xref: /dflybsd-src/sys/kern/uipc_socket.c (revision c3c96e4421a1087a390825eac6c01c9ed9182387)
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 	}
158 	return so;
159 }
160 
161 int
162 socreate(int dom, struct socket **aso, int type,
163 	int proto, struct thread *td)
164 {
165 	struct proc *p = td->td_proc;
166 	struct protosw *prp;
167 	struct socket *so;
168 	struct pru_attach_info ai;
169 	int error;
170 
171 	if (proto)
172 		prp = pffindproto(dom, proto, type);
173 	else
174 		prp = pffindtype(dom, type);
175 
176 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
177 		return (EPROTONOSUPPORT);
178 
179 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
180 	    prp->pr_domain->dom_family != PF_LOCAL &&
181 	    prp->pr_domain->dom_family != PF_INET &&
182 	    prp->pr_domain->dom_family != PF_INET6 &&
183 	    prp->pr_domain->dom_family != PF_ROUTE) {
184 		return (EPROTONOSUPPORT);
185 	}
186 
187 	if (prp->pr_type != type)
188 		return (EPROTOTYPE);
189 	so = soalloc(p != 0);
190 	if (so == 0)
191 		return (ENOBUFS);
192 
193 	/*
194 	 * Set a default port for protocol processing.  No action will occur
195 	 * on the socket on this port until an inpcb is attached to it and
196 	 * is able to match incoming packets, or until the socket becomes
197 	 * available to userland.
198 	 */
199 	so->so_port = cpu0_soport(so, NULL, NULL);
200 
201 	TAILQ_INIT(&so->so_incomp);
202 	TAILQ_INIT(&so->so_comp);
203 	so->so_type = type;
204 	so->so_cred = crhold(p->p_ucred);
205 	so->so_proto = prp;
206 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
207 	ai.p_ucred = p->p_ucred;
208 	ai.fd_rdir = p->p_fd->fd_rdir;
209 
210 	/*
211 	 * Auto-sizing of socket buffers is managed by the protocols and
212 	 * the appropriate flags must be set in the pru_attach function.
213 	 */
214 	error = so_pru_attach(so, proto, &ai);
215 	if (error) {
216 		so->so_state |= SS_NOFDREF;
217 		sofree(so);
218 		return (error);
219 	}
220 
221 	*aso = so;
222 	return (0);
223 }
224 
225 int
226 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
227 {
228 	int error;
229 
230 	crit_enter();
231 	error = so_pru_bind(so, nam, td);
232 	crit_exit();
233 	return (error);
234 }
235 
236 void
237 sodealloc(struct socket *so)
238 {
239 	if (so->so_rcv.ssb_hiwat)
240 		(void)chgsbsize(so->so_cred->cr_uidinfo,
241 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
242 	if (so->so_snd.ssb_hiwat)
243 		(void)chgsbsize(so->so_cred->cr_uidinfo,
244 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
245 #ifdef INET
246 	/* remove accept filter if present */
247 	if (so->so_accf != NULL)
248 		do_setopt_accept_filter(so, NULL);
249 #endif /* INET */
250 	crfree(so->so_cred);
251 	kfree(so, M_SOCKET);
252 }
253 
254 int
255 solisten(struct socket *so, int backlog, struct thread *td)
256 {
257 	int error;
258 #ifdef SCTP
259 	short oldopt, oldqlimit;
260 #endif /* SCTP */
261 
262 	crit_enter();
263 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
264 		crit_exit();
265 		return (EINVAL);
266 	}
267 
268 #ifdef SCTP
269 	oldopt = so->so_options;
270 	oldqlimit = so->so_qlimit;
271 #endif /* SCTP */
272 
273 	if (TAILQ_EMPTY(&so->so_comp))
274 		so->so_options |= SO_ACCEPTCONN;
275 	if (backlog < 0 || backlog > somaxconn)
276 		backlog = somaxconn;
277 	so->so_qlimit = backlog;
278 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
279 	 * the so_options (UDP model both connect's and gets inbound
280 	 * connections .. implicitly).
281 	 */
282 	error = so_pru_listen(so, td);
283 	if (error) {
284 #ifdef SCTP
285 		/* Restore the params */
286 		so->so_options = oldopt;
287 		so->so_qlimit = oldqlimit;
288 #endif /* SCTP */
289 		crit_exit();
290 		return (error);
291 	}
292 	crit_exit();
293 	return (0);
294 }
295 
296 /*
297  * Destroy a disconnected socket.  This routine is a NOP if entities
298  * still have a reference on the socket:
299  *
300  *	so_pcb -	The protocol stack still has a reference
301  *	SS_NOFDREF -	There is no longer a file pointer reference
302  *	SS_ABORTING -	An abort netmsg is in-flight
303  */
304 void
305 sofree(struct socket *so)
306 {
307 	struct socket *head = so->so_head;
308 
309 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
310 		return;
311 	if (so->so_state & SS_ABORTING)
312 		return;
313 	if (head != NULL) {
314 		if (so->so_state & SS_INCOMP) {
315 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
316 			head->so_incqlen--;
317 		} else if (so->so_state & SS_COMP) {
318 			/*
319 			 * We must not decommission a socket that's
320 			 * on the accept(2) queue.  If we do, then
321 			 * accept(2) may hang after select(2) indicated
322 			 * that the listening socket was ready.
323 			 */
324 			return;
325 		} else {
326 			panic("sofree: not queued");
327 		}
328 		so->so_state &= ~SS_INCOMP;
329 		so->so_head = NULL;
330 	}
331 	ssb_release(&so->so_snd, so);
332 	sorflush(so);
333 	sodealloc(so);
334 }
335 
336 /*
337  * Close a socket on last file table reference removal.
338  * Initiate disconnect if connected.
339  * Free socket when disconnect complete.
340  */
341 int
342 soclose(struct socket *so, int fflag)
343 {
344 	int error = 0;
345 
346 	crit_enter();
347 	funsetown(so->so_sigio);
348 	if (so->so_pcb == NULL)
349 		goto discard;
350 	if (so->so_state & SS_ISCONNECTED) {
351 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
352 			error = sodisconnect(so);
353 			if (error)
354 				goto drop;
355 		}
356 		if (so->so_options & SO_LINGER) {
357 			if ((so->so_state & SS_ISDISCONNECTING) &&
358 			    (fflag & FNONBLOCK))
359 				goto drop;
360 			while (so->so_state & SS_ISCONNECTED) {
361 				error = tsleep((caddr_t)&so->so_timeo,
362 				    PCATCH, "soclos", so->so_linger * hz);
363 				if (error)
364 					break;
365 			}
366 		}
367 	}
368 drop:
369 	if (so->so_pcb) {
370 		int error2;
371 
372 		error2 = so_pru_detach(so);
373 		if (error == 0)
374 			error = error2;
375 	}
376 discard:
377 	if (so->so_options & SO_ACCEPTCONN) {
378 		struct socket *sp;
379 
380 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
381 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
382 			sp->so_state &= ~SS_INCOMP;
383 			sp->so_head = NULL;
384 			so->so_incqlen--;
385 			soaborta(sp);
386 		}
387 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
388 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
389 			sp->so_state &= ~SS_COMP;
390 			sp->so_head = NULL;
391 			so->so_qlen--;
392 			soaborta(sp);
393 		}
394 	}
395 	if (so->so_state & SS_NOFDREF)
396 		panic("soclose: NOFDREF");
397 	so->so_state |= SS_NOFDREF;
398 	sofree(so);
399 	crit_exit();
400 	return (error);
401 }
402 
403 /*
404  * Abort and destroy a socket.  Only one abort can be in progress
405  * at any given moment.
406  */
407 void
408 soabort(struct socket *so)
409 {
410 	if ((so->so_state & SS_ABORTING) == 0) {
411 		so->so_state |= SS_ABORTING;
412 		so_pru_abort(so);
413 	}
414 }
415 
416 void
417 soaborta(struct socket *so)
418 {
419 	if ((so->so_state & SS_ABORTING) == 0) {
420 		so->so_state |= SS_ABORTING;
421 		so_pru_aborta(so);
422 	}
423 }
424 
425 void
426 soabort_oncpu(struct socket *so)
427 {
428 	if ((so->so_state & SS_ABORTING) == 0) {
429 		so->so_state |= SS_ABORTING;
430 		so_pru_abort_oncpu(so);
431 	}
432 }
433 
434 int
435 soaccept(struct socket *so, struct sockaddr **nam)
436 {
437 	int error;
438 
439 	crit_enter();
440 	if ((so->so_state & SS_NOFDREF) == 0)
441 		panic("soaccept: !NOFDREF");
442 	so->so_state &= ~SS_NOFDREF;
443 	error = so_pru_accept(so, nam);
444 	crit_exit();
445 	return (error);
446 }
447 
448 int
449 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
450 {
451 	int error;
452 
453 	if (so->so_options & SO_ACCEPTCONN)
454 		return (EOPNOTSUPP);
455 	crit_enter();
456 	/*
457 	 * If protocol is connection-based, can only connect once.
458 	 * Otherwise, if connected, try to disconnect first.
459 	 * This allows user to disconnect by connecting to, e.g.,
460 	 * a null address.
461 	 */
462 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
463 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
464 	    (error = sodisconnect(so)))) {
465 		error = EISCONN;
466 	} else {
467 		/*
468 		 * Prevent accumulated error from previous connection
469 		 * from biting us.
470 		 */
471 		so->so_error = 0;
472 		error = so_pru_connect(so, nam, td);
473 	}
474 	crit_exit();
475 	return (error);
476 }
477 
478 int
479 soconnect2(struct socket *so1, struct socket *so2)
480 {
481 	int error;
482 
483 	crit_enter();
484 	error = so_pru_connect2(so1, so2);
485 	crit_exit();
486 	return (error);
487 }
488 
489 int
490 sodisconnect(struct socket *so)
491 {
492 	int error;
493 
494 	crit_enter();
495 	if ((so->so_state & SS_ISCONNECTED) == 0) {
496 		error = ENOTCONN;
497 		goto bad;
498 	}
499 	if (so->so_state & SS_ISDISCONNECTING) {
500 		error = EALREADY;
501 		goto bad;
502 	}
503 	error = so_pru_disconnect(so);
504 bad:
505 	crit_exit();
506 	return (error);
507 }
508 
509 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
510 /*
511  * Send on a socket.
512  * If send must go all at once and message is larger than
513  * send buffering, then hard error.
514  * Lock against other senders.
515  * If must go all at once and not enough room now, then
516  * inform user that this would block and do nothing.
517  * Otherwise, if nonblocking, send as much as possible.
518  * The data to be sent is described by "uio" if nonzero,
519  * otherwise by the mbuf chain "top" (which must be null
520  * if uio is not).  Data provided in mbuf chain must be small
521  * enough to send all at once.
522  *
523  * Returns nonzero on error, timeout or signal; callers
524  * must check for short counts if EINTR/ERESTART are returned.
525  * Data and control buffers are freed on return.
526  */
527 int
528 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
529 	struct mbuf *top, struct mbuf *control, int flags,
530 	struct thread *td)
531 {
532 	struct mbuf **mp;
533 	struct mbuf *m;
534 	size_t resid;
535 	int space, len;
536 	int clen = 0, error, dontroute, mlen;
537 	int atomic = sosendallatonce(so) || top;
538 	int pru_flags;
539 
540 	if (uio) {
541 		resid = uio->uio_resid;
542 	} else {
543 		resid = (size_t)top->m_pkthdr.len;
544 #ifdef INVARIANTS
545 		len = 0;
546 		for (m = top; m; m = m->m_next)
547 			len += m->m_len;
548 		KKASSERT(top->m_pkthdr.len == len);
549 #endif
550 	}
551 
552 	/*
553 	 * WARNING!  resid is unsigned, space and len are signed.  space
554 	 * 	     can wind up negative if the sockbuf is overcommitted.
555 	 *
556 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
557 	 * type sockets since that's an error.
558 	 */
559 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
560 		error = EINVAL;
561 		goto out;
562 	}
563 
564 	dontroute =
565 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
566 	    (so->so_proto->pr_flags & PR_ATOMIC);
567 	if (td->td_lwp != NULL)
568 		td->td_lwp->lwp_ru.ru_msgsnd++;
569 	if (control)
570 		clen = control->m_len;
571 #define	gotoerr(errcode)	{ error = errcode; crit_exit(); goto release; }
572 
573 restart:
574 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
575 	if (error)
576 		goto out;
577 
578 	do {
579 		crit_enter();
580 		if (so->so_state & SS_CANTSENDMORE)
581 			gotoerr(EPIPE);
582 		if (so->so_error) {
583 			error = so->so_error;
584 			so->so_error = 0;
585 			crit_exit();
586 			goto release;
587 		}
588 		if ((so->so_state & SS_ISCONNECTED) == 0) {
589 			/*
590 			 * `sendto' and `sendmsg' is allowed on a connection-
591 			 * based socket if it supports implied connect.
592 			 * Return ENOTCONN if not connected and no address is
593 			 * supplied.
594 			 */
595 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
596 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
597 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
598 				    !(resid == 0 && clen != 0))
599 					gotoerr(ENOTCONN);
600 			} else if (addr == 0)
601 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
602 				   ENOTCONN : EDESTADDRREQ);
603 		}
604 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
605 		    clen > so->so_snd.ssb_hiwat) {
606 			gotoerr(EMSGSIZE);
607 		}
608 		space = ssb_space(&so->so_snd);
609 		if (flags & MSG_OOB)
610 			space += 1024;
611 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
612 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
613 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
614 				gotoerr(EWOULDBLOCK);
615 			ssb_unlock(&so->so_snd);
616 			error = ssb_wait(&so->so_snd);
617 			crit_exit();
618 			if (error)
619 				goto out;
620 			goto restart;
621 		}
622 		crit_exit();
623 		mp = &top;
624 		space -= clen;
625 		do {
626 		    if (uio == NULL) {
627 			/*
628 			 * Data is prepackaged in "top".
629 			 */
630 			resid = 0;
631 			if (flags & MSG_EOR)
632 				top->m_flags |= M_EOR;
633 		    } else do {
634 			if (resid > INT_MAX)
635 				resid = INT_MAX;
636 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
637 				   top == NULL ? M_PKTHDR : 0, &mlen);
638 			if (top == NULL) {
639 				m->m_pkthdr.len = 0;
640 				m->m_pkthdr.rcvif = NULL;
641 			}
642 			len = imin((int)szmin(mlen, resid), space);
643 			if (resid < MINCLSIZE) {
644 				/*
645 				 * For datagram protocols, leave room
646 				 * for protocol headers in first mbuf.
647 				 */
648 				if (atomic && top == 0 && len < mlen)
649 					MH_ALIGN(m, len);
650 			}
651 			space -= len;
652 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
653 			resid = uio->uio_resid;
654 			m->m_len = len;
655 			*mp = m;
656 			top->m_pkthdr.len += len;
657 			if (error)
658 				goto release;
659 			mp = &m->m_next;
660 			if (resid == 0) {
661 				if (flags & MSG_EOR)
662 					top->m_flags |= M_EOR;
663 				break;
664 			}
665 		    } while (space > 0 && atomic);
666 		    if (dontroute)
667 			    so->so_options |= SO_DONTROUTE;
668 		    if (flags & MSG_OOB) {
669 		    	    pru_flags = PRUS_OOB;
670 		    } else if ((flags & MSG_EOF) &&
671 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
672 			       (resid == 0)) {
673 			    /*
674 			     * If the user set MSG_EOF, the protocol
675 			     * understands this flag and nothing left to
676 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
677 			     */
678 		    	    pru_flags = PRUS_EOF;
679 		    } else if (resid > 0 && space > 0) {
680 			    /* If there is more to send, set PRUS_MORETOCOME */
681 		    	    pru_flags = PRUS_MORETOCOME;
682 		    } else {
683 		    	    pru_flags = 0;
684 		    }
685 		    crit_enter();
686 		    /*
687 		     * XXX all the SS_CANTSENDMORE checks previously
688 		     * done could be out of date.  We could have recieved
689 		     * a reset packet in an interrupt or maybe we slept
690 		     * while doing page faults in uiomove() etc. We could
691 		     * probably recheck again inside the splnet() protection
692 		     * here, but there are probably other places that this
693 		     * also happens.  We must rethink this.
694 		     */
695 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
696 		    crit_exit();
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 	crit_enter();
753 	if (so->so_state & SS_CANTSENDMORE)
754 		gotoerr(EPIPE);
755 	if (so->so_error) {
756 		error = so->so_error;
757 		so->so_error = 0;
758 		crit_exit();
759 		goto release;
760 	}
761 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
762 		gotoerr(EDESTADDRREQ);
763 	if (resid > so->so_snd.ssb_hiwat)
764 		gotoerr(EMSGSIZE);
765 	space = ssb_space(&so->so_snd);
766 	if (uio && (space < 0 || (size_t)space < resid)) {
767 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
768 			gotoerr(EWOULDBLOCK);
769 		ssb_unlock(&so->so_snd);
770 		error = ssb_wait(&so->so_snd);
771 		crit_exit();
772 		if (error)
773 			goto out;
774 		goto restart;
775 	}
776 	crit_exit();
777 
778 	if (uio) {
779 		top = m_uiomove(uio);
780 		if (top == NULL)
781 			goto release;
782 	}
783 
784 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
785 	if (dontroute)
786 		so->so_options |= SO_DONTROUTE;
787 
788 	error = so_pru_send(so, 0, top, addr, NULL, td);
789 	top = NULL;		/* sent or freed in lower layer */
790 
791 	if (dontroute)
792 		so->so_options &= ~SO_DONTROUTE;
793 
794 release:
795 	ssb_unlock(&so->so_snd);
796 out:
797 	if (top)
798 		m_freem(top);
799 	return (error);
800 }
801 
802 /*
803  * Implement receive operations on a socket.
804  * We depend on the way that records are added to the signalsockbuf
805  * by sbappend*.  In particular, each record (mbufs linked through m_next)
806  * must begin with an address if the protocol so specifies,
807  * followed by an optional mbuf or mbufs containing ancillary data,
808  * and then zero or more mbufs of data.
809  * In order to avoid blocking network interrupts for the entire time here,
810  * we exit the critical section while doing the actual copy to user space.
811  * Although the signalsockbuf is locked, new data may still be appended,
812  * and thus we must maintain consistency of the signalsockbuf during that time.
813  *
814  * The caller may receive the data as a single mbuf chain by supplying
815  * an mbuf **mp0 for use in returning the chain.  The uio is then used
816  * only for the count in uio_resid.
817  */
818 int
819 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
820 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
821 {
822 	struct mbuf *m, *n;
823 	struct mbuf *free_chain = NULL;
824 	int flags, len, error, offset;
825 	struct protosw *pr = so->so_proto;
826 	int moff, type = 0;
827 	size_t resid, orig_resid;
828 
829 	if (uio)
830 		resid = uio->uio_resid;
831 	else
832 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
833 	orig_resid = resid;
834 
835 	if (psa)
836 		*psa = NULL;
837 	if (controlp)
838 		*controlp = NULL;
839 	if (flagsp)
840 		flags = *flagsp &~ MSG_EOR;
841 	else
842 		flags = 0;
843 	if (flags & MSG_OOB) {
844 		m = m_get(MB_WAIT, MT_DATA);
845 		if (m == NULL)
846 			return (ENOBUFS);
847 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
848 		if (error)
849 			goto bad;
850 		if (sio) {
851 			do {
852 				sbappend(sio, m);
853 				KKASSERT(resid >= (size_t)m->m_len);
854 				resid -= (size_t)m->m_len;
855 			} while (resid > 0 && m);
856 		} else {
857 			do {
858 				uio->uio_resid = resid;
859 				error = uiomove(mtod(m, caddr_t),
860 						(int)szmin(resid, m->m_len),
861 						uio);
862 				resid = uio->uio_resid;
863 				m = m_free(m);
864 			} while (uio->uio_resid && error == 0 && m);
865 		}
866 bad:
867 		if (m)
868 			m_freem(m);
869 		return (error);
870 	}
871 	if ((so->so_state & SS_ISCONFIRMING) && resid)
872 		so_pru_rcvd(so, 0);
873 
874 restart:
875 	crit_enter();
876 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
877 	if (error)
878 		goto done;
879 
880 	m = so->so_rcv.ssb_mb;
881 	/*
882 	 * If we have less data than requested, block awaiting more
883 	 * (subject to any timeout) if:
884 	 *   1. the current count is less than the low water mark, or
885 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
886 	 *	receive operation at once if we block (resid <= hiwat).
887 	 *   3. MSG_DONTWAIT is not set
888 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
889 	 * we have to do the receive in sections, and thus risk returning
890 	 * a short count if a timeout or signal occurs after we start.
891 	 */
892 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
893 	    (size_t)so->so_rcv.ssb_cc < resid) &&
894 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
895 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
896 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
897 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
898 		if (so->so_error) {
899 			if (m)
900 				goto dontblock;
901 			error = so->so_error;
902 			if ((flags & MSG_PEEK) == 0)
903 				so->so_error = 0;
904 			goto release;
905 		}
906 		if (so->so_state & SS_CANTRCVMORE) {
907 			if (m)
908 				goto dontblock;
909 			else
910 				goto release;
911 		}
912 		for (; m; m = m->m_next) {
913 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
914 				m = so->so_rcv.ssb_mb;
915 				goto dontblock;
916 			}
917 		}
918 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
919 		    (pr->pr_flags & PR_CONNREQUIRED)) {
920 			error = ENOTCONN;
921 			goto release;
922 		}
923 		if (resid == 0)
924 			goto release;
925 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
926 			error = EWOULDBLOCK;
927 			goto release;
928 		}
929 		ssb_unlock(&so->so_rcv);
930 		error = ssb_wait(&so->so_rcv);
931 		if (error)
932 			goto done;
933 		crit_exit();
934 		goto restart;
935 	}
936 dontblock:
937 	if (uio && uio->uio_td && uio->uio_td->td_proc)
938 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
939 
940 	/*
941 	 * note: m should be == sb_mb here.  Cache the next record while
942 	 * cleaning up.  Note that calling m_free*() will break out critical
943 	 * section.
944 	 */
945 	KKASSERT(m == so->so_rcv.ssb_mb);
946 
947 	/*
948 	 * Skip any address mbufs prepending the record.
949 	 */
950 	if (pr->pr_flags & PR_ADDR) {
951 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
952 		orig_resid = 0;
953 		if (psa)
954 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
955 		if (flags & MSG_PEEK)
956 			m = m->m_next;
957 		else
958 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
959 	}
960 
961 	/*
962 	 * Skip any control mbufs prepending the record.
963 	 */
964 #ifdef SCTP
965 	if (pr->pr_flags & PR_ADDR_OPT) {
966 		/*
967 		 * For SCTP we may be getting a
968 		 * whole message OR a partial delivery.
969 		 */
970 		if (m && m->m_type == MT_SONAME) {
971 			orig_resid = 0;
972 			if (psa)
973 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
974 			if (flags & MSG_PEEK)
975 				m = m->m_next;
976 			else
977 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
978 		}
979 	}
980 #endif /* SCTP */
981 	while (m && m->m_type == MT_CONTROL && error == 0) {
982 		if (flags & MSG_PEEK) {
983 			if (controlp)
984 				*controlp = m_copy(m, 0, m->m_len);
985 			m = m->m_next;	/* XXX race */
986 		} else {
987 			if (controlp) {
988 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
989 				if (pr->pr_domain->dom_externalize &&
990 				    mtod(m, struct cmsghdr *)->cmsg_type ==
991 				    SCM_RIGHTS)
992 				   error = (*pr->pr_domain->dom_externalize)(m);
993 				*controlp = m;
994 				m = n;
995 			} else {
996 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
997 			}
998 		}
999 		if (controlp && *controlp) {
1000 			orig_resid = 0;
1001 			controlp = &(*controlp)->m_next;
1002 		}
1003 	}
1004 
1005 	/*
1006 	 * flag OOB data.
1007 	 */
1008 	if (m) {
1009 		type = m->m_type;
1010 		if (type == MT_OOBDATA)
1011 			flags |= MSG_OOB;
1012 	}
1013 
1014 	/*
1015 	 * Copy to the UIO or mbuf return chain (*mp).
1016 	 */
1017 	moff = 0;
1018 	offset = 0;
1019 	while (m && resid > 0 && error == 0) {
1020 		if (m->m_type == MT_OOBDATA) {
1021 			if (type != MT_OOBDATA)
1022 				break;
1023 		} else if (type == MT_OOBDATA)
1024 			break;
1025 		else
1026 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1027 			("receive 3"));
1028 		so->so_state &= ~SS_RCVATMARK;
1029 		len = (resid > INT_MAX) ? INT_MAX : resid;
1030 		if (so->so_oobmark && len > so->so_oobmark - offset)
1031 			len = so->so_oobmark - offset;
1032 		if (len > m->m_len - moff)
1033 			len = m->m_len - moff;
1034 
1035 		/*
1036 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1037 		 * The SIO is dealt with when we eat the mbuf, but deal
1038 		 * with the resid here either way.
1039 		 */
1040 		if (uio) {
1041 			crit_exit();
1042 			uio->uio_resid = resid;
1043 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1044 			resid = uio->uio_resid;
1045 			crit_enter();
1046 			if (error)
1047 				goto release;
1048 		} else {
1049 			resid -= (size_t)len;
1050 		}
1051 
1052 		/*
1053 		 * Eat the entire mbuf or just a piece of it
1054 		 */
1055 		if (len == m->m_len - moff) {
1056 			if (m->m_flags & M_EOR)
1057 				flags |= MSG_EOR;
1058 #ifdef SCTP
1059 			if (m->m_flags & M_NOTIFICATION)
1060 				flags |= MSG_NOTIFICATION;
1061 #endif /* SCTP */
1062 			if (flags & MSG_PEEK) {
1063 				m = m->m_next;
1064 				moff = 0;
1065 			} else {
1066 				if (sio) {
1067 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1068 					sbappend(sio, m);
1069 					m = n;
1070 				} else {
1071 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1072 				}
1073 			}
1074 		} else {
1075 			if (flags & MSG_PEEK) {
1076 				moff += len;
1077 			} else {
1078 				if (sio) {
1079 					n = m_copym(m, 0, len, MB_WAIT);
1080 					if (n)
1081 						sbappend(sio, n);
1082 				}
1083 				m->m_data += len;
1084 				m->m_len -= len;
1085 				so->so_rcv.ssb_cc -= len;
1086 			}
1087 		}
1088 		if (so->so_oobmark) {
1089 			if ((flags & MSG_PEEK) == 0) {
1090 				so->so_oobmark -= len;
1091 				if (so->so_oobmark == 0) {
1092 					so->so_state |= SS_RCVATMARK;
1093 					break;
1094 				}
1095 			} else {
1096 				offset += len;
1097 				if (offset == so->so_oobmark)
1098 					break;
1099 			}
1100 		}
1101 		if (flags & MSG_EOR)
1102 			break;
1103 		/*
1104 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1105 		 * we must not quit until resid == 0 or an error
1106 		 * termination.  If a signal/timeout occurs, return
1107 		 * with a short count but without error.
1108 		 * Keep signalsockbuf locked against other readers.
1109 		 */
1110 		while ((flags & MSG_WAITALL) && m == NULL &&
1111 		       resid > 0 && !sosendallatonce(so) &&
1112 		       so->so_rcv.ssb_mb == NULL) {
1113 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1114 				break;
1115 			/*
1116 			 * The window might have closed to zero, make
1117 			 * sure we send an ack now that we've drained
1118 			 * the buffer or we might end up blocking until
1119 			 * the idle takes over (5 seconds).
1120 			 */
1121 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1122 				so_pru_rcvd(so, flags);
1123 			error = ssb_wait(&so->so_rcv);
1124 			if (error) {
1125 				ssb_unlock(&so->so_rcv);
1126 				error = 0;
1127 				goto done;
1128 			}
1129 			m = so->so_rcv.ssb_mb;
1130 		}
1131 	}
1132 
1133 	/*
1134 	 * If an atomic read was requested but unread data still remains
1135 	 * in the record, set MSG_TRUNC.
1136 	 */
1137 	if (m && pr->pr_flags & PR_ATOMIC)
1138 		flags |= MSG_TRUNC;
1139 
1140 	/*
1141 	 * Cleanup.  If an atomic read was requested drop any unread data.
1142 	 */
1143 	if ((flags & MSG_PEEK) == 0) {
1144 		if (m && (pr->pr_flags & PR_ATOMIC))
1145 			sbdroprecord(&so->so_rcv.sb);
1146 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1147 			so_pru_rcvd(so, flags);
1148 	}
1149 
1150 	if (orig_resid == resid && orig_resid &&
1151 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1152 		ssb_unlock(&so->so_rcv);
1153 		crit_exit();
1154 		goto restart;
1155 	}
1156 
1157 	if (flagsp)
1158 		*flagsp |= flags;
1159 release:
1160 	ssb_unlock(&so->so_rcv);
1161 done:
1162 	crit_exit();
1163 	if (free_chain)
1164 		m_freem(free_chain);
1165 	return (error);
1166 }
1167 
1168 int
1169 soshutdown(struct socket *so, int how)
1170 {
1171 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1172 		return (EINVAL);
1173 
1174 	if (how != SHUT_WR)
1175 		sorflush(so);
1176 	if (how != SHUT_RD)
1177 		return (so_pru_shutdown(so));
1178 	return (0);
1179 }
1180 
1181 void
1182 sorflush(struct socket *so)
1183 {
1184 	struct signalsockbuf *ssb = &so->so_rcv;
1185 	struct protosw *pr = so->so_proto;
1186 	struct signalsockbuf asb;
1187 
1188 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1189 
1190 	ssb_lock(ssb, M_WAITOK);
1191 	socantrcvmore(so);
1192 	asb = *ssb;
1193 
1194 	/*
1195 	 * Can't just blow up the ssb structure here
1196 	 */
1197 	ssb->ssb_timeo = 0;
1198 	ssb->ssb_unused01 = 0;
1199 	ssb->ssb_lowat = 0;
1200 	ssb->ssb_hiwat = 0;
1201 	ssb->ssb_mbmax = 0;
1202 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1203 
1204 	ssb_unlock(ssb);
1205 
1206 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1207 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1208 	ssb_release(&asb, so);
1209 }
1210 
1211 #ifdef INET
1212 static int
1213 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1214 {
1215 	struct accept_filter_arg	*afap = NULL;
1216 	struct accept_filter	*afp;
1217 	struct so_accf	*af = so->so_accf;
1218 	int	error = 0;
1219 
1220 	/* do not set/remove accept filters on non listen sockets */
1221 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1222 		error = EINVAL;
1223 		goto out;
1224 	}
1225 
1226 	/* removing the filter */
1227 	if (sopt == NULL) {
1228 		if (af != NULL) {
1229 			if (af->so_accept_filter != NULL &&
1230 				af->so_accept_filter->accf_destroy != NULL) {
1231 				af->so_accept_filter->accf_destroy(so);
1232 			}
1233 			if (af->so_accept_filter_str != NULL) {
1234 				FREE(af->so_accept_filter_str, M_ACCF);
1235 			}
1236 			FREE(af, M_ACCF);
1237 			so->so_accf = NULL;
1238 		}
1239 		so->so_options &= ~SO_ACCEPTFILTER;
1240 		return (0);
1241 	}
1242 	/* adding a filter */
1243 	/* must remove previous filter first */
1244 	if (af != NULL) {
1245 		error = EINVAL;
1246 		goto out;
1247 	}
1248 	/* don't put large objects on the kernel stack */
1249 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1250 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1251 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1252 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1253 	if (error)
1254 		goto out;
1255 	afp = accept_filt_get(afap->af_name);
1256 	if (afp == NULL) {
1257 		error = ENOENT;
1258 		goto out;
1259 	}
1260 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1261 	if (afp->accf_create != NULL) {
1262 		if (afap->af_name[0] != '\0') {
1263 			int len = strlen(afap->af_name) + 1;
1264 
1265 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1266 			strcpy(af->so_accept_filter_str, afap->af_name);
1267 		}
1268 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1269 		if (af->so_accept_filter_arg == NULL) {
1270 			FREE(af->so_accept_filter_str, M_ACCF);
1271 			FREE(af, M_ACCF);
1272 			so->so_accf = NULL;
1273 			error = EINVAL;
1274 			goto out;
1275 		}
1276 	}
1277 	af->so_accept_filter = afp;
1278 	so->so_accf = af;
1279 	so->so_options |= SO_ACCEPTFILTER;
1280 out:
1281 	if (afap != NULL)
1282 		FREE(afap, M_TEMP);
1283 	return (error);
1284 }
1285 #endif /* INET */
1286 
1287 /*
1288  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1289  * an additional variant to handle the case where the option value needs
1290  * to be some kind of integer, but not a specific size.
1291  * In addition to their use here, these functions are also called by the
1292  * protocol-level pr_ctloutput() routines.
1293  */
1294 int
1295 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1296 {
1297 	return soopt_to_kbuf(sopt, buf, len, minlen);
1298 }
1299 
1300 int
1301 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1302 {
1303 	size_t	valsize;
1304 
1305 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1306 	KKASSERT(kva_p(buf));
1307 
1308 	/*
1309 	 * If the user gives us more than we wanted, we ignore it,
1310 	 * but if we don't get the minimum length the caller
1311 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1312 	 * is set to however much we actually retrieved.
1313 	 */
1314 	if ((valsize = sopt->sopt_valsize) < minlen)
1315 		return EINVAL;
1316 	if (valsize > len)
1317 		sopt->sopt_valsize = valsize = len;
1318 
1319 	bcopy(sopt->sopt_val, buf, valsize);
1320 	return 0;
1321 }
1322 
1323 
1324 int
1325 sosetopt(struct socket *so, struct sockopt *sopt)
1326 {
1327 	int	error, optval;
1328 	struct	linger l;
1329 	struct	timeval tv;
1330 	u_long  val;
1331 	struct signalsockbuf *sotmp;
1332 
1333 	error = 0;
1334 	sopt->sopt_dir = SOPT_SET;
1335 	if (sopt->sopt_level != SOL_SOCKET) {
1336 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1337 			return (so_pru_ctloutput(so, sopt));
1338 		}
1339 		error = ENOPROTOOPT;
1340 	} else {
1341 		switch (sopt->sopt_name) {
1342 #ifdef INET
1343 		case SO_ACCEPTFILTER:
1344 			error = do_setopt_accept_filter(so, sopt);
1345 			if (error)
1346 				goto bad;
1347 			break;
1348 #endif /* INET */
1349 		case SO_LINGER:
1350 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1351 			if (error)
1352 				goto bad;
1353 
1354 			so->so_linger = l.l_linger;
1355 			if (l.l_onoff)
1356 				so->so_options |= SO_LINGER;
1357 			else
1358 				so->so_options &= ~SO_LINGER;
1359 			break;
1360 
1361 		case SO_DEBUG:
1362 		case SO_KEEPALIVE:
1363 		case SO_DONTROUTE:
1364 		case SO_USELOOPBACK:
1365 		case SO_BROADCAST:
1366 		case SO_REUSEADDR:
1367 		case SO_REUSEPORT:
1368 		case SO_OOBINLINE:
1369 		case SO_TIMESTAMP:
1370 			error = sooptcopyin(sopt, &optval, sizeof optval,
1371 					    sizeof optval);
1372 			if (error)
1373 				goto bad;
1374 			if (optval)
1375 				so->so_options |= sopt->sopt_name;
1376 			else
1377 				so->so_options &= ~sopt->sopt_name;
1378 			break;
1379 
1380 		case SO_SNDBUF:
1381 		case SO_RCVBUF:
1382 		case SO_SNDLOWAT:
1383 		case SO_RCVLOWAT:
1384 			error = sooptcopyin(sopt, &optval, sizeof optval,
1385 					    sizeof optval);
1386 			if (error)
1387 				goto bad;
1388 
1389 			/*
1390 			 * Values < 1 make no sense for any of these
1391 			 * options, so disallow them.
1392 			 */
1393 			if (optval < 1) {
1394 				error = EINVAL;
1395 				goto bad;
1396 			}
1397 
1398 			switch (sopt->sopt_name) {
1399 			case SO_SNDBUF:
1400 			case SO_RCVBUF:
1401 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1402 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1403 				    so,
1404 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1405 					error = ENOBUFS;
1406 					goto bad;
1407 				}
1408 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1409 						&so->so_snd : &so->so_rcv;
1410 				atomic_clear_int(&sotmp->ssb_flags,
1411 						 SSB_AUTOSIZE);
1412 				break;
1413 
1414 			/*
1415 			 * Make sure the low-water is never greater than
1416 			 * the high-water.
1417 			 */
1418 			case SO_SNDLOWAT:
1419 				so->so_snd.ssb_lowat =
1420 				    (optval > so->so_snd.ssb_hiwat) ?
1421 				    so->so_snd.ssb_hiwat : optval;
1422 				atomic_clear_int(&so->so_snd.ssb_flags,
1423 						 SSB_AUTOLOWAT);
1424 				break;
1425 			case SO_RCVLOWAT:
1426 				so->so_rcv.ssb_lowat =
1427 				    (optval > so->so_rcv.ssb_hiwat) ?
1428 				    so->so_rcv.ssb_hiwat : optval;
1429 				atomic_clear_int(&so->so_rcv.ssb_flags,
1430 						 SSB_AUTOLOWAT);
1431 				break;
1432 			}
1433 			break;
1434 
1435 		case SO_SNDTIMEO:
1436 		case SO_RCVTIMEO:
1437 			error = sooptcopyin(sopt, &tv, sizeof tv,
1438 					    sizeof tv);
1439 			if (error)
1440 				goto bad;
1441 
1442 			/* assert(hz > 0); */
1443 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1444 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1445 				error = EDOM;
1446 				goto bad;
1447 			}
1448 			/* assert(tick > 0); */
1449 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1450 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1451 			if (val > SHRT_MAX) {
1452 				error = EDOM;
1453 				goto bad;
1454 			}
1455 			if (val == 0 && tv.tv_usec != 0)
1456 				val = 1;
1457 
1458 			switch (sopt->sopt_name) {
1459 			case SO_SNDTIMEO:
1460 				so->so_snd.ssb_timeo = val;
1461 				break;
1462 			case SO_RCVTIMEO:
1463 				so->so_rcv.ssb_timeo = val;
1464 				break;
1465 			}
1466 			break;
1467 		default:
1468 			error = ENOPROTOOPT;
1469 			break;
1470 		}
1471 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1472 			(void) so_pru_ctloutput(so, sopt);
1473 		}
1474 	}
1475 bad:
1476 	return (error);
1477 }
1478 
1479 /* Helper routine for getsockopt */
1480 int
1481 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1482 {
1483 	soopt_from_kbuf(sopt, buf, len);
1484 	return 0;
1485 }
1486 
1487 void
1488 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1489 {
1490 	size_t	valsize;
1491 
1492 	if (len == 0) {
1493 		sopt->sopt_valsize = 0;
1494 		return;
1495 	}
1496 
1497 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1498 	KKASSERT(kva_p(buf));
1499 
1500 	/*
1501 	 * Documented get behavior is that we always return a value,
1502 	 * possibly truncated to fit in the user's buffer.
1503 	 * Traditional behavior is that we always tell the user
1504 	 * precisely how much we copied, rather than something useful
1505 	 * like the total amount we had available for her.
1506 	 * Note that this interface is not idempotent; the entire answer must
1507 	 * generated ahead of time.
1508 	 */
1509 	valsize = szmin(len, sopt->sopt_valsize);
1510 	sopt->sopt_valsize = valsize;
1511 	if (sopt->sopt_val != 0) {
1512 		bcopy(buf, sopt->sopt_val, valsize);
1513 	}
1514 }
1515 
1516 int
1517 sogetopt(struct socket *so, struct sockopt *sopt)
1518 {
1519 	int	error, optval;
1520 	struct	linger l;
1521 	struct	timeval tv;
1522 #ifdef INET
1523 	struct accept_filter_arg *afap;
1524 #endif
1525 
1526 	error = 0;
1527 	sopt->sopt_dir = SOPT_GET;
1528 	if (sopt->sopt_level != SOL_SOCKET) {
1529 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1530 			return (so_pru_ctloutput(so, sopt));
1531 		} else
1532 			return (ENOPROTOOPT);
1533 	} else {
1534 		switch (sopt->sopt_name) {
1535 #ifdef INET
1536 		case SO_ACCEPTFILTER:
1537 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1538 				return (EINVAL);
1539 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1540 				M_TEMP, M_WAITOK | M_ZERO);
1541 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1542 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1543 				if (so->so_accf->so_accept_filter_str != NULL)
1544 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1545 			}
1546 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1547 			FREE(afap, M_TEMP);
1548 			break;
1549 #endif /* INET */
1550 
1551 		case SO_LINGER:
1552 			l.l_onoff = so->so_options & SO_LINGER;
1553 			l.l_linger = so->so_linger;
1554 			error = sooptcopyout(sopt, &l, sizeof l);
1555 			break;
1556 
1557 		case SO_USELOOPBACK:
1558 		case SO_DONTROUTE:
1559 		case SO_DEBUG:
1560 		case SO_KEEPALIVE:
1561 		case SO_REUSEADDR:
1562 		case SO_REUSEPORT:
1563 		case SO_BROADCAST:
1564 		case SO_OOBINLINE:
1565 		case SO_TIMESTAMP:
1566 			optval = so->so_options & sopt->sopt_name;
1567 integer:
1568 			error = sooptcopyout(sopt, &optval, sizeof optval);
1569 			break;
1570 
1571 		case SO_TYPE:
1572 			optval = so->so_type;
1573 			goto integer;
1574 
1575 		case SO_ERROR:
1576 			optval = so->so_error;
1577 			so->so_error = 0;
1578 			goto integer;
1579 
1580 		case SO_SNDBUF:
1581 			optval = so->so_snd.ssb_hiwat;
1582 			goto integer;
1583 
1584 		case SO_RCVBUF:
1585 			optval = so->so_rcv.ssb_hiwat;
1586 			goto integer;
1587 
1588 		case SO_SNDLOWAT:
1589 			optval = so->so_snd.ssb_lowat;
1590 			goto integer;
1591 
1592 		case SO_RCVLOWAT:
1593 			optval = so->so_rcv.ssb_lowat;
1594 			goto integer;
1595 
1596 		case SO_SNDTIMEO:
1597 		case SO_RCVTIMEO:
1598 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1599 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1600 
1601 			tv.tv_sec = optval / hz;
1602 			tv.tv_usec = (optval % hz) * ustick;
1603 			error = sooptcopyout(sopt, &tv, sizeof tv);
1604 			break;
1605 
1606 		default:
1607 			error = ENOPROTOOPT;
1608 			break;
1609 		}
1610 		return (error);
1611 	}
1612 }
1613 
1614 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1615 int
1616 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1617 {
1618 	struct mbuf *m, *m_prev;
1619 	int sopt_size = sopt->sopt_valsize, msize;
1620 
1621 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1622 		   0, &msize);
1623 	if (m == NULL)
1624 		return (ENOBUFS);
1625 	m->m_len = min(msize, sopt_size);
1626 	sopt_size -= m->m_len;
1627 	*mp = m;
1628 	m_prev = m;
1629 
1630 	while (sopt_size > 0) {
1631 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1632 			   MT_DATA, 0, &msize);
1633 		if (m == NULL) {
1634 			m_freem(*mp);
1635 			return (ENOBUFS);
1636 		}
1637 		m->m_len = min(msize, sopt_size);
1638 		sopt_size -= m->m_len;
1639 		m_prev->m_next = m;
1640 		m_prev = m;
1641 	}
1642 	return (0);
1643 }
1644 
1645 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1646 int
1647 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1648 {
1649 	soopt_to_mbuf(sopt, m);
1650 	return 0;
1651 }
1652 
1653 void
1654 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1655 {
1656 	size_t valsize;
1657 	void *val;
1658 
1659 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1660 	KKASSERT(kva_p(m));
1661 	if (sopt->sopt_val == NULL)
1662 		return;
1663 	val = sopt->sopt_val;
1664 	valsize = sopt->sopt_valsize;
1665 	while (m != NULL && valsize >= m->m_len) {
1666 		bcopy(val, mtod(m, char *), m->m_len);
1667 		valsize -= m->m_len;
1668 		val = (caddr_t)val + m->m_len;
1669 		m = m->m_next;
1670 	}
1671 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1672 		panic("ip6_sooptmcopyin");
1673 }
1674 
1675 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1676 int
1677 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1678 {
1679 	return soopt_from_mbuf(sopt, m);
1680 }
1681 
1682 int
1683 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1684 {
1685 	struct mbuf *m0 = m;
1686 	size_t valsize = 0;
1687 	size_t maxsize;
1688 	void *val;
1689 
1690 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1691 	KKASSERT(kva_p(m));
1692 	if (sopt->sopt_val == NULL)
1693 		return 0;
1694 	val = sopt->sopt_val;
1695 	maxsize = sopt->sopt_valsize;
1696 	while (m != NULL && maxsize >= m->m_len) {
1697 		bcopy(mtod(m, char *), val, m->m_len);
1698 	       maxsize -= m->m_len;
1699 	       val = (caddr_t)val + m->m_len;
1700 	       valsize += m->m_len;
1701 	       m = m->m_next;
1702 	}
1703 	if (m != NULL) {
1704 		/* enough soopt buffer should be given from user-land */
1705 		m_freem(m0);
1706 		return (EINVAL);
1707 	}
1708 	sopt->sopt_valsize = valsize;
1709 	return 0;
1710 }
1711 
1712 void
1713 sohasoutofband(struct socket *so)
1714 {
1715 	if (so->so_sigio != NULL)
1716 		pgsigio(so->so_sigio, SIGURG, 0);
1717 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1718 }
1719 
1720 int
1721 sokqfilter(struct file *fp, struct knote *kn)
1722 {
1723 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1724 	struct signalsockbuf *ssb;
1725 
1726 	switch (kn->kn_filter) {
1727 	case EVFILT_READ:
1728 		if (so->so_options & SO_ACCEPTCONN)
1729 			kn->kn_fop = &solisten_filtops;
1730 		else
1731 			kn->kn_fop = &soread_filtops;
1732 		ssb = &so->so_rcv;
1733 		break;
1734 	case EVFILT_WRITE:
1735 		kn->kn_fop = &sowrite_filtops;
1736 		ssb = &so->so_snd;
1737 		break;
1738 	case EVFILT_EXCEPT:
1739 		kn->kn_fop = &soexcept_filtops;
1740 		ssb = &so->so_rcv;
1741 		break;
1742 	default:
1743 		return (EOPNOTSUPP);
1744 	}
1745 
1746 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1747 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1748 	return (0);
1749 }
1750 
1751 static void
1752 filt_sordetach(struct knote *kn)
1753 {
1754 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1755 
1756 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1757 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1758 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1759 }
1760 
1761 /*ARGSUSED*/
1762 static int
1763 filt_soread(struct knote *kn, long hint)
1764 {
1765 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1766 
1767 	if (kn->kn_sfflags & NOTE_OOB) {
1768 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1769 			kn->kn_fflags |= NOTE_OOB;
1770 			return (1);
1771 		}
1772 		return (0);
1773 	}
1774 	kn->kn_data = so->so_rcv.ssb_cc;
1775 
1776 	/*
1777 	 * Only set EOF if all data has been exhausted.
1778 	 */
1779 	if ((so->so_state & SS_CANTRCVMORE) && kn->kn_data == 0) {
1780 		kn->kn_flags |= EV_EOF;
1781 		kn->kn_fflags = so->so_error;
1782 		return (1);
1783 	}
1784 	if (so->so_error)	/* temporary udp error */
1785 		return (1);
1786 	if (kn->kn_sfflags & NOTE_LOWAT)
1787 		return (kn->kn_data >= kn->kn_sdata);
1788 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1789 	    !TAILQ_EMPTY(&so->so_comp));
1790 }
1791 
1792 static void
1793 filt_sowdetach(struct knote *kn)
1794 {
1795 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1796 
1797 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
1798 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
1799 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
1800 }
1801 
1802 /*ARGSUSED*/
1803 static int
1804 filt_sowrite(struct knote *kn, long hint)
1805 {
1806 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1807 
1808 	kn->kn_data = ssb_space(&so->so_snd);
1809 	if (so->so_state & SS_CANTSENDMORE) {
1810 		kn->kn_flags |= EV_EOF;
1811 		kn->kn_fflags = so->so_error;
1812 		return (1);
1813 	}
1814 	if (so->so_error)	/* temporary udp error */
1815 		return (1);
1816 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1817 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1818 		return (0);
1819 	if (kn->kn_sfflags & NOTE_LOWAT)
1820 		return (kn->kn_data >= kn->kn_sdata);
1821 	return (kn->kn_data >= so->so_snd.ssb_lowat);
1822 }
1823 
1824 /*ARGSUSED*/
1825 static int
1826 filt_solisten(struct knote *kn, long hint)
1827 {
1828 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1829 
1830 	kn->kn_data = so->so_qlen;
1831 	return (! TAILQ_EMPTY(&so->so_comp));
1832 }
1833