xref: /dflybsd-src/sys/kern/uipc_socket.c (revision bc76a771df54af7e361532b257cecc26227736b4)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 1982, 1986, 1988, 1990, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 3. All advertising materials mentioning features or use of this software
15  *    must display the following acknowledgement:
16  *	This product includes software developed by the University of
17  *	California, Berkeley and its contributors.
18  * 4. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
35  * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
36  * $DragonFly: src/sys/kern/uipc_socket.c,v 1.17 2004/04/10 00:48:06 hsu Exp $
37  */
38 
39 #include "opt_inet.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/fcntl.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/domain.h>
47 #include <sys/file.h>			/* for struct knote */
48 #include <sys/kernel.h>
49 #include <sys/malloc.h>
50 #include <sys/event.h>
51 #include <sys/poll.h>
52 #include <sys/proc.h>
53 #include <sys/protosw.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/socketops.h>
57 #include <sys/resourcevar.h>
58 #include <sys/signalvar.h>
59 #include <sys/sysctl.h>
60 #include <sys/uio.h>
61 #include <sys/jail.h>
62 #include <vm/vm_zone.h>
63 
64 #include <machine/limits.h>
65 
66 #ifdef INET
67 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
68 #endif /* INET */
69 
70 static void 	filt_sordetach(struct knote *kn);
71 static int 	filt_soread(struct knote *kn, long hint);
72 static void 	filt_sowdetach(struct knote *kn);
73 static int	filt_sowrite(struct knote *kn, long hint);
74 static int	filt_solisten(struct knote *kn, long hint);
75 
76 static struct filterops solisten_filtops =
77 	{ 1, NULL, filt_sordetach, filt_solisten };
78 static struct filterops soread_filtops =
79 	{ 1, NULL, filt_sordetach, filt_soread };
80 static struct filterops sowrite_filtops =
81 	{ 1, NULL, filt_sowdetach, filt_sowrite };
82 
83 struct	vm_zone *socket_zone;
84 so_gen_t	so_gencnt;	/* generation count for sockets */
85 
86 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
87 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
88 
89 SYSCTL_DECL(_kern_ipc);
90 
91 static int somaxconn = SOMAXCONN;
92 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
93     &somaxconn, 0, "Maximum pending socket connection queue size");
94 
95 /*
96  * Socket operation routines.
97  * These routines are called by the routines in
98  * sys_socket.c or from a system process, and
99  * implement the semantics of socket operations by
100  * switching out to the protocol specific routines.
101  */
102 
103 /*
104  * Get a socket structure from our zone, and initialize it.
105  * We don't implement `waitok' yet (see comments in uipc_domain.c).
106  * Note that it would probably be better to allocate socket
107  * and PCB at the same time, but I'm not convinced that all
108  * the protocols can be easily modified to do this.
109  */
110 struct socket *
111 soalloc(waitok)
112 	int waitok;
113 {
114 	struct socket *so;
115 
116 	so = zalloc(socket_zone);
117 	if (so) {
118 		/* XXX race condition for reentrant kernel */
119 		bzero(so, sizeof *so);
120 		so->so_gencnt = ++so_gencnt;
121 		TAILQ_INIT(&so->so_aiojobq);
122 		TAILQ_INIT(&so->so_rcv.sb_sel.si_mlist);
123 		TAILQ_INIT(&so->so_snd.sb_sel.si_mlist);
124 	}
125 	return so;
126 }
127 
128 int
129 socreate(int dom, struct socket **aso, int type,
130 	int proto, struct thread *td)
131 {
132 	struct proc *p = td->td_proc;
133 	struct protosw *prp;
134 	struct socket *so;
135 	struct pru_attach_info ai;
136 	int error;
137 
138 	if (proto)
139 		prp = pffindproto(dom, proto, type);
140 	else
141 		prp = pffindtype(dom, type);
142 
143 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
144 		return (EPROTONOSUPPORT);
145 
146 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
147 	    prp->pr_domain->dom_family != PF_LOCAL &&
148 	    prp->pr_domain->dom_family != PF_INET &&
149 	    prp->pr_domain->dom_family != PF_ROUTE) {
150 		return (EPROTONOSUPPORT);
151 	}
152 
153 	if (prp->pr_type != type)
154 		return (EPROTOTYPE);
155 	so = soalloc(p != 0);
156 	if (so == 0)
157 		return (ENOBUFS);
158 
159 	TAILQ_INIT(&so->so_incomp);
160 	TAILQ_INIT(&so->so_comp);
161 	so->so_type = type;
162 	so->so_cred = crhold(p->p_ucred);
163 	so->so_proto = prp;
164 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
165 	ai.p_ucred = p->p_ucred;
166 	ai.fd_rdir = p->p_fd->fd_rdir;
167 	error = so_pru_attach(so, proto, &ai);
168 	if (error) {
169 		so->so_state |= SS_NOFDREF;
170 		sofree(so);
171 		return (error);
172 	}
173 	*aso = so;
174 	return (0);
175 }
176 
177 int
178 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
179 {
180 	int s = splnet();
181 	int error;
182 
183 	error = so_pru_bind(so, nam, td);
184 	splx(s);
185 	return (error);
186 }
187 
188 void
189 sodealloc(struct socket *so)
190 {
191 
192 	so->so_gencnt = ++so_gencnt;
193 	if (so->so_rcv.sb_hiwat)
194 		(void)chgsbsize(so->so_cred->cr_uidinfo,
195 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
196 	if (so->so_snd.sb_hiwat)
197 		(void)chgsbsize(so->so_cred->cr_uidinfo,
198 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
199 #ifdef INET
200 	if (so->so_accf != NULL) {
201 		if (so->so_accf->so_accept_filter != NULL &&
202 			so->so_accf->so_accept_filter->accf_destroy != NULL) {
203 			so->so_accf->so_accept_filter->accf_destroy(so);
204 		}
205 		if (so->so_accf->so_accept_filter_str != NULL)
206 			FREE(so->so_accf->so_accept_filter_str, M_ACCF);
207 		FREE(so->so_accf, M_ACCF);
208 	}
209 #endif /* INET */
210 	crfree(so->so_cred);
211 	zfree(socket_zone, so);
212 }
213 
214 int
215 solisten(struct socket *so, int backlog, struct thread *td)
216 {
217 	int s, error;
218 
219 	s = splnet();
220  	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
221  		splx(s);
222  		return (EINVAL);
223  	}
224 
225 	error = so_pru_listen(so, td);
226 	if (error) {
227 		splx(s);
228 		return (error);
229 	}
230 	if (TAILQ_EMPTY(&so->so_comp))
231 		so->so_options |= SO_ACCEPTCONN;
232 	if (backlog < 0 || backlog > somaxconn)
233 		backlog = somaxconn;
234 	so->so_qlimit = backlog;
235 	splx(s);
236 	return (0);
237 }
238 
239 void
240 sofree(struct socket *so)
241 {
242 	struct socket *head = so->so_head;
243 
244 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
245 		return;
246 	if (head != NULL) {
247 		if (so->so_state & SS_INCOMP) {
248 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
249 			head->so_incqlen--;
250 		} else if (so->so_state & SS_COMP) {
251 			/*
252 			 * We must not decommission a socket that's
253 			 * on the accept(2) queue.  If we do, then
254 			 * accept(2) may hang after select(2) indicated
255 			 * that the listening socket was ready.
256 			 */
257 			return;
258 		} else {
259 			panic("sofree: not queued");
260 		}
261 		so->so_state &= ~SS_INCOMP;
262 		so->so_head = NULL;
263 	}
264 	sbrelease(&so->so_snd, so);
265 	sorflush(so);
266 	sodealloc(so);
267 }
268 
269 /*
270  * Close a socket on last file table reference removal.
271  * Initiate disconnect if connected.
272  * Free socket when disconnect complete.
273  */
274 int
275 soclose(struct socket *so)
276 {
277 	int s = splnet();		/* conservative */
278 	int error = 0;
279 
280 	funsetown(so->so_sigio);
281 	if (so->so_options & SO_ACCEPTCONN) {
282 		struct socket *sp, *sonext;
283 
284 		sp = TAILQ_FIRST(&so->so_incomp);
285 		for (; sp != NULL; sp = sonext) {
286 			sonext = TAILQ_NEXT(sp, so_list);
287 			(void) soabort(sp);
288 		}
289 		for (sp = TAILQ_FIRST(&so->so_comp); sp != NULL; sp = sonext) {
290 			sonext = TAILQ_NEXT(sp, so_list);
291 			/* Dequeue from so_comp since sofree() won't do it */
292 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
293 			so->so_qlen--;
294 			sp->so_state &= ~SS_COMP;
295 			sp->so_head = NULL;
296 			(void) soabort(sp);
297 		}
298 	}
299 	if (so->so_pcb == 0)
300 		goto discard;
301 	if (so->so_state & SS_ISCONNECTED) {
302 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
303 			error = sodisconnect(so);
304 			if (error)
305 				goto drop;
306 		}
307 		if (so->so_options & SO_LINGER) {
308 			if ((so->so_state & SS_ISDISCONNECTING) &&
309 			    (so->so_state & SS_NBIO))
310 				goto drop;
311 			while (so->so_state & SS_ISCONNECTED) {
312 				error = tsleep((caddr_t)&so->so_timeo,
313 				    PCATCH, "soclos", so->so_linger * hz);
314 				if (error)
315 					break;
316 			}
317 		}
318 	}
319 drop:
320 	if (so->so_pcb) {
321 		int error2;
322 
323 		error2 = so_pru_detach(so);
324 		if (error == 0)
325 			error = error2;
326 	}
327 discard:
328 	if (so->so_state & SS_NOFDREF)
329 		panic("soclose: NOFDREF");
330 	so->so_state |= SS_NOFDREF;
331 	sofree(so);
332 	splx(s);
333 	return (error);
334 }
335 
336 /*
337  * Must be called at splnet...
338  */
339 int
340 soabort(so)
341 	struct socket *so;
342 {
343 	int error;
344 
345 	error = so_pru_abort(so);
346 	if (error) {
347 		sofree(so);
348 		return error;
349 	}
350 	return (0);
351 }
352 
353 int
354 soaccept(struct socket *so, struct sockaddr **nam)
355 {
356 	int s = splnet();
357 	int error;
358 
359 	if ((so->so_state & SS_NOFDREF) == 0)
360 		panic("soaccept: !NOFDREF");
361 	so->so_state &= ~SS_NOFDREF;
362 	error = so_pru_accept(so, nam);
363 	splx(s);
364 	return (error);
365 }
366 
367 int
368 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
369 {
370 	int s;
371 	int error;
372 
373 	if (so->so_options & SO_ACCEPTCONN)
374 		return (EOPNOTSUPP);
375 	s = splnet();
376 	/*
377 	 * If protocol is connection-based, can only connect once.
378 	 * Otherwise, if connected, try to disconnect first.
379 	 * This allows user to disconnect by connecting to, e.g.,
380 	 * a null address.
381 	 */
382 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
383 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
384 	    (error = sodisconnect(so))))
385 		error = EISCONN;
386 	else
387 		error = so_pru_connect(so, nam, td);
388 	splx(s);
389 	return (error);
390 }
391 
392 int
393 soconnect2(struct socket *so1, struct socket *so2)
394 {
395 	int s = splnet();
396 	int error;
397 
398 	error = so_pru_connect2(so1, so2);
399 	splx(s);
400 	return (error);
401 }
402 
403 int
404 sodisconnect(struct socket *so)
405 {
406 	int s = splnet();
407 	int error;
408 
409 	if ((so->so_state & SS_ISCONNECTED) == 0) {
410 		error = ENOTCONN;
411 		goto bad;
412 	}
413 	if (so->so_state & SS_ISDISCONNECTING) {
414 		error = EALREADY;
415 		goto bad;
416 	}
417 	error = so_pru_disconnect(so);
418 bad:
419 	splx(s);
420 	return (error);
421 }
422 
423 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
424 /*
425  * Send on a socket.
426  * If send must go all at once and message is larger than
427  * send buffering, then hard error.
428  * Lock against other senders.
429  * If must go all at once and not enough room now, then
430  * inform user that this would block and do nothing.
431  * Otherwise, if nonblocking, send as much as possible.
432  * The data to be sent is described by "uio" if nonzero,
433  * otherwise by the mbuf chain "top" (which must be null
434  * if uio is not).  Data provided in mbuf chain must be small
435  * enough to send all at once.
436  *
437  * Returns nonzero on error, timeout or signal; callers
438  * must check for short counts if EINTR/ERESTART are returned.
439  * Data and control buffers are freed on return.
440  */
441 int
442 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
443 	struct mbuf *top, struct mbuf *control, int flags,
444 	struct thread *td)
445 {
446 	struct mbuf **mp;
447 	struct mbuf *m;
448 	long space, len, resid;
449 	int clen = 0, error, s, dontroute, mlen;
450 	int atomic = sosendallatonce(so) || top;
451 	int pru_flags;
452 
453 	if (uio)
454 		resid = uio->uio_resid;
455 	else
456 		resid = top->m_pkthdr.len;
457 	/*
458 	 * In theory resid should be unsigned.
459 	 * However, space must be signed, as it might be less than 0
460 	 * if we over-committed, and we must use a signed comparison
461 	 * of space and resid.  On the other hand, a negative resid
462 	 * causes us to loop sending 0-length segments to the protocol.
463 	 *
464 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
465 	 * type sockets since that's an error.
466 	 */
467 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
468 		error = EINVAL;
469 		goto out;
470 	}
471 
472 	dontroute =
473 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
474 	    (so->so_proto->pr_flags & PR_ATOMIC);
475 	if (td->td_proc && td->td_proc->p_stats)
476 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
477 	if (control)
478 		clen = control->m_len;
479 #define	gotoerr(errno)	{ error = errno; splx(s); goto release; }
480 
481 restart:
482 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
483 	if (error)
484 		goto out;
485 	do {
486 		s = splnet();
487 		if (so->so_state & SS_CANTSENDMORE)
488 			gotoerr(EPIPE);
489 		if (so->so_error) {
490 			error = so->so_error;
491 			so->so_error = 0;
492 			splx(s);
493 			goto release;
494 		}
495 		if ((so->so_state & SS_ISCONNECTED) == 0) {
496 			/*
497 			 * `sendto' and `sendmsg' is allowed on a connection-
498 			 * based socket if it supports implied connect.
499 			 * Return ENOTCONN if not connected and no address is
500 			 * supplied.
501 			 */
502 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
503 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
504 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
505 				    !(resid == 0 && clen != 0))
506 					gotoerr(ENOTCONN);
507 			} else if (addr == 0)
508 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
509 				   ENOTCONN : EDESTADDRREQ);
510 		}
511 		space = sbspace(&so->so_snd);
512 		if (flags & MSG_OOB)
513 			space += 1024;
514 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
515 		    clen > so->so_snd.sb_hiwat)
516 			gotoerr(EMSGSIZE);
517 		if (space < resid + clen && uio &&
518 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
519 			if (so->so_state & SS_NBIO)
520 				gotoerr(EWOULDBLOCK);
521 			sbunlock(&so->so_snd);
522 			error = sbwait(&so->so_snd);
523 			splx(s);
524 			if (error)
525 				goto out;
526 			goto restart;
527 		}
528 		splx(s);
529 		mp = &top;
530 		space -= clen;
531 		do {
532 		    if (uio == NULL) {
533 			/*
534 			 * Data is prepackaged in "top".
535 			 */
536 			resid = 0;
537 			if (flags & MSG_EOR)
538 				top->m_flags |= M_EOR;
539 		    } else do {
540 			if (top == 0) {
541 				MGETHDR(m, M_WAIT, MT_DATA);
542 				if (m == NULL) {
543 					error = ENOBUFS;
544 					goto release;
545 				}
546 				mlen = MHLEN;
547 				m->m_pkthdr.len = 0;
548 				m->m_pkthdr.rcvif = (struct ifnet *)0;
549 			} else {
550 				MGET(m, M_WAIT, MT_DATA);
551 				if (m == NULL) {
552 					error = ENOBUFS;
553 					goto release;
554 				}
555 				mlen = MLEN;
556 			}
557 			if (resid >= MINCLSIZE) {
558 				MCLGET(m, M_WAIT);
559 				if ((m->m_flags & M_EXT) == 0)
560 					goto nopages;
561 				mlen = MCLBYTES;
562 				len = min(min(mlen, resid), space);
563 			} else {
564 nopages:
565 				len = min(min(mlen, resid), space);
566 				/*
567 				 * For datagram protocols, leave room
568 				 * for protocol headers in first mbuf.
569 				 */
570 				if (atomic && top == 0 && len < mlen)
571 					MH_ALIGN(m, len);
572 			}
573 			space -= len;
574 			error = uiomove(mtod(m, caddr_t), (int)len, uio);
575 			resid = uio->uio_resid;
576 			m->m_len = len;
577 			*mp = m;
578 			top->m_pkthdr.len += len;
579 			if (error)
580 				goto release;
581 			mp = &m->m_next;
582 			if (resid <= 0) {
583 				if (flags & MSG_EOR)
584 					top->m_flags |= M_EOR;
585 				break;
586 			}
587 		    } while (space > 0 && atomic);
588 		    if (dontroute)
589 			    so->so_options |= SO_DONTROUTE;
590 		    if (flags & MSG_OOB) {
591 		    	    pru_flags = PRUS_OOB;
592 		    } else if ((flags & MSG_EOF) &&
593 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
594 		    	       (resid <= 0)) {
595 			    /*
596 			     * If the user set MSG_EOF, the protocol
597 			     * understands this flag and nothing left to
598 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
599 			     */
600 		    	    pru_flags = PRUS_EOF;
601 		    } else if (resid > 0 && space > 0) {
602 			    /* If there is more to send, set PRUS_MORETOCOME */
603 		    	    pru_flags = PRUS_MORETOCOME;
604 		    } else {
605 		    	    pru_flags = 0;
606 		    }
607 		    s = splnet();				/* XXX */
608 		    /*
609 		     * XXX all the SS_CANTSENDMORE checks previously
610 		     * done could be out of date.  We could have recieved
611 		     * a reset packet in an interrupt or maybe we slept
612 		     * while doing page faults in uiomove() etc. We could
613 		     * probably recheck again inside the splnet() protection
614 		     * here, but there are probably other places that this
615 		     * also happens.  We must rethink this.
616 		     */
617 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
618 		    splx(s);
619 		    if (dontroute)
620 			    so->so_options &= ~SO_DONTROUTE;
621 		    clen = 0;
622 		    control = 0;
623 		    top = 0;
624 		    mp = &top;
625 		    if (error)
626 			    goto release;
627 		} while (resid && space > 0);
628 	} while (resid);
629 
630 release:
631 	sbunlock(&so->so_snd);
632 out:
633 	if (top)
634 		m_freem(top);
635 	if (control)
636 		m_freem(control);
637 	return (error);
638 }
639 
640 /*
641  * A specialization of sosend() for UDP based on protocol-specific knowledge:
642  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
643  *	sosendallatonce() returns true,
644  *	the "atomic" variable is true,
645  *	and sosendudp() blocks until space is available for the entire send.
646  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
647  *	PR_IMPLOPCL flags set.
648  *   UDP has no out-of-band data.
649  *   UDP has no control data.
650  *   UDP does not support MSG_EOR.
651  */
652 int
653 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
654 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
655 {
656 	int resid, error, s;
657 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
658 
659 	if (td->td_proc && td->td_proc->p_stats)
660 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
661 	if (control)
662 		m_freem(control);
663 
664 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
665 	resid = uio ? uio->uio_resid : top->m_pkthdr.len;
666 
667 restart:
668 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
669 	if (error)
670 		goto out;
671 
672 	s = splnet();
673 	if (so->so_state & SS_CANTSENDMORE)
674 		gotoerr(EPIPE);
675 	if (so->so_error) {
676 		error = so->so_error;
677 		so->so_error = 0;
678 		splx(s);
679 		goto release;
680 	}
681 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
682 		gotoerr(EDESTADDRREQ);
683 	if (resid > so->so_snd.sb_hiwat)
684 		gotoerr(EMSGSIZE);
685 	if (uio && sbspace(&so->so_snd) < resid) {
686 		if (so->so_state & SS_NBIO)
687 			gotoerr(EWOULDBLOCK);
688 		sbunlock(&so->so_snd);
689 		error = sbwait(&so->so_snd);
690 		splx(s);
691 		if (error)
692 			goto out;
693 		goto restart;
694 	}
695 	splx(s);
696 
697 	if (uio) {
698 		top = m_uiomove(uio, M_WAIT, 0);
699 		if (top == NULL)
700 			goto release;
701 	}
702 
703 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
704 	if (dontroute)
705 		so->so_options |= SO_DONTROUTE;
706 
707 	error = so_pru_send(so, 0, top, addr, NULL, td);
708 	top = NULL;		/* sent or freed in lower layer */
709 
710 	if (dontroute)
711 		so->so_options &= ~SO_DONTROUTE;
712 
713 release:
714 	sbunlock(&so->so_snd);
715 out:
716 	if (top)
717 		m_freem(top);
718 	return (error);
719 }
720 
721 /*
722  * Implement receive operations on a socket.
723  * We depend on the way that records are added to the sockbuf
724  * by sbappend*.  In particular, each record (mbufs linked through m_next)
725  * must begin with an address if the protocol so specifies,
726  * followed by an optional mbuf or mbufs containing ancillary data,
727  * and then zero or more mbufs of data.
728  * In order to avoid blocking network interrupts for the entire time here,
729  * we splx() while doing the actual copy to user space.
730  * Although the sockbuf is locked, new data may still be appended,
731  * and thus we must maintain consistency of the sockbuf during that time.
732  *
733  * The caller may receive the data as a single mbuf chain by supplying
734  * an mbuf **mp0 for use in returning the chain.  The uio is then used
735  * only for the count in uio_resid.
736  */
737 int
738 soreceive(so, psa, uio, mp0, controlp, flagsp)
739 	struct socket *so;
740 	struct sockaddr **psa;
741 	struct uio *uio;
742 	struct mbuf **mp0;
743 	struct mbuf **controlp;
744 	int *flagsp;
745 {
746 	struct mbuf *m, **mp;
747 	int flags, len, error, s, offset;
748 	struct protosw *pr = so->so_proto;
749 	struct mbuf *nextrecord;
750 	int moff, type = 0;
751 	int orig_resid = uio->uio_resid;
752 
753 	mp = mp0;
754 	if (psa)
755 		*psa = 0;
756 	if (controlp)
757 		*controlp = 0;
758 	if (flagsp)
759 		flags = *flagsp &~ MSG_EOR;
760 	else
761 		flags = 0;
762 	if (flags & MSG_OOB) {
763 		m = m_get(M_WAIT, MT_DATA);
764 		if (m == NULL)
765 			return (ENOBUFS);
766 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
767 		if (error)
768 			goto bad;
769 		do {
770 			error = uiomove(mtod(m, caddr_t),
771 			    (int) min(uio->uio_resid, m->m_len), uio);
772 			m = m_free(m);
773 		} while (uio->uio_resid && error == 0 && m);
774 bad:
775 		if (m)
776 			m_freem(m);
777 		return (error);
778 	}
779 	if (mp)
780 		*mp = (struct mbuf *)0;
781 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
782 		so_pru_rcvd(so, 0);
783 
784 restart:
785 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
786 	if (error)
787 		return (error);
788 	s = splnet();
789 
790 	m = so->so_rcv.sb_mb;
791 	/*
792 	 * If we have less data than requested, block awaiting more
793 	 * (subject to any timeout) if:
794 	 *   1. the current count is less than the low water mark, or
795 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
796 	 *	receive operation at once if we block (resid <= hiwat).
797 	 *   3. MSG_DONTWAIT is not set
798 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
799 	 * we have to do the receive in sections, and thus risk returning
800 	 * a short count if a timeout or signal occurs after we start.
801 	 */
802 	if (m == 0 || (((flags & MSG_DONTWAIT) == 0 &&
803 	    so->so_rcv.sb_cc < uio->uio_resid) &&
804 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
805 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
806 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
807 		KASSERT(m != 0 || !so->so_rcv.sb_cc, ("receive 1"));
808 		if (so->so_error) {
809 			if (m)
810 				goto dontblock;
811 			error = so->so_error;
812 			if ((flags & MSG_PEEK) == 0)
813 				so->so_error = 0;
814 			goto release;
815 		}
816 		if (so->so_state & SS_CANTRCVMORE) {
817 			if (m)
818 				goto dontblock;
819 			else
820 				goto release;
821 		}
822 		for (; m; m = m->m_next)
823 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
824 				m = so->so_rcv.sb_mb;
825 				goto dontblock;
826 			}
827 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
828 		    (pr->pr_flags & PR_CONNREQUIRED)) {
829 			error = ENOTCONN;
830 			goto release;
831 		}
832 		if (uio->uio_resid == 0)
833 			goto release;
834 		if ((so->so_state & SS_NBIO) || (flags & MSG_DONTWAIT)) {
835 			error = EWOULDBLOCK;
836 			goto release;
837 		}
838 		sbunlock(&so->so_rcv);
839 		error = sbwait(&so->so_rcv);
840 		splx(s);
841 		if (error)
842 			return (error);
843 		goto restart;
844 	}
845 dontblock:
846 	if (uio->uio_td && uio->uio_td->td_proc)
847 		uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
848 	nextrecord = m->m_nextpkt;
849 	if (pr->pr_flags & PR_ADDR) {
850 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
851 		orig_resid = 0;
852 		if (psa)
853 			*psa = dup_sockaddr(mtod(m, struct sockaddr *),
854 					    mp0 == 0);
855 		if (flags & MSG_PEEK) {
856 			m = m->m_next;
857 		} else {
858 			sbfree(&so->so_rcv, m);
859 			so->so_rcv.sb_mb = m_free(m);
860 			m = so->so_rcv.sb_mb;
861 		}
862 	}
863 	while (m && m->m_type == MT_CONTROL && error == 0) {
864 		if (flags & MSG_PEEK) {
865 			if (controlp)
866 				*controlp = m_copy(m, 0, m->m_len);
867 			m = m->m_next;
868 		} else {
869 			sbfree(&so->so_rcv, m);
870 			if (controlp) {
871 				if (pr->pr_domain->dom_externalize &&
872 				    mtod(m, struct cmsghdr *)->cmsg_type ==
873 				    SCM_RIGHTS)
874 				   error = (*pr->pr_domain->dom_externalize)(m);
875 				*controlp = m;
876 				so->so_rcv.sb_mb = m->m_next;
877 				m->m_next = 0;
878 				m = so->so_rcv.sb_mb;
879 			} else {
880 				so->so_rcv.sb_mb = m_free(m);
881 				m = so->so_rcv.sb_mb;
882 			}
883 		}
884 		if (controlp) {
885 			orig_resid = 0;
886 			controlp = &(*controlp)->m_next;
887 		}
888 	}
889 	if (m) {
890 		if ((flags & MSG_PEEK) == 0)
891 			m->m_nextpkt = nextrecord;
892 		type = m->m_type;
893 		if (type == MT_OOBDATA)
894 			flags |= MSG_OOB;
895 	}
896 	moff = 0;
897 	offset = 0;
898 	while (m && uio->uio_resid > 0 && error == 0) {
899 		if (m->m_type == MT_OOBDATA) {
900 			if (type != MT_OOBDATA)
901 				break;
902 		} else if (type == MT_OOBDATA)
903 			break;
904 		else
905 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
906 			("receive 3"));
907 		so->so_state &= ~SS_RCVATMARK;
908 		len = uio->uio_resid;
909 		if (so->so_oobmark && len > so->so_oobmark - offset)
910 			len = so->so_oobmark - offset;
911 		if (len > m->m_len - moff)
912 			len = m->m_len - moff;
913 		/*
914 		 * If mp is set, just pass back the mbufs.
915 		 * Otherwise copy them out via the uio, then free.
916 		 * Sockbuf must be consistent here (points to current mbuf,
917 		 * it points to next record) when we drop priority;
918 		 * we must note any additions to the sockbuf when we
919 		 * block interrupts again.
920 		 */
921 		if (mp == 0) {
922 			splx(s);
923 			error = uiomove(mtod(m, caddr_t) + moff, (int)len, uio);
924 			s = splnet();
925 			if (error)
926 				goto release;
927 		} else
928 			uio->uio_resid -= len;
929 		if (len == m->m_len - moff) {
930 			if (m->m_flags & M_EOR)
931 				flags |= MSG_EOR;
932 			if (flags & MSG_PEEK) {
933 				m = m->m_next;
934 				moff = 0;
935 			} else {
936 				nextrecord = m->m_nextpkt;
937 				sbfree(&so->so_rcv, m);
938 				if (mp) {
939 					*mp = m;
940 					mp = &m->m_next;
941 					so->so_rcv.sb_mb = m = m->m_next;
942 					*mp = (struct mbuf *)0;
943 				} else {
944 					so->so_rcv.sb_mb = m = m_free(m);
945 				}
946 				if (m)
947 					m->m_nextpkt = nextrecord;
948 			}
949 		} else {
950 			if (flags & MSG_PEEK)
951 				moff += len;
952 			else {
953 				if (mp)
954 					*mp = m_copym(m, 0, len, M_WAIT);
955 				m->m_data += len;
956 				m->m_len -= len;
957 				so->so_rcv.sb_cc -= len;
958 			}
959 		}
960 		if (so->so_oobmark) {
961 			if ((flags & MSG_PEEK) == 0) {
962 				so->so_oobmark -= len;
963 				if (so->so_oobmark == 0) {
964 					so->so_state |= SS_RCVATMARK;
965 					break;
966 				}
967 			} else {
968 				offset += len;
969 				if (offset == so->so_oobmark)
970 					break;
971 			}
972 		}
973 		if (flags & MSG_EOR)
974 			break;
975 		/*
976 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
977 		 * we must not quit until "uio->uio_resid == 0" or an error
978 		 * termination.  If a signal/timeout occurs, return
979 		 * with a short count but without error.
980 		 * Keep sockbuf locked against other readers.
981 		 */
982 		while (flags & MSG_WAITALL && m == 0 && uio->uio_resid > 0 &&
983 		    !sosendallatonce(so) && !nextrecord) {
984 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
985 				break;
986 			/*
987 			 * The window might have closed to zero, make
988 			 * sure we send an ack now that we've drained
989 			 * the buffer or we might end up blocking until
990 			 * the idle takes over (5 seconds).
991 			 */
992 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
993 				so_pru_rcvd(so, flags);
994 			error = sbwait(&so->so_rcv);
995 			if (error) {
996 				sbunlock(&so->so_rcv);
997 				splx(s);
998 				return (0);
999 			}
1000 			m = so->so_rcv.sb_mb;
1001 			if (m)
1002 				nextrecord = m->m_nextpkt;
1003 		}
1004 	}
1005 
1006 	if (m && pr->pr_flags & PR_ATOMIC) {
1007 		flags |= MSG_TRUNC;
1008 		if ((flags & MSG_PEEK) == 0)
1009 			(void) sbdroprecord(&so->so_rcv);
1010 	}
1011 	if ((flags & MSG_PEEK) == 0) {
1012 		if (m == 0)
1013 			so->so_rcv.sb_mb = nextrecord;
1014 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1015 			so_pru_rcvd(so, flags);
1016 	}
1017 	if (orig_resid == uio->uio_resid && orig_resid &&
1018 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1019 		sbunlock(&so->so_rcv);
1020 		splx(s);
1021 		goto restart;
1022 	}
1023 
1024 	if (flagsp)
1025 		*flagsp |= flags;
1026 release:
1027 	sbunlock(&so->so_rcv);
1028 	splx(s);
1029 	return (error);
1030 }
1031 
1032 int
1033 soshutdown(so, how)
1034 	struct socket *so;
1035 	int how;
1036 {
1037 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1038 		return (EINVAL);
1039 
1040 	if (how != SHUT_WR)
1041 		sorflush(so);
1042 	if (how != SHUT_RD)
1043 		return (so_pru_shutdown(so));
1044 	return (0);
1045 }
1046 
1047 void
1048 sorflush(so)
1049 	struct socket *so;
1050 {
1051 	struct sockbuf *sb = &so->so_rcv;
1052 	struct protosw *pr = so->so_proto;
1053 	int s;
1054 	struct sockbuf asb;
1055 
1056 	sb->sb_flags |= SB_NOINTR;
1057 	(void) sblock(sb, M_WAITOK);
1058 	s = splimp();
1059 	socantrcvmore(so);
1060 	sbunlock(sb);
1061 	asb = *sb;
1062 	bzero((caddr_t)sb, sizeof (*sb));
1063 	if (asb.sb_flags & SB_KNOTE) {
1064 		sb->sb_sel.si_note = asb.sb_sel.si_note;
1065 		sb->sb_flags = SB_KNOTE;
1066 	}
1067 	splx(s);
1068 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1069 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1070 	sbrelease(&asb, so);
1071 }
1072 
1073 #ifdef INET
1074 static int
1075 do_setopt_accept_filter(so, sopt)
1076 	struct	socket *so;
1077 	struct	sockopt *sopt;
1078 {
1079 	struct accept_filter_arg	*afap = NULL;
1080 	struct accept_filter	*afp;
1081 	struct so_accf	*af = so->so_accf;
1082 	int	error = 0;
1083 
1084 	/* do not set/remove accept filters on non listen sockets */
1085 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1086 		error = EINVAL;
1087 		goto out;
1088 	}
1089 
1090 	/* removing the filter */
1091 	if (sopt == NULL) {
1092 		if (af != NULL) {
1093 			if (af->so_accept_filter != NULL &&
1094 				af->so_accept_filter->accf_destroy != NULL) {
1095 				af->so_accept_filter->accf_destroy(so);
1096 			}
1097 			if (af->so_accept_filter_str != NULL) {
1098 				FREE(af->so_accept_filter_str, M_ACCF);
1099 			}
1100 			FREE(af, M_ACCF);
1101 			so->so_accf = NULL;
1102 		}
1103 		so->so_options &= ~SO_ACCEPTFILTER;
1104 		return (0);
1105 	}
1106 	/* adding a filter */
1107 	/* must remove previous filter first */
1108 	if (af != NULL) {
1109 		error = EINVAL;
1110 		goto out;
1111 	}
1112 	/* don't put large objects on the kernel stack */
1113 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1114 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1115 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1116 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1117 	if (error)
1118 		goto out;
1119 	afp = accept_filt_get(afap->af_name);
1120 	if (afp == NULL) {
1121 		error = ENOENT;
1122 		goto out;
1123 	}
1124 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK);
1125 	bzero(af, sizeof(*af));
1126 	if (afp->accf_create != NULL) {
1127 		if (afap->af_name[0] != '\0') {
1128 			int len = strlen(afap->af_name) + 1;
1129 
1130 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1131 			strcpy(af->so_accept_filter_str, afap->af_name);
1132 		}
1133 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1134 		if (af->so_accept_filter_arg == NULL) {
1135 			FREE(af->so_accept_filter_str, M_ACCF);
1136 			FREE(af, M_ACCF);
1137 			so->so_accf = NULL;
1138 			error = EINVAL;
1139 			goto out;
1140 		}
1141 	}
1142 	af->so_accept_filter = afp;
1143 	so->so_accf = af;
1144 	so->so_options |= SO_ACCEPTFILTER;
1145 out:
1146 	if (afap != NULL)
1147 		FREE(afap, M_TEMP);
1148 	return (error);
1149 }
1150 #endif /* INET */
1151 
1152 /*
1153  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1154  * an additional variant to handle the case where the option value needs
1155  * to be some kind of integer, but not a specific size.
1156  * In addition to their use here, these functions are also called by the
1157  * protocol-level pr_ctloutput() routines.
1158  */
1159 int
1160 sooptcopyin(sopt, buf, len, minlen)
1161 	struct	sockopt *sopt;
1162 	void	*buf;
1163 	size_t	len;
1164 	size_t	minlen;
1165 {
1166 	size_t	valsize;
1167 
1168 	/*
1169 	 * If the user gives us more than we wanted, we ignore it,
1170 	 * but if we don't get the minimum length the caller
1171 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1172 	 * is set to however much we actually retrieved.
1173 	 */
1174 	if ((valsize = sopt->sopt_valsize) < minlen)
1175 		return EINVAL;
1176 	if (valsize > len)
1177 		sopt->sopt_valsize = valsize = len;
1178 
1179 	if (sopt->sopt_td != NULL)
1180 		return (copyin(sopt->sopt_val, buf, valsize));
1181 
1182 	bcopy(sopt->sopt_val, buf, valsize);
1183 	return 0;
1184 }
1185 
1186 int
1187 sosetopt(so, sopt)
1188 	struct socket *so;
1189 	struct sockopt *sopt;
1190 {
1191 	int	error, optval;
1192 	struct	linger l;
1193 	struct	timeval tv;
1194 	u_long  val;
1195 
1196 	error = 0;
1197 	if (sopt->sopt_level != SOL_SOCKET) {
1198 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1199 			return (so_pr_ctloutput(so, sopt));
1200 		}
1201 		error = ENOPROTOOPT;
1202 	} else {
1203 		switch (sopt->sopt_name) {
1204 #ifdef INET
1205 		case SO_ACCEPTFILTER:
1206 			error = do_setopt_accept_filter(so, sopt);
1207 			if (error)
1208 				goto bad;
1209 			break;
1210 #endif /* INET */
1211 		case SO_LINGER:
1212 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1213 			if (error)
1214 				goto bad;
1215 
1216 			so->so_linger = l.l_linger;
1217 			if (l.l_onoff)
1218 				so->so_options |= SO_LINGER;
1219 			else
1220 				so->so_options &= ~SO_LINGER;
1221 			break;
1222 
1223 		case SO_DEBUG:
1224 		case SO_KEEPALIVE:
1225 		case SO_DONTROUTE:
1226 		case SO_USELOOPBACK:
1227 		case SO_BROADCAST:
1228 		case SO_REUSEADDR:
1229 		case SO_REUSEPORT:
1230 		case SO_OOBINLINE:
1231 		case SO_TIMESTAMP:
1232 			error = sooptcopyin(sopt, &optval, sizeof optval,
1233 					    sizeof optval);
1234 			if (error)
1235 				goto bad;
1236 			if (optval)
1237 				so->so_options |= sopt->sopt_name;
1238 			else
1239 				so->so_options &= ~sopt->sopt_name;
1240 			break;
1241 
1242 		case SO_SNDBUF:
1243 		case SO_RCVBUF:
1244 		case SO_SNDLOWAT:
1245 		case SO_RCVLOWAT:
1246 			error = sooptcopyin(sopt, &optval, sizeof optval,
1247 					    sizeof optval);
1248 			if (error)
1249 				goto bad;
1250 
1251 			/*
1252 			 * Values < 1 make no sense for any of these
1253 			 * options, so disallow them.
1254 			 */
1255 			if (optval < 1) {
1256 				error = EINVAL;
1257 				goto bad;
1258 			}
1259 
1260 			switch (sopt->sopt_name) {
1261 			case SO_SNDBUF:
1262 			case SO_RCVBUF:
1263 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1264 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1265 				    so,
1266 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1267 					error = ENOBUFS;
1268 					goto bad;
1269 				}
1270 				break;
1271 
1272 			/*
1273 			 * Make sure the low-water is never greater than
1274 			 * the high-water.
1275 			 */
1276 			case SO_SNDLOWAT:
1277 				so->so_snd.sb_lowat =
1278 				    (optval > so->so_snd.sb_hiwat) ?
1279 				    so->so_snd.sb_hiwat : optval;
1280 				break;
1281 			case SO_RCVLOWAT:
1282 				so->so_rcv.sb_lowat =
1283 				    (optval > so->so_rcv.sb_hiwat) ?
1284 				    so->so_rcv.sb_hiwat : optval;
1285 				break;
1286 			}
1287 			break;
1288 
1289 		case SO_SNDTIMEO:
1290 		case SO_RCVTIMEO:
1291 			error = sooptcopyin(sopt, &tv, sizeof tv,
1292 					    sizeof tv);
1293 			if (error)
1294 				goto bad;
1295 
1296 			/* assert(hz > 0); */
1297 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1298 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1299 				error = EDOM;
1300 				goto bad;
1301 			}
1302 			/* assert(tick > 0); */
1303 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1304 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1305 			if (val > SHRT_MAX) {
1306 				error = EDOM;
1307 				goto bad;
1308 			}
1309 			if (val == 0 && tv.tv_usec != 0)
1310 				val = 1;
1311 
1312 			switch (sopt->sopt_name) {
1313 			case SO_SNDTIMEO:
1314 				so->so_snd.sb_timeo = val;
1315 				break;
1316 			case SO_RCVTIMEO:
1317 				so->so_rcv.sb_timeo = val;
1318 				break;
1319 			}
1320 			break;
1321 		default:
1322 			error = ENOPROTOOPT;
1323 			break;
1324 		}
1325 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1326 			(void) so_pr_ctloutput(so, sopt);
1327 		}
1328 	}
1329 bad:
1330 	return (error);
1331 }
1332 
1333 /* Helper routine for getsockopt */
1334 int
1335 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1336 {
1337 	int	error;
1338 	size_t	valsize;
1339 
1340 	error = 0;
1341 
1342 	/*
1343 	 * Documented get behavior is that we always return a value,
1344 	 * possibly truncated to fit in the user's buffer.
1345 	 * Traditional behavior is that we always tell the user
1346 	 * precisely how much we copied, rather than something useful
1347 	 * like the total amount we had available for her.
1348 	 * Note that this interface is not idempotent; the entire answer must
1349 	 * generated ahead of time.
1350 	 */
1351 	valsize = min(len, sopt->sopt_valsize);
1352 	sopt->sopt_valsize = valsize;
1353 	if (sopt->sopt_val != 0) {
1354 		if (sopt->sopt_td != NULL)
1355 			error = copyout(buf, sopt->sopt_val, valsize);
1356 		else
1357 			bcopy(buf, sopt->sopt_val, valsize);
1358 	}
1359 	return error;
1360 }
1361 
1362 int
1363 sogetopt(so, sopt)
1364 	struct socket *so;
1365 	struct sockopt *sopt;
1366 {
1367 	int	error, optval;
1368 	struct	linger l;
1369 	struct	timeval tv;
1370 	struct accept_filter_arg *afap;
1371 
1372 	error = 0;
1373 	if (sopt->sopt_level != SOL_SOCKET) {
1374 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1375 			return (so_pr_ctloutput(so, sopt));
1376 		} else
1377 			return (ENOPROTOOPT);
1378 	} else {
1379 		switch (sopt->sopt_name) {
1380 #ifdef INET
1381 		case SO_ACCEPTFILTER:
1382 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1383 				return (EINVAL);
1384 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1385 				M_TEMP, M_WAITOK);
1386 			bzero(afap, sizeof(*afap));
1387 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1388 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1389 				if (so->so_accf->so_accept_filter_str != NULL)
1390 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1391 			}
1392 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1393 			FREE(afap, M_TEMP);
1394 			break;
1395 #endif /* INET */
1396 
1397 		case SO_LINGER:
1398 			l.l_onoff = so->so_options & SO_LINGER;
1399 			l.l_linger = so->so_linger;
1400 			error = sooptcopyout(sopt, &l, sizeof l);
1401 			break;
1402 
1403 		case SO_USELOOPBACK:
1404 		case SO_DONTROUTE:
1405 		case SO_DEBUG:
1406 		case SO_KEEPALIVE:
1407 		case SO_REUSEADDR:
1408 		case SO_REUSEPORT:
1409 		case SO_BROADCAST:
1410 		case SO_OOBINLINE:
1411 		case SO_TIMESTAMP:
1412 			optval = so->so_options & sopt->sopt_name;
1413 integer:
1414 			error = sooptcopyout(sopt, &optval, sizeof optval);
1415 			break;
1416 
1417 		case SO_TYPE:
1418 			optval = so->so_type;
1419 			goto integer;
1420 
1421 		case SO_ERROR:
1422 			optval = so->so_error;
1423 			so->so_error = 0;
1424 			goto integer;
1425 
1426 		case SO_SNDBUF:
1427 			optval = so->so_snd.sb_hiwat;
1428 			goto integer;
1429 
1430 		case SO_RCVBUF:
1431 			optval = so->so_rcv.sb_hiwat;
1432 			goto integer;
1433 
1434 		case SO_SNDLOWAT:
1435 			optval = so->so_snd.sb_lowat;
1436 			goto integer;
1437 
1438 		case SO_RCVLOWAT:
1439 			optval = so->so_rcv.sb_lowat;
1440 			goto integer;
1441 
1442 		case SO_SNDTIMEO:
1443 		case SO_RCVTIMEO:
1444 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1445 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1446 
1447 			tv.tv_sec = optval / hz;
1448 			tv.tv_usec = (optval % hz) * tick;
1449 			error = sooptcopyout(sopt, &tv, sizeof tv);
1450 			break;
1451 
1452 		default:
1453 			error = ENOPROTOOPT;
1454 			break;
1455 		}
1456 		return (error);
1457 	}
1458 }
1459 
1460 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1461 int
1462 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1463 {
1464 	struct mbuf *m, *m_prev;
1465 	int sopt_size = sopt->sopt_valsize;
1466 
1467 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
1468 	if (m == 0)
1469 		return ENOBUFS;
1470 	if (sopt_size > MLEN) {
1471 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
1472 		if ((m->m_flags & M_EXT) == 0) {
1473 			m_free(m);
1474 			return ENOBUFS;
1475 		}
1476 		m->m_len = min(MCLBYTES, sopt_size);
1477 	} else {
1478 		m->m_len = min(MLEN, sopt_size);
1479 	}
1480 	sopt_size -= m->m_len;
1481 	*mp = m;
1482 	m_prev = m;
1483 
1484 	while (sopt_size) {
1485 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
1486 		if (m == 0) {
1487 			m_freem(*mp);
1488 			return ENOBUFS;
1489 		}
1490 		if (sopt_size > MLEN) {
1491 			MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
1492 			if ((m->m_flags & M_EXT) == 0) {
1493 				m_freem(*mp);
1494 				return ENOBUFS;
1495 			}
1496 			m->m_len = min(MCLBYTES, sopt_size);
1497 		} else {
1498 			m->m_len = min(MLEN, sopt_size);
1499 		}
1500 		sopt_size -= m->m_len;
1501 		m_prev->m_next = m;
1502 		m_prev = m;
1503 	}
1504 	return 0;
1505 }
1506 
1507 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1508 int
1509 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1510 {
1511 	struct mbuf *m0 = m;
1512 
1513 	if (sopt->sopt_val == NULL)
1514 		return 0;
1515 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1516 		if (sopt->sopt_td != NULL) {
1517 			int error;
1518 
1519 			error = copyin(sopt->sopt_val, mtod(m, char *),
1520 				       m->m_len);
1521 			if (error != 0) {
1522 				m_freem(m0);
1523 				return(error);
1524 			}
1525 		} else
1526 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1527 		sopt->sopt_valsize -= m->m_len;
1528 		(caddr_t)sopt->sopt_val += m->m_len;
1529 		m = m->m_next;
1530 	}
1531 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1532 		panic("ip6_sooptmcopyin");
1533 	return 0;
1534 }
1535 
1536 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1537 int
1538 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1539 {
1540 	struct mbuf *m0 = m;
1541 	size_t valsize = 0;
1542 
1543 	if (sopt->sopt_val == NULL)
1544 		return 0;
1545 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1546 		if (sopt->sopt_td != NULL) {
1547 			int error;
1548 
1549 			error = copyout(mtod(m, char *), sopt->sopt_val,
1550 				       m->m_len);
1551 			if (error != 0) {
1552 				m_freem(m0);
1553 				return(error);
1554 			}
1555 		} else
1556 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
1557 	       sopt->sopt_valsize -= m->m_len;
1558 	       (caddr_t)sopt->sopt_val += m->m_len;
1559 	       valsize += m->m_len;
1560 	       m = m->m_next;
1561 	}
1562 	if (m != NULL) {
1563 		/* enough soopt buffer should be given from user-land */
1564 		m_freem(m0);
1565 		return(EINVAL);
1566 	}
1567 	sopt->sopt_valsize = valsize;
1568 	return 0;
1569 }
1570 
1571 void
1572 sohasoutofband(so)
1573 	struct socket *so;
1574 {
1575 	if (so->so_sigio != NULL)
1576 		pgsigio(so->so_sigio, SIGURG, 0);
1577 	selwakeup(&so->so_rcv.sb_sel);
1578 }
1579 
1580 int
1581 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1582 {
1583 	int revents = 0;
1584 	int s = splnet();
1585 
1586 	if (events & (POLLIN | POLLRDNORM))
1587 		if (soreadable(so))
1588 			revents |= events & (POLLIN | POLLRDNORM);
1589 
1590 	if (events & POLLINIGNEOF)
1591 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
1592 			!TAILQ_EMPTY(&so->so_comp) || so->so_error)
1593 			revents |= POLLINIGNEOF;
1594 
1595 	if (events & (POLLOUT | POLLWRNORM))
1596 		if (sowriteable(so))
1597 			revents |= events & (POLLOUT | POLLWRNORM);
1598 
1599 	if (events & (POLLPRI | POLLRDBAND))
1600 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1601 			revents |= events & (POLLPRI | POLLRDBAND);
1602 
1603 	if (revents == 0) {
1604 		if (events &
1605 			(POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1606 			 POLLRDBAND)) {
1607 			selrecord(td, &so->so_rcv.sb_sel);
1608 			so->so_rcv.sb_flags |= SB_SEL;
1609 		}
1610 
1611 		if (events & (POLLOUT | POLLWRNORM)) {
1612 			selrecord(td, &so->so_snd.sb_sel);
1613 			so->so_snd.sb_flags |= SB_SEL;
1614 		}
1615 	}
1616 
1617 	splx(s);
1618 	return (revents);
1619 }
1620 
1621 int
1622 sokqfilter(struct file *fp, struct knote *kn)
1623 {
1624 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1625 	struct sockbuf *sb;
1626 	int s;
1627 
1628 	switch (kn->kn_filter) {
1629 	case EVFILT_READ:
1630 		if (so->so_options & SO_ACCEPTCONN)
1631 			kn->kn_fop = &solisten_filtops;
1632 		else
1633 			kn->kn_fop = &soread_filtops;
1634 		sb = &so->so_rcv;
1635 		break;
1636 	case EVFILT_WRITE:
1637 		kn->kn_fop = &sowrite_filtops;
1638 		sb = &so->so_snd;
1639 		break;
1640 	default:
1641 		return (1);
1642 	}
1643 
1644 	s = splnet();
1645 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
1646 	sb->sb_flags |= SB_KNOTE;
1647 	splx(s);
1648 	return (0);
1649 }
1650 
1651 static void
1652 filt_sordetach(struct knote *kn)
1653 {
1654 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1655 	int s = splnet();
1656 
1657 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
1658 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
1659 		so->so_rcv.sb_flags &= ~SB_KNOTE;
1660 	splx(s);
1661 }
1662 
1663 /*ARGSUSED*/
1664 static int
1665 filt_soread(struct knote *kn, long hint)
1666 {
1667 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1668 
1669 	kn->kn_data = so->so_rcv.sb_cc;
1670 	if (so->so_state & SS_CANTRCVMORE) {
1671 		kn->kn_flags |= EV_EOF;
1672 		kn->kn_fflags = so->so_error;
1673 		return (1);
1674 	}
1675 	if (so->so_error)	/* temporary udp error */
1676 		return (1);
1677 	if (kn->kn_sfflags & NOTE_LOWAT)
1678 		return (kn->kn_data >= kn->kn_sdata);
1679 	return (kn->kn_data >= so->so_rcv.sb_lowat);
1680 }
1681 
1682 static void
1683 filt_sowdetach(struct knote *kn)
1684 {
1685 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1686 	int s = splnet();
1687 
1688 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
1689 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
1690 		so->so_snd.sb_flags &= ~SB_KNOTE;
1691 	splx(s);
1692 }
1693 
1694 /*ARGSUSED*/
1695 static int
1696 filt_sowrite(struct knote *kn, long hint)
1697 {
1698 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1699 
1700 	kn->kn_data = sbspace(&so->so_snd);
1701 	if (so->so_state & SS_CANTSENDMORE) {
1702 		kn->kn_flags |= EV_EOF;
1703 		kn->kn_fflags = so->so_error;
1704 		return (1);
1705 	}
1706 	if (so->so_error)	/* temporary udp error */
1707 		return (1);
1708 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1709 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1710 		return (0);
1711 	if (kn->kn_sfflags & NOTE_LOWAT)
1712 		return (kn->kn_data >= kn->kn_sdata);
1713 	return (kn->kn_data >= so->so_snd.sb_lowat);
1714 }
1715 
1716 /*ARGSUSED*/
1717 static int
1718 filt_solisten(struct knote *kn, long hint)
1719 {
1720 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1721 
1722 	kn->kn_data = so->so_qlen;
1723 	return (! TAILQ_EMPTY(&so->so_comp));
1724 }
1725