xref: /freebsd/sys/kern/uipc_socket.c (revision a3e8fd0b7f663db7eafff527d5c3ca3bcfa8a537)
1 /*
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
34  * $FreeBSD$
35  */
36 
37 #include "opt_inet.h"
38 #include "opt_mac.h"
39 #include "opt_zero.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/fcntl.h>
44 #include <sys/lock.h>
45 #include <sys/mac.h>
46 #include <sys/malloc.h>
47 #include <sys/mbuf.h>
48 #include <sys/mutex.h>
49 #include <sys/domain.h>
50 #include <sys/file.h>			/* for struct knote */
51 #include <sys/kernel.h>
52 #include <sys/malloc.h>
53 #include <sys/event.h>
54 #include <sys/poll.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/resourcevar.h>
60 #include <sys/signalvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/uio.h>
63 #include <sys/jail.h>
64 
65 #include <vm/uma.h>
66 
67 #include <machine/limits.h>
68 
69 #ifdef INET
70 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
71 #endif
72 
73 static void 	filt_sordetach(struct knote *kn);
74 static int 	filt_soread(struct knote *kn, long hint);
75 static void 	filt_sowdetach(struct knote *kn);
76 static int	filt_sowrite(struct knote *kn, long hint);
77 static int	filt_solisten(struct knote *kn, long hint);
78 
79 static struct filterops solisten_filtops =
80 	{ 1, NULL, filt_sordetach, filt_solisten };
81 static struct filterops soread_filtops =
82 	{ 1, NULL, filt_sordetach, filt_soread };
83 static struct filterops sowrite_filtops =
84 	{ 1, NULL, filt_sowdetach, filt_sowrite };
85 
86 uma_zone_t socket_zone;
87 so_gen_t	so_gencnt;	/* generation count for sockets */
88 
89 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
90 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
91 
92 SYSCTL_DECL(_kern_ipc);
93 
94 static int somaxconn = SOMAXCONN;
95 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
96     &somaxconn, 0, "Maximum pending socket connection queue size");
97 static int numopensockets;
98 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
99     &numopensockets, 0, "Number of open sockets");
100 #ifdef ZERO_COPY_SOCKETS
101 /* These aren't static because they're used in other files. */
102 int so_zero_copy_send = 1;
103 int so_zero_copy_receive = 1;
104 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
105     "Zero copy controls");
106 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
107     &so_zero_copy_receive, 0, "Enable zero copy receive");
108 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
109     &so_zero_copy_send, 0, "Enable zero copy send");
110 #endif /* ZERO_COPY_SOCKETS */
111 
112 
113 /*
114  * Socket operation routines.
115  * These routines are called by the routines in
116  * sys_socket.c or from a system process, and
117  * implement the semantics of socket operations by
118  * switching out to the protocol specific routines.
119  */
120 
121 /*
122  * Get a socket structure from our zone, and initialize it.
123  * Note that it would probably be better to allocate socket
124  * and PCB at the same time, but I'm not convinced that all
125  * the protocols can be easily modified to do this.
126  *
127  * soalloc() returns a socket with a ref count of 0.
128  */
129 struct socket *
130 soalloc(waitok)
131 	int waitok;
132 {
133 	struct socket *so;
134 #ifdef MAC
135 	int error;
136 #endif
137 	int flag;
138 
139 	if (waitok == 1)
140 		flag = M_WAITOK;
141 	else
142 		flag = M_NOWAIT;
143 	flag |= M_ZERO;
144 	so = uma_zalloc(socket_zone, flag);
145 	if (so) {
146 #ifdef MAC
147 		error = mac_init_socket(so, flag);
148 		if (error != 0) {
149 			uma_zfree(socket_zone, so);
150 			so = NULL;
151 			return so;
152 		}
153 #endif
154 		/* XXX race condition for reentrant kernel */
155 		so->so_gencnt = ++so_gencnt;
156 		/* sx_init(&so->so_sxlock, "socket sxlock"); */
157 		TAILQ_INIT(&so->so_aiojobq);
158 		++numopensockets;
159 	}
160 	return so;
161 }
162 
163 /*
164  * socreate returns a socket with a ref count of 1.  The socket should be
165  * closed with soclose().
166  */
167 int
168 socreate(dom, aso, type, proto, cred, td)
169 	int dom;
170 	struct socket **aso;
171 	register int type;
172 	int proto;
173 	struct ucred *cred;
174 	struct thread *td;
175 {
176 	register struct protosw *prp;
177 	register struct socket *so;
178 	register int error;
179 
180 	if (proto)
181 		prp = pffindproto(dom, proto, type);
182 	else
183 		prp = pffindtype(dom, type);
184 
185 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
186 		return (EPROTONOSUPPORT);
187 
188 	if (jailed(cred) && jail_socket_unixiproute_only &&
189 	    prp->pr_domain->dom_family != PF_LOCAL &&
190 	    prp->pr_domain->dom_family != PF_INET &&
191 	    prp->pr_domain->dom_family != PF_ROUTE) {
192 		return (EPROTONOSUPPORT);
193 	}
194 
195 	if (prp->pr_type != type)
196 		return (EPROTOTYPE);
197 	so = soalloc(M_NOWAIT);
198 	if (so == NULL)
199 		return (ENOBUFS);
200 
201 	TAILQ_INIT(&so->so_incomp);
202 	TAILQ_INIT(&so->so_comp);
203 	so->so_type = type;
204 	so->so_cred = crhold(cred);
205 	so->so_proto = prp;
206 #ifdef MAC
207 	mac_create_socket(cred, so);
208 #endif
209 	soref(so);
210 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
211 	if (error) {
212 		so->so_state |= SS_NOFDREF;
213 		sorele(so);
214 		return (error);
215 	}
216 	*aso = so;
217 	return (0);
218 }
219 
220 int
221 sobind(so, nam, td)
222 	struct socket *so;
223 	struct sockaddr *nam;
224 	struct thread *td;
225 {
226 	int s = splnet();
227 	int error;
228 
229 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
230 	splx(s);
231 	return (error);
232 }
233 
234 static void
235 sodealloc(struct socket *so)
236 {
237 
238 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
239 	so->so_gencnt = ++so_gencnt;
240 	if (so->so_rcv.sb_hiwat)
241 		(void)chgsbsize(so->so_cred->cr_uidinfo,
242 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
243 	if (so->so_snd.sb_hiwat)
244 		(void)chgsbsize(so->so_cred->cr_uidinfo,
245 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
246 #ifdef INET
247 	if (so->so_accf != NULL) {
248 		if (so->so_accf->so_accept_filter != NULL &&
249 			so->so_accf->so_accept_filter->accf_destroy != NULL) {
250 			so->so_accf->so_accept_filter->accf_destroy(so);
251 		}
252 		if (so->so_accf->so_accept_filter_str != NULL)
253 			FREE(so->so_accf->so_accept_filter_str, M_ACCF);
254 		FREE(so->so_accf, M_ACCF);
255 	}
256 #endif
257 #ifdef MAC
258 	mac_destroy_socket(so);
259 #endif
260 	crfree(so->so_cred);
261 	/* sx_destroy(&so->so_sxlock); */
262 	uma_zfree(socket_zone, so);
263 	--numopensockets;
264 }
265 
266 int
267 solisten(so, backlog, td)
268 	register struct socket *so;
269 	int backlog;
270 	struct thread *td;
271 {
272 	int s, error;
273 
274 	s = splnet();
275 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td);
276 	if (error) {
277 		splx(s);
278 		return (error);
279 	}
280 	if (TAILQ_EMPTY(&so->so_comp))
281 		so->so_options |= SO_ACCEPTCONN;
282 	if (backlog < 0 || backlog > somaxconn)
283 		backlog = somaxconn;
284 	so->so_qlimit = backlog;
285 	splx(s);
286 	return (0);
287 }
288 
289 void
290 sofree(so)
291 	register struct socket *so;
292 {
293 	struct socket *head = so->so_head;
294 
295 	KASSERT(so->so_count == 0, ("socket %p so_count not 0", so));
296 
297 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
298 		return;
299 	if (head != NULL) {
300 		if (so->so_state & SS_INCOMP) {
301 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
302 			head->so_incqlen--;
303 		} else if (so->so_state & SS_COMP) {
304 			/*
305 			 * We must not decommission a socket that's
306 			 * on the accept(2) queue.  If we do, then
307 			 * accept(2) may hang after select(2) indicated
308 			 * that the listening socket was ready.
309 			 */
310 			return;
311 		} else {
312 			panic("sofree: not queued");
313 		}
314 		so->so_state &= ~SS_INCOMP;
315 		so->so_head = NULL;
316 	}
317 	sbrelease(&so->so_snd, so);
318 	sorflush(so);
319 	sodealloc(so);
320 }
321 
322 /*
323  * Close a socket on last file table reference removal.
324  * Initiate disconnect if connected.
325  * Free socket when disconnect complete.
326  *
327  * This function will sorele() the socket.  Note that soclose() may be
328  * called prior to the ref count reaching zero.  The actual socket
329  * structure will not be freed until the ref count reaches zero.
330  */
331 int
332 soclose(so)
333 	register struct socket *so;
334 {
335 	int s = splnet();		/* conservative */
336 	int error = 0;
337 
338 	funsetown(&so->so_sigio);
339 	if (so->so_options & SO_ACCEPTCONN) {
340 		struct socket *sp, *sonext;
341 
342 		sp = TAILQ_FIRST(&so->so_incomp);
343 		for (; sp != NULL; sp = sonext) {
344 			sonext = TAILQ_NEXT(sp, so_list);
345 			(void) soabort(sp);
346 		}
347 		for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) {
348 			sonext = TAILQ_NEXT(sp, so_list);
349 			/* Dequeue from so_comp since sofree() won't do it */
350 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
351 			so->so_qlen--;
352 			sp->so_state &= ~SS_COMP;
353 			sp->so_head = NULL;
354 			(void) soabort(sp);
355 		}
356 	}
357 	if (so->so_pcb == 0)
358 		goto discard;
359 	if (so->so_state & SS_ISCONNECTED) {
360 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
361 			error = sodisconnect(so);
362 			if (error)
363 				goto drop;
364 		}
365 		if (so->so_options & SO_LINGER) {
366 			if ((so->so_state & SS_ISDISCONNECTING) &&
367 			    (so->so_state & SS_NBIO))
368 				goto drop;
369 			while (so->so_state & SS_ISCONNECTED) {
370 				error = tsleep(&so->so_timeo,
371 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
372 				if (error)
373 					break;
374 			}
375 		}
376 	}
377 drop:
378 	if (so->so_pcb) {
379 		int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
380 		if (error == 0)
381 			error = error2;
382 	}
383 discard:
384 	if (so->so_state & SS_NOFDREF)
385 		panic("soclose: NOFDREF");
386 	so->so_state |= SS_NOFDREF;
387 	sorele(so);
388 	splx(s);
389 	return (error);
390 }
391 
392 /*
393  * Must be called at splnet...
394  */
395 int
396 soabort(so)
397 	struct socket *so;
398 {
399 	int error;
400 
401 	error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
402 	if (error) {
403 		sotryfree(so);	/* note: does not decrement the ref count */
404 		return error;
405 	}
406 	return (0);
407 }
408 
409 int
410 soaccept(so, nam)
411 	register struct socket *so;
412 	struct sockaddr **nam;
413 {
414 	int s = splnet();
415 	int error;
416 
417 	if ((so->so_state & SS_NOFDREF) == 0)
418 		panic("soaccept: !NOFDREF");
419 	so->so_state &= ~SS_NOFDREF;
420 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
421 	splx(s);
422 	return (error);
423 }
424 
425 int
426 soconnect(so, nam, td)
427 	register struct socket *so;
428 	struct sockaddr *nam;
429 	struct thread *td;
430 {
431 	int s;
432 	int error;
433 
434 	if (so->so_options & SO_ACCEPTCONN)
435 		return (EOPNOTSUPP);
436 	s = splnet();
437 	/*
438 	 * If protocol is connection-based, can only connect once.
439 	 * Otherwise, if connected, try to disconnect first.
440 	 * This allows user to disconnect by connecting to, e.g.,
441 	 * a null address.
442 	 */
443 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
444 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
445 	    (error = sodisconnect(so))))
446 		error = EISCONN;
447 	else
448 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
449 	splx(s);
450 	return (error);
451 }
452 
453 int
454 soconnect2(so1, so2)
455 	register struct socket *so1;
456 	struct socket *so2;
457 {
458 	int s = splnet();
459 	int error;
460 
461 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
462 	splx(s);
463 	return (error);
464 }
465 
466 int
467 sodisconnect(so)
468 	register struct socket *so;
469 {
470 	int s = splnet();
471 	int error;
472 
473 	if ((so->so_state & SS_ISCONNECTED) == 0) {
474 		error = ENOTCONN;
475 		goto bad;
476 	}
477 	if (so->so_state & SS_ISDISCONNECTING) {
478 		error = EALREADY;
479 		goto bad;
480 	}
481 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
482 bad:
483 	splx(s);
484 	return (error);
485 }
486 
487 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
488 /*
489  * Send on a socket.
490  * If send must go all at once and message is larger than
491  * send buffering, then hard error.
492  * Lock against other senders.
493  * If must go all at once and not enough room now, then
494  * inform user that this would block and do nothing.
495  * Otherwise, if nonblocking, send as much as possible.
496  * The data to be sent is described by "uio" if nonzero,
497  * otherwise by the mbuf chain "top" (which must be null
498  * if uio is not).  Data provided in mbuf chain must be small
499  * enough to send all at once.
500  *
501  * Returns nonzero on error, timeout or signal; callers
502  * must check for short counts if EINTR/ERESTART are returned.
503  * Data and control buffers are freed on return.
504  */
505 
506 #ifdef ZERO_COPY_SOCKETS
507 struct so_zerocopy_stats{
508 	int size_ok;
509 	int align_ok;
510 	int found_ifp;
511 };
512 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
513 #include <netinet/in.h>
514 #include <net/route.h>
515 #include <netinet/in_pcb.h>
516 #include <vm/vm.h>
517 #include <vm/vm_page.h>
518 #include <vm/vm_object.h>
519 #endif /*ZERO_COPY_SOCKETS*/
520 
521 int
522 sosend(so, addr, uio, top, control, flags, td)
523 	register struct socket *so;
524 	struct sockaddr *addr;
525 	struct uio *uio;
526 	struct mbuf *top;
527 	struct mbuf *control;
528 	int flags;
529 	struct thread *td;
530 {
531 	struct mbuf **mp;
532 	register struct mbuf *m;
533 	register long space, len, resid;
534 	int clen = 0, error, s, dontroute, mlen;
535 	int atomic = sosendallatonce(so) || top;
536 #ifdef ZERO_COPY_SOCKETS
537 	int cow_send;
538 #endif /* ZERO_COPY_SOCKETS */
539 
540 	if (uio)
541 		resid = uio->uio_resid;
542 	else
543 		resid = top->m_pkthdr.len;
544 	/*
545 	 * In theory resid should be unsigned.
546 	 * However, space must be signed, as it might be less than 0
547 	 * if we over-committed, and we must use a signed comparison
548 	 * of space and resid.  On the other hand, a negative resid
549 	 * causes us to loop sending 0-length segments to the protocol.
550 	 *
551 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
552 	 * type sockets since that's an error.
553 	 */
554 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
555 		error = EINVAL;
556 		goto out;
557 	}
558 
559 	dontroute =
560 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
561 	    (so->so_proto->pr_flags & PR_ATOMIC);
562 	if (td)
563 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
564 	if (control)
565 		clen = control->m_len;
566 #define	snderr(errno)	{ error = errno; splx(s); goto release; }
567 
568 restart:
569 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
570 	if (error)
571 		goto out;
572 	do {
573 		s = splnet();
574 		if (so->so_state & SS_CANTSENDMORE)
575 			snderr(EPIPE);
576 		if (so->so_error) {
577 			error = so->so_error;
578 			so->so_error = 0;
579 			splx(s);
580 			goto release;
581 		}
582 		if ((so->so_state & SS_ISCONNECTED) == 0) {
583 			/*
584 			 * `sendto' and `sendmsg' is allowed on a connection-
585 			 * based socket if it supports implied connect.
586 			 * Return ENOTCONN if not connected and no address is
587 			 * supplied.
588 			 */
589 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
590 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
591 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
592 				    !(resid == 0 && clen != 0))
593 					snderr(ENOTCONN);
594 			} else if (addr == 0)
595 			    snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
596 				   ENOTCONN : EDESTADDRREQ);
597 		}
598 		space = sbspace(&so->so_snd);
599 		if (flags & MSG_OOB)
600 			space += 1024;
601 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
602 		    clen > so->so_snd.sb_hiwat)
603 			snderr(EMSGSIZE);
604 		if (space < resid + clen &&
605 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
606 			if (so->so_state & SS_NBIO)
607 				snderr(EWOULDBLOCK);
608 			sbunlock(&so->so_snd);
609 			error = sbwait(&so->so_snd);
610 			splx(s);
611 			if (error)
612 				goto out;
613 			goto restart;
614 		}
615 		splx(s);
616 		mp = &top;
617 		space -= clen;
618 		do {
619 		    if (uio == NULL) {
620 			/*
621 			 * Data is prepackaged in "top".
622 			 */
623 			resid = 0;
624 			if (flags & MSG_EOR)
625 				top->m_flags |= M_EOR;
626 		    } else do {
627 #ifdef ZERO_COPY_SOCKETS
628 			cow_send = 0;
629 #endif /* ZERO_COPY_SOCKETS */
630 			if (top == 0) {
631 				MGETHDR(m, M_TRYWAIT, MT_DATA);
632 				if (m == NULL) {
633 					error = ENOBUFS;
634 					goto release;
635 				}
636 				mlen = MHLEN;
637 				m->m_pkthdr.len = 0;
638 				m->m_pkthdr.rcvif = (struct ifnet *)0;
639 			} else {
640 				MGET(m, M_TRYWAIT, MT_DATA);
641 				if (m == NULL) {
642 					error = ENOBUFS;
643 					goto release;
644 				}
645 				mlen = MLEN;
646 			}
647 			if (resid >= MINCLSIZE) {
648 #ifdef ZERO_COPY_SOCKETS
649 				if (so_zero_copy_send &&
650 				    resid>=PAGE_SIZE &&
651 				    space>=PAGE_SIZE &&
652 				    uio->uio_iov->iov_len>=PAGE_SIZE) {
653 					so_zerocp_stats.size_ok++;
654 					if (!((vm_offset_t)
655 					  uio->uio_iov->iov_base & PAGE_MASK)){
656 						so_zerocp_stats.align_ok++;
657 						cow_send = socow_setup(m, uio);
658 					}
659 				}
660 				if (!cow_send){
661 #endif /* ZERO_COPY_SOCKETS */
662 				MCLGET(m, M_TRYWAIT);
663 				if ((m->m_flags & M_EXT) == 0)
664 					goto nopages;
665 				mlen = MCLBYTES;
666 				len = min(min(mlen, resid), space);
667 			} else {
668 #ifdef ZERO_COPY_SOCKETS
669 					len = PAGE_SIZE;
670 				}
671 
672 			} else {
673 #endif /* ZERO_COPY_SOCKETS */
674 nopages:
675 				len = min(min(mlen, resid), space);
676 				/*
677 				 * For datagram protocols, leave room
678 				 * for protocol headers in first mbuf.
679 				 */
680 				if (atomic && top == 0 && len < mlen)
681 					MH_ALIGN(m, len);
682 			}
683 			space -= len;
684 #ifdef ZERO_COPY_SOCKETS
685 			if (cow_send)
686 				error = 0;
687 			else
688 #endif /* ZERO_COPY_SOCKETS */
689 			error = uiomove(mtod(m, caddr_t), (int)len, uio);
690 			resid = uio->uio_resid;
691 			m->m_len = len;
692 			*mp = m;
693 			top->m_pkthdr.len += len;
694 			if (error)
695 				goto release;
696 			mp = &m->m_next;
697 			if (resid <= 0) {
698 				if (flags & MSG_EOR)
699 					top->m_flags |= M_EOR;
700 				break;
701 			}
702 		    } while (space > 0 && atomic);
703 		    if (dontroute)
704 			    so->so_options |= SO_DONTROUTE;
705 		    s = splnet();				/* XXX */
706 		    /*
707 		     * XXX all the SS_CANTSENDMORE checks previously
708 		     * done could be out of date.  We could have recieved
709 		     * a reset packet in an interrupt or maybe we slept
710 		     * while doing page faults in uiomove() etc. We could
711 		     * probably recheck again inside the splnet() protection
712 		     * here, but there are probably other places that this
713 		     * also happens.  We must rethink this.
714 		     */
715 		    error = (*so->so_proto->pr_usrreqs->pru_send)(so,
716 			(flags & MSG_OOB) ? PRUS_OOB :
717 			/*
718 			 * If the user set MSG_EOF, the protocol
719 			 * understands this flag and nothing left to
720 			 * send then use PRU_SEND_EOF instead of PRU_SEND.
721 			 */
722 			((flags & MSG_EOF) &&
723 			 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
724 			 (resid <= 0)) ?
725 				PRUS_EOF :
726 			/* If there is more to send set PRUS_MORETOCOME */
727 			(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
728 			top, addr, control, td);
729 		    splx(s);
730 		    if (dontroute)
731 			    so->so_options &= ~SO_DONTROUTE;
732 		    clen = 0;
733 		    control = 0;
734 		    top = 0;
735 		    mp = &top;
736 		    if (error)
737 			goto release;
738 		} while (resid && space > 0);
739 	} while (resid);
740 
741 release:
742 	sbunlock(&so->so_snd);
743 out:
744 	if (top)
745 		m_freem(top);
746 	if (control)
747 		m_freem(control);
748 	return (error);
749 }
750 
751 /*
752  * Implement receive operations on a socket.
753  * We depend on the way that records are added to the sockbuf
754  * by sbappend*.  In particular, each record (mbufs linked through m_next)
755  * must begin with an address if the protocol so specifies,
756  * followed by an optional mbuf or mbufs containing ancillary data,
757  * and then zero or more mbufs of data.
758  * In order to avoid blocking network interrupts for the entire time here,
759  * we splx() while doing the actual copy to user space.
760  * Although the sockbuf is locked, new data may still be appended,
761  * and thus we must maintain consistency of the sockbuf during that time.
762  *
763  * The caller may receive the data as a single mbuf chain by supplying
764  * an mbuf **mp0 for use in returning the chain.  The uio is then used
765  * only for the count in uio_resid.
766  */
767 int
768 soreceive(so, psa, uio, mp0, controlp, flagsp)
769 	register struct socket *so;
770 	struct sockaddr **psa;
771 	struct uio *uio;
772 	struct mbuf **mp0;
773 	struct mbuf **controlp;
774 	int *flagsp;
775 {
776 	struct mbuf *m, **mp;
777 	register int flags, len, error, s, offset;
778 	struct protosw *pr = so->so_proto;
779 	struct mbuf *nextrecord;
780 	int moff, type = 0;
781 	int orig_resid = uio->uio_resid;
782 
783 	mp = mp0;
784 	if (psa)
785 		*psa = 0;
786 	if (controlp)
787 		*controlp = 0;
788 	if (flagsp)
789 		flags = *flagsp &~ MSG_EOR;
790 	else
791 		flags = 0;
792 	if (flags & MSG_OOB) {
793 		m = m_get(M_TRYWAIT, MT_DATA);
794 		if (m == NULL)
795 			return (ENOBUFS);
796 		error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
797 		if (error)
798 			goto bad;
799 		do {
800 #ifdef ZERO_COPY_SOCKETS
801 			if (so_zero_copy_receive) {
802 				vm_page_t pg;
803 				int disposable;
804 
805 				if ((m->m_flags & M_EXT)
806 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
807 					disposable = 1;
808 				else
809 					disposable = 0;
810 
811 				pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t)));
812 				if (uio->uio_offset == -1)
813 					uio->uio_offset =IDX_TO_OFF(pg->pindex);
814 
815 				error = uiomoveco(mtod(m, caddr_t),
816 						  min(uio->uio_resid, m->m_len),
817 						  uio, pg->object,
818 						  disposable);
819 			} else
820 #endif /* ZERO_COPY_SOCKETS */
821 			error = uiomove(mtod(m, caddr_t),
822 			    (int) min(uio->uio_resid, m->m_len), uio);
823 			m = m_free(m);
824 		} while (uio->uio_resid && error == 0 && m);
825 bad:
826 		if (m)
827 			m_freem(m);
828 		return (error);
829 	}
830 	if (mp)
831 		*mp = (struct mbuf *)0;
832 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
833 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
834 
835 restart:
836 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
837 	if (error)
838 		return (error);
839 	s = splnet();
840 
841 	m = so->so_rcv.sb_mb;
842 	/*
843 	 * If we have less data than requested, block awaiting more
844 	 * (subject to any timeout) if:
845 	 *   1. the current count is less than the low water mark, or
846 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
847 	 *	receive operation at once if we block (resid <= hiwat).
848 	 *   3. MSG_DONTWAIT is not set
849 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
850 	 * we have to do the receive in sections, and thus risk returning
851 	 * a short count if a timeout or signal occurs after we start.
852 	 */
853 	if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
854 	    so->so_rcv.sb_cc < uio->uio_resid) &&
855 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
856 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
857 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
858 		KASSERT(m != 0 || !so->so_rcv.sb_cc,
859 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
860 		    m, so->so_rcv.sb_cc));
861 		if (so->so_error) {
862 			if (m)
863 				goto dontblock;
864 			error = so->so_error;
865 			if ((flags & MSG_PEEK) == 0)
866 				so->so_error = 0;
867 			goto release;
868 		}
869 		if (so->so_state & SS_CANTRCVMORE) {
870 			if (m)
871 				goto dontblock;
872 			else
873 				goto release;
874 		}
875 		for (; m; m = m->m_next)
876 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
877 				m = so->so_rcv.sb_mb;
878 				goto dontblock;
879 			}
880 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
881 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
882 			error = ENOTCONN;
883 			goto release;
884 		}
885 		if (uio->uio_resid == 0)
886 			goto release;
887 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
888 			error = EWOULDBLOCK;
889 			goto release;
890 		}
891 		sbunlock(&so->so_rcv);
892 		error = sbwait(&so->so_rcv);
893 		splx(s);
894 		if (error)
895 			return (error);
896 		goto restart;
897 	}
898 dontblock:
899 	if (uio->uio_td)
900 		uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
901 	nextrecord = m->m_nextpkt;
902 	if (pr->pr_flags & PR_ADDR) {
903 		KASSERT(m->m_type == MT_SONAME,
904 		    ("m->m_type == %d", m->m_type));
905 		orig_resid = 0;
906 		if (psa)
907 			*psa = dup_sockaddr(mtod(m, struct sockaddr *),
908 					    mp0 == 0);
909 		if (flags & MSG_PEEK) {
910 			m = m->m_next;
911 		} else {
912 			sbfree(&so->so_rcv, m);
913 			so->so_rcv.sb_mb = m_free(m);
914 			m = so->so_rcv.sb_mb;
915 		}
916 	}
917 	while (m && m->m_type == MT_CONTROL && error == 0) {
918 		if (flags & MSG_PEEK) {
919 			if (controlp)
920 				*controlp = m_copy(m, 0, m->m_len);
921 			m = m->m_next;
922 		} else {
923 			sbfree(&so->so_rcv, m);
924 			so->so_rcv.sb_mb = m->m_next;
925 			m->m_next = NULL;
926 			if (pr->pr_domain->dom_externalize)
927 				error =
928 				(*pr->pr_domain->dom_externalize)(m, controlp);
929 			else if (controlp)
930 				*controlp = m;
931 			else
932 				m_freem(m);
933 			m = so->so_rcv.sb_mb;
934 		}
935 		if (controlp) {
936 			orig_resid = 0;
937 			do
938 				controlp = &(*controlp)->m_next;
939 			while (*controlp != NULL);
940 		}
941 	}
942 	if (m) {
943 		if ((flags & MSG_PEEK) == 0)
944 			m->m_nextpkt = nextrecord;
945 		type = m->m_type;
946 		if (type == MT_OOBDATA)
947 			flags |= MSG_OOB;
948 	}
949 	moff = 0;
950 	offset = 0;
951 	while (m && uio->uio_resid > 0 && error == 0) {
952 		if (m->m_type == MT_OOBDATA) {
953 			if (type != MT_OOBDATA)
954 				break;
955 		} else if (type == MT_OOBDATA)
956 			break;
957 		else
958 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
959 			("m->m_type == %d", m->m_type));
960 		so->so_state &= ~SS_RCVATMARK;
961 		len = uio->uio_resid;
962 		if (so->so_oobmark && len > so->so_oobmark - offset)
963 			len = so->so_oobmark - offset;
964 		if (len > m->m_len - moff)
965 			len = m->m_len - moff;
966 		/*
967 		 * If mp is set, just pass back the mbufs.
968 		 * Otherwise copy them out via the uio, then free.
969 		 * Sockbuf must be consistent here (points to current mbuf,
970 		 * it points to next record) when we drop priority;
971 		 * we must note any additions to the sockbuf when we
972 		 * block interrupts again.
973 		 */
974 		if (mp == 0) {
975 			splx(s);
976 #ifdef ZERO_COPY_SOCKETS
977 			if (so_zero_copy_receive) {
978 				vm_page_t pg;
979 				int disposable;
980 
981 				if ((m->m_flags & M_EXT)
982 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
983 					disposable = 1;
984 				else
985 					disposable = 0;
986 
987 				pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) +
988 					moff));
989 
990 				if (uio->uio_offset == -1)
991 					uio->uio_offset =IDX_TO_OFF(pg->pindex);
992 
993 				error = uiomoveco(mtod(m, caddr_t) + moff,
994 						  (int)len, uio,pg->object,
995 						  disposable);
996 			} else
997 #endif /* ZERO_COPY_SOCKETS */
998 			error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
999 			s = splnet();
1000 			if (error)
1001 				goto release;
1002 		} else
1003 			uio->uio_resid -= len;
1004 		if (len == m->m_len - moff) {
1005 			if (m->m_flags & M_EOR)
1006 				flags |= MSG_EOR;
1007 			if (flags & MSG_PEEK) {
1008 				m = m->m_next;
1009 				moff = 0;
1010 			} else {
1011 				nextrecord = m->m_nextpkt;
1012 				sbfree(&so->so_rcv, m);
1013 				if (mp) {
1014 					*mp = m;
1015 					mp = &m->m_next;
1016 					so->so_rcv.sb_mb = m = m->m_next;
1017 					*mp = (struct mbuf *)0;
1018 				} else {
1019 					so->so_rcv.sb_mb = m_free(m);
1020 					m = so->so_rcv.sb_mb;
1021 				}
1022 				if (m)
1023 					m->m_nextpkt = nextrecord;
1024 			}
1025 		} else {
1026 			if (flags & MSG_PEEK)
1027 				moff += len;
1028 			else {
1029 				if (mp)
1030 					*mp = m_copym(m, 0, len, M_TRYWAIT);
1031 				m->m_data += len;
1032 				m->m_len -= len;
1033 				so->so_rcv.sb_cc -= len;
1034 			}
1035 		}
1036 		if (so->so_oobmark) {
1037 			if ((flags & MSG_PEEK) == 0) {
1038 				so->so_oobmark -= len;
1039 				if (so->so_oobmark == 0) {
1040 					so->so_state |= SS_RCVATMARK;
1041 					break;
1042 				}
1043 			} else {
1044 				offset += len;
1045 				if (offset == so->so_oobmark)
1046 					break;
1047 			}
1048 		}
1049 		if (flags & MSG_EOR)
1050 			break;
1051 		/*
1052 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1053 		 * we must not quit until "uio->uio_resid == 0" or an error
1054 		 * termination.  If a signal/timeout occurs, return
1055 		 * with a short count but without error.
1056 		 * Keep sockbuf locked against other readers.
1057 		 */
1058 		while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 &&
1059 		    !sosendallatonce(so) && !nextrecord) {
1060 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1061 				break;
1062 			/*
1063 			 * Notify the protocol that some data has been
1064 			 * drained before blocking.
1065 			 */
1066 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1067 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1068 			error = sbwait(&so->so_rcv);
1069 			if (error) {
1070 				sbunlock(&so->so_rcv);
1071 				splx(s);
1072 				return (0);
1073 			}
1074 			m = so->so_rcv.sb_mb;
1075 			if (m)
1076 				nextrecord = m->m_nextpkt;
1077 		}
1078 	}
1079 
1080 	if (m && pr->pr_flags & PR_ATOMIC) {
1081 		flags |= MSG_TRUNC;
1082 		if ((flags & MSG_PEEK) == 0)
1083 			(void) sbdroprecord(&so->so_rcv);
1084 	}
1085 	if ((flags & MSG_PEEK) == 0) {
1086 		if (m == 0)
1087 			so->so_rcv.sb_mb = nextrecord;
1088 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1089 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1090 	}
1091 	if (orig_resid == uio->uio_resid && orig_resid &&
1092 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1093 		sbunlock(&so->so_rcv);
1094 		splx(s);
1095 		goto restart;
1096 	}
1097 
1098 	if (flagsp)
1099 		*flagsp |= flags;
1100 release:
1101 	sbunlock(&so->so_rcv);
1102 	splx(s);
1103 	return (error);
1104 }
1105 
1106 int
1107 soshutdown(so, how)
1108 	register struct socket *so;
1109 	register int how;
1110 {
1111 	register struct protosw *pr = so->so_proto;
1112 
1113 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1114 		return (EINVAL);
1115 
1116 	if (how != SHUT_WR)
1117 		sorflush(so);
1118 	if (how != SHUT_RD)
1119 		return ((*pr->pr_usrreqs->pru_shutdown)(so));
1120 	return (0);
1121 }
1122 
1123 void
1124 sorflush(so)
1125 	register struct socket *so;
1126 {
1127 	register struct sockbuf *sb = &so->so_rcv;
1128 	register struct protosw *pr = so->so_proto;
1129 	register int s;
1130 	struct sockbuf asb;
1131 
1132 	sb->sb_flags |= SB_NOINTR;
1133 	(void) sblock(sb, M_WAITOK);
1134 	s = splimp();
1135 	socantrcvmore(so);
1136 	sbunlock(sb);
1137 	asb = *sb;
1138 	bzero(sb, sizeof (*sb));
1139 	splx(s);
1140 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1141 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1142 	sbrelease(&asb, so);
1143 }
1144 
1145 #ifdef INET
1146 static int
1147 do_setopt_accept_filter(so, sopt)
1148 	struct	socket *so;
1149 	struct	sockopt *sopt;
1150 {
1151 	struct accept_filter_arg	*afap = NULL;
1152 	struct accept_filter	*afp;
1153 	struct so_accf	*af = so->so_accf;
1154 	int	error = 0;
1155 
1156 	/* do not set/remove accept filters on non listen sockets */
1157 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1158 		error = EINVAL;
1159 		goto out;
1160 	}
1161 
1162 	/* removing the filter */
1163 	if (sopt == NULL) {
1164 		if (af != NULL) {
1165 			if (af->so_accept_filter != NULL &&
1166 				af->so_accept_filter->accf_destroy != NULL) {
1167 				af->so_accept_filter->accf_destroy(so);
1168 			}
1169 			if (af->so_accept_filter_str != NULL) {
1170 				FREE(af->so_accept_filter_str, M_ACCF);
1171 			}
1172 			FREE(af, M_ACCF);
1173 			so->so_accf = NULL;
1174 		}
1175 		so->so_options &= ~SO_ACCEPTFILTER;
1176 		return (0);
1177 	}
1178 	/* adding a filter */
1179 	/* must remove previous filter first */
1180 	if (af != NULL) {
1181 		error = EINVAL;
1182 		goto out;
1183 	}
1184 	/* don't put large objects on the kernel stack */
1185 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1186 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1187 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1188 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1189 	if (error)
1190 		goto out;
1191 	afp = accept_filt_get(afap->af_name);
1192 	if (afp == NULL) {
1193 		error = ENOENT;
1194 		goto out;
1195 	}
1196 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1197 	if (afp->accf_create != NULL) {
1198 		if (afap->af_name[0] != '\0') {
1199 			int len = strlen(afap->af_name) + 1;
1200 
1201 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1202 			strcpy(af->so_accept_filter_str, afap->af_name);
1203 		}
1204 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1205 		if (af->so_accept_filter_arg == NULL) {
1206 			FREE(af->so_accept_filter_str, M_ACCF);
1207 			FREE(af, M_ACCF);
1208 			so->so_accf = NULL;
1209 			error = EINVAL;
1210 			goto out;
1211 		}
1212 	}
1213 	af->so_accept_filter = afp;
1214 	so->so_accf = af;
1215 	so->so_options |= SO_ACCEPTFILTER;
1216 out:
1217 	if (afap != NULL)
1218 		FREE(afap, M_TEMP);
1219 	return (error);
1220 }
1221 #endif /* INET */
1222 
1223 /*
1224  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1225  * an additional variant to handle the case where the option value needs
1226  * to be some kind of integer, but not a specific size.
1227  * In addition to their use here, these functions are also called by the
1228  * protocol-level pr_ctloutput() routines.
1229  */
1230 int
1231 sooptcopyin(sopt, buf, len, minlen)
1232 	struct	sockopt *sopt;
1233 	void	*buf;
1234 	size_t	len;
1235 	size_t	minlen;
1236 {
1237 	size_t	valsize;
1238 
1239 	/*
1240 	 * If the user gives us more than we wanted, we ignore it,
1241 	 * but if we don't get the minimum length the caller
1242 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1243 	 * is set to however much we actually retrieved.
1244 	 */
1245 	if ((valsize = sopt->sopt_valsize) < minlen)
1246 		return EINVAL;
1247 	if (valsize > len)
1248 		sopt->sopt_valsize = valsize = len;
1249 
1250 	if (sopt->sopt_td != 0)
1251 		return (copyin(sopt->sopt_val, buf, valsize));
1252 
1253 	bcopy(sopt->sopt_val, buf, valsize);
1254 	return 0;
1255 }
1256 
1257 int
1258 sosetopt(so, sopt)
1259 	struct socket *so;
1260 	struct sockopt *sopt;
1261 {
1262 	int	error, optval;
1263 	struct	linger l;
1264 	struct	timeval tv;
1265 	u_long  val;
1266 #ifdef MAC
1267 	struct mac extmac;
1268 #endif /* MAC */
1269 
1270 	error = 0;
1271 	if (sopt->sopt_level != SOL_SOCKET) {
1272 		if (so->so_proto && so->so_proto->pr_ctloutput)
1273 			return ((*so->so_proto->pr_ctloutput)
1274 				  (so, sopt));
1275 		error = ENOPROTOOPT;
1276 	} else {
1277 		switch (sopt->sopt_name) {
1278 #ifdef INET
1279 		case SO_ACCEPTFILTER:
1280 			error = do_setopt_accept_filter(so, sopt);
1281 			if (error)
1282 				goto bad;
1283 			break;
1284 #endif
1285 		case SO_LINGER:
1286 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1287 			if (error)
1288 				goto bad;
1289 
1290 			so->so_linger = l.l_linger;
1291 			if (l.l_onoff)
1292 				so->so_options |= SO_LINGER;
1293 			else
1294 				so->so_options &= ~SO_LINGER;
1295 			break;
1296 
1297 		case SO_DEBUG:
1298 		case SO_KEEPALIVE:
1299 		case SO_DONTROUTE:
1300 		case SO_USELOOPBACK:
1301 		case SO_BROADCAST:
1302 		case SO_REUSEADDR:
1303 		case SO_REUSEPORT:
1304 		case SO_OOBINLINE:
1305 		case SO_TIMESTAMP:
1306 		case SO_NOSIGPIPE:
1307 			error = sooptcopyin(sopt, &optval, sizeof optval,
1308 					    sizeof optval);
1309 			if (error)
1310 				goto bad;
1311 			if (optval)
1312 				so->so_options |= sopt->sopt_name;
1313 			else
1314 				so->so_options &= ~sopt->sopt_name;
1315 			break;
1316 
1317 		case SO_SNDBUF:
1318 		case SO_RCVBUF:
1319 		case SO_SNDLOWAT:
1320 		case SO_RCVLOWAT:
1321 			error = sooptcopyin(sopt, &optval, sizeof optval,
1322 					    sizeof optval);
1323 			if (error)
1324 				goto bad;
1325 
1326 			/*
1327 			 * Values < 1 make no sense for any of these
1328 			 * options, so disallow them.
1329 			 */
1330 			if (optval < 1) {
1331 				error = EINVAL;
1332 				goto bad;
1333 			}
1334 
1335 			switch (sopt->sopt_name) {
1336 			case SO_SNDBUF:
1337 			case SO_RCVBUF:
1338 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1339 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1340 				    so, curthread) == 0) {
1341 					error = ENOBUFS;
1342 					goto bad;
1343 				}
1344 				break;
1345 
1346 			/*
1347 			 * Make sure the low-water is never greater than
1348 			 * the high-water.
1349 			 */
1350 			case SO_SNDLOWAT:
1351 				so->so_snd.sb_lowat =
1352 				    (optval > so->so_snd.sb_hiwat) ?
1353 				    so->so_snd.sb_hiwat : optval;
1354 				break;
1355 			case SO_RCVLOWAT:
1356 				so->so_rcv.sb_lowat =
1357 				    (optval > so->so_rcv.sb_hiwat) ?
1358 				    so->so_rcv.sb_hiwat : optval;
1359 				break;
1360 			}
1361 			break;
1362 
1363 		case SO_SNDTIMEO:
1364 		case SO_RCVTIMEO:
1365 			error = sooptcopyin(sopt, &tv, sizeof tv,
1366 					    sizeof tv);
1367 			if (error)
1368 				goto bad;
1369 
1370 			/* assert(hz > 0); */
1371 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1372 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1373 				error = EDOM;
1374 				goto bad;
1375 			}
1376 			/* assert(tick > 0); */
1377 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1378 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1379 			if (val > SHRT_MAX) {
1380 				error = EDOM;
1381 				goto bad;
1382 			}
1383 
1384 			switch (sopt->sopt_name) {
1385 			case SO_SNDTIMEO:
1386 				so->so_snd.sb_timeo = val;
1387 				break;
1388 			case SO_RCVTIMEO:
1389 				so->so_rcv.sb_timeo = val;
1390 				break;
1391 			}
1392 			break;
1393 		case SO_LABEL:
1394 #ifdef MAC
1395 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
1396 			    sizeof extmac);
1397 			if (error)
1398 				goto bad;
1399 
1400 			error = mac_setsockopt_label_set(
1401 			    sopt->sopt_td->td_ucred, so, &extmac);
1402 
1403 #else /* MAC */
1404 			error = EOPNOTSUPP;
1405 #endif /* MAC */
1406 			break;
1407 		default:
1408 			error = ENOPROTOOPT;
1409 			break;
1410 		}
1411 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1412 			(void) ((*so->so_proto->pr_ctloutput)
1413 				  (so, sopt));
1414 		}
1415 	}
1416 bad:
1417 	return (error);
1418 }
1419 
1420 /* Helper routine for getsockopt */
1421 int
1422 sooptcopyout(sopt, buf, len)
1423 	struct	sockopt *sopt;
1424 	void	*buf;
1425 	size_t	len;
1426 {
1427 	int	error;
1428 	size_t	valsize;
1429 
1430 	error = 0;
1431 
1432 	/*
1433 	 * Documented get behavior is that we always return a value,
1434 	 * possibly truncated to fit in the user's buffer.
1435 	 * Traditional behavior is that we always tell the user
1436 	 * precisely how much we copied, rather than something useful
1437 	 * like the total amount we had available for her.
1438 	 * Note that this interface is not idempotent; the entire answer must
1439 	 * generated ahead of time.
1440 	 */
1441 	valsize = min(len, sopt->sopt_valsize);
1442 	sopt->sopt_valsize = valsize;
1443 	if (sopt->sopt_val != 0) {
1444 		if (sopt->sopt_td != 0)
1445 			error = copyout(buf, sopt->sopt_val, valsize);
1446 		else
1447 			bcopy(buf, sopt->sopt_val, valsize);
1448 	}
1449 	return error;
1450 }
1451 
1452 int
1453 sogetopt(so, sopt)
1454 	struct socket *so;
1455 	struct sockopt *sopt;
1456 {
1457 	int	error, optval;
1458 	struct	linger l;
1459 	struct	timeval tv;
1460 #ifdef INET
1461 	struct accept_filter_arg *afap;
1462 #endif
1463 #ifdef MAC
1464 	struct mac extmac;
1465 #endif /* MAC */
1466 
1467 	error = 0;
1468 	if (sopt->sopt_level != SOL_SOCKET) {
1469 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1470 			return ((*so->so_proto->pr_ctloutput)
1471 				  (so, sopt));
1472 		} else
1473 			return (ENOPROTOOPT);
1474 	} else {
1475 		switch (sopt->sopt_name) {
1476 #ifdef INET
1477 		case SO_ACCEPTFILTER:
1478 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1479 				return (EINVAL);
1480 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1481 				M_TEMP, M_WAITOK | M_ZERO);
1482 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1483 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1484 				if (so->so_accf->so_accept_filter_str != NULL)
1485 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1486 			}
1487 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1488 			FREE(afap, M_TEMP);
1489 			break;
1490 #endif
1491 
1492 		case SO_LINGER:
1493 			l.l_onoff = so->so_options & SO_LINGER;
1494 			l.l_linger = so->so_linger;
1495 			error = sooptcopyout(sopt, &l, sizeof l);
1496 			break;
1497 
1498 		case SO_USELOOPBACK:
1499 		case SO_DONTROUTE:
1500 		case SO_DEBUG:
1501 		case SO_KEEPALIVE:
1502 		case SO_REUSEADDR:
1503 		case SO_REUSEPORT:
1504 		case SO_BROADCAST:
1505 		case SO_OOBINLINE:
1506 		case SO_TIMESTAMP:
1507 		case SO_NOSIGPIPE:
1508 			optval = so->so_options & sopt->sopt_name;
1509 integer:
1510 			error = sooptcopyout(sopt, &optval, sizeof optval);
1511 			break;
1512 
1513 		case SO_TYPE:
1514 			optval = so->so_type;
1515 			goto integer;
1516 
1517 		case SO_ERROR:
1518 			optval = so->so_error;
1519 			so->so_error = 0;
1520 			goto integer;
1521 
1522 		case SO_SNDBUF:
1523 			optval = so->so_snd.sb_hiwat;
1524 			goto integer;
1525 
1526 		case SO_RCVBUF:
1527 			optval = so->so_rcv.sb_hiwat;
1528 			goto integer;
1529 
1530 		case SO_SNDLOWAT:
1531 			optval = so->so_snd.sb_lowat;
1532 			goto integer;
1533 
1534 		case SO_RCVLOWAT:
1535 			optval = so->so_rcv.sb_lowat;
1536 			goto integer;
1537 
1538 		case SO_SNDTIMEO:
1539 		case SO_RCVTIMEO:
1540 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1541 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1542 
1543 			tv.tv_sec = optval / hz;
1544 			tv.tv_usec = (optval % hz) * tick;
1545 			error = sooptcopyout(sopt, &tv, sizeof tv);
1546 			break;
1547 		case SO_LABEL:
1548 #ifdef MAC
1549 			error = mac_getsockopt_label_get(
1550 			    sopt->sopt_td->td_ucred, so, &extmac);
1551 			if (error)
1552 				return (error);
1553 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
1554 #else /* MAC */
1555 			error = EOPNOTSUPP;
1556 #endif /* MAC */
1557 			break;
1558 		case SO_PEERLABEL:
1559 #ifdef MAC
1560 			error = mac_getsockopt_peerlabel_get(
1561 			    sopt->sopt_td->td_ucred, so, &extmac);
1562 			if (error)
1563 				return (error);
1564 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
1565 #else /* MAC */
1566 			error = EOPNOTSUPP;
1567 #endif /* MAC */
1568 			break;
1569 		default:
1570 			error = ENOPROTOOPT;
1571 			break;
1572 		}
1573 		return (error);
1574 	}
1575 }
1576 
1577 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1578 int
1579 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1580 {
1581 	struct mbuf *m, *m_prev;
1582 	int sopt_size = sopt->sopt_valsize;
1583 
1584 	MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1585 	if (m == 0)
1586 		return ENOBUFS;
1587 	if (sopt_size > MLEN) {
1588 		MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
1589 		if ((m->m_flags & M_EXT) == 0) {
1590 			m_free(m);
1591 			return ENOBUFS;
1592 		}
1593 		m->m_len = min(MCLBYTES, sopt_size);
1594 	} else {
1595 		m->m_len = min(MLEN, sopt_size);
1596 	}
1597 	sopt_size -= m->m_len;
1598 	*mp = m;
1599 	m_prev = m;
1600 
1601 	while (sopt_size) {
1602 		MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1603 		if (m == 0) {
1604 			m_freem(*mp);
1605 			return ENOBUFS;
1606 		}
1607 		if (sopt_size > MLEN) {
1608 			MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
1609 			if ((m->m_flags & M_EXT) == 0) {
1610 				m_freem(*mp);
1611 				return ENOBUFS;
1612 			}
1613 			m->m_len = min(MCLBYTES, sopt_size);
1614 		} else {
1615 			m->m_len = min(MLEN, sopt_size);
1616 		}
1617 		sopt_size -= m->m_len;
1618 		m_prev->m_next = m;
1619 		m_prev = m;
1620 	}
1621 	return 0;
1622 }
1623 
1624 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1625 int
1626 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1627 {
1628 	struct mbuf *m0 = m;
1629 
1630 	if (sopt->sopt_val == NULL)
1631 		return 0;
1632 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1633 		if (sopt->sopt_td != NULL) {
1634 			int error;
1635 
1636 			error = copyin(sopt->sopt_val, mtod(m, char *),
1637 				       m->m_len);
1638 			if (error != 0) {
1639 				m_freem(m0);
1640 				return(error);
1641 			}
1642 		} else
1643 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1644 		sopt->sopt_valsize -= m->m_len;
1645 		(caddr_t)sopt->sopt_val += m->m_len;
1646 		m = m->m_next;
1647 	}
1648 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1649 		panic("ip6_sooptmcopyin");
1650 	return 0;
1651 }
1652 
1653 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1654 int
1655 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1656 {
1657 	struct mbuf *m0 = m;
1658 	size_t valsize = 0;
1659 
1660 	if (sopt->sopt_val == NULL)
1661 		return 0;
1662 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1663 		if (sopt->sopt_td != NULL) {
1664 			int error;
1665 
1666 			error = copyout(mtod(m, char *), sopt->sopt_val,
1667 				       m->m_len);
1668 			if (error != 0) {
1669 				m_freem(m0);
1670 				return(error);
1671 			}
1672 		} else
1673 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
1674 	       sopt->sopt_valsize -= m->m_len;
1675 	       (caddr_t)sopt->sopt_val += m->m_len;
1676 	       valsize += m->m_len;
1677 	       m = m->m_next;
1678 	}
1679 	if (m != NULL) {
1680 		/* enough soopt buffer should be given from user-land */
1681 		m_freem(m0);
1682 		return(EINVAL);
1683 	}
1684 	sopt->sopt_valsize = valsize;
1685 	return 0;
1686 }
1687 
1688 void
1689 sohasoutofband(so)
1690 	register struct socket *so;
1691 {
1692 	if (so->so_sigio != NULL)
1693 		pgsigio(&so->so_sigio, SIGURG, 0);
1694 	selwakeup(&so->so_rcv.sb_sel);
1695 }
1696 
1697 int
1698 sopoll(struct socket *so, int events, struct ucred *active_cred,
1699     struct thread *td)
1700 {
1701 	int revents = 0;
1702 	int s = splnet();
1703 
1704 	if (events & (POLLIN | POLLRDNORM))
1705 		if (soreadable(so))
1706 			revents |= events & (POLLIN | POLLRDNORM);
1707 
1708 	if (events & POLLINIGNEOF)
1709 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
1710 		    !TAILQ_EMPTY(&so->so_comp) || so->so_error)
1711 			revents |= POLLINIGNEOF;
1712 
1713 	if (events & (POLLOUT | POLLWRNORM))
1714 		if (sowriteable(so))
1715 			revents |= events & (POLLOUT | POLLWRNORM);
1716 
1717 	if (events & (POLLPRI | POLLRDBAND))
1718 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1719 			revents |= events & (POLLPRI | POLLRDBAND);
1720 
1721 	if (revents == 0) {
1722 		if (events &
1723 		    (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1724 		     POLLRDBAND)) {
1725 			selrecord(td, &so->so_rcv.sb_sel);
1726 			so->so_rcv.sb_flags |= SB_SEL;
1727 		}
1728 
1729 		if (events & (POLLOUT | POLLWRNORM)) {
1730 			selrecord(td, &so->so_snd.sb_sel);
1731 			so->so_snd.sb_flags |= SB_SEL;
1732 		}
1733 	}
1734 
1735 	splx(s);
1736 	return (revents);
1737 }
1738 
1739 int
1740 sokqfilter(struct file *fp, struct knote *kn)
1741 {
1742 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1743 	struct sockbuf *sb;
1744 	int s;
1745 
1746 	switch (kn->kn_filter) {
1747 	case EVFILT_READ:
1748 		if (so->so_options & SO_ACCEPTCONN)
1749 			kn->kn_fop = &solisten_filtops;
1750 		else
1751 			kn->kn_fop = &soread_filtops;
1752 		sb = &so->so_rcv;
1753 		break;
1754 	case EVFILT_WRITE:
1755 		kn->kn_fop = &sowrite_filtops;
1756 		sb = &so->so_snd;
1757 		break;
1758 	default:
1759 		return (1);
1760 	}
1761 
1762 	s = splnet();
1763 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1764 	sb->sb_flags |= SB_KNOTE;
1765 	splx(s);
1766 	return (0);
1767 }
1768 
1769 static void
1770 filt_sordetach(struct knote *kn)
1771 {
1772 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1773 	int s = splnet();
1774 
1775 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1776 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1777 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1778 	splx(s);
1779 }
1780 
1781 /*ARGSUSED*/
1782 static int
1783 filt_soread(struct knote *kn, long hint)
1784 {
1785 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1786 
1787 	kn->kn_data = so->so_rcv.sb_cc;
1788 	if (so->so_state & SS_CANTRCVMORE) {
1789 		kn->kn_flags |= EV_EOF;
1790 		kn->kn_fflags = so->so_error;
1791 		return (1);
1792 	}
1793 	if (so->so_error)	/* temporary udp error */
1794 		return (1);
1795 	if (kn->kn_sfflags & NOTE_LOWAT)
1796 		return (kn->kn_data >= kn->kn_sdata);
1797 	return (kn->kn_data >= so->so_rcv.sb_lowat);
1798 }
1799 
1800 static void
1801 filt_sowdetach(struct knote *kn)
1802 {
1803 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1804 	int s = splnet();
1805 
1806 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1807 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1808 		so->so_snd.sb_flags &= ~SB_KNOTE;
1809 	splx(s);
1810 }
1811 
1812 /*ARGSUSED*/
1813 static int
1814 filt_sowrite(struct knote *kn, long hint)
1815 {
1816 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1817 
1818 	kn->kn_data = sbspace(&so->so_snd);
1819 	if (so->so_state & SS_CANTSENDMORE) {
1820 		kn->kn_flags |= EV_EOF;
1821 		kn->kn_fflags = so->so_error;
1822 		return (1);
1823 	}
1824 	if (so->so_error)	/* temporary udp error */
1825 		return (1);
1826 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1827 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1828 		return (0);
1829 	if (kn->kn_sfflags & NOTE_LOWAT)
1830 		return (kn->kn_data >= kn->kn_sdata);
1831 	return (kn->kn_data >= so->so_snd.sb_lowat);
1832 }
1833 
1834 /*ARGSUSED*/
1835 static int
1836 filt_solisten(struct knote *kn, long hint)
1837 {
1838 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1839 
1840 	kn->kn_data = so->so_qlen;
1841 	return (! TAILQ_EMPTY(&so->so_comp));
1842 }
1843 
1844 int
1845 socheckuid(struct socket *so, uid_t uid)
1846 {
1847 
1848 	if (so == NULL)
1849 		return (EPERM);
1850 	if (so->so_cred->cr_uid == uid)
1851 		return (0);
1852 	return (EPERM);
1853 }
1854