endif /* SCTP */
1361 	while (m && m->m_type == MT_CONTROL && error == 0) {
1362 		if (flags & MSG_PEEK) {
1363 			if (controlp)
1364 				*controlp = m_copy(m, 0, m->m_len);
1365 			m = m->m_next;	/* XXX race */
1366 		} else {
1367 			if (controlp) {
1368 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1369 				if (pr->pr_domain->dom_externalize &&
1370 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1371 				    SCM_RIGHTS)
1372 				   error = (*pr->pr_domain->dom_externalize)(m);
1373 				*controlp = m;
1374 				m = n;
1375 			} else {
1376 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1377 			}
1378 		}
1379 		if (controlp && *controlp) {
1380 			orig_resid = 0;
1381 			controlp = &(*controlp)->m_next;
1382 		}
1383 	}
1384 
1385 	/*
1386 	 * flag OOB data.
1387 	 */
1388 	if (m) {
1389 		type = m->m_type;
1390 		if (type == MT_OOBDATA)
1391 			flags |= MSG_OOB;
1392 	}
1393 
1394 	/*
1395 	 * Copy to the UIO or mbuf return chain (*mp).
1396 	 */
1397 	moff = 0;
1398 	offset = 0;
1399 	while (m && resid > 0 && error == 0) {
1400 		if (m->m_type == MT_OOBDATA) {
1401 			if (type != MT_OOBDATA)
1402 				break;
1403 		} else if (type == MT_OOBDATA)
1404 			break;
1405 		else
1406 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1407 			("receive 3"));
1408 		soclrstate(so, SS_RCVATMARK);
1409 		len = (resid > INT_MAX) ? INT_MAX : resid;
1410 		if (so->so_oobmark && len > so->so_oobmark - offset)
1411 			len = so->so_oobmark - offset;
1412 		if (len > m->m_len - moff)
1413 			len = m->m_len - moff;
1414 
1415 		/*
1416 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1417 		 * The SIO is dealt with when we eat the mbuf, but deal
1418 		 * with the resid here either way.
1419 		 */
1420 		if (uio) {
1421 			uio->uio_resid = resid;
1422 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1423 			resid = uio->uio_resid;
1424 			if (error)
1425 				goto release;
1426 		} else {
1427 			resid -= (size_t)len;
1428 		}
1429 
1430 		/*
1431 		 * Eat the entire mbuf or just a piece of it
1432 		 */
1433 		if (len == m->m_len - moff) {
1434 			if (m->m_flags & M_EOR)
1435 				flags |= MSG_EOR;
1436 #ifdef SCTP
1437 			if (m->m_flags & M_NOTIFICATION)
1438 				flags |= MSG_NOTIFICATION;
1439 #endif /* SCTP */
1440 			if (flags & MSG_PEEK) {
1441 				m = m->m_next;
1442 				moff = 0;
1443 			} else {
1444 				if (sio) {
1445 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1446 					sbappend(sio, m);
1447 					m = n;
1448 				} else {
1449 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1450 				}
1451 			}
1452 		} else {
1453 			if (flags & MSG_PEEK) {
1454 				moff += len;
1455 			} else {
1456 				if (sio) {
1457 					n = m_copym(m, 0, len, MB_WAIT);
1458 					if (n)
1459 						sbappend(sio, n);
1460 				}
1461 				m->m_data += len;
1462 				m->m_len -= len;
1463 				so->so_rcv.ssb_cc -= len;
1464 			}
1465 		}
1466 		if (so->so_oobmark) {
1467 			if ((flags & MSG_PEEK) == 0) {
1468 				so->so_oobmark -= len;
1469 				if (so->so_oobmark == 0) {
1470 					sosetstate(so, SS_RCVATMARK);
1471 					break;
1472 				}
1473 			} else {
1474 				offset += len;
1475 				if (offset == so->so_oobmark)
1476 					break;
1477 			}
1478 		}
1479 		if (flags & MSG_EOR)
1480 			break;
1481 		/*
1482 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1483 		 * we must not quit until resid == 0 or an error
1484 		 * termination.  If a signal/timeout occurs, return
1485 		 * with a short count but without error.
1486 		 * Keep signalsockbuf locked against other readers.
1487 		 */
1488 		while ((flags & MSG_WAITALL) && m == NULL &&
1489 		       resid > 0 && !sosendallatonce(so) &&
1490 		       so->so_rcv.ssb_mb == NULL) {
1491 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1492 				break;
1493 			/*
1494 			 * The window might have closed to zero, make
1495 			 * sure we send an ack now that we've drained
1496 			 * the buffer or we might end up blocking until
1497 			 * the idle takes over (5 seconds).
1498 			 */
1499 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1500 				so_pru_rcvd(so, flags);
1501 			error = ssb_wait(&so->so_rcv);
1502 			if (error) {
1503 				ssb_unlock(&so->so_rcv);
1504 				error = 0;
1505 				goto done;
1506 			}
1507 			m = so->so_rcv.ssb_mb;
1508 		}
1509 	}
1510 
1511 	/*
1512 	 * If an atomic read was requested but unread data still remains
1513 	 * in the record, set MSG_TRUNC.
1514 	 */
1515 	if (m && pr->pr_flags & PR_ATOMIC)
1516 		flags |= MSG_TRUNC;
1517 
1518 	/*
1519 	 * Cleanup.  If an atomic read was requested drop any unread data.
1520 	 */
1521 	if ((flags & MSG_PEEK) == 0) {
1522 		if (m && (pr->pr_flags & PR_ATOMIC))
1523 			sbdroprecord(&so->so_rcv.sb);
1524 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1525 			so_pru_rcvd(so, flags);
1526 	}
1527 
1528 	if (orig_resid == resid && orig_resid &&
1529 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1530 		ssb_unlock(&so->so_rcv);
1531 		goto restart;
1532 	}
1533 
1534 	if (flagsp)
1535 		*flagsp |= flags;
1536 release:
1537 	ssb_unlock(&so->so_rcv);
1538 done:
1539 	lwkt_reltoken(&so->so_rcv.ssb_token);
1540 	if (free_chain)
1541 		m_freem(free_chain);
1542 	return (error);
1543 }
1544 
1545 int
1546 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1547 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1548 {
1549 	struct mbuf *m, *n;
1550 	struct mbuf *free_chain = NULL;
1551 	int flags, len, error, offset;
1552 	struct protosw *pr = so->so_proto;
1553 	int moff;
1554 	size_t resid, orig_resid;
1555 
1556 	if (uio)
1557 		resid = uio->uio_resid;
1558 	else
1559 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1560 	orig_resid = resid;
1561 
1562 	if (psa)
1563 		*psa = NULL;
1564 	if (controlp)
1565 		*controlp = NULL;
1566 	if (flagsp)
1567 		flags = *flagsp &~ MSG_EOR;
1568 	else
1569 		flags = 0;
1570 	if (flags & MSG_OOB) {
1571 		m = m_get(MB_WAIT, MT_DATA);
1572 		if (m == NULL)
1573 			return (ENOBUFS);
1574 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1575 		if (error)
1576 			goto bad;
1577 		if (sio) {
1578 			do {
1579 				sbappend(sio, m);
1580 				KKASSERT(resid >= (size_t)m->m_len);
1581 				resid -= (size_t)m->m_len;
1582 			} while (resid > 0 && m);
1583 		} else {
1584 			do {
1585 				uio->uio_resid = resid;
1586 				error = uiomove(mtod(m, caddr_t),
1587 						(int)szmin(resid, m->m_len),
1588 						uio);
1589 				resid = uio->uio_resid;
1590 				m = m_free(m);
1591 			} while (uio->uio_resid && error == 0 && m);
1592 		}
1593 bad:
1594 		if (m)
1595 			m_freem(m);
1596 		return (error);
1597 	}
1598 
1599 	/*
1600 	 * The token interlocks against the protocol thread while
1601 	 * ssb_lock is a blocking lock against other userland entities.
1602 	 */
1603 	lwkt_gettoken(&so->so_rcv.ssb_token);
1604 restart:
1605 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1606 	if (error)
1607 		goto done;
1608 
1609 	m = so->so_rcv.ssb_mb;
1610 	/*
1611 	 * If we have less data than requested, block awaiting more
1612 	 * (subject to any timeout) if:
1613 	 *   1. the current count is less than the low water mark, or
1614 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1615 	 *	receive operation at once if we block (resid <= hiwat).
1616 	 *   3. MSG_DONTWAIT is not set
1617 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1618 	 * we have to do the receive in sections, and thus risk returning
1619 	 * a short count if a timeout or signal occurs after we start.
1620 	 */
1621 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1622 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1623 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1624 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1625 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1626 		if (so->so_error) {
1627 			if (m)
1628 				goto dontblock;
1629 			error = so->so_error;
1630 			if ((flags & MSG_PEEK) == 0)
1631 				so->so_error = 0;
1632 			goto release;
1633 		}
1634 		if (so->so_state & SS_CANTRCVMORE) {
1635 			if (m)
1636 				goto dontblock;
1637 			else
1638 				goto release;
1639 		}
1640 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1641 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1642 			error = ENOTCONN;
1643 			goto release;
1644 		}
1645 		if (resid == 0)
1646 			goto release;
1647 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1648 			error = EWOULDBLOCK;
1649 			goto release;
1650 		}
1651 		ssb_unlock(&so->so_rcv);
1652 		error = ssb_wait(&so->so_rcv);
1653 		if (error)
1654 			goto done;
1655 		goto restart;
1656 	}
1657 dontblock:
1658 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1659 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1660 
1661 	/*
1662 	 * note: m should be == sb_mb here.  Cache the next record while
1663 	 * cleaning up.  Note that calling m_free*() will break out critical
1664 	 * section.
1665 	 */
1666 	KKASSERT(m == so->so_rcv.ssb_mb);
1667 
1668 	/*
1669 	 * Copy to the UIO or mbuf return chain (*mp).
1670 	 */
1671 	moff = 0;
1672 	offset = 0;
1673 	while (m && resid > 0 && error == 0) {
1674 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1675 		    ("receive 3"));
1676 
1677 		soclrstate(so, SS_RCVATMARK);
1678 		len = (resid > INT_MAX) ? INT_MAX : resid;
1679 		if (so->so_oobmark && len > so->so_oobmark - offset)
1680 			len = so->so_oobmark - offset;
1681 		if (len > m->m_len - moff)
1682 			len = m->m_len - moff;
1683 
1684 		/*
1685 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1686 		 * The SIO is dealt with when we eat the mbuf, but deal
1687 		 * with the resid here either way.
1688 		 */
1689 		if (uio) {
1690 			uio->uio_resid = resid;
1691 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1692 			resid = uio->uio_resid;
1693 			if (error)
1694 				goto release;
1695 		} else {
1696 			resid -= (size_t)len;
1697 		}
1698 
1699 		/*
1700 		 * Eat the entire mbuf or just a piece of it
1701 		 */
1702 		if (len == m->m_len - moff) {
1703 			if (flags & MSG_PEEK) {
1704 				m = m->m_next;
1705 				moff = 0;
1706 			} else {
1707 				if (sio) {
1708 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1709 					sbappend(sio, m);
1710 					m = n;
1711 				} else {
1712 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1713 				}
1714 			}
1715 		} else {
1716 			if (flags & MSG_PEEK) {
1717 				moff += len;
1718 			} else {
1719 				if (sio) {
1720 					n = m_copym(m, 0, len, MB_WAIT);
1721 					if (n)
1722 						sbappend(sio, n);
1723 				}
1724 				m->m_data += len;
1725 				m->m_len -= len;
1726 				so->so_rcv.ssb_cc -= len;
1727 			}
1728 		}
1729 		if (so->so_oobmark) {
1730 			if ((flags & MSG_PEEK) == 0) {
1731 				so->so_oobmark -= len;
1732 				if (so->so_oobmark == 0) {
1733 					sosetstate(so, SS_RCVATMARK);
1734 					break;
1735 				}
1736 			} else {
1737 				offset += len;
1738 				if (offset == so->so_oobmark)
1739 					break;
1740 			}
1741 		}
1742 		/*
1743 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1744 		 * we must not quit until resid == 0 or an error
1745 		 * termination.  If a signal/timeout occurs, return
1746 		 * with a short count but without error.
1747 		 * Keep signalsockbuf locked against other readers.
1748 		 */
1749 		while ((flags & MSG_WAITALL) && m == NULL &&
1750 		       resid > 0 && !sosendallatonce(so) &&
1751 		       so->so_rcv.ssb_mb == NULL) {
1752 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1753 				break;
1754 			/*
1755 			 * The window might have closed to zero, make
1756 			 * sure we send an ack now that we've drained
1757 			 * the buffer or we might end up blocking until
1758 			 * the idle takes over (5 seconds).
1759 			 */
1760 			if (so->so_pcb)
1761 				so_pru_rcvd_async(so);
1762 			error = ssb_wait(&so->so_rcv);
1763 			if (error) {
1764 				ssb_unlock(&so->so_rcv);
1765 				error = 0;
1766 				goto done;
1767 			}
1768 			m = so->so_rcv.ssb_mb;
1769 		}
1770 	}
1771 
1772 	/*
1773 	 * Cleanup.  If an atomic read was requested drop any unread data.
1774 	 */
1775 	if ((flags & MSG_PEEK) == 0) {
1776 		if (so->so_pcb)
1777 			so_pru_rcvd_async(so);
1778 	}
1779 
1780 	if (orig_resid == resid && orig_resid &&
1781 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1782 		ssb_unlock(&so->so_rcv);
1783 		goto restart;
1784 	}
1785 
1786 	if (flagsp)
1787 		*flagsp |= flags;
1788 release:
1789 	ssb_unlock(&so->so_rcv);
1790 done:
1791 	lwkt_reltoken(&so->so_rcv.ssb_token);
1792 	if (free_chain)
1793 		m_freem(free_chain);
1794 	return (error);
1795 }
1796 
1797 /*
1798  * Shut a socket down.  Note that we do not get a frontend lock as we
1799  * want to be able to shut the socket down even if another thread is
1800  * blocked in a read(), thus waking it up.
1801  */
1802 int
1803 soshutdown(struct socket *so, int how)
1804 {
1805 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1806 		return (EINVAL);
1807 
1808 	if (how != SHUT_WR) {
1809 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1810 		sorflush(so);
1811 		/*ssb_unlock(&so->so_rcv);*/
1812 	}
1813 	if (how != SHUT_RD)
1814 		return (so_pru_shutdown(so));
1815 	return (0);
1816 }
1817 
1818 void
1819 sorflush(struct socket *so)
1820 {
1821 	struct signalsockbuf *ssb = &so->so_rcv;
1822 	struct protosw *pr = so->so_proto;
1823 	struct signalsockbuf asb;
1824 
1825 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1826 
1827 	lwkt_gettoken(&ssb->ssb_token);
1828 	socantrcvmore(so);
1829 	asb = *ssb;
1830 
1831 	/*
1832 	 * Can't just blow up the ssb structure here
1833 	 */
1834 	bzero(&ssb->sb, sizeof(ssb->sb));
1835 	ssb->ssb_timeo = 0;
1836 	ssb->ssb_lowat = 0;
1837 	ssb->ssb_hiwat = 0;
1838 	ssb->ssb_mbmax = 0;
1839 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1840 
1841 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1842 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1843 	ssb_release(&asb, so);
1844 
1845 	lwkt_reltoken(&ssb->ssb_token);
1846 }
1847 
1848 #ifdef INET
1849 static int
1850 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1851 {
1852 	struct accept_filter_arg	*afap = NULL;
1853 	struct accept_filter	*afp;
1854 	struct so_accf	*af = so->so_accf;
1855 	int	error = 0;
1856 
1857 	/* do not set/remove accept filters on non listen sockets */
1858 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1859 		error = EINVAL;
1860 		goto out;
1861 	}
1862 
1863 	/* removing the filter */
1864 	if (sopt == NULL) {
1865 		if (af != NULL) {
1866 			if (af->so_accept_filter != NULL &&
1867 				af->so_accept_filter->accf_destroy != NULL) {
1868 				af->so_accept_filter->accf_destroy(so);
1869 			}
1870 			if (af->so_accept_filter_str != NULL) {
1871 				kfree(af->so_accept_filter_str, M_ACCF);
1872 			}
1873 			kfree(af, M_ACCF);
1874 			so->so_accf = NULL;
1875 		}
1876 		so->so_options &= ~SO_ACCEPTFILTER;
1877 		return (0);
1878 	}
1879 	/* adding a filter */
1880 	/* must remove previous filter first */
1881 	if (af != NULL) {
1882 		error = EINVAL;
1883 		goto out;
1884 	}
1885 	/* don't put large objects on the kernel stack */
1886 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
1887 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1888 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1889 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1890 	if (error)
1891 		goto out;
1892 	afp = accept_filt_get(afap->af_name);
1893 	if (afp == NULL) {
1894 		error = ENOENT;
1895 		goto out;
1896 	}
1897 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1898 	if (afp->accf_create != NULL) {
1899 		if (afap->af_name[0] != '\0') {
1900 			int len = strlen(afap->af_name) + 1;
1901 
1902 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
1903 							   M_WAITOK);
1904 			strcpy(af->so_accept_filter_str, afap->af_name);
1905 		}
1906 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1907 		if (af->so_accept_filter_arg == NULL) {
1908 			kfree(af->so_accept_filter_str, M_ACCF);
1909 			kfree(af, M_ACCF);
1910 			so->so_accf = NULL;
1911 			error = EINVAL;
1912 			goto out;
1913 		}
1914 	}
1915 	af->so_accept_filter = afp;
1916 	so->so_accf = af;
1917 	so->so_options |= SO_ACCEPTFILTER;
1918 out:
1919 	if (afap != NULL)
1920 		kfree(afap, M_TEMP);
1921 	return (error);
1922 }
1923 #endif /* INET */
1924 
1925 /*
1926  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1927  * an additional variant to handle the case where the option value needs
1928  * to be some kind of integer, but not a specific size.
1929  * In addition to their use here, these functions are also called by the
1930  * protocol-level pr_ctloutput() routines.
1931  */
1932 int
1933 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1934 {
1935 	return soopt_to_kbuf(sopt, buf, len, minlen);
1936 }
1937 
1938 int
1939 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1940 {
1941 	size_t	valsize;
1942 
1943 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1944 	KKASSERT(kva_p(buf));
1945 
1946 	/*
1947 	 * If the user gives us more than we wanted, we ignore it,
1948 	 * but if we don't get the minimum length the caller
1949 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1950 	 * is set to however much we actually retrieved.
1951 	 */
1952 	if ((valsize = sopt->sopt_valsize) < minlen)
1953 		return EINVAL;
1954 	if (valsize > len)
1955 		sopt->sopt_valsize = valsize = len;
1956 
1957 	bcopy(sopt->sopt_val, buf, valsize);
1958 	return 0;
1959 }
1960 
1961 
1962 int
1963 sosetopt(struct socket *so, struct sockopt *sopt)
1964 {
1965 	int	error, optval;
1966 	struct	linger l;
1967 	struct	timeval tv;
1968 	u_long  val;
1969 	struct signalsockbuf *sotmp;
1970 
1971 	error = 0;
1972 	sopt->sopt_dir = SOPT_SET;
1973 	if (sopt->sopt_level != SOL_SOCKET) {
1974 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1975 			return (so_pr_ctloutput(so, sopt));
1976 		}
1977 		error = ENOPROTOOPT;
1978 	} else {
1979 		switch (sopt->sopt_name) {
1980 #ifdef INET
1981 		case SO_ACCEPTFILTER:
1982 			error = do_setopt_accept_filter(so, sopt);
1983 			if (error)
1984 				goto bad;
1985 			break;
1986 #endif /* INET */
1987 		case SO_LINGER:
1988 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1989 			if (error)
1990 				goto bad;
1991 
1992 			so->so_linger = l.l_linger;
1993 			if (l.l_onoff)
1994 				so->so_options |= SO_LINGER;
1995 			else
1996 				so->so_options &= ~SO_LINGER;
1997 			break;
1998 
1999 		case SO_DEBUG:
2000 		case SO_KEEPALIVE:
2001 		case SO_DONTROUTE:
2002 		case SO_USELOOPBACK:
2003 		case SO_BROADCAST:
2004 		case SO_REUSEADDR:
2005 		case SO_REUSEPORT:
2006 		case SO_OOBINLINE:
2007 		case SO_TIMESTAMP:
2008 		case SO_NOSIGPIPE:
2009 			error = sooptcopyin(sopt, &optval, sizeof optval,
2010 					    sizeof optval);
2011 			if (error)
2012 				goto bad;
2013 			if (optval)
2014 				so->so_options |= sopt->sopt_name;
2015 			else
2016 				so->so_options &= ~sopt->sopt_name;
2017 			break;
2018 
2019 		case SO_SNDBUF:
2020 		case SO_RCVBUF:
2021 		case SO_SNDLOWAT:
2022 		case SO_RCVLOWAT:
2023 			error = sooptcopyin(sopt, &optval, sizeof optval,
2024 					    sizeof optval);
2025 			if (error)
2026 				goto bad;
2027 
2028 			/*
2029 			 * Values < 1 make no sense for any of these
2030 			 * options, so disallow them.
2031 			 */
2032 			if (optval < 1) {
2033 				error = EINVAL;
2034 				goto bad;
2035 			}
2036 
2037 			switch (sopt->sopt_name) {
2038 			case SO_SNDBUF:
2039 			case SO_RCVBUF:
2040 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2041 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2042 				    so,
2043 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2044 					error = ENOBUFS;
2045 					goto bad;
2046 				}
2047 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2048 						&so->so_snd : &so->so_rcv;
2049 				atomic_clear_int(&sotmp->ssb_flags,
2050 						 SSB_AUTOSIZE);
2051 				break;
2052 
2053 			/*
2054 			 * Make sure the low-water is never greater than
2055 			 * the high-water.
2056 			 */
2057 			case SO_SNDLOWAT:
2058 				so->so_snd.ssb_lowat =
2059 				    (optval > so->so_snd.ssb_hiwat) ?
2060 				    so->so_snd.ssb_hiwat : optval;
2061 				atomic_clear_int(&so->so_snd.ssb_flags,
2062 						 SSB_AUTOLOWAT);
2063 				break;
2064 			case SO_RCVLOWAT:
2065 				so->so_rcv.ssb_lowat =
2066 				    (optval > so->so_rcv.ssb_hiwat) ?
2067 				    so->so_rcv.ssb_hiwat : optval;
2068 				atomic_clear_int(&so->so_rcv.ssb_flags,
2069 						 SSB_AUTOLOWAT);
2070 				break;
2071 			}
2072 			break;
2073 
2074 		case SO_SNDTIMEO:
2075 		case SO_RCVTIMEO:
2076 			error = sooptcopyin(sopt, &tv, sizeof tv,
2077 					    sizeof tv);
2078 			if (error)
2079 				goto bad;
2080 
2081 			/* assert(hz > 0); */
2082 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2083 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2084 				error = EDOM;
2085 				goto bad;
2086 			}
2087 			/* assert(tick > 0); */
2088 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2089 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2090 			if (val > INT_MAX) {
2091 				error = EDOM;
2092 				goto bad;
2093 			}
2094 			if (val == 0 && tv.tv_usec != 0)
2095 				val = 1;
2096 
2097 			switch (sopt->sopt_name) {
2098 			case SO_SNDTIMEO:
2099 				so->so_snd.ssb_timeo = val;
2100 				break;
2101 			case SO_RCVTIMEO:
2102 				so->so_rcv.ssb_timeo = val;
2103 				break;
2104 			}
2105 			break;
2106 		default:
2107 			error = ENOPROTOOPT;
2108 			break;
2109 		}
2110 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2111 			(void) so_pr_ctloutput(so, sopt);
2112 		}
2113 	}
2114 bad:
2115 	return (error);
2116 }
2117 
2118 /* Helper routine for getsockopt */
2119 int
2120 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2121 {
2122 	soopt_from_kbuf(sopt, buf, len);
2123 	return 0;
2124 }
2125 
2126 void
2127 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2128 {
2129 	size_t	valsize;
2130 
2131 	if (len == 0) {
2132 		sopt->sopt_valsize = 0;
2133 		return;
2134 	}
2135 
2136 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2137 	KKASSERT(kva_p(buf));
2138 
2139 	/*
2140 	 * Documented get behavior is that we always return a value,
2141 	 * possibly truncated to fit in the user's buffer.
2142 	 * Traditional behavior is that we always tell the user
2143 	 * precisely how much we copied, rather than something useful
2144 	 * like the total amount we had available for her.
2145 	 * Note that this interface is not idempotent; the entire answer must
2146 	 * generated ahead of time.
2147 	 */
2148 	valsize = szmin(len, sopt->sopt_valsize);
2149 	sopt->sopt_valsize = valsize;
2150 	if (sopt->sopt_val != 0) {
2151 		bcopy(buf, sopt->sopt_val, valsize);
2152 	}
2153 }
2154 
2155 int
2156 sogetopt(struct socket *so, struct sockopt *sopt)
2157 {
2158 	int	error, optval;
2159 	long	optval_l;
2160 	struct	linger l;
2161 	struct	timeval tv;
2162 #ifdef INET
2163 	struct accept_filter_arg *afap;
2164 #endif
2165 
2166 	error = 0;
2167 	sopt->sopt_dir = SOPT_GET;
2168 	if (sopt->sopt_level != SOL_SOCKET) {
2169 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2170 			return (so_pr_ctloutput(so, sopt));
2171 		} else
2172 			return (ENOPROTOOPT);
2173 	} else {
2174 		switch (sopt->sopt_name) {
2175 #ifdef INET
2176 		case SO_ACCEPTFILTER:
2177 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2178 				return (EINVAL);
2179 			afap = kmalloc(sizeof(*afap), M_TEMP,
2180 				       M_WAITOK | M_ZERO);
2181 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2182 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2183 				if (so->so_accf->so_accept_filter_str != NULL)
2184 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2185 			}
2186 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2187 			kfree(afap, M_TEMP);
2188 			break;
2189 #endif /* INET */
2190 
2191 		case SO_LINGER:
2192 			l.l_onoff = so->so_options & SO_LINGER;
2193 			l.l_linger = so->so_linger;
2194 			error = sooptcopyout(sopt, &l, sizeof l);
2195 			break;
2196 
2197 		case SO_USELOOPBACK:
2198 		case SO_DONTROUTE:
2199 		case SO_DEBUG:
2200 		case SO_KEEPALIVE:
2201 		case SO_REUSEADDR:
2202 		case SO_REUSEPORT:
2203 		case SO_BROADCAST:
2204 		case SO_OOBINLINE:
2205 		case SO_TIMESTAMP:
2206 		case SO_NOSIGPIPE:
2207 			optval = so->so_options & sopt->sopt_name;
2208 integer:
2209 			error = sooptcopyout(sopt, &optval, sizeof optval);
2210 			break;
2211 
2212 		case SO_TYPE:
2213 			optval = so->so_type;
2214 			goto integer;
2215 
2216 		case SO_ERROR:
2217 			optval = so->so_error;
2218 			so->so_error = 0;
2219 			goto integer;
2220 
2221 		case SO_SNDBUF:
2222 			optval = so->so_snd.ssb_hiwat;
2223 			goto integer;
2224 
2225 		case SO_RCVBUF:
2226 			optval = so->so_rcv.ssb_hiwat;
2227 			goto integer;
2228 
2229 		case SO_SNDLOWAT:
2230 			optval = so->so_snd.ssb_lowat;
2231 			goto integer;
2232 
2233 		case SO_RCVLOWAT:
2234 			optval = so->so_rcv.ssb_lowat;
2235 			goto integer;
2236 
2237 		case SO_SNDTIMEO:
2238 		case SO_RCVTIMEO:
2239 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2240 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2241 
2242 			tv.tv_sec = optval / hz;
2243 			tv.tv_usec = (optval % hz) * ustick;
2244 			error = sooptcopyout(sopt, &tv, sizeof tv);
2245 			break;
2246 
2247 		case SO_SNDSPACE:
2248 			optval_l = ssb_space(&so->so_snd);
2249 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2250 			break;
2251 
2252 		default:
2253 			error = ENOPROTOOPT;
2254 			break;
2255 		}
2256 		return (error);
2257 	}
2258 }
2259 
2260 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2261 int
2262 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2263 {
2264 	struct mbuf *m, *m_prev;
2265 	int sopt_size = sopt->sopt_valsize, msize;
2266 
2267 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
2268 		   0, &msize);
2269 	if (m == NULL)
2270 		return (ENOBUFS);
2271 	m->m_len = min(msize, sopt_size);
2272 	sopt_size -= m->m_len;
2273 	*mp = m;
2274 	m_prev = m;
2275 
2276 	while (sopt_size > 0) {
2277 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
2278 			   MT_DATA, 0, &msize);
2279 		if (m == NULL) {
2280 			m_freem(*mp);
2281 			return (ENOBUFS);
2282 		}
2283 		m->m_len = min(msize, sopt_size);
2284 		sopt_size -= m->m_len;
2285 		m_prev->m_next = m;
2286 		m_prev = m;
2287 	}
2288 	return (0);
2289 }
2290 
2291 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2292 int
2293 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2294 {
2295 	soopt_to_mbuf(sopt, m);
2296 	return 0;
2297 }
2298 
2299 void
2300 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2301 {
2302 	size_t valsize;
2303 	void *val;
2304 
2305 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2306 	KKASSERT(kva_p(m));
2307 	if (sopt->sopt_val == NULL)
2308 		return;
2309 	val = sopt->sopt_val;
2310 	valsize = sopt->sopt_valsize;
2311 	while (m != NULL && valsize >= m->m_len) {
2312 		bcopy(val, mtod(m, char *), m->m_len);
2313 		valsize -= m->m_len;
2314 		val = (caddr_t)val + m->m_len;
2315 		m = m->m_next;
2316 	}
2317 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2318 		panic("ip6_sooptmcopyin");
2319 }
2320 
2321 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2322 int
2323 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2324 {
2325 	return soopt_from_mbuf(sopt, m);
2326 }
2327 
2328 int
2329 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2330 {
2331 	struct mbuf *m0 = m;
2332 	size_t valsize = 0;
2333 	size_t maxsize;
2334 	void *val;
2335 
2336 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2337 	KKASSERT(kva_p(m));
2338 	if (sopt->sopt_val == NULL)
2339 		return 0;
2340 	val = sopt->sopt_val;
2341 	maxsize = sopt->sopt_valsize;
2342 	while (m != NULL && maxsize >= m->m_len) {
2343 		bcopy(mtod(m, char *), val, m->m_len);
2344 	       maxsize -= m->m_len;
2345 	       val = (caddr_t)val + m->m_len;
2346 	       valsize += m->m_len;
2347 	       m = m->m_next;
2348 	}
2349 	if (m != NULL) {
2350 		/* enough soopt buffer should be given from user-land */
2351 		m_freem(m0);
2352 		return (EINVAL);
2353 	}
2354 	sopt->sopt_valsize = valsize;
2355 	return 0;
2356 }
2357 
2358 void
2359 sohasoutofband(struct socket *so)
2360 {
2361 	if (so->so_sigio != NULL)
2362 		pgsigio(so->so_sigio, SIGURG, 0);
2363 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2364 }
2365 
2366 int
2367 sokqfilter(struct file *fp, struct knote *kn)
2368 {
2369 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2370 	struct signalsockbuf *ssb;
2371 
2372 	switch (kn->kn_filter) {
2373 	case EVFILT_READ:
2374 		if (so->so_options & SO_ACCEPTCONN)
2375 			kn->kn_fop = &solisten_filtops;
2376 		else
2377 			kn->kn_fop = &soread_filtops;
2378 		ssb = &so->so_rcv;
2379 		break;
2380 	case EVFILT_WRITE:
2381 		kn->kn_fop = &sowrite_filtops;
2382 		ssb = &so->so_snd;
2383 		break;
2384 	case EVFILT_EXCEPT:
2385 		kn->kn_fop = &soexcept_filtops;
2386 		ssb = &so->so_rcv;
2387 		break;
2388 	default:
2389 		return (EOPNOTSUPP);
2390 	}
2391 
2392 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2393 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2394 	return (0);
2395 }
2396 
2397 static void
2398 filt_sordetach(struct knote *kn)
2399 {
2400 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2401 
2402 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2403 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2404 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2405 }
2406 
2407 /*ARGSUSED*/
2408 static int
2409 filt_soread(struct knote *kn, long hint)
2410 {
2411 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2412 
2413 	if (kn->kn_sfflags & NOTE_OOB) {
2414 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2415 			kn->kn_fflags |= NOTE_OOB;
2416 			return (1);
2417 		}
2418 		return (0);
2419 	}
2420 	kn->kn_data = so->so_rcv.ssb_cc;
2421 
2422 	if (so->so_state & SS_CANTRCVMORE) {
2423 		/*
2424 		 * Only set NODATA if all data has been exhausted.
2425 		 */
2426 		if (kn->kn_data == 0)
2427 			kn->kn_flags |= EV_NODATA;
2428 		kn->kn_flags |= EV_EOF;
2429 		kn->kn_fflags = so->so_error;
2430 		return (1);
2431 	}
2432 	if (so->so_error)	/* temporary udp error */
2433 		return (1);
2434 	if (kn->kn_sfflags & NOTE_LOWAT)
2435 		return (kn->kn_data >= kn->kn_sdata);
2436 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2437 		!TAILQ_EMPTY(&so->so_comp));
2438 }
2439 
2440 static void
2441 filt_sowdetach(struct knote *kn)
2442 {
2443 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2444 
2445 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2446 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2447 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2448 }
2449 
2450 /*ARGSUSED*/
2451 static int
2452 filt_sowrite(struct knote *kn, long hint)
2453 {
2454 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2455 
2456 	kn->kn_data = ssb_space(&so->so_snd);
2457 	if (so->so_state & SS_CANTSENDMORE) {
2458 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2459 		kn->kn_fflags = so->so_error;
2460 		return (1);
2461 	}
2462 	if (so->so_error)	/* temporary udp error */
2463 		return (1);
2464 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2465 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2466 		return (0);
2467 	if (kn->kn_sfflags & NOTE_LOWAT)
2468 		return (kn->kn_data >= kn->kn_sdata);
2469 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2470 }
2471 
2472 /*ARGSUSED*/
2473 static int
2474 filt_solisten(struct knote *kn, long hint)
2475 {
2476 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2477 
2478 	kn->kn_data = so->so_qlen;
2479 	return (! TAILQ_EMPTY(&so->so_comp));
2480 }
2481