xref: /dflybsd-src/sys/kern/uipc_socket.c (revision 6cef7136f04e2b24a6db289e78720d6d8c60274e)
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 	so->so_port = cpu0_soport(so, NULL, NULL);
211 
212 	TAILQ_INIT(&so->so_incomp);
213 	TAILQ_INIT(&so->so_comp);
214 	so->so_type = type;
215 	so->so_cred = crhold(p->p_ucred);
216 	so->so_proto = prp;
217 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
218 	ai.p_ucred = p->p_ucred;
219 	ai.fd_rdir = p->p_fd->fd_rdir;
220 
221 	/*
222 	 * Auto-sizing of socket buffers is managed by the protocols and
223 	 * the appropriate flags must be set in the pru_attach function.
224 	 */
225 	error = so_pru_attach(so, proto, &ai);
226 	if (error) {
227 		sosetstate(so, SS_NOFDREF);
228 		sofree(so);	/* from soalloc */
229 		return error;
230 	}
231 
232 	/*
233 	 * NOTE: Returns referenced socket.
234 	 */
235 	*aso = so;
236 	return (0);
237 }
238 
239 int
240 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
241 {
242 	int error;
243 
244 	error = so_pru_bind(so, nam, td);
245 	return (error);
246 }
247 
248 static void
249 sodealloc(struct socket *so)
250 {
251 	if (so->so_rcv.ssb_hiwat)
252 		(void)chgsbsize(so->so_cred->cr_uidinfo,
253 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
254 	if (so->so_snd.ssb_hiwat)
255 		(void)chgsbsize(so->so_cred->cr_uidinfo,
256 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
257 #ifdef INET
258 	/* remove accept filter if present */
259 	if (so->so_accf != NULL)
260 		do_setopt_accept_filter(so, NULL);
261 #endif /* INET */
262 	crfree(so->so_cred);
263 	kfree(so, M_SOCKET);
264 }
265 
266 int
267 solisten(struct socket *so, int backlog, struct thread *td)
268 {
269 	int error;
270 #ifdef SCTP
271 	short oldopt, oldqlimit;
272 #endif /* SCTP */
273 
274 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
275 		return (EINVAL);
276 
277 #ifdef SCTP
278 	oldopt = so->so_options;
279 	oldqlimit = so->so_qlimit;
280 #endif /* SCTP */
281 
282 	lwkt_gettoken(&so->so_rcv.ssb_token);
283 	if (TAILQ_EMPTY(&so->so_comp))
284 		so->so_options |= SO_ACCEPTCONN;
285 	lwkt_reltoken(&so->so_rcv.ssb_token);
286 	if (backlog < 0 || backlog > somaxconn)
287 		backlog = somaxconn;
288 	so->so_qlimit = backlog;
289 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
290 	 * the so_options (UDP model both connect's and gets inbound
291 	 * connections .. implicitly).
292 	 */
293 	error = so_pru_listen(so, td);
294 	if (error) {
295 #ifdef SCTP
296 		/* Restore the params */
297 		so->so_options = oldopt;
298 		so->so_qlimit = oldqlimit;
299 #endif /* SCTP */
300 		return (error);
301 	}
302 	return (0);
303 }
304 
305 /*
306  * Destroy a disconnected socket.  This routine is a NOP if entities
307  * still have a reference on the socket:
308  *
309  *	so_pcb -	The protocol stack still has a reference
310  *	SS_NOFDREF -	There is no longer a file pointer reference
311  */
312 void
313 sofree(struct socket *so)
314 {
315 	struct socket *head = so->so_head;
316 
317 	/*
318 	 * Arbitrage the last free.
319 	 */
320 	KKASSERT(so->so_refs > 0);
321 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1)
322 		return;
323 
324 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
325 
326 	/*
327 	 * We're done, clean up
328 	 */
329 	if (head != NULL) {
330 		lwkt_gettoken(&head->so_rcv.ssb_token);
331 		if (so->so_state & SS_INCOMP) {
332 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
333 			head->so_incqlen--;
334 		} else if (so->so_state & SS_COMP) {
335 			/*
336 			 * We must not decommission a socket that's
337 			 * on the accept(2) queue.  If we do, then
338 			 * accept(2) may hang after select(2) indicated
339 			 * that the listening socket was ready.
340 			 */
341 			lwkt_reltoken(&head->so_rcv.ssb_token);
342 			return;
343 		} else {
344 			panic("sofree: not queued");
345 		}
346 		soclrstate(so, SS_INCOMP);
347 		so->so_head = NULL;
348 		lwkt_reltoken(&head->so_rcv.ssb_token);
349 	}
350 	ssb_release(&so->so_snd, so);
351 	sorflush(so);
352 	sodealloc(so);
353 }
354 
355 /*
356  * Close a socket on last file table reference removal.
357  * Initiate disconnect if connected.
358  * Free socket when disconnect complete.
359  */
360 int
361 soclose(struct socket *so, int fflag)
362 {
363 	int error = 0;
364 
365 	funsetown(so->so_sigio);
366 	if (so->so_pcb == NULL)
367 		goto discard;
368 	if (so->so_state & SS_ISCONNECTED) {
369 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
370 			error = sodisconnect(so);
371 			if (error)
372 				goto drop;
373 		}
374 		if (so->so_options & SO_LINGER) {
375 			if ((so->so_state & SS_ISDISCONNECTING) &&
376 			    (fflag & FNONBLOCK))
377 				goto drop;
378 			while (so->so_state & SS_ISCONNECTED) {
379 				error = tsleep(&so->so_timeo, PCATCH,
380 					       "soclos", so->so_linger * hz);
381 				if (error)
382 					break;
383 			}
384 		}
385 	}
386 drop:
387 	if (so->so_pcb) {
388 		int error2;
389 
390 		error2 = so_pru_detach(so);
391 		if (error == 0)
392 			error = error2;
393 	}
394 discard:
395 	lwkt_gettoken(&so->so_rcv.ssb_token);
396 	if (so->so_options & SO_ACCEPTCONN) {
397 		struct socket *sp;
398 
399 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
400 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
401 			soclrstate(sp, SS_INCOMP);
402 			sp->so_head = NULL;
403 			so->so_incqlen--;
404 			soaborta(sp);
405 		}
406 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
407 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
408 			soclrstate(sp, SS_COMP);
409 			sp->so_head = NULL;
410 			so->so_qlen--;
411 			soaborta(sp);
412 		}
413 	}
414 	lwkt_reltoken(&so->so_rcv.ssb_token);
415 	if (so->so_state & SS_NOFDREF)
416 		panic("soclose: NOFDREF");
417 	sosetstate(so, SS_NOFDREF);	/* take ref */
418 	sofree(so);			/* dispose of ref */
419 	return (error);
420 }
421 
422 /*
423  * Abort and destroy a socket.  Only one abort can be in progress
424  * at any given moment.
425  */
426 void
427 soabort(struct socket *so)
428 {
429 	soreference(so);
430 	so_pru_abort(so);
431 }
432 
433 void
434 soaborta(struct socket *so)
435 {
436 	soreference(so);
437 	so_pru_aborta(so);
438 }
439 
440 void
441 soabort_oncpu(struct socket *so)
442 {
443 	soreference(so);
444 	so_pru_abort_oncpu(so);
445 }
446 
447 int
448 soaccept(struct socket *so, struct sockaddr **nam)
449 {
450 	int error;
451 
452 	if ((so->so_state & SS_NOFDREF) == 0)
453 		panic("soaccept: !NOFDREF");
454 	soreference(so);		/* create ref */
455 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
456 	error = so_pru_accept(so, nam);
457 	return (error);
458 }
459 
460 int
461 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
462 {
463 	int error;
464 
465 	if (so->so_options & SO_ACCEPTCONN)
466 		return (EOPNOTSUPP);
467 	/*
468 	 * If protocol is connection-based, can only connect once.
469 	 * Otherwise, if connected, try to disconnect first.
470 	 * This allows user to disconnect by connecting to, e.g.,
471 	 * a null address.
472 	 */
473 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
474 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
475 	    (error = sodisconnect(so)))) {
476 		error = EISCONN;
477 	} else {
478 		/*
479 		 * Prevent accumulated error from previous connection
480 		 * from biting us.
481 		 */
482 		so->so_error = 0;
483 		error = so_pru_connect(so, nam, td);
484 	}
485 	return (error);
486 }
487 
488 int
489 soconnect2(struct socket *so1, struct socket *so2)
490 {
491 	int error;
492 
493 	error = so_pru_connect2(so1, so2);
494 	return (error);
495 }
496 
497 int
498 sodisconnect(struct socket *so)
499 {
500 	int error;
501 
502 	if ((so->so_state & SS_ISCONNECTED) == 0) {
503 		error = ENOTCONN;
504 		goto bad;
505 	}
506 	if (so->so_state & SS_ISDISCONNECTING) {
507 		error = EALREADY;
508 		goto bad;
509 	}
510 	error = so_pru_disconnect(so);
511 bad:
512 	return (error);
513 }
514 
515 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
516 /*
517  * Send on a socket.
518  * If send must go all at once and message is larger than
519  * send buffering, then hard error.
520  * Lock against other senders.
521  * If must go all at once and not enough room now, then
522  * inform user that this would block and do nothing.
523  * Otherwise, if nonblocking, send as much as possible.
524  * The data to be sent is described by "uio" if nonzero,
525  * otherwise by the mbuf chain "top" (which must be null
526  * if uio is not).  Data provided in mbuf chain must be small
527  * enough to send all at once.
528  *
529  * Returns nonzero on error, timeout or signal; callers
530  * must check for short counts if EINTR/ERESTART are returned.
531  * Data and control buffers are freed on return.
532  */
533 int
534 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
535 	struct mbuf *top, struct mbuf *control, int flags,
536 	struct thread *td)
537 {
538 	struct mbuf **mp;
539 	struct mbuf *m;
540 	size_t resid;
541 	int space, len;
542 	int clen = 0, error, dontroute, mlen;
543 	int atomic = sosendallatonce(so) || top;
544 	int pru_flags;
545 
546 	if (uio) {
547 		resid = uio->uio_resid;
548 	} else {
549 		resid = (size_t)top->m_pkthdr.len;
550 #ifdef INVARIANTS
551 		len = 0;
552 		for (m = top; m; m = m->m_next)
553 			len += m->m_len;
554 		KKASSERT(top->m_pkthdr.len == len);
555 #endif
556 	}
557 
558 	/*
559 	 * WARNING!  resid is unsigned, space and len are signed.  space
560 	 * 	     can wind up negative if the sockbuf is overcommitted.
561 	 *
562 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
563 	 * type sockets since that's an error.
564 	 */
565 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
566 		error = EINVAL;
567 		goto out;
568 	}
569 
570 	dontroute =
571 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
572 	    (so->so_proto->pr_flags & PR_ATOMIC);
573 	if (td->td_lwp != NULL)
574 		td->td_lwp->lwp_ru.ru_msgsnd++;
575 	if (control)
576 		clen = control->m_len;
577 #define	gotoerr(errcode)	{ error = errcode; goto release; }
578 
579 restart:
580 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
581 	if (error)
582 		goto out;
583 
584 	do {
585 		if (so->so_state & SS_CANTSENDMORE)
586 			gotoerr(EPIPE);
587 		if (so->so_error) {
588 			error = so->so_error;
589 			so->so_error = 0;
590 			goto release;
591 		}
592 		if ((so->so_state & SS_ISCONNECTED) == 0) {
593 			/*
594 			 * `sendto' and `sendmsg' is allowed on a connection-
595 			 * based socket if it supports implied connect.
596 			 * Return ENOTCONN if not connected and no address is
597 			 * supplied.
598 			 */
599 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
600 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
601 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
602 				    !(resid == 0 && clen != 0))
603 					gotoerr(ENOTCONN);
604 			} else if (addr == 0)
605 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
606 				   ENOTCONN : EDESTADDRREQ);
607 		}
608 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
609 		    clen > so->so_snd.ssb_hiwat) {
610 			gotoerr(EMSGSIZE);
611 		}
612 		space = ssb_space(&so->so_snd);
613 		if (flags & MSG_OOB)
614 			space += 1024;
615 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
616 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
617 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
618 				gotoerr(EWOULDBLOCK);
619 			ssb_unlock(&so->so_snd);
620 			error = ssb_wait(&so->so_snd);
621 			if (error)
622 				goto out;
623 			goto restart;
624 		}
625 		mp = &top;
626 		space -= clen;
627 		do {
628 		    if (uio == NULL) {
629 			/*
630 			 * Data is prepackaged in "top".
631 			 */
632 			resid = 0;
633 			if (flags & MSG_EOR)
634 				top->m_flags |= M_EOR;
635 		    } else do {
636 			if (resid > INT_MAX)
637 				resid = INT_MAX;
638 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
639 				   top == NULL ? M_PKTHDR : 0, &mlen);
640 			if (top == NULL) {
641 				m->m_pkthdr.len = 0;
642 				m->m_pkthdr.rcvif = NULL;
643 			}
644 			len = imin((int)szmin(mlen, resid), space);
645 			if (resid < MINCLSIZE) {
646 				/*
647 				 * For datagram protocols, leave room
648 				 * for protocol headers in first mbuf.
649 				 */
650 				if (atomic && top == 0 && len < mlen)
651 					MH_ALIGN(m, len);
652 			}
653 			space -= len;
654 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
655 			resid = uio->uio_resid;
656 			m->m_len = len;
657 			*mp = m;
658 			top->m_pkthdr.len += len;
659 			if (error)
660 				goto release;
661 			mp = &m->m_next;
662 			if (resid == 0) {
663 				if (flags & MSG_EOR)
664 					top->m_flags |= M_EOR;
665 				break;
666 			}
667 		    } while (space > 0 && atomic);
668 		    if (dontroute)
669 			    so->so_options |= SO_DONTROUTE;
670 		    if (flags & MSG_OOB) {
671 		    	    pru_flags = PRUS_OOB;
672 		    } else if ((flags & MSG_EOF) &&
673 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
674 			       (resid == 0)) {
675 			    /*
676 			     * If the user set MSG_EOF, the protocol
677 			     * understands this flag and nothing left to
678 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
679 			     */
680 		    	    pru_flags = PRUS_EOF;
681 		    } else if (resid > 0 && space > 0) {
682 			    /* If there is more to send, set PRUS_MORETOCOME */
683 		    	    pru_flags = PRUS_MORETOCOME;
684 		    } else {
685 		    	    pru_flags = 0;
686 		    }
687 		    /*
688 		     * XXX all the SS_CANTSENDMORE checks previously
689 		     * done could be out of date.  We could have recieved
690 		     * a reset packet in an interrupt or maybe we slept
691 		     * while doing page faults in uiomove() etc. We could
692 		     * probably recheck again inside the splnet() protection
693 		     * here, but there are probably other places that this
694 		     * also happens.  We must rethink this.
695 		     */
696 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
697 		    if (dontroute)
698 			    so->so_options &= ~SO_DONTROUTE;
699 		    clen = 0;
700 		    control = 0;
701 		    top = 0;
702 		    mp = &top;
703 		    if (error)
704 			    goto release;
705 		} while (resid && space > 0);
706 	} while (resid);
707 
708 release:
709 	ssb_unlock(&so->so_snd);
710 out:
711 	if (top)
712 		m_freem(top);
713 	if (control)
714 		m_freem(control);
715 	return (error);
716 }
717 
718 /*
719  * A specialization of sosend() for UDP based on protocol-specific knowledge:
720  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
721  *	sosendallatonce() returns true,
722  *	the "atomic" variable is true,
723  *	and sosendudp() blocks until space is available for the entire send.
724  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
725  *	PR_IMPLOPCL flags set.
726  *   UDP has no out-of-band data.
727  *   UDP has no control data.
728  *   UDP does not support MSG_EOR.
729  */
730 int
731 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
732 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
733 {
734 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
735 	size_t resid;
736 	int error;
737 	int space;
738 
739 	if (td->td_lwp != NULL)
740 		td->td_lwp->lwp_ru.ru_msgsnd++;
741 	if (control)
742 		m_freem(control);
743 
744 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
745 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
746 
747 restart:
748 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
749 	if (error)
750 		goto out;
751 
752 	if (so->so_state & SS_CANTSENDMORE)
753 		gotoerr(EPIPE);
754 	if (so->so_error) {
755 		error = so->so_error;
756 		so->so_error = 0;
757 		goto release;
758 	}
759 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
760 		gotoerr(EDESTADDRREQ);
761 	if (resid > so->so_snd.ssb_hiwat)
762 		gotoerr(EMSGSIZE);
763 	space = ssb_space(&so->so_snd);
764 	if (uio && (space < 0 || (size_t)space < resid)) {
765 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
766 			gotoerr(EWOULDBLOCK);
767 		ssb_unlock(&so->so_snd);
768 		error = ssb_wait(&so->so_snd);
769 		if (error)
770 			goto out;
771 		goto restart;
772 	}
773 
774 	if (uio) {
775 		top = m_uiomove(uio);
776 		if (top == NULL)
777 			goto release;
778 	}
779 
780 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
781 	if (dontroute)
782 		so->so_options |= SO_DONTROUTE;
783 
784 	error = so_pru_send(so, 0, top, addr, NULL, td);
785 	top = NULL;		/* sent or freed in lower layer */
786 
787 	if (dontroute)
788 		so->so_options &= ~SO_DONTROUTE;
789 
790 release:
791 	ssb_unlock(&so->so_snd);
792 out:
793 	if (top)
794 		m_freem(top);
795 	return (error);
796 }
797 
798 /*
799  * Implement receive operations on a socket.
800  *
801  * We depend on the way that records are added to the signalsockbuf
802  * by sbappend*.  In particular, each record (mbufs linked through m_next)
803  * must begin with an address if the protocol so specifies,
804  * followed by an optional mbuf or mbufs containing ancillary data,
805  * and then zero or more mbufs of data.
806  *
807  * Although the signalsockbuf is locked, new data may still be appended.
808  * A token inside the ssb_lock deals with MP issues and still allows
809  * the network to access the socket if we block in a uio.
810  *
811  * The caller may receive the data as a single mbuf chain by supplying
812  * an mbuf **mp0 for use in returning the chain.  The uio is then used
813  * only for the count in uio_resid.
814  */
815 int
816 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
817 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
818 {
819 	struct mbuf *m, *n;
820 	struct mbuf *free_chain = NULL;
821 	int flags, len, error, offset;
822 	struct protosw *pr = so->so_proto;
823 	int moff, type = 0;
824 	size_t resid, orig_resid;
825 
826 	if (uio)
827 		resid = uio->uio_resid;
828 	else
829 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
830 	orig_resid = resid;
831 
832 	if (psa)
833 		*psa = NULL;
834 	if (controlp)
835 		*controlp = NULL;
836 	if (flagsp)
837 		flags = *flagsp &~ MSG_EOR;
838 	else
839 		flags = 0;
840 	if (flags & MSG_OOB) {
841 		m = m_get(MB_WAIT, MT_DATA);
842 		if (m == NULL)
843 			return (ENOBUFS);
844 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
845 		if (error)
846 			goto bad;
847 		if (sio) {
848 			do {
849 				sbappend(sio, m);
850 				KKASSERT(resid >= (size_t)m->m_len);
851 				resid -= (size_t)m->m_len;
852 			} while (resid > 0 && m);
853 		} else {
854 			do {
855 				uio->uio_resid = resid;
856 				error = uiomove(mtod(m, caddr_t),
857 						(int)szmin(resid, m->m_len),
858 						uio);
859 				resid = uio->uio_resid;
860 				m = m_free(m);
861 			} while (uio->uio_resid && error == 0 && m);
862 		}
863 bad:
864 		if (m)
865 			m_freem(m);
866 		return (error);
867 	}
868 	if ((so->so_state & SS_ISCONFIRMING) && resid)
869 		so_pru_rcvd(so, 0);
870 
871 restart:
872 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
873 	if (error)
874 		goto done;
875 
876 	m = so->so_rcv.ssb_mb;
877 	/*
878 	 * If we have less data than requested, block awaiting more
879 	 * (subject to any timeout) if:
880 	 *   1. the current count is less than the low water mark, or
881 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
882 	 *	receive operation at once if we block (resid <= hiwat).
883 	 *   3. MSG_DONTWAIT is not set
884 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
885 	 * we have to do the receive in sections, and thus risk returning
886 	 * a short count if a timeout or signal occurs after we start.
887 	 */
888 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
889 	    (size_t)so->so_rcv.ssb_cc < resid) &&
890 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
891 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
892 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
893 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
894 		if (so->so_error) {
895 			if (m)
896 				goto dontblock;
897 			error = so->so_error;
898 			if ((flags & MSG_PEEK) == 0)
899 				so->so_error = 0;
900 			goto release;
901 		}
902 		if (so->so_state & SS_CANTRCVMORE) {
903 			if (m)
904 				goto dontblock;
905 			else
906 				goto release;
907 		}
908 		for (; m; m = m->m_next) {
909 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
910 				m = so->so_rcv.ssb_mb;
911 				goto dontblock;
912 			}
913 		}
914 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
915 		    (pr->pr_flags & PR_CONNREQUIRED)) {
916 			error = ENOTCONN;
917 			goto release;
918 		}
919 		if (resid == 0)
920 			goto release;
921 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
922 			error = EWOULDBLOCK;
923 			goto release;
924 		}
925 		ssb_unlock(&so->so_rcv);
926 		error = ssb_wait(&so->so_rcv);
927 		if (error)
928 			goto done;
929 		goto restart;
930 	}
931 dontblock:
932 	if (uio && uio->uio_td && uio->uio_td->td_proc)
933 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
934 
935 	/*
936 	 * note: m should be == sb_mb here.  Cache the next record while
937 	 * cleaning up.  Note that calling m_free*() will break out critical
938 	 * section.
939 	 */
940 	KKASSERT(m == so->so_rcv.ssb_mb);
941 
942 	/*
943 	 * Skip any address mbufs prepending the record.
944 	 */
945 	if (pr->pr_flags & PR_ADDR) {
946 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
947 		orig_resid = 0;
948 		if (psa)
949 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
950 		if (flags & MSG_PEEK)
951 			m = m->m_next;
952 		else
953 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
954 	}
955 
956 	/*
957 	 * Skip any control mbufs prepending the record.
958 	 */
959 #ifdef SCTP
960 	if (pr->pr_flags & PR_ADDR_OPT) {
961 		/*
962 		 * For SCTP we may be getting a
963 		 * whole message OR a partial delivery.
964 		 */
965 		if (m && m->m_type == MT_SONAME) {
966 			orig_resid = 0;
967 			if (psa)
968 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
969 			if (flags & MSG_PEEK)
970 				m = m->m_next;
971 			else
972 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
973 		}
974 	}
975 #endif /* SCTP */
976 	while (m && m->m_type == MT_CONTROL && error == 0) {
977 		if (flags & MSG_PEEK) {
978 			if (controlp)
979 				*controlp = m_copy(m, 0, m->m_len);
980 			m = m->m_next;	/* XXX race */
981 		} else {
982 			if (controlp) {
983 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
984 				if (pr->pr_domain->dom_externalize &&
985 				    mtod(m, struct cmsghdr *)->cmsg_type ==
986 				    SCM_RIGHTS)
987 				   error = (*pr->pr_domain->dom_externalize)(m);
988 				*controlp = m;
989 				m = n;
990 			} else {
991 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
992 			}
993 		}
994 		if (controlp && *controlp) {
995 			orig_resid = 0;
996 			controlp = &(*controlp)->m_next;
997 		}
998 	}
999 
1000 	/*
1001 	 * flag OOB data.
1002 	 */
1003 	if (m) {
1004 		type = m->m_type;
1005 		if (type == MT_OOBDATA)
1006 			flags |= MSG_OOB;
1007 	}
1008 
1009 	/*
1010 	 * Copy to the UIO or mbuf return chain (*mp).
1011 	 */
1012 	moff = 0;
1013 	offset = 0;
1014 	while (m && resid > 0 && error == 0) {
1015 		if (m->m_type == MT_OOBDATA) {
1016 			if (type != MT_OOBDATA)
1017 				break;
1018 		} else if (type == MT_OOBDATA)
1019 			break;
1020 		else
1021 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1022 			("receive 3"));
1023 		soclrstate(so, SS_RCVATMARK);
1024 		len = (resid > INT_MAX) ? INT_MAX : resid;
1025 		if (so->so_oobmark && len > so->so_oobmark - offset)
1026 			len = so->so_oobmark - offset;
1027 		if (len > m->m_len - moff)
1028 			len = m->m_len - moff;
1029 
1030 		/*
1031 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1032 		 * The SIO is dealt with when we eat the mbuf, but deal
1033 		 * with the resid here either way.
1034 		 */
1035 		if (uio) {
1036 			uio->uio_resid = resid;
1037 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1038 			resid = uio->uio_resid;
1039 			if (error)
1040 				goto release;
1041 		} else {
1042 			resid -= (size_t)len;
1043 		}
1044 
1045 		/*
1046 		 * Eat the entire mbuf or just a piece of it
1047 		 */
1048 		if (len == m->m_len - moff) {
1049 			if (m->m_flags & M_EOR)
1050 				flags |= MSG_EOR;
1051 #ifdef SCTP
1052 			if (m->m_flags & M_NOTIFICATION)
1053 				flags |= MSG_NOTIFICATION;
1054 #endif /* SCTP */
1055 			if (flags & MSG_PEEK) {
1056 				m = m->m_next;
1057 				moff = 0;
1058 			} else {
1059 				if (sio) {
1060 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1061 					sbappend(sio, m);
1062 					m = n;
1063 				} else {
1064 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1065 				}
1066 			}
1067 		} else {
1068 			if (flags & MSG_PEEK) {
1069 				moff += len;
1070 			} else {
1071 				if (sio) {
1072 					n = m_copym(m, 0, len, MB_WAIT);
1073 					if (n)
1074 						sbappend(sio, n);
1075 				}
1076 				m->m_data += len;
1077 				m->m_len -= len;
1078 				so->so_rcv.ssb_cc -= len;
1079 			}
1080 		}
1081 		if (so->so_oobmark) {
1082 			if ((flags & MSG_PEEK) == 0) {
1083 				so->so_oobmark -= len;
1084 				if (so->so_oobmark == 0) {
1085 					sosetstate(so, SS_RCVATMARK);
1086 					break;
1087 				}
1088 			} else {
1089 				offset += len;
1090 				if (offset == so->so_oobmark)
1091 					break;
1092 			}
1093 		}
1094 		if (flags & MSG_EOR)
1095 			break;
1096 		/*
1097 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1098 		 * we must not quit until resid == 0 or an error
1099 		 * termination.  If a signal/timeout occurs, return
1100 		 * with a short count but without error.
1101 		 * Keep signalsockbuf locked against other readers.
1102 		 */
1103 		while ((flags & MSG_WAITALL) && m == NULL &&
1104 		       resid > 0 && !sosendallatonce(so) &&
1105 		       so->so_rcv.ssb_mb == NULL) {
1106 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1107 				break;
1108 			/*
1109 			 * The window might have closed to zero, make
1110 			 * sure we send an ack now that we've drained
1111 			 * the buffer or we might end up blocking until
1112 			 * the idle takes over (5 seconds).
1113 			 */
1114 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1115 				so_pru_rcvd(so, flags);
1116 			error = ssb_wait(&so->so_rcv);
1117 			if (error) {
1118 				ssb_unlock(&so->so_rcv);
1119 				error = 0;
1120 				goto done;
1121 			}
1122 			m = so->so_rcv.ssb_mb;
1123 		}
1124 	}
1125 
1126 	/*
1127 	 * If an atomic read was requested but unread data still remains
1128 	 * in the record, set MSG_TRUNC.
1129 	 */
1130 	if (m && pr->pr_flags & PR_ATOMIC)
1131 		flags |= MSG_TRUNC;
1132 
1133 	/*
1134 	 * Cleanup.  If an atomic read was requested drop any unread data.
1135 	 */
1136 	if ((flags & MSG_PEEK) == 0) {
1137 		if (m && (pr->pr_flags & PR_ATOMIC))
1138 			sbdroprecord(&so->so_rcv.sb);
1139 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1140 			so_pru_rcvd(so, flags);
1141 	}
1142 
1143 	if (orig_resid == resid && orig_resid &&
1144 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1145 		ssb_unlock(&so->so_rcv);
1146 		goto restart;
1147 	}
1148 
1149 	if (flagsp)
1150 		*flagsp |= flags;
1151 release:
1152 	ssb_unlock(&so->so_rcv);
1153 done:
1154 	if (free_chain)
1155 		m_freem(free_chain);
1156 	return (error);
1157 }
1158 
1159 int
1160 soshutdown(struct socket *so, int how)
1161 {
1162 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1163 		return (EINVAL);
1164 
1165 	if (how != SHUT_WR)
1166 		sorflush(so);
1167 	if (how != SHUT_RD)
1168 		return (so_pru_shutdown(so));
1169 	return (0);
1170 }
1171 
1172 void
1173 sorflush(struct socket *so)
1174 {
1175 	struct signalsockbuf *ssb = &so->so_rcv;
1176 	struct protosw *pr = so->so_proto;
1177 	struct signalsockbuf asb;
1178 
1179 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1180 
1181 	ssb_lock(ssb, M_WAITOK);
1182 	socantrcvmore(so);
1183 	asb = *ssb;
1184 
1185 	/*
1186 	 * Can't just blow up the ssb structure here
1187 	 */
1188 	ssb->ssb_timeo = 0;
1189 	ssb->ssb_unused01 = 0;
1190 	ssb->ssb_lowat = 0;
1191 	ssb->ssb_hiwat = 0;
1192 	ssb->ssb_mbmax = 0;
1193 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1194 
1195 	ssb_unlock(ssb);
1196 
1197 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1198 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1199 	ssb_release(&asb, so);
1200 }
1201 
1202 #ifdef INET
1203 static int
1204 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1205 {
1206 	struct accept_filter_arg	*afap = NULL;
1207 	struct accept_filter	*afp;
1208 	struct so_accf	*af = so->so_accf;
1209 	int	error = 0;
1210 
1211 	/* do not set/remove accept filters on non listen sockets */
1212 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1213 		error = EINVAL;
1214 		goto out;
1215 	}
1216 
1217 	/* removing the filter */
1218 	if (sopt == NULL) {
1219 		if (af != NULL) {
1220 			if (af->so_accept_filter != NULL &&
1221 				af->so_accept_filter->accf_destroy != NULL) {
1222 				af->so_accept_filter->accf_destroy(so);
1223 			}
1224 			if (af->so_accept_filter_str != NULL) {
1225 				FREE(af->so_accept_filter_str, M_ACCF);
1226 			}
1227 			FREE(af, M_ACCF);
1228 			so->so_accf = NULL;
1229 		}
1230 		so->so_options &= ~SO_ACCEPTFILTER;
1231 		return (0);
1232 	}
1233 	/* adding a filter */
1234 	/* must remove previous filter first */
1235 	if (af != NULL) {
1236 		error = EINVAL;
1237 		goto out;
1238 	}
1239 	/* don't put large objects on the kernel stack */
1240 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1241 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1242 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1243 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1244 	if (error)
1245 		goto out;
1246 	afp = accept_filt_get(afap->af_name);
1247 	if (afp == NULL) {
1248 		error = ENOENT;
1249 		goto out;
1250 	}
1251 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1252 	if (afp->accf_create != NULL) {
1253 		if (afap->af_name[0] != '\0') {
1254 			int len = strlen(afap->af_name) + 1;
1255 
1256 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1257 			strcpy(af->so_accept_filter_str, afap->af_name);
1258 		}
1259 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1260 		if (af->so_accept_filter_arg == NULL) {
1261 			FREE(af->so_accept_filter_str, M_ACCF);
1262 			FREE(af, M_ACCF);
1263 			so->so_accf = NULL;
1264 			error = EINVAL;
1265 			goto out;
1266 		}
1267 	}
1268 	af->so_accept_filter = afp;
1269 	so->so_accf = af;
1270 	so->so_options |= SO_ACCEPTFILTER;
1271 out:
1272 	if (afap != NULL)
1273 		FREE(afap, M_TEMP);
1274 	return (error);
1275 }
1276 #endif /* INET */
1277 
1278 /*
1279  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1280  * an additional variant to handle the case where the option value needs
1281  * to be some kind of integer, but not a specific size.
1282  * In addition to their use here, these functions are also called by the
1283  * protocol-level pr_ctloutput() routines.
1284  */
1285 int
1286 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1287 {
1288 	return soopt_to_kbuf(sopt, buf, len, minlen);
1289 }
1290 
1291 int
1292 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1293 {
1294 	size_t	valsize;
1295 
1296 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1297 	KKASSERT(kva_p(buf));
1298 
1299 	/*
1300 	 * If the user gives us more than we wanted, we ignore it,
1301 	 * but if we don't get the minimum length the caller
1302 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1303 	 * is set to however much we actually retrieved.
1304 	 */
1305 	if ((valsize = sopt->sopt_valsize) < minlen)
1306 		return EINVAL;
1307 	if (valsize > len)
1308 		sopt->sopt_valsize = valsize = len;
1309 
1310 	bcopy(sopt->sopt_val, buf, valsize);
1311 	return 0;
1312 }
1313 
1314 
1315 int
1316 sosetopt(struct socket *so, struct sockopt *sopt)
1317 {
1318 	int	error, optval;
1319 	struct	linger l;
1320 	struct	timeval tv;
1321 	u_long  val;
1322 	struct signalsockbuf *sotmp;
1323 
1324 	error = 0;
1325 	sopt->sopt_dir = SOPT_SET;
1326 	if (sopt->sopt_level != SOL_SOCKET) {
1327 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1328 			return (so_pru_ctloutput(so, sopt));
1329 		}
1330 		error = ENOPROTOOPT;
1331 	} else {
1332 		switch (sopt->sopt_name) {
1333 #ifdef INET
1334 		case SO_ACCEPTFILTER:
1335 			error = do_setopt_accept_filter(so, sopt);
1336 			if (error)
1337 				goto bad;
1338 			break;
1339 #endif /* INET */
1340 		case SO_LINGER:
1341 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1342 			if (error)
1343 				goto bad;
1344 
1345 			so->so_linger = l.l_linger;
1346 			if (l.l_onoff)
1347 				so->so_options |= SO_LINGER;
1348 			else
1349 				so->so_options &= ~SO_LINGER;
1350 			break;
1351 
1352 		case SO_DEBUG:
1353 		case SO_KEEPALIVE:
1354 		case SO_DONTROUTE:
1355 		case SO_USELOOPBACK:
1356 		case SO_BROADCAST:
1357 		case SO_REUSEADDR:
1358 		case SO_REUSEPORT:
1359 		case SO_OOBINLINE:
1360 		case SO_TIMESTAMP:
1361 			error = sooptcopyin(sopt, &optval, sizeof optval,
1362 					    sizeof optval);
1363 			if (error)
1364 				goto bad;
1365 			if (optval)
1366 				so->so_options |= sopt->sopt_name;
1367 			else
1368 				so->so_options &= ~sopt->sopt_name;
1369 			break;
1370 
1371 		case SO_SNDBUF:
1372 		case SO_RCVBUF:
1373 		case SO_SNDLOWAT:
1374 		case SO_RCVLOWAT:
1375 			error = sooptcopyin(sopt, &optval, sizeof optval,
1376 					    sizeof optval);
1377 			if (error)
1378 				goto bad;
1379 
1380 			/*
1381 			 * Values < 1 make no sense for any of these
1382 			 * options, so disallow them.
1383 			 */
1384 			if (optval < 1) {
1385 				error = EINVAL;
1386 				goto bad;
1387 			}
1388 
1389 			switch (sopt->sopt_name) {
1390 			case SO_SNDBUF:
1391 			case SO_RCVBUF:
1392 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1393 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1394 				    so,
1395 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1396 					error = ENOBUFS;
1397 					goto bad;
1398 				}
1399 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1400 						&so->so_snd : &so->so_rcv;
1401 				atomic_clear_int(&sotmp->ssb_flags,
1402 						 SSB_AUTOSIZE);
1403 				break;
1404 
1405 			/*
1406 			 * Make sure the low-water is never greater than
1407 			 * the high-water.
1408 			 */
1409 			case SO_SNDLOWAT:
1410 				so->so_snd.ssb_lowat =
1411 				    (optval > so->so_snd.ssb_hiwat) ?
1412 				    so->so_snd.ssb_hiwat : optval;
1413 				atomic_clear_int(&so->so_snd.ssb_flags,
1414 						 SSB_AUTOLOWAT);
1415 				break;
1416 			case SO_RCVLOWAT:
1417 				so->so_rcv.ssb_lowat =
1418 				    (optval > so->so_rcv.ssb_hiwat) ?
1419 				    so->so_rcv.ssb_hiwat : optval;
1420 				atomic_clear_int(&so->so_rcv.ssb_flags,
1421 						 SSB_AUTOLOWAT);
1422 				break;
1423 			}
1424 			break;
1425 
1426 		case SO_SNDTIMEO:
1427 		case SO_RCVTIMEO:
1428 			error = sooptcopyin(sopt, &tv, sizeof tv,
1429 					    sizeof tv);
1430 			if (error)
1431 				goto bad;
1432 
1433 			/* assert(hz > 0); */
1434 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1435 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1436 				error = EDOM;
1437 				goto bad;
1438 			}
1439 			/* assert(tick > 0); */
1440 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1441 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1442 			if (val > SHRT_MAX) {
1443 				error = EDOM;
1444 				goto bad;
1445 			}
1446 			if (val == 0 && tv.tv_usec != 0)
1447 				val = 1;
1448 
1449 			switch (sopt->sopt_name) {
1450 			case SO_SNDTIMEO:
1451 				so->so_snd.ssb_timeo = val;
1452 				break;
1453 			case SO_RCVTIMEO:
1454 				so->so_rcv.ssb_timeo = val;
1455 				break;
1456 			}
1457 			break;
1458 		default:
1459 			error = ENOPROTOOPT;
1460 			break;
1461 		}
1462 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1463 			(void) so_pru_ctloutput(so, sopt);
1464 		}
1465 	}
1466 bad:
1467 	return (error);
1468 }
1469 
1470 /* Helper routine for getsockopt */
1471 int
1472 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1473 {
1474 	soopt_from_kbuf(sopt, buf, len);
1475 	return 0;
1476 }
1477 
1478 void
1479 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1480 {
1481 	size_t	valsize;
1482 
1483 	if (len == 0) {
1484 		sopt->sopt_valsize = 0;
1485 		return;
1486 	}
1487 
1488 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1489 	KKASSERT(kva_p(buf));
1490 
1491 	/*
1492 	 * Documented get behavior is that we always return a value,
1493 	 * possibly truncated to fit in the user's buffer.
1494 	 * Traditional behavior is that we always tell the user
1495 	 * precisely how much we copied, rather than something useful
1496 	 * like the total amount we had available for her.
1497 	 * Note that this interface is not idempotent; the entire answer must
1498 	 * generated ahead of time.
1499 	 */
1500 	valsize = szmin(len, sopt->sopt_valsize);
1501 	sopt->sopt_valsize = valsize;
1502 	if (sopt->sopt_val != 0) {
1503 		bcopy(buf, sopt->sopt_val, valsize);
1504 	}
1505 }
1506 
1507 int
1508 sogetopt(struct socket *so, struct sockopt *sopt)
1509 {
1510 	int	error, optval;
1511 	struct	linger l;
1512 	struct	timeval tv;
1513 #ifdef INET
1514 	struct accept_filter_arg *afap;
1515 #endif
1516 
1517 	error = 0;
1518 	sopt->sopt_dir = SOPT_GET;
1519 	if (sopt->sopt_level != SOL_SOCKET) {
1520 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1521 			return (so_pru_ctloutput(so, sopt));
1522 		} else
1523 			return (ENOPROTOOPT);
1524 	} else {
1525 		switch (sopt->sopt_name) {
1526 #ifdef INET
1527 		case SO_ACCEPTFILTER:
1528 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1529 				return (EINVAL);
1530 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1531 				M_TEMP, M_WAITOK | M_ZERO);
1532 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1533 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1534 				if (so->so_accf->so_accept_filter_str != NULL)
1535 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1536 			}
1537 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1538 			FREE(afap, M_TEMP);
1539 			break;
1540 #endif /* INET */
1541 
1542 		case SO_LINGER:
1543 			l.l_onoff = so->so_options & SO_LINGER;
1544 			l.l_linger = so->so_linger;
1545 			error = sooptcopyout(sopt, &l, sizeof l);
1546 			break;
1547 
1548 		case SO_USELOOPBACK:
1549 		case SO_DONTROUTE:
1550 		case SO_DEBUG:
1551 		case SO_KEEPALIVE:
1552 		case SO_REUSEADDR:
1553 		case SO_REUSEPORT:
1554 		case SO_BROADCAST:
1555 		case SO_OOBINLINE:
1556 		case SO_TIMESTAMP:
1557 			optval = so->so_options & sopt->sopt_name;
1558 integer:
1559 			error = sooptcopyout(sopt, &optval, sizeof optval);
1560 			break;
1561 
1562 		case SO_TYPE:
1563 			optval = so->so_type;
1564 			goto integer;
1565 
1566 		case SO_ERROR:
1567 			optval = so->so_error;
1568 			so->so_error = 0;
1569 			goto integer;
1570 
1571 		case SO_SNDBUF:
1572 			optval = so->so_snd.ssb_hiwat;
1573 			goto integer;
1574 
1575 		case SO_RCVBUF:
1576 			optval = so->so_rcv.ssb_hiwat;
1577 			goto integer;
1578 
1579 		case SO_SNDLOWAT:
1580 			optval = so->so_snd.ssb_lowat;
1581 			goto integer;
1582 
1583 		case SO_RCVLOWAT:
1584 			optval = so->so_rcv.ssb_lowat;
1585 			goto integer;
1586 
1587 		case SO_SNDTIMEO:
1588 		case SO_RCVTIMEO:
1589 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1590 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1591 
1592 			tv.tv_sec = optval / hz;
1593 			tv.tv_usec = (optval % hz) * ustick;
1594 			error = sooptcopyout(sopt, &tv, sizeof tv);
1595 			break;
1596 
1597 		default:
1598 			error = ENOPROTOOPT;
1599 			break;
1600 		}
1601 		return (error);
1602 	}
1603 }
1604 
1605 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1606 int
1607 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1608 {
1609 	struct mbuf *m, *m_prev;
1610 	int sopt_size = sopt->sopt_valsize, msize;
1611 
1612 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1613 		   0, &msize);
1614 	if (m == NULL)
1615 		return (ENOBUFS);
1616 	m->m_len = min(msize, sopt_size);
1617 	sopt_size -= m->m_len;
1618 	*mp = m;
1619 	m_prev = m;
1620 
1621 	while (sopt_size > 0) {
1622 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1623 			   MT_DATA, 0, &msize);
1624 		if (m == NULL) {
1625 			m_freem(*mp);
1626 			return (ENOBUFS);
1627 		}
1628 		m->m_len = min(msize, sopt_size);
1629 		sopt_size -= m->m_len;
1630 		m_prev->m_next = m;
1631 		m_prev = m;
1632 	}
1633 	return (0);
1634 }
1635 
1636 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1637 int
1638 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1639 {
1640 	soopt_to_mbuf(sopt, m);
1641 	return 0;
1642 }
1643 
1644 void
1645 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1646 {
1647 	size_t valsize;
1648 	void *val;
1649 
1650 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1651 	KKASSERT(kva_p(m));
1652 	if (sopt->sopt_val == NULL)
1653 		return;
1654 	val = sopt->sopt_val;
1655 	valsize = sopt->sopt_valsize;
1656 	while (m != NULL && valsize >= m->m_len) {
1657 		bcopy(val, mtod(m, char *), m->m_len);
1658 		valsize -= m->m_len;
1659 		val = (caddr_t)val + m->m_len;
1660 		m = m->m_next;
1661 	}
1662 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1663 		panic("ip6_sooptmcopyin");
1664 }
1665 
1666 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1667 int
1668 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1669 {
1670 	return soopt_from_mbuf(sopt, m);
1671 }
1672 
1673 int
1674 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1675 {
1676 	struct mbuf *m0 = m;
1677 	size_t valsize = 0;
1678 	size_t maxsize;
1679 	void *val;
1680 
1681 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1682 	KKASSERT(kva_p(m));
1683 	if (sopt->sopt_val == NULL)
1684 		return 0;
1685 	val = sopt->sopt_val;
1686 	maxsize = sopt->sopt_valsize;
1687 	while (m != NULL && maxsize >= m->m_len) {
1688 		bcopy(mtod(m, char *), val, m->m_len);
1689 	       maxsize -= m->m_len;
1690 	       val = (caddr_t)val + m->m_len;
1691 	       valsize += m->m_len;
1692 	       m = m->m_next;
1693 	}
1694 	if (m != NULL) {
1695 		/* enough soopt buffer should be given from user-land */
1696 		m_freem(m0);
1697 		return (EINVAL);
1698 	}
1699 	sopt->sopt_valsize = valsize;
1700 	return 0;
1701 }
1702 
1703 void
1704 sohasoutofband(struct socket *so)
1705 {
1706 	if (so->so_sigio != NULL)
1707 		pgsigio(so->so_sigio, SIGURG, 0);
1708 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1709 }
1710 
1711 int
1712 sokqfilter(struct file *fp, struct knote *kn)
1713 {
1714 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1715 	struct signalsockbuf *ssb;
1716 
1717 	switch (kn->kn_filter) {
1718 	case EVFILT_READ:
1719 		if (so->so_options & SO_ACCEPTCONN)
1720 			kn->kn_fop = &solisten_filtops;
1721 		else
1722 			kn->kn_fop = &soread_filtops;
1723 		ssb = &so->so_rcv;
1724 		break;
1725 	case EVFILT_WRITE:
1726 		kn->kn_fop = &sowrite_filtops;
1727 		ssb = &so->so_snd;
1728 		break;
1729 	case EVFILT_EXCEPT:
1730 		kn->kn_fop = &soexcept_filtops;
1731 		ssb = &so->so_rcv;
1732 		break;
1733 	default:
1734 		return (EOPNOTSUPP);
1735 	}
1736 
1737 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1738 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1739 	return (0);
1740 }
1741 
1742 static void
1743 filt_sordetach(struct knote *kn)
1744 {
1745 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1746 
1747 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1748 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1749 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1750 }
1751 
1752 /*ARGSUSED*/
1753 static int
1754 filt_soread(struct knote *kn, long hint)
1755 {
1756 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1757 
1758 	if (kn->kn_sfflags & NOTE_OOB) {
1759 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1760 			kn->kn_fflags |= NOTE_OOB;
1761 			return (1);
1762 		}
1763 		return (0);
1764 	}
1765 	kn->kn_data = so->so_rcv.ssb_cc;
1766 
1767 	/*
1768 	 * Only set EOF if all data has been exhausted.
1769 	 */
1770 	if ((so->so_state & SS_CANTRCVMORE) && kn->kn_data == 0) {
1771 		kn->kn_flags |= EV_EOF;
1772 		kn->kn_fflags = so->so_error;
1773 		return (1);
1774 	}
1775 	if (so->so_error)	/* temporary udp error */
1776 		return (1);
1777 	if (kn->kn_sfflags & NOTE_LOWAT)
1778 		return (kn->kn_data >= kn->kn_sdata);
1779 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1780 		!TAILQ_EMPTY(&so->so_comp));
1781 }
1782 
1783 static void
1784 filt_sowdetach(struct knote *kn)
1785 {
1786 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1787 
1788 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
1789 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
1790 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
1791 }
1792 
1793 /*ARGSUSED*/
1794 static int
1795 filt_sowrite(struct knote *kn, long hint)
1796 {
1797 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1798 
1799 	kn->kn_data = ssb_space(&so->so_snd);
1800 	if (so->so_state & SS_CANTSENDMORE) {
1801 		kn->kn_flags |= EV_EOF;
1802 		kn->kn_fflags = so->so_error;
1803 		return (1);
1804 	}
1805 	if (so->so_error)	/* temporary udp error */
1806 		return (1);
1807 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1808 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1809 		return (0);
1810 	if (kn->kn_sfflags & NOTE_LOWAT)
1811 		return (kn->kn_data >= kn->kn_sdata);
1812 	return (kn->kn_data >= so->so_snd.ssb_lowat);
1813 }
1814 
1815 /*ARGSUSED*/
1816 static int
1817 filt_solisten(struct knote *kn, long hint)
1818 {
1819 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1820 
1821 	kn->kn_data = so->so_qlen;
1822 	return (! TAILQ_EMPTY(&so->so_comp));
1823 }
1824