xref: /dflybsd-src/sys/kern/uipc_socket.c (revision 67bf99c4e3c62e257027c8f0d3b312f44cfe622f)
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  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
67  * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
68  * $DragonFly: src/sys/kern/uipc_socket.c,v 1.55 2008/09/02 16:17:52 dillon Exp $
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_sctp.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/fcntl.h>
77 #include <sys/malloc.h>
78 #include <sys/mbuf.h>
79 #include <sys/domain.h>
80 #include <sys/file.h>			/* for struct knote */
81 #include <sys/kernel.h>
82 #include <sys/malloc.h>
83 #include <sys/event.h>
84 #include <sys/proc.h>
85 #include <sys/protosw.h>
86 #include <sys/socket.h>
87 #include <sys/socketvar.h>
88 #include <sys/socketops.h>
89 #include <sys/resourcevar.h>
90 #include <sys/signalvar.h>
91 #include <sys/sysctl.h>
92 #include <sys/uio.h>
93 #include <sys/jail.h>
94 #include <vm/vm_zone.h>
95 #include <vm/pmap.h>
96 
97 #include <sys/thread2.h>
98 #include <sys/socketvar2.h>
99 
100 #include <machine/limits.h>
101 
102 #ifdef INET
103 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
104 #endif /* INET */
105 
106 static void 	filt_sordetach(struct knote *kn);
107 static int 	filt_soread(struct knote *kn, long hint);
108 static void 	filt_sowdetach(struct knote *kn);
109 static int	filt_sowrite(struct knote *kn, long hint);
110 static int	filt_solisten(struct knote *kn, long hint);
111 
112 static struct filterops solisten_filtops =
113 	{ FILTEROP_ISFD, NULL, filt_sordetach, filt_solisten };
114 static struct filterops soread_filtops =
115 	{ FILTEROP_ISFD, NULL, filt_sordetach, filt_soread };
116 static struct filterops sowrite_filtops =
117 	{ FILTEROP_ISFD, NULL, filt_sowdetach, filt_sowrite };
118 static struct filterops soexcept_filtops =
119 	{ FILTEROP_ISFD, NULL, filt_sordetach, filt_soread };
120 
121 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
122 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
123 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
124 
125 
126 static int somaxconn = SOMAXCONN;
127 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
128     &somaxconn, 0, "Maximum pending socket connection queue size");
129 
130 /*
131  * Socket operation routines.
132  * These routines are called by the routines in
133  * sys_socket.c or from a system process, and
134  * implement the semantics of socket operations by
135  * switching out to the protocol specific routines.
136  */
137 
138 /*
139  * Get a socket structure, and initialize it.
140  * Note that it would probably be better to allocate socket
141  * and PCB at the same time, but I'm not convinced that all
142  * the protocols can be easily modified to do this.
143  */
144 struct socket *
145 soalloc(int waitok)
146 {
147 	struct socket *so;
148 	unsigned waitmask;
149 
150 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
151 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
152 	if (so) {
153 		/* XXX race condition for reentrant kernel */
154 		TAILQ_INIT(&so->so_aiojobq);
155 		TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
156 		TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
157 		lwkt_token_init(&so->so_rcv.ssb_token, 1, "rcvtok");
158 		lwkt_token_init(&so->so_snd.ssb_token, 1, "rcvtok");
159 		so->so_state = SS_NOFDREF;
160 		so->so_refs = 1;
161 	}
162 	return so;
163 }
164 
165 int
166 socreate(int dom, struct socket **aso, int type,
167 	int proto, struct thread *td)
168 {
169 	struct proc *p = td->td_proc;
170 	struct protosw *prp;
171 	struct socket *so;
172 	struct pru_attach_info ai;
173 	int error;
174 
175 	if (proto)
176 		prp = pffindproto(dom, proto, type);
177 	else
178 		prp = pffindtype(dom, type);
179 
180 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
181 		return (EPROTONOSUPPORT);
182 
183 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
184 	    prp->pr_domain->dom_family != PF_LOCAL &&
185 	    prp->pr_domain->dom_family != PF_INET &&
186 	    prp->pr_domain->dom_family != PF_INET6 &&
187 	    prp->pr_domain->dom_family != PF_ROUTE) {
188 		return (EPROTONOSUPPORT);
189 	}
190 
191 	if (prp->pr_type != type)
192 		return (EPROTOTYPE);
193 	so = soalloc(p != 0);
194 	if (so == NULL)
195 		return (ENOBUFS);
196 
197 	/*
198 	 * Callers of socreate() presumably will connect up a descriptor
199 	 * and call soclose() if they cannot.  This represents our so_refs
200 	 * (which should be 1) from soalloc().
201 	 */
202 	soclrstate(so, SS_NOFDREF);
203 
204 	/*
205 	 * Set a default port for protocol processing.  No action will occur
206 	 * on the socket on this port until an inpcb is attached to it and
207 	 * is able to match incoming packets, or until the socket becomes
208 	 * available to userland.
209 	 *
210 	 * We normally default the socket to the protocol thread on cpu 0.
211 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
212 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
213 	 */
214 	if (prp->pr_flags & PR_SYNC_PORT)
215 		so->so_port = &netisr_sync_port;
216 	else
217 		so->so_port = cpu_portfn(0);
218 
219 	TAILQ_INIT(&so->so_incomp);
220 	TAILQ_INIT(&so->so_comp);
221 	so->so_type = type;
222 	so->so_cred = crhold(p->p_ucred);
223 	so->so_proto = prp;
224 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
225 	ai.p_ucred = p->p_ucred;
226 	ai.fd_rdir = p->p_fd->fd_rdir;
227 
228 	/*
229 	 * Auto-sizing of socket buffers is managed by the protocols and
230 	 * the appropriate flags must be set in the pru_attach function.
231 	 */
232 	error = so_pru_attach(so, proto, &ai);
233 	if (error) {
234 		sosetstate(so, SS_NOFDREF);
235 		sofree(so);	/* from soalloc */
236 		return error;
237 	}
238 
239 	/*
240 	 * NOTE: Returns referenced socket.
241 	 */
242 	*aso = so;
243 	return (0);
244 }
245 
246 int
247 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
248 {
249 	int error;
250 
251 	error = so_pru_bind(so, nam, td);
252 	return (error);
253 }
254 
255 static void
256 sodealloc(struct socket *so)
257 {
258 	if (so->so_rcv.ssb_hiwat)
259 		(void)chgsbsize(so->so_cred->cr_uidinfo,
260 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
261 	if (so->so_snd.ssb_hiwat)
262 		(void)chgsbsize(so->so_cred->cr_uidinfo,
263 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
264 #ifdef INET
265 	/* remove accept filter if present */
266 	if (so->so_accf != NULL)
267 		do_setopt_accept_filter(so, NULL);
268 #endif /* INET */
269 	crfree(so->so_cred);
270 	kfree(so, M_SOCKET);
271 }
272 
273 int
274 solisten(struct socket *so, int backlog, struct thread *td)
275 {
276 	int error;
277 #ifdef SCTP
278 	short oldopt, oldqlimit;
279 #endif /* SCTP */
280 
281 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
282 		return (EINVAL);
283 
284 #ifdef SCTP
285 	oldopt = so->so_options;
286 	oldqlimit = so->so_qlimit;
287 #endif /* SCTP */
288 
289 	lwkt_gettoken(&so->so_rcv.ssb_token);
290 	if (TAILQ_EMPTY(&so->so_comp))
291 		so->so_options |= SO_ACCEPTCONN;
292 	lwkt_reltoken(&so->so_rcv.ssb_token);
293 	if (backlog < 0 || backlog > somaxconn)
294 		backlog = somaxconn;
295 	so->so_qlimit = backlog;
296 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
297 	 * the so_options (UDP model both connect's and gets inbound
298 	 * connections .. implicitly).
299 	 */
300 	error = so_pru_listen(so, td);
301 	if (error) {
302 #ifdef SCTP
303 		/* Restore the params */
304 		so->so_options = oldopt;
305 		so->so_qlimit = oldqlimit;
306 #endif /* SCTP */
307 		return (error);
308 	}
309 	return (0);
310 }
311 
312 /*
313  * Destroy a disconnected socket.  This routine is a NOP if entities
314  * still have a reference on the socket:
315  *
316  *	so_pcb -	The protocol stack still has a reference
317  *	SS_NOFDREF -	There is no longer a file pointer reference
318  */
319 void
320 sofree(struct socket *so)
321 {
322 	struct socket *head = so->so_head;
323 
324 	/*
325 	 * Arbitrage the last free.
326 	 */
327 	KKASSERT(so->so_refs > 0);
328 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1)
329 		return;
330 
331 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
332 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
333 
334 	/*
335 	 * We're done, clean up
336 	 */
337 	if (head != NULL) {
338 		lwkt_gettoken(&head->so_rcv.ssb_token);
339 		if (so->so_state & SS_INCOMP) {
340 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
341 			head->so_incqlen--;
342 		} else if (so->so_state & SS_COMP) {
343 			/*
344 			 * We must not decommission a socket that's
345 			 * on the accept(2) queue.  If we do, then
346 			 * accept(2) may hang after select(2) indicated
347 			 * that the listening socket was ready.
348 			 */
349 			lwkt_reltoken(&head->so_rcv.ssb_token);
350 			return;
351 		} else {
352 			panic("sofree: not queued");
353 		}
354 		soclrstate(so, SS_INCOMP);
355 		so->so_head = NULL;
356 		lwkt_reltoken(&head->so_rcv.ssb_token);
357 	}
358 	ssb_release(&so->so_snd, so);
359 	sorflush(so);
360 	sodealloc(so);
361 }
362 
363 /*
364  * Close a socket on last file table reference removal.
365  * Initiate disconnect if connected.
366  * Free socket when disconnect complete.
367  */
368 int
369 soclose(struct socket *so, int fflag)
370 {
371 	int error = 0;
372 
373 	funsetown(so->so_sigio);
374 	if (so->so_pcb == NULL)
375 		goto discard;
376 	if (so->so_state & SS_ISCONNECTED) {
377 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
378 			error = sodisconnect(so);
379 			if (error)
380 				goto drop;
381 		}
382 		if (so->so_options & SO_LINGER) {
383 			if ((so->so_state & SS_ISDISCONNECTING) &&
384 			    (fflag & FNONBLOCK))
385 				goto drop;
386 			while (so->so_state & SS_ISCONNECTED) {
387 				error = tsleep(&so->so_timeo, PCATCH,
388 					       "soclos", so->so_linger * hz);
389 				if (error)
390 					break;
391 			}
392 		}
393 	}
394 drop:
395 	if (so->so_pcb) {
396 		int error2;
397 
398 		error2 = so_pru_detach(so);
399 		if (error == 0)
400 			error = error2;
401 	}
402 discard:
403 	lwkt_gettoken(&so->so_rcv.ssb_token);
404 	if (so->so_options & SO_ACCEPTCONN) {
405 		struct socket *sp;
406 
407 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
408 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
409 			soclrstate(sp, SS_INCOMP);
410 			sp->so_head = NULL;
411 			so->so_incqlen--;
412 			soaborta(sp);
413 		}
414 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
415 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
416 			soclrstate(sp, SS_COMP);
417 			sp->so_head = NULL;
418 			so->so_qlen--;
419 			soaborta(sp);
420 		}
421 	}
422 	lwkt_reltoken(&so->so_rcv.ssb_token);
423 	if (so->so_state & SS_NOFDREF)
424 		panic("soclose: NOFDREF");
425 	sosetstate(so, SS_NOFDREF);	/* take ref */
426 	sofree(so);			/* dispose of ref */
427 	return (error);
428 }
429 
430 /*
431  * Abort and destroy a socket.  Only one abort can be in progress
432  * at any given moment.
433  */
434 void
435 soabort(struct socket *so)
436 {
437 	soreference(so);
438 	so_pru_abort(so);
439 }
440 
441 void
442 soaborta(struct socket *so)
443 {
444 	soreference(so);
445 	so_pru_aborta(so);
446 }
447 
448 void
449 soabort_oncpu(struct socket *so)
450 {
451 	soreference(so);
452 	so_pru_abort_oncpu(so);
453 }
454 
455 int
456 soaccept(struct socket *so, struct sockaddr **nam)
457 {
458 	int error;
459 
460 	if ((so->so_state & SS_NOFDREF) == 0)
461 		panic("soaccept: !NOFDREF");
462 	soreference(so);		/* create ref */
463 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
464 	error = so_pru_accept_direct(so, nam);
465 	return (error);
466 }
467 
468 int
469 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
470 {
471 	int error;
472 
473 	if (so->so_options & SO_ACCEPTCONN)
474 		return (EOPNOTSUPP);
475 	/*
476 	 * If protocol is connection-based, can only connect once.
477 	 * Otherwise, if connected, try to disconnect first.
478 	 * This allows user to disconnect by connecting to, e.g.,
479 	 * a null address.
480 	 */
481 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
482 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
483 	    (error = sodisconnect(so)))) {
484 		error = EISCONN;
485 	} else {
486 		/*
487 		 * Prevent accumulated error from previous connection
488 		 * from biting us.
489 		 */
490 		so->so_error = 0;
491 		error = so_pru_connect(so, nam, td);
492 	}
493 	return (error);
494 }
495 
496 int
497 soconnect2(struct socket *so1, struct socket *so2)
498 {
499 	int error;
500 
501 	error = so_pru_connect2(so1, so2);
502 	return (error);
503 }
504 
505 int
506 sodisconnect(struct socket *so)
507 {
508 	int error;
509 
510 	if ((so->so_state & SS_ISCONNECTED) == 0) {
511 		error = ENOTCONN;
512 		goto bad;
513 	}
514 	if (so->so_state & SS_ISDISCONNECTING) {
515 		error = EALREADY;
516 		goto bad;
517 	}
518 	error = so_pru_disconnect(so);
519 bad:
520 	return (error);
521 }
522 
523 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
524 /*
525  * Send on a socket.
526  * If send must go all at once and message is larger than
527  * send buffering, then hard error.
528  * Lock against other senders.
529  * If must go all at once and not enough room now, then
530  * inform user that this would block and do nothing.
531  * Otherwise, if nonblocking, send as much as possible.
532  * The data to be sent is described by "uio" if nonzero,
533  * otherwise by the mbuf chain "top" (which must be null
534  * if uio is not).  Data provided in mbuf chain must be small
535  * enough to send all at once.
536  *
537  * Returns nonzero on error, timeout or signal; callers
538  * must check for short counts if EINTR/ERESTART are returned.
539  * Data and control buffers are freed on return.
540  */
541 int
542 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
543 	struct mbuf *top, struct mbuf *control, int flags,
544 	struct thread *td)
545 {
546 	struct mbuf **mp;
547 	struct mbuf *m;
548 	size_t resid;
549 	int space, len;
550 	int clen = 0, error, dontroute, mlen;
551 	int atomic = sosendallatonce(so) || top;
552 	int pru_flags;
553 
554 	if (uio) {
555 		resid = uio->uio_resid;
556 	} else {
557 		resid = (size_t)top->m_pkthdr.len;
558 #ifdef INVARIANTS
559 		len = 0;
560 		for (m = top; m; m = m->m_next)
561 			len += m->m_len;
562 		KKASSERT(top->m_pkthdr.len == len);
563 #endif
564 	}
565 
566 	/*
567 	 * WARNING!  resid is unsigned, space and len are signed.  space
568 	 * 	     can wind up negative if the sockbuf is overcommitted.
569 	 *
570 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
571 	 * type sockets since that's an error.
572 	 */
573 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
574 		error = EINVAL;
575 		goto out;
576 	}
577 
578 	dontroute =
579 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
580 	    (so->so_proto->pr_flags & PR_ATOMIC);
581 	if (td->td_lwp != NULL)
582 		td->td_lwp->lwp_ru.ru_msgsnd++;
583 	if (control)
584 		clen = control->m_len;
585 #define	gotoerr(errcode)	{ error = errcode; goto release; }
586 
587 restart:
588 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
589 	if (error)
590 		goto out;
591 
592 	do {
593 		if (so->so_state & SS_CANTSENDMORE)
594 			gotoerr(EPIPE);
595 		if (so->so_error) {
596 			error = so->so_error;
597 			so->so_error = 0;
598 			goto release;
599 		}
600 		if ((so->so_state & SS_ISCONNECTED) == 0) {
601 			/*
602 			 * `sendto' and `sendmsg' is allowed on a connection-
603 			 * based socket if it supports implied connect.
604 			 * Return ENOTCONN if not connected and no address is
605 			 * supplied.
606 			 */
607 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
608 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
609 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
610 				    !(resid == 0 && clen != 0))
611 					gotoerr(ENOTCONN);
612 			} else if (addr == 0)
613 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
614 				   ENOTCONN : EDESTADDRREQ);
615 		}
616 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
617 		    clen > so->so_snd.ssb_hiwat) {
618 			gotoerr(EMSGSIZE);
619 		}
620 		space = ssb_space(&so->so_snd);
621 		if (flags & MSG_OOB)
622 			space += 1024;
623 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
624 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
625 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
626 				gotoerr(EWOULDBLOCK);
627 			ssb_unlock(&so->so_snd);
628 			error = ssb_wait(&so->so_snd);
629 			if (error)
630 				goto out;
631 			goto restart;
632 		}
633 		mp = &top;
634 		space -= clen;
635 		do {
636 		    if (uio == NULL) {
637 			/*
638 			 * Data is prepackaged in "top".
639 			 */
640 			resid = 0;
641 			if (flags & MSG_EOR)
642 				top->m_flags |= M_EOR;
643 		    } else do {
644 			if (resid > INT_MAX)
645 				resid = INT_MAX;
646 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
647 				   top == NULL ? M_PKTHDR : 0, &mlen);
648 			if (top == NULL) {
649 				m->m_pkthdr.len = 0;
650 				m->m_pkthdr.rcvif = NULL;
651 			}
652 			len = imin((int)szmin(mlen, resid), space);
653 			if (resid < MINCLSIZE) {
654 				/*
655 				 * For datagram protocols, leave room
656 				 * for protocol headers in first mbuf.
657 				 */
658 				if (atomic && top == 0 && len < mlen)
659 					MH_ALIGN(m, len);
660 			}
661 			space -= len;
662 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
663 			resid = uio->uio_resid;
664 			m->m_len = len;
665 			*mp = m;
666 			top->m_pkthdr.len += len;
667 			if (error)
668 				goto release;
669 			mp = &m->m_next;
670 			if (resid == 0) {
671 				if (flags & MSG_EOR)
672 					top->m_flags |= M_EOR;
673 				break;
674 			}
675 		    } while (space > 0 && atomic);
676 		    if (dontroute)
677 			    so->so_options |= SO_DONTROUTE;
678 		    if (flags & MSG_OOB) {
679 		    	    pru_flags = PRUS_OOB;
680 		    } else if ((flags & MSG_EOF) &&
681 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
682 			       (resid == 0)) {
683 			    /*
684 			     * If the user set MSG_EOF, the protocol
685 			     * understands this flag and nothing left to
686 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
687 			     */
688 		    	    pru_flags = PRUS_EOF;
689 		    } else if (resid > 0 && space > 0) {
690 			    /* If there is more to send, set PRUS_MORETOCOME */
691 		    	    pru_flags = PRUS_MORETOCOME;
692 		    } else {
693 		    	    pru_flags = 0;
694 		    }
695 		    /*
696 		     * XXX all the SS_CANTSENDMORE checks previously
697 		     * done could be out of date.  We could have recieved
698 		     * a reset packet in an interrupt or maybe we slept
699 		     * while doing page faults in uiomove() etc. We could
700 		     * probably recheck again inside the splnet() protection
701 		     * here, but there are probably other places that this
702 		     * also happens.  We must rethink this.
703 		     */
704 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
705 		    if (dontroute)
706 			    so->so_options &= ~SO_DONTROUTE;
707 		    clen = 0;
708 		    control = 0;
709 		    top = NULL;
710 		    mp = &top;
711 		    if (error)
712 			    goto release;
713 		} while (resid && space > 0);
714 	} while (resid);
715 
716 release:
717 	ssb_unlock(&so->so_snd);
718 out:
719 	if (top)
720 		m_freem(top);
721 	if (control)
722 		m_freem(control);
723 	return (error);
724 }
725 
726 /*
727  * A specialization of sosend() for UDP based on protocol-specific knowledge:
728  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
729  *	sosendallatonce() returns true,
730  *	the "atomic" variable is true,
731  *	and sosendudp() blocks until space is available for the entire send.
732  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
733  *	PR_IMPLOPCL flags set.
734  *   UDP has no out-of-band data.
735  *   UDP has no control data.
736  *   UDP does not support MSG_EOR.
737  */
738 int
739 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
740 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
741 {
742 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
743 	size_t resid;
744 	int error;
745 	int space;
746 
747 	if (td->td_lwp != NULL)
748 		td->td_lwp->lwp_ru.ru_msgsnd++;
749 	if (control)
750 		m_freem(control);
751 
752 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
753 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
754 
755 restart:
756 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
757 	if (error)
758 		goto out;
759 
760 	if (so->so_state & SS_CANTSENDMORE)
761 		gotoerr(EPIPE);
762 	if (so->so_error) {
763 		error = so->so_error;
764 		so->so_error = 0;
765 		goto release;
766 	}
767 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
768 		gotoerr(EDESTADDRREQ);
769 	if (resid > so->so_snd.ssb_hiwat)
770 		gotoerr(EMSGSIZE);
771 	space = ssb_space(&so->so_snd);
772 	if (uio && (space < 0 || (size_t)space < resid)) {
773 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
774 			gotoerr(EWOULDBLOCK);
775 		ssb_unlock(&so->so_snd);
776 		error = ssb_wait(&so->so_snd);
777 		if (error)
778 			goto out;
779 		goto restart;
780 	}
781 
782 	if (uio) {
783 		top = m_uiomove(uio);
784 		if (top == NULL)
785 			goto release;
786 	}
787 
788 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
789 	if (dontroute)
790 		so->so_options |= SO_DONTROUTE;
791 
792 	error = so_pru_send(so, 0, top, addr, NULL, td);
793 	top = NULL;		/* sent or freed in lower layer */
794 
795 	if (dontroute)
796 		so->so_options &= ~SO_DONTROUTE;
797 
798 release:
799 	ssb_unlock(&so->so_snd);
800 out:
801 	if (top)
802 		m_freem(top);
803 	return (error);
804 }
805 
806 /*
807  * Implement receive operations on a socket.
808  *
809  * We depend on the way that records are added to the signalsockbuf
810  * by sbappend*.  In particular, each record (mbufs linked through m_next)
811  * must begin with an address if the protocol so specifies,
812  * followed by an optional mbuf or mbufs containing ancillary data,
813  * and then zero or more mbufs of data.
814  *
815  * Although the signalsockbuf is locked, new data may still be appended.
816  * A token inside the ssb_lock deals with MP issues and still allows
817  * the network to access the socket if we block in a uio.
818  *
819  * The caller may receive the data as a single mbuf chain by supplying
820  * an mbuf **mp0 for use in returning the chain.  The uio is then used
821  * only for the count in uio_resid.
822  */
823 int
824 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
825 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
826 {
827 	struct mbuf *m, *n;
828 	struct mbuf *free_chain = NULL;
829 	int flags, len, error, offset;
830 	struct protosw *pr = so->so_proto;
831 	int moff, type = 0;
832 	size_t resid, orig_resid;
833 
834 	if (uio)
835 		resid = uio->uio_resid;
836 	else
837 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
838 	orig_resid = resid;
839 
840 	if (psa)
841 		*psa = NULL;
842 	if (controlp)
843 		*controlp = NULL;
844 	if (flagsp)
845 		flags = *flagsp &~ MSG_EOR;
846 	else
847 		flags = 0;
848 	if (flags & MSG_OOB) {
849 		m = m_get(MB_WAIT, MT_DATA);
850 		if (m == NULL)
851 			return (ENOBUFS);
852 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
853 		if (error)
854 			goto bad;
855 		if (sio) {
856 			do {
857 				sbappend(sio, m);
858 				KKASSERT(resid >= (size_t)m->m_len);
859 				resid -= (size_t)m->m_len;
860 			} while (resid > 0 && m);
861 		} else {
862 			do {
863 				uio->uio_resid = resid;
864 				error = uiomove(mtod(m, caddr_t),
865 						(int)szmin(resid, m->m_len),
866 						uio);
867 				resid = uio->uio_resid;
868 				m = m_free(m);
869 			} while (uio->uio_resid && error == 0 && m);
870 		}
871 bad:
872 		if (m)
873 			m_freem(m);
874 		return (error);
875 	}
876 	if ((so->so_state & SS_ISCONFIRMING) && resid)
877 		so_pru_rcvd(so, 0);
878 
879 	/*
880 	 * The token interlocks against the protocol thread while
881 	 * ssb_lock is a blocking lock against other userland entities.
882 	 */
883 	lwkt_gettoken(&so->so_rcv.ssb_token);
884 restart:
885 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
886 	if (error)
887 		goto done;
888 
889 	m = so->so_rcv.ssb_mb;
890 	/*
891 	 * If we have less data than requested, block awaiting more
892 	 * (subject to any timeout) if:
893 	 *   1. the current count is less than the low water mark, or
894 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
895 	 *	receive operation at once if we block (resid <= hiwat).
896 	 *   3. MSG_DONTWAIT is not set
897 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
898 	 * we have to do the receive in sections, and thus risk returning
899 	 * a short count if a timeout or signal occurs after we start.
900 	 */
901 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
902 	    (size_t)so->so_rcv.ssb_cc < resid) &&
903 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
904 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
905 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
906 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
907 		if (so->so_error) {
908 			if (m)
909 				goto dontblock;
910 			error = so->so_error;
911 			if ((flags & MSG_PEEK) == 0)
912 				so->so_error = 0;
913 			goto release;
914 		}
915 		if (so->so_state & SS_CANTRCVMORE) {
916 			if (m)
917 				goto dontblock;
918 			else
919 				goto release;
920 		}
921 		for (; m; m = m->m_next) {
922 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
923 				m = so->so_rcv.ssb_mb;
924 				goto dontblock;
925 			}
926 		}
927 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
928 		    (pr->pr_flags & PR_CONNREQUIRED)) {
929 			error = ENOTCONN;
930 			goto release;
931 		}
932 		if (resid == 0)
933 			goto release;
934 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
935 			error = EWOULDBLOCK;
936 			goto release;
937 		}
938 		ssb_unlock(&so->so_rcv);
939 		error = ssb_wait(&so->so_rcv);
940 		if (error)
941 			goto done;
942 		goto restart;
943 	}
944 dontblock:
945 	if (uio && uio->uio_td && uio->uio_td->td_proc)
946 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
947 
948 	/*
949 	 * note: m should be == sb_mb here.  Cache the next record while
950 	 * cleaning up.  Note that calling m_free*() will break out critical
951 	 * section.
952 	 */
953 	KKASSERT(m == so->so_rcv.ssb_mb);
954 
955 	/*
956 	 * Skip any address mbufs prepending the record.
957 	 */
958 	if (pr->pr_flags & PR_ADDR) {
959 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
960 		orig_resid = 0;
961 		if (psa)
962 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
963 		if (flags & MSG_PEEK)
964 			m = m->m_next;
965 		else
966 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
967 	}
968 
969 	/*
970 	 * Skip any control mbufs prepending the record.
971 	 */
972 #ifdef SCTP
973 	if (pr->pr_flags & PR_ADDR_OPT) {
974 		/*
975 		 * For SCTP we may be getting a
976 		 * whole message OR a partial delivery.
977 		 */
978 		if (m && m->m_type == MT_SONAME) {
979 			orig_resid = 0;
980 			if (psa)
981 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
982 			if (flags & MSG_PEEK)
983 				m = m->m_next;
984 			else
985 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
986 		}
987 	}
988 #endif /* SCTP */
989 	while (m && m->m_type == MT_CONTROL && error == 0) {
990 		if (flags & MSG_PEEK) {
991 			if (controlp)
992 				*controlp = m_copy(m, 0, m->m_len);
993 			m = m->m_next;	/* XXX race */
994 		} else {
995 			if (controlp) {
996 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
997 				if (pr->pr_domain->dom_externalize &&
998 				    mtod(m, struct cmsghdr *)->cmsg_type ==
999 				    SCM_RIGHTS)
1000 				   error = (*pr->pr_domain->dom_externalize)(m);
1001 				*controlp = m;
1002 				m = n;
1003 			} else {
1004 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1005 			}
1006 		}
1007 		if (controlp && *controlp) {
1008 			orig_resid = 0;
1009 			controlp = &(*controlp)->m_next;
1010 		}
1011 	}
1012 
1013 	/*
1014 	 * flag OOB data.
1015 	 */
1016 	if (m) {
1017 		type = m->m_type;
1018 		if (type == MT_OOBDATA)
1019 			flags |= MSG_OOB;
1020 	}
1021 
1022 	/*
1023 	 * Copy to the UIO or mbuf return chain (*mp).
1024 	 */
1025 	moff = 0;
1026 	offset = 0;
1027 	while (m && resid > 0 && error == 0) {
1028 		if (m->m_type == MT_OOBDATA) {
1029 			if (type != MT_OOBDATA)
1030 				break;
1031 		} else if (type == MT_OOBDATA)
1032 			break;
1033 		else
1034 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1035 			("receive 3"));
1036 		soclrstate(so, SS_RCVATMARK);
1037 		len = (resid > INT_MAX) ? INT_MAX : resid;
1038 		if (so->so_oobmark && len > so->so_oobmark - offset)
1039 			len = so->so_oobmark - offset;
1040 		if (len > m->m_len - moff)
1041 			len = m->m_len - moff;
1042 
1043 		/*
1044 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1045 		 * The SIO is dealt with when we eat the mbuf, but deal
1046 		 * with the resid here either way.
1047 		 */
1048 		if (uio) {
1049 			uio->uio_resid = resid;
1050 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1051 			resid = uio->uio_resid;
1052 			if (error)
1053 				goto release;
1054 		} else {
1055 			resid -= (size_t)len;
1056 		}
1057 
1058 		/*
1059 		 * Eat the entire mbuf or just a piece of it
1060 		 */
1061 		if (len == m->m_len - moff) {
1062 			if (m->m_flags & M_EOR)
1063 				flags |= MSG_EOR;
1064 #ifdef SCTP
1065 			if (m->m_flags & M_NOTIFICATION)
1066 				flags |= MSG_NOTIFICATION;
1067 #endif /* SCTP */
1068 			if (flags & MSG_PEEK) {
1069 				m = m->m_next;
1070 				moff = 0;
1071 			} else {
1072 				if (sio) {
1073 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1074 					sbappend(sio, m);
1075 					m = n;
1076 				} else {
1077 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1078 				}
1079 			}
1080 		} else {
1081 			if (flags & MSG_PEEK) {
1082 				moff += len;
1083 			} else {
1084 				if (sio) {
1085 					n = m_copym(m, 0, len, MB_WAIT);
1086 					if (n)
1087 						sbappend(sio, n);
1088 				}
1089 				m->m_data += len;
1090 				m->m_len -= len;
1091 				so->so_rcv.ssb_cc -= len;
1092 			}
1093 		}
1094 		if (so->so_oobmark) {
1095 			if ((flags & MSG_PEEK) == 0) {
1096 				so->so_oobmark -= len;
1097 				if (so->so_oobmark == 0) {
1098 					sosetstate(so, SS_RCVATMARK);
1099 					break;
1100 				}
1101 			} else {
1102 				offset += len;
1103 				if (offset == so->so_oobmark)
1104 					break;
1105 			}
1106 		}
1107 		if (flags & MSG_EOR)
1108 			break;
1109 		/*
1110 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1111 		 * we must not quit until resid == 0 or an error
1112 		 * termination.  If a signal/timeout occurs, return
1113 		 * with a short count but without error.
1114 		 * Keep signalsockbuf locked against other readers.
1115 		 */
1116 		while ((flags & MSG_WAITALL) && m == NULL &&
1117 		       resid > 0 && !sosendallatonce(so) &&
1118 		       so->so_rcv.ssb_mb == NULL) {
1119 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1120 				break;
1121 			/*
1122 			 * The window might have closed to zero, make
1123 			 * sure we send an ack now that we've drained
1124 			 * the buffer or we might end up blocking until
1125 			 * the idle takes over (5 seconds).
1126 			 */
1127 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1128 				so_pru_rcvd(so, flags);
1129 			error = ssb_wait(&so->so_rcv);
1130 			if (error) {
1131 				ssb_unlock(&so->so_rcv);
1132 				error = 0;
1133 				goto done;
1134 			}
1135 			m = so->so_rcv.ssb_mb;
1136 		}
1137 	}
1138 
1139 	/*
1140 	 * If an atomic read was requested but unread data still remains
1141 	 * in the record, set MSG_TRUNC.
1142 	 */
1143 	if (m && pr->pr_flags & PR_ATOMIC)
1144 		flags |= MSG_TRUNC;
1145 
1146 	/*
1147 	 * Cleanup.  If an atomic read was requested drop any unread data.
1148 	 */
1149 	if ((flags & MSG_PEEK) == 0) {
1150 		if (m && (pr->pr_flags & PR_ATOMIC))
1151 			sbdroprecord(&so->so_rcv.sb);
1152 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1153 			so_pru_rcvd(so, flags);
1154 	}
1155 
1156 	if (orig_resid == resid && orig_resid &&
1157 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1158 		ssb_unlock(&so->so_rcv);
1159 		goto restart;
1160 	}
1161 
1162 	if (flagsp)
1163 		*flagsp |= flags;
1164 release:
1165 	ssb_unlock(&so->so_rcv);
1166 done:
1167 	lwkt_reltoken(&so->so_rcv.ssb_token);
1168 	if (free_chain)
1169 		m_freem(free_chain);
1170 	return (error);
1171 }
1172 
1173 int
1174 soshutdown(struct socket *so, int how)
1175 {
1176 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1177 		return (EINVAL);
1178 
1179 	if (how != SHUT_WR) {
1180 		ssb_lock(&so->so_rcv, M_WAITOK);	/* frontend lock */
1181 		sorflush(so);
1182 		ssb_unlock(&so->so_rcv);
1183 	}
1184 	if (how != SHUT_RD)
1185 		return (so_pru_shutdown(so));
1186 	return (0);
1187 }
1188 
1189 void
1190 sorflush(struct socket *so)
1191 {
1192 	struct signalsockbuf *ssb = &so->so_rcv;
1193 	struct protosw *pr = so->so_proto;
1194 	struct signalsockbuf asb;
1195 
1196 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1197 
1198 	lwkt_gettoken(&ssb->ssb_token);
1199 	socantrcvmore(so);
1200 	asb = *ssb;
1201 
1202 	/*
1203 	 * Can't just blow up the ssb structure here
1204 	 */
1205 	bzero(&ssb->sb, sizeof(ssb->sb));
1206 	ssb->ssb_timeo = 0;
1207 	ssb->ssb_unused01 = 0;
1208 	ssb->ssb_lowat = 0;
1209 	ssb->ssb_hiwat = 0;
1210 	ssb->ssb_mbmax = 0;
1211 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1212 
1213 	lwkt_reltoken(&ssb->ssb_token);
1214 
1215 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1216 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1217 	ssb_release(&asb, so);
1218 }
1219 
1220 #ifdef INET
1221 static int
1222 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1223 {
1224 	struct accept_filter_arg	*afap = NULL;
1225 	struct accept_filter	*afp;
1226 	struct so_accf	*af = so->so_accf;
1227 	int	error = 0;
1228 
1229 	/* do not set/remove accept filters on non listen sockets */
1230 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1231 		error = EINVAL;
1232 		goto out;
1233 	}
1234 
1235 	/* removing the filter */
1236 	if (sopt == NULL) {
1237 		if (af != NULL) {
1238 			if (af->so_accept_filter != NULL &&
1239 				af->so_accept_filter->accf_destroy != NULL) {
1240 				af->so_accept_filter->accf_destroy(so);
1241 			}
1242 			if (af->so_accept_filter_str != NULL) {
1243 				FREE(af->so_accept_filter_str, M_ACCF);
1244 			}
1245 			FREE(af, M_ACCF);
1246 			so->so_accf = NULL;
1247 		}
1248 		so->so_options &= ~SO_ACCEPTFILTER;
1249 		return (0);
1250 	}
1251 	/* adding a filter */
1252 	/* must remove previous filter first */
1253 	if (af != NULL) {
1254 		error = EINVAL;
1255 		goto out;
1256 	}
1257 	/* don't put large objects on the kernel stack */
1258 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1259 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1260 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1261 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1262 	if (error)
1263 		goto out;
1264 	afp = accept_filt_get(afap->af_name);
1265 	if (afp == NULL) {
1266 		error = ENOENT;
1267 		goto out;
1268 	}
1269 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1270 	if (afp->accf_create != NULL) {
1271 		if (afap->af_name[0] != '\0') {
1272 			int len = strlen(afap->af_name) + 1;
1273 
1274 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1275 			strcpy(af->so_accept_filter_str, afap->af_name);
1276 		}
1277 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1278 		if (af->so_accept_filter_arg == NULL) {
1279 			FREE(af->so_accept_filter_str, M_ACCF);
1280 			FREE(af, M_ACCF);
1281 			so->so_accf = NULL;
1282 			error = EINVAL;
1283 			goto out;
1284 		}
1285 	}
1286 	af->so_accept_filter = afp;
1287 	so->so_accf = af;
1288 	so->so_options |= SO_ACCEPTFILTER;
1289 out:
1290 	if (afap != NULL)
1291 		FREE(afap, M_TEMP);
1292 	return (error);
1293 }
1294 #endif /* INET */
1295 
1296 /*
1297  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1298  * an additional variant to handle the case where the option value needs
1299  * to be some kind of integer, but not a specific size.
1300  * In addition to their use here, these functions are also called by the
1301  * protocol-level pr_ctloutput() routines.
1302  */
1303 int
1304 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1305 {
1306 	return soopt_to_kbuf(sopt, buf, len, minlen);
1307 }
1308 
1309 int
1310 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1311 {
1312 	size_t	valsize;
1313 
1314 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1315 	KKASSERT(kva_p(buf));
1316 
1317 	/*
1318 	 * If the user gives us more than we wanted, we ignore it,
1319 	 * but if we don't get the minimum length the caller
1320 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1321 	 * is set to however much we actually retrieved.
1322 	 */
1323 	if ((valsize = sopt->sopt_valsize) < minlen)
1324 		return EINVAL;
1325 	if (valsize > len)
1326 		sopt->sopt_valsize = valsize = len;
1327 
1328 	bcopy(sopt->sopt_val, buf, valsize);
1329 	return 0;
1330 }
1331 
1332 
1333 int
1334 sosetopt(struct socket *so, struct sockopt *sopt)
1335 {
1336 	int	error, optval;
1337 	struct	linger l;
1338 	struct	timeval tv;
1339 	u_long  val;
1340 	struct signalsockbuf *sotmp;
1341 
1342 	error = 0;
1343 	sopt->sopt_dir = SOPT_SET;
1344 	if (sopt->sopt_level != SOL_SOCKET) {
1345 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1346 			return (so_pr_ctloutput(so, sopt));
1347 		}
1348 		error = ENOPROTOOPT;
1349 	} else {
1350 		switch (sopt->sopt_name) {
1351 #ifdef INET
1352 		case SO_ACCEPTFILTER:
1353 			error = do_setopt_accept_filter(so, sopt);
1354 			if (error)
1355 				goto bad;
1356 			break;
1357 #endif /* INET */
1358 		case SO_LINGER:
1359 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1360 			if (error)
1361 				goto bad;
1362 
1363 			so->so_linger = l.l_linger;
1364 			if (l.l_onoff)
1365 				so->so_options |= SO_LINGER;
1366 			else
1367 				so->so_options &= ~SO_LINGER;
1368 			break;
1369 
1370 		case SO_DEBUG:
1371 		case SO_KEEPALIVE:
1372 		case SO_DONTROUTE:
1373 		case SO_USELOOPBACK:
1374 		case SO_BROADCAST:
1375 		case SO_REUSEADDR:
1376 		case SO_REUSEPORT:
1377 		case SO_OOBINLINE:
1378 		case SO_TIMESTAMP:
1379 			error = sooptcopyin(sopt, &optval, sizeof optval,
1380 					    sizeof optval);
1381 			if (error)
1382 				goto bad;
1383 			if (optval)
1384 				so->so_options |= sopt->sopt_name;
1385 			else
1386 				so->so_options &= ~sopt->sopt_name;
1387 			break;
1388 
1389 		case SO_SNDBUF:
1390 		case SO_RCVBUF:
1391 		case SO_SNDLOWAT:
1392 		case SO_RCVLOWAT:
1393 			error = sooptcopyin(sopt, &optval, sizeof optval,
1394 					    sizeof optval);
1395 			if (error)
1396 				goto bad;
1397 
1398 			/*
1399 			 * Values < 1 make no sense for any of these
1400 			 * options, so disallow them.
1401 			 */
1402 			if (optval < 1) {
1403 				error = EINVAL;
1404 				goto bad;
1405 			}
1406 
1407 			switch (sopt->sopt_name) {
1408 			case SO_SNDBUF:
1409 			case SO_RCVBUF:
1410 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1411 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1412 				    so,
1413 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1414 					error = ENOBUFS;
1415 					goto bad;
1416 				}
1417 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1418 						&so->so_snd : &so->so_rcv;
1419 				atomic_clear_int(&sotmp->ssb_flags,
1420 						 SSB_AUTOSIZE);
1421 				break;
1422 
1423 			/*
1424 			 * Make sure the low-water is never greater than
1425 			 * the high-water.
1426 			 */
1427 			case SO_SNDLOWAT:
1428 				so->so_snd.ssb_lowat =
1429 				    (optval > so->so_snd.ssb_hiwat) ?
1430 				    so->so_snd.ssb_hiwat : optval;
1431 				atomic_clear_int(&so->so_snd.ssb_flags,
1432 						 SSB_AUTOLOWAT);
1433 				break;
1434 			case SO_RCVLOWAT:
1435 				so->so_rcv.ssb_lowat =
1436 				    (optval > so->so_rcv.ssb_hiwat) ?
1437 				    so->so_rcv.ssb_hiwat : optval;
1438 				atomic_clear_int(&so->so_rcv.ssb_flags,
1439 						 SSB_AUTOLOWAT);
1440 				break;
1441 			}
1442 			break;
1443 
1444 		case SO_SNDTIMEO:
1445 		case SO_RCVTIMEO:
1446 			error = sooptcopyin(sopt, &tv, sizeof tv,
1447 					    sizeof tv);
1448 			if (error)
1449 				goto bad;
1450 
1451 			/* assert(hz > 0); */
1452 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1453 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1454 				error = EDOM;
1455 				goto bad;
1456 			}
1457 			/* assert(tick > 0); */
1458 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1459 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1460 			if (val > SHRT_MAX) {
1461 				error = EDOM;
1462 				goto bad;
1463 			}
1464 			if (val == 0 && tv.tv_usec != 0)
1465 				val = 1;
1466 
1467 			switch (sopt->sopt_name) {
1468 			case SO_SNDTIMEO:
1469 				so->so_snd.ssb_timeo = val;
1470 				break;
1471 			case SO_RCVTIMEO:
1472 				so->so_rcv.ssb_timeo = val;
1473 				break;
1474 			}
1475 			break;
1476 		default:
1477 			error = ENOPROTOOPT;
1478 			break;
1479 		}
1480 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1481 			(void) so_pr_ctloutput(so, sopt);
1482 		}
1483 	}
1484 bad:
1485 	return (error);
1486 }
1487 
1488 /* Helper routine for getsockopt */
1489 int
1490 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1491 {
1492 	soopt_from_kbuf(sopt, buf, len);
1493 	return 0;
1494 }
1495 
1496 void
1497 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1498 {
1499 	size_t	valsize;
1500 
1501 	if (len == 0) {
1502 		sopt->sopt_valsize = 0;
1503 		return;
1504 	}
1505 
1506 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1507 	KKASSERT(kva_p(buf));
1508 
1509 	/*
1510 	 * Documented get behavior is that we always return a value,
1511 	 * possibly truncated to fit in the user's buffer.
1512 	 * Traditional behavior is that we always tell the user
1513 	 * precisely how much we copied, rather than something useful
1514 	 * like the total amount we had available for her.
1515 	 * Note that this interface is not idempotent; the entire answer must
1516 	 * generated ahead of time.
1517 	 */
1518 	valsize = szmin(len, sopt->sopt_valsize);
1519 	sopt->sopt_valsize = valsize;
1520 	if (sopt->sopt_val != 0) {
1521 		bcopy(buf, sopt->sopt_val, valsize);
1522 	}
1523 }
1524 
1525 int
1526 sogetopt(struct socket *so, struct sockopt *sopt)
1527 {
1528 	int	error, optval;
1529 	struct	linger l;
1530 	struct	timeval tv;
1531 #ifdef INET
1532 	struct accept_filter_arg *afap;
1533 #endif
1534 
1535 	error = 0;
1536 	sopt->sopt_dir = SOPT_GET;
1537 	if (sopt->sopt_level != SOL_SOCKET) {
1538 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1539 			return (so_pr_ctloutput(so, sopt));
1540 		} else
1541 			return (ENOPROTOOPT);
1542 	} else {
1543 		switch (sopt->sopt_name) {
1544 #ifdef INET
1545 		case SO_ACCEPTFILTER:
1546 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1547 				return (EINVAL);
1548 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1549 				M_TEMP, M_WAITOK | M_ZERO);
1550 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1551 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1552 				if (so->so_accf->so_accept_filter_str != NULL)
1553 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1554 			}
1555 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1556 			FREE(afap, M_TEMP);
1557 			break;
1558 #endif /* INET */
1559 
1560 		case SO_LINGER:
1561 			l.l_onoff = so->so_options & SO_LINGER;
1562 			l.l_linger = so->so_linger;
1563 			error = sooptcopyout(sopt, &l, sizeof l);
1564 			break;
1565 
1566 		case SO_USELOOPBACK:
1567 		case SO_DONTROUTE:
1568 		case SO_DEBUG:
1569 		case SO_KEEPALIVE:
1570 		case SO_REUSEADDR:
1571 		case SO_REUSEPORT:
1572 		case SO_BROADCAST:
1573 		case SO_OOBINLINE:
1574 		case SO_TIMESTAMP:
1575 			optval = so->so_options & sopt->sopt_name;
1576 integer:
1577 			error = sooptcopyout(sopt, &optval, sizeof optval);
1578 			break;
1579 
1580 		case SO_TYPE:
1581 			optval = so->so_type;
1582 			goto integer;
1583 
1584 		case SO_ERROR:
1585 			optval = so->so_error;
1586 			so->so_error = 0;
1587 			goto integer;
1588 
1589 		case SO_SNDBUF:
1590 			optval = so->so_snd.ssb_hiwat;
1591 			goto integer;
1592 
1593 		case SO_RCVBUF:
1594 			optval = so->so_rcv.ssb_hiwat;
1595 			goto integer;
1596 
1597 		case SO_SNDLOWAT:
1598 			optval = so->so_snd.ssb_lowat;
1599 			goto integer;
1600 
1601 		case SO_RCVLOWAT:
1602 			optval = so->so_rcv.ssb_lowat;
1603 			goto integer;
1604 
1605 		case SO_SNDTIMEO:
1606 		case SO_RCVTIMEO:
1607 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1608 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1609 
1610 			tv.tv_sec = optval / hz;
1611 			tv.tv_usec = (optval % hz) * ustick;
1612 			error = sooptcopyout(sopt, &tv, sizeof tv);
1613 			break;
1614 
1615 		default:
1616 			error = ENOPROTOOPT;
1617 			break;
1618 		}
1619 		return (error);
1620 	}
1621 }
1622 
1623 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1624 int
1625 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1626 {
1627 	struct mbuf *m, *m_prev;
1628 	int sopt_size = sopt->sopt_valsize, msize;
1629 
1630 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1631 		   0, &msize);
1632 	if (m == NULL)
1633 		return (ENOBUFS);
1634 	m->m_len = min(msize, sopt_size);
1635 	sopt_size -= m->m_len;
1636 	*mp = m;
1637 	m_prev = m;
1638 
1639 	while (sopt_size > 0) {
1640 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1641 			   MT_DATA, 0, &msize);
1642 		if (m == NULL) {
1643 			m_freem(*mp);
1644 			return (ENOBUFS);
1645 		}
1646 		m->m_len = min(msize, sopt_size);
1647 		sopt_size -= m->m_len;
1648 		m_prev->m_next = m;
1649 		m_prev = m;
1650 	}
1651 	return (0);
1652 }
1653 
1654 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1655 int
1656 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1657 {
1658 	soopt_to_mbuf(sopt, m);
1659 	return 0;
1660 }
1661 
1662 void
1663 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1664 {
1665 	size_t valsize;
1666 	void *val;
1667 
1668 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1669 	KKASSERT(kva_p(m));
1670 	if (sopt->sopt_val == NULL)
1671 		return;
1672 	val = sopt->sopt_val;
1673 	valsize = sopt->sopt_valsize;
1674 	while (m != NULL && valsize >= m->m_len) {
1675 		bcopy(val, mtod(m, char *), m->m_len);
1676 		valsize -= m->m_len;
1677 		val = (caddr_t)val + m->m_len;
1678 		m = m->m_next;
1679 	}
1680 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1681 		panic("ip6_sooptmcopyin");
1682 }
1683 
1684 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1685 int
1686 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1687 {
1688 	return soopt_from_mbuf(sopt, m);
1689 }
1690 
1691 int
1692 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1693 {
1694 	struct mbuf *m0 = m;
1695 	size_t valsize = 0;
1696 	size_t maxsize;
1697 	void *val;
1698 
1699 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1700 	KKASSERT(kva_p(m));
1701 	if (sopt->sopt_val == NULL)
1702 		return 0;
1703 	val = sopt->sopt_val;
1704 	maxsize = sopt->sopt_valsize;
1705 	while (m != NULL && maxsize >= m->m_len) {
1706 		bcopy(mtod(m, char *), val, m->m_len);
1707 	       maxsize -= m->m_len;
1708 	       val = (caddr_t)val + m->m_len;
1709 	       valsize += m->m_len;
1710 	       m = m->m_next;
1711 	}
1712 	if (m != NULL) {
1713 		/* enough soopt buffer should be given from user-land */
1714 		m_freem(m0);
1715 		return (EINVAL);
1716 	}
1717 	sopt->sopt_valsize = valsize;
1718 	return 0;
1719 }
1720 
1721 void
1722 sohasoutofband(struct socket *so)
1723 {
1724 	if (so->so_sigio != NULL)
1725 		pgsigio(so->so_sigio, SIGURG, 0);
1726 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1727 }
1728 
1729 int
1730 sokqfilter(struct file *fp, struct knote *kn)
1731 {
1732 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1733 	struct signalsockbuf *ssb;
1734 
1735 	switch (kn->kn_filter) {
1736 	case EVFILT_READ:
1737 		if (so->so_options & SO_ACCEPTCONN)
1738 			kn->kn_fop = &solisten_filtops;
1739 		else
1740 			kn->kn_fop = &soread_filtops;
1741 		ssb = &so->so_rcv;
1742 		break;
1743 	case EVFILT_WRITE:
1744 		kn->kn_fop = &sowrite_filtops;
1745 		ssb = &so->so_snd;
1746 		break;
1747 	case EVFILT_EXCEPT:
1748 		kn->kn_fop = &soexcept_filtops;
1749 		ssb = &so->so_rcv;
1750 		break;
1751 	default:
1752 		return (EOPNOTSUPP);
1753 	}
1754 
1755 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1756 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1757 	return (0);
1758 }
1759 
1760 static void
1761 filt_sordetach(struct knote *kn)
1762 {
1763 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1764 
1765 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1766 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1767 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1768 }
1769 
1770 /*ARGSUSED*/
1771 static int
1772 filt_soread(struct knote *kn, long hint)
1773 {
1774 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1775 
1776 	if (kn->kn_sfflags & NOTE_OOB) {
1777 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1778 			kn->kn_fflags |= NOTE_OOB;
1779 			return (1);
1780 		}
1781 		return (0);
1782 	}
1783 	kn->kn_data = so->so_rcv.ssb_cc;
1784 
1785 	/*
1786 	 * Only set EOF if all data has been exhausted.
1787 	 */
1788 	if ((so->so_state & SS_CANTRCVMORE) && kn->kn_data == 0) {
1789 		kn->kn_flags |= EV_EOF;
1790 		kn->kn_fflags = so->so_error;
1791 		return (1);
1792 	}
1793 	if (so->so_error)	/* temporary udp error */
1794 		return (1);
1795 	if (kn->kn_sfflags & NOTE_LOWAT)
1796 		return (kn->kn_data >= kn->kn_sdata);
1797 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1798 		!TAILQ_EMPTY(&so->so_comp));
1799 }
1800 
1801 static void
1802 filt_sowdetach(struct knote *kn)
1803 {
1804 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1805 
1806 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
1807 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
1808 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
1809 }
1810 
1811 /*ARGSUSED*/
1812 static int
1813 filt_sowrite(struct knote *kn, long hint)
1814 {
1815 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1816 
1817 	kn->kn_data = ssb_space(&so->so_snd);
1818 	if (so->so_state & SS_CANTSENDMORE) {
1819 		kn->kn_flags |= EV_EOF;
1820 		kn->kn_fflags = so->so_error;
1821 		return (1);
1822 	}
1823 	if (so->so_error)	/* temporary udp error */
1824 		return (1);
1825 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1826 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1827 		return (0);
1828 	if (kn->kn_sfflags & NOTE_LOWAT)
1829 		return (kn->kn_data >= kn->kn_sdata);
1830 	return (kn->kn_data >= so->so_snd.ssb_lowat);
1831 }
1832 
1833 /*ARGSUSED*/
1834 static int
1835 filt_solisten(struct knote *kn, long hint)
1836 {
1837 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1838 
1839 	kn->kn_data = so->so_qlen;
1840 	return (! TAILQ_EMPTY(&so->so_comp));
1841 }
1842