xref: /dflybsd-src/sys/kern/uipc_socket.c (revision cd29885abfb8f68adb0c082e313b891156d66964)
1 /*
2  * Copyright (c) 2004 Jeffrey M. Hsu.  All rights reserved.
3  * Copyright (c) 2004 The DragonFly Project.  All rights reserved.
4  *
5  * This code is derived from software contributed to The DragonFly Project
6  * by Jeffrey M. Hsu.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of The DragonFly Project nor the names of its
17  *    contributors may be used to endorse or promote products derived
18  *    from this software without specific, prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
23  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
24  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
25  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
26  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
27  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
28  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
29  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
30  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * Copyright (c) 1982, 1986, 1988, 1990, 1993
36  *	The Regents of the University of California.  All rights reserved.
37  *
38  * Redistribution and use in source and binary forms, with or without
39  * modification, are permitted provided that the following conditions
40  * are met:
41  * 1. Redistributions of source code must retain the above copyright
42  *    notice, this list of conditions and the following disclaimer.
43  * 2. Redistributions in binary form must reproduce the above copyright
44  *    notice, this list of conditions and the following disclaimer in the
45  *    documentation and/or other materials provided with the distribution.
46  * 3. All advertising materials mentioning features or use of this software
47  *    must display the following acknowledgement:
48  *	This product includes software developed by the University of
49  *	California, Berkeley and its contributors.
50  * 4. Neither the name of the University nor the names of its contributors
51  *    may be used to endorse or promote products derived from this software
52  *    without specific prior written permission.
53  *
54  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
55  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
56  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
57  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
58  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
59  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
60  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
61  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
62  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
63  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
64  * SUCH DAMAGE.
65  *
66  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
67  * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
68  * $DragonFly: src/sys/kern/uipc_socket.c,v 1.55 2008/09/02 16:17:52 dillon Exp $
69  */
70 
71 #include "opt_inet.h"
72 #include "opt_sctp.h"
73 
74 #include <sys/param.h>
75 #include <sys/systm.h>
76 #include <sys/fcntl.h>
77 #include <sys/malloc.h>
78 #include <sys/mbuf.h>
79 #include <sys/domain.h>
80 #include <sys/file.h>			/* for struct knote */
81 #include <sys/kernel.h>
82 #include <sys/malloc.h>
83 #include <sys/event.h>
84 #include <sys/poll.h>
85 #include <sys/proc.h>
86 #include <sys/protosw.h>
87 #include <sys/socket.h>
88 #include <sys/socketvar.h>
89 #include <sys/socketops.h>
90 #include <sys/resourcevar.h>
91 #include <sys/signalvar.h>
92 #include <sys/sysctl.h>
93 #include <sys/uio.h>
94 #include <sys/jail.h>
95 #include <vm/vm_zone.h>
96 #include <vm/pmap.h>
97 
98 #include <sys/thread2.h>
99 #include <sys/socketvar2.h>
100 
101 #include <machine/limits.h>
102 
103 #ifdef INET
104 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
105 #endif /* INET */
106 
107 static void 	filt_sordetach(struct knote *kn);
108 static int 	filt_soread(struct knote *kn, long hint);
109 static void 	filt_sowdetach(struct knote *kn);
110 static int	filt_sowrite(struct knote *kn, long hint);
111 static int	filt_solisten(struct knote *kn, long hint);
112 
113 static struct filterops solisten_filtops =
114 	{ 1, NULL, filt_sordetach, filt_solisten };
115 static struct filterops soread_filtops =
116 	{ 1, NULL, filt_sordetach, filt_soread };
117 static struct filterops sowrite_filtops =
118 	{ 1, NULL, filt_sowdetach, filt_sowrite };
119 
120 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
121 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
122 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
123 
124 
125 static int somaxconn = SOMAXCONN;
126 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
127     &somaxconn, 0, "Maximum pending socket connection queue size");
128 
129 /*
130  * Socket operation routines.
131  * These routines are called by the routines in
132  * sys_socket.c or from a system process, and
133  * implement the semantics of socket operations by
134  * switching out to the protocol specific routines.
135  */
136 
137 /*
138  * Get a socket structure, and initialize it.
139  * Note that it would probably be better to allocate socket
140  * and PCB at the same time, but I'm not convinced that all
141  * the protocols can be easily modified to do this.
142  */
143 struct socket *
144 soalloc(int waitok)
145 {
146 	struct socket *so;
147 	unsigned waitmask;
148 
149 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
150 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
151 	if (so) {
152 		/* XXX race condition for reentrant kernel */
153 		TAILQ_INIT(&so->so_aiojobq);
154 		TAILQ_INIT(&so->so_rcv.ssb_sel.si_mlist);
155 		TAILQ_INIT(&so->so_snd.ssb_sel.si_mlist);
156 	}
157 	return so;
158 }
159 
160 int
161 socreate(int dom, struct socket **aso, int type,
162 	int proto, struct thread *td)
163 {
164 	struct proc *p = td->td_proc;
165 	struct protosw *prp;
166 	struct socket *so;
167 	struct pru_attach_info ai;
168 	int error;
169 
170 	if (proto)
171 		prp = pffindproto(dom, proto, type);
172 	else
173 		prp = pffindtype(dom, type);
174 
175 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
176 		return (EPROTONOSUPPORT);
177 
178 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
179 	    prp->pr_domain->dom_family != PF_LOCAL &&
180 	    prp->pr_domain->dom_family != PF_INET &&
181 	    prp->pr_domain->dom_family != PF_INET6 &&
182 	    prp->pr_domain->dom_family != PF_ROUTE) {
183 		return (EPROTONOSUPPORT);
184 	}
185 
186 	if (prp->pr_type != type)
187 		return (EPROTOTYPE);
188 	so = soalloc(p != 0);
189 	if (so == 0)
190 		return (ENOBUFS);
191 
192 	TAILQ_INIT(&so->so_incomp);
193 	TAILQ_INIT(&so->so_comp);
194 	so->so_type = type;
195 	so->so_cred = crhold(p->p_ucred);
196 	so->so_proto = prp;
197 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
198 	ai.p_ucred = p->p_ucred;
199 	ai.fd_rdir = p->p_fd->fd_rdir;
200 	/*
201 	 * Auto-sizing of socket buffers is managed by the protocols and
202 	 * the appropriate flags must be set in the pru_attach function.
203 	 */
204 	error = so_pru_attach(so, proto, &ai);
205 	if (error) {
206 		so->so_state |= SS_NOFDREF;
207 		sofree(so);
208 		return (error);
209 	}
210 	*aso = so;
211 	return (0);
212 }
213 
214 int
215 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
216 {
217 	int error;
218 
219 	crit_enter();
220 	error = so_pru_bind(so, nam, td);
221 	crit_exit();
222 	return (error);
223 }
224 
225 void
226 sodealloc(struct socket *so)
227 {
228 	if (so->so_rcv.ssb_hiwat)
229 		(void)chgsbsize(so->so_cred->cr_uidinfo,
230 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
231 	if (so->so_snd.ssb_hiwat)
232 		(void)chgsbsize(so->so_cred->cr_uidinfo,
233 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
234 #ifdef INET
235 	/* remove accept filter if present */
236 	if (so->so_accf != NULL)
237 		do_setopt_accept_filter(so, NULL);
238 #endif /* INET */
239 	crfree(so->so_cred);
240 	kfree(so, M_SOCKET);
241 }
242 
243 int
244 solisten(struct socket *so, int backlog, struct thread *td)
245 {
246 	int error;
247 #ifdef SCTP
248 	short oldopt, oldqlimit;
249 #endif /* SCTP */
250 
251 	crit_enter();
252 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
253 		crit_exit();
254 		return (EINVAL);
255 	}
256 
257 #ifdef SCTP
258 	oldopt = so->so_options;
259 	oldqlimit = so->so_qlimit;
260 #endif /* SCTP */
261 
262 	if (TAILQ_EMPTY(&so->so_comp))
263 		so->so_options |= SO_ACCEPTCONN;
264 	if (backlog < 0 || backlog > somaxconn)
265 		backlog = somaxconn;
266 	so->so_qlimit = backlog;
267 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
268 	 * the so_options (UDP model both connect's and gets inbound
269 	 * connections .. implicitly).
270 	 */
271 	error = so_pru_listen(so, td);
272 	if (error) {
273 #ifdef SCTP
274 		/* Restore the params */
275 		so->so_options = oldopt;
276 		so->so_qlimit = oldqlimit;
277 #endif /* SCTP */
278 		crit_exit();
279 		return (error);
280 	}
281 	crit_exit();
282 	return (0);
283 }
284 
285 /*
286  * Destroy a disconnected socket.  This routine is a NOP if entities
287  * still have a reference on the socket:
288  *
289  *	so_pcb -	The protocol stack still has a reference
290  *	SS_NOFDREF -	There is no longer a file pointer reference
291  *	SS_ABORTING -	An abort netmsg is in-flight
292  */
293 void
294 sofree(struct socket *so)
295 {
296 	struct socket *head = so->so_head;
297 
298 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
299 		return;
300 	if (so->so_state & SS_ABORTING)
301 		return;
302 	if (head != NULL) {
303 		if (so->so_state & SS_INCOMP) {
304 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
305 			head->so_incqlen--;
306 		} else if (so->so_state & SS_COMP) {
307 			/*
308 			 * We must not decommission a socket that's
309 			 * on the accept(2) queue.  If we do, then
310 			 * accept(2) may hang after select(2) indicated
311 			 * that the listening socket was ready.
312 			 */
313 			return;
314 		} else {
315 			panic("sofree: not queued");
316 		}
317 		so->so_state &= ~SS_INCOMP;
318 		so->so_head = NULL;
319 	}
320 	ssb_release(&so->so_snd, so);
321 	sorflush(so);
322 	sodealloc(so);
323 }
324 
325 /*
326  * Close a socket on last file table reference removal.
327  * Initiate disconnect if connected.
328  * Free socket when disconnect complete.
329  */
330 int
331 soclose(struct socket *so, int fflag)
332 {
333 	int error = 0;
334 
335 	crit_enter();
336 	funsetown(so->so_sigio);
337 	if (so->so_pcb == NULL)
338 		goto discard;
339 	if (so->so_state & SS_ISCONNECTED) {
340 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
341 			error = sodisconnect(so);
342 			if (error)
343 				goto drop;
344 		}
345 		if (so->so_options & SO_LINGER) {
346 			if ((so->so_state & SS_ISDISCONNECTING) &&
347 			    (fflag & FNONBLOCK))
348 				goto drop;
349 			while (so->so_state & SS_ISCONNECTED) {
350 				error = tsleep((caddr_t)&so->so_timeo,
351 				    PCATCH, "soclos", so->so_linger * hz);
352 				if (error)
353 					break;
354 			}
355 		}
356 	}
357 drop:
358 	if (so->so_pcb) {
359 		int error2;
360 
361 		error2 = so_pru_detach(so);
362 		if (error == 0)
363 			error = error2;
364 	}
365 discard:
366 	if (so->so_options & SO_ACCEPTCONN) {
367 		struct socket *sp;
368 
369 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
370 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
371 			sp->so_state &= ~SS_INCOMP;
372 			sp->so_head = NULL;
373 			so->so_incqlen--;
374 			soaborta(sp);
375 		}
376 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
377 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
378 			sp->so_state &= ~SS_COMP;
379 			sp->so_head = NULL;
380 			so->so_qlen--;
381 			soaborta(sp);
382 		}
383 	}
384 	if (so->so_state & SS_NOFDREF)
385 		panic("soclose: NOFDREF");
386 	so->so_state |= SS_NOFDREF;
387 	sofree(so);
388 	crit_exit();
389 	return (error);
390 }
391 
392 /*
393  * Abort and destroy a socket.  Only one abort can be in progress
394  * at any given moment.
395  */
396 void
397 soabort(struct socket *so)
398 {
399 	if ((so->so_state & SS_ABORTING) == 0) {
400 		so->so_state |= SS_ABORTING;
401 		so_pru_abort(so);
402 	}
403 }
404 
405 void
406 soaborta(struct socket *so)
407 {
408 	if ((so->so_state & SS_ABORTING) == 0) {
409 		so->so_state |= SS_ABORTING;
410 		so_pru_aborta(so);
411 	}
412 }
413 
414 void
415 soabort_oncpu(struct socket *so)
416 {
417 	if ((so->so_state & SS_ABORTING) == 0) {
418 		so->so_state |= SS_ABORTING;
419 		so_pru_abort_oncpu(so);
420 	}
421 }
422 
423 int
424 soaccept(struct socket *so, struct sockaddr **nam)
425 {
426 	int error;
427 
428 	crit_enter();
429 	if ((so->so_state & SS_NOFDREF) == 0)
430 		panic("soaccept: !NOFDREF");
431 	so->so_state &= ~SS_NOFDREF;
432 	error = so_pru_accept(so, nam);
433 	crit_exit();
434 	return (error);
435 }
436 
437 int
438 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
439 {
440 	int error;
441 
442 	if (so->so_options & SO_ACCEPTCONN)
443 		return (EOPNOTSUPP);
444 	crit_enter();
445 	/*
446 	 * If protocol is connection-based, can only connect once.
447 	 * Otherwise, if connected, try to disconnect first.
448 	 * This allows user to disconnect by connecting to, e.g.,
449 	 * a null address.
450 	 */
451 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
452 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
453 	    (error = sodisconnect(so)))) {
454 		error = EISCONN;
455 	} else {
456 		/*
457 		 * Prevent accumulated error from previous connection
458 		 * from biting us.
459 		 */
460 		so->so_error = 0;
461 		error = so_pru_connect(so, nam, td);
462 	}
463 	crit_exit();
464 	return (error);
465 }
466 
467 int
468 soconnect2(struct socket *so1, struct socket *so2)
469 {
470 	int error;
471 
472 	crit_enter();
473 	error = so_pru_connect2(so1, so2);
474 	crit_exit();
475 	return (error);
476 }
477 
478 int
479 sodisconnect(struct socket *so)
480 {
481 	int error;
482 
483 	crit_enter();
484 	if ((so->so_state & SS_ISCONNECTED) == 0) {
485 		error = ENOTCONN;
486 		goto bad;
487 	}
488 	if (so->so_state & SS_ISDISCONNECTING) {
489 		error = EALREADY;
490 		goto bad;
491 	}
492 	error = so_pru_disconnect(so);
493 bad:
494 	crit_exit();
495 	return (error);
496 }
497 
498 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
499 /*
500  * Send on a socket.
501  * If send must go all at once and message is larger than
502  * send buffering, then hard error.
503  * Lock against other senders.
504  * If must go all at once and not enough room now, then
505  * inform user that this would block and do nothing.
506  * Otherwise, if nonblocking, send as much as possible.
507  * The data to be sent is described by "uio" if nonzero,
508  * otherwise by the mbuf chain "top" (which must be null
509  * if uio is not).  Data provided in mbuf chain must be small
510  * enough to send all at once.
511  *
512  * Returns nonzero on error, timeout or signal; callers
513  * must check for short counts if EINTR/ERESTART are returned.
514  * Data and control buffers are freed on return.
515  */
516 int
517 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
518 	struct mbuf *top, struct mbuf *control, int flags,
519 	struct thread *td)
520 {
521 	struct mbuf **mp;
522 	struct mbuf *m;
523 	long space, len, resid;
524 	int clen = 0, error, dontroute, mlen;
525 	int atomic = sosendallatonce(so) || top;
526 	int pru_flags;
527 
528 	if (uio)
529 		resid = uio->uio_resid;
530 	else
531 		resid = top->m_pkthdr.len;
532 	/*
533 	 * In theory resid should be unsigned.
534 	 * However, space must be signed, as it might be less than 0
535 	 * if we over-committed, and we must use a signed comparison
536 	 * of space and resid.  On the other hand, a negative resid
537 	 * causes us to loop sending 0-length segments to the protocol.
538 	 *
539 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
540 	 * type sockets since that's an error.
541 	 */
542 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
543 		error = EINVAL;
544 		goto out;
545 	}
546 
547 	dontroute =
548 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
549 	    (so->so_proto->pr_flags & PR_ATOMIC);
550 	if (td->td_lwp != NULL)
551 		td->td_lwp->lwp_ru.ru_msgsnd++;
552 	if (control)
553 		clen = control->m_len;
554 #define	gotoerr(errcode)	{ error = errcode; crit_exit(); goto release; }
555 
556 restart:
557 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
558 	if (error)
559 		goto out;
560 	do {
561 		crit_enter();
562 		if (so->so_state & SS_CANTSENDMORE)
563 			gotoerr(EPIPE);
564 		if (so->so_error) {
565 			error = so->so_error;
566 			so->so_error = 0;
567 			crit_exit();
568 			goto release;
569 		}
570 		if ((so->so_state & SS_ISCONNECTED) == 0) {
571 			/*
572 			 * `sendto' and `sendmsg' is allowed on a connection-
573 			 * based socket if it supports implied connect.
574 			 * Return ENOTCONN if not connected and no address is
575 			 * supplied.
576 			 */
577 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
578 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
579 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
580 				    !(resid == 0 && clen != 0))
581 					gotoerr(ENOTCONN);
582 			} else if (addr == 0)
583 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
584 				   ENOTCONN : EDESTADDRREQ);
585 		}
586 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
587 		    clen > so->so_snd.ssb_hiwat) {
588 			gotoerr(EMSGSIZE);
589 		}
590 		space = ssb_space(&so->so_snd);
591 		if (flags & MSG_OOB)
592 			space += 1024;
593 		if (space < resid + clen && uio &&
594 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
595 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
596 				gotoerr(EWOULDBLOCK);
597 			ssb_unlock(&so->so_snd);
598 			error = ssb_wait(&so->so_snd);
599 			crit_exit();
600 			if (error)
601 				goto out;
602 			goto restart;
603 		}
604 		crit_exit();
605 		mp = &top;
606 		space -= clen;
607 		do {
608 		    if (uio == NULL) {
609 			/*
610 			 * Data is prepackaged in "top".
611 			 */
612 			resid = 0;
613 			if (flags & MSG_EOR)
614 				top->m_flags |= M_EOR;
615 		    } else do {
616 			m = m_getl(resid, MB_WAIT, MT_DATA,
617 				   top == NULL ? M_PKTHDR : 0, &mlen);
618 			if (top == NULL) {
619 				m->m_pkthdr.len = 0;
620 				m->m_pkthdr.rcvif = NULL;
621 			}
622 			len = min(min(mlen, resid), space);
623 			if (resid < MINCLSIZE) {
624 				/*
625 				 * For datagram protocols, leave room
626 				 * for protocol headers in first mbuf.
627 				 */
628 				if (atomic && top == 0 && len < mlen)
629 					MH_ALIGN(m, len);
630 			}
631 			space -= len;
632 			error = uiomove(mtod(m, caddr_t), (int)len, uio);
633 			resid = uio->uio_resid;
634 			m->m_len = len;
635 			*mp = m;
636 			top->m_pkthdr.len += len;
637 			if (error)
638 				goto release;
639 			mp = &m->m_next;
640 			if (resid <= 0) {
641 				if (flags & MSG_EOR)
642 					top->m_flags |= M_EOR;
643 				break;
644 			}
645 		    } while (space > 0 && atomic);
646 		    if (dontroute)
647 			    so->so_options |= SO_DONTROUTE;
648 		    if (flags & MSG_OOB) {
649 		    	    pru_flags = PRUS_OOB;
650 		    } else if ((flags & MSG_EOF) &&
651 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
652 		    	       (resid <= 0)) {
653 			    /*
654 			     * If the user set MSG_EOF, the protocol
655 			     * understands this flag and nothing left to
656 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
657 			     */
658 		    	    pru_flags = PRUS_EOF;
659 		    } else if (resid > 0 && space > 0) {
660 			    /* If there is more to send, set PRUS_MORETOCOME */
661 		    	    pru_flags = PRUS_MORETOCOME;
662 		    } else {
663 		    	    pru_flags = 0;
664 		    }
665 		    crit_enter();
666 		    /*
667 		     * XXX all the SS_CANTSENDMORE checks previously
668 		     * done could be out of date.  We could have recieved
669 		     * a reset packet in an interrupt or maybe we slept
670 		     * while doing page faults in uiomove() etc. We could
671 		     * probably recheck again inside the splnet() protection
672 		     * here, but there are probably other places that this
673 		     * also happens.  We must rethink this.
674 		     */
675 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
676 		    crit_exit();
677 		    if (dontroute)
678 			    so->so_options &= ~SO_DONTROUTE;
679 		    clen = 0;
680 		    control = 0;
681 		    top = 0;
682 		    mp = &top;
683 		    if (error)
684 			    goto release;
685 		} while (resid && space > 0);
686 	} while (resid);
687 
688 release:
689 	ssb_unlock(&so->so_snd);
690 out:
691 	if (top)
692 		m_freem(top);
693 	if (control)
694 		m_freem(control);
695 	return (error);
696 }
697 
698 /*
699  * A specialization of sosend() for UDP based on protocol-specific knowledge:
700  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
701  *	sosendallatonce() returns true,
702  *	the "atomic" variable is true,
703  *	and sosendudp() blocks until space is available for the entire send.
704  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
705  *	PR_IMPLOPCL flags set.
706  *   UDP has no out-of-band data.
707  *   UDP has no control data.
708  *   UDP does not support MSG_EOR.
709  */
710 int
711 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
712 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
713 {
714 	int resid, error;
715 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
716 
717 	if (td->td_lwp != NULL)
718 		td->td_lwp->lwp_ru.ru_msgsnd++;
719 	if (control)
720 		m_freem(control);
721 
722 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
723 	resid = uio ? uio->uio_resid : top->m_pkthdr.len;
724 
725 restart:
726 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
727 	if (error)
728 		goto out;
729 
730 	crit_enter();
731 	if (so->so_state & SS_CANTSENDMORE)
732 		gotoerr(EPIPE);
733 	if (so->so_error) {
734 		error = so->so_error;
735 		so->so_error = 0;
736 		crit_exit();
737 		goto release;
738 	}
739 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
740 		gotoerr(EDESTADDRREQ);
741 	if (resid > so->so_snd.ssb_hiwat)
742 		gotoerr(EMSGSIZE);
743 	if (uio && ssb_space(&so->so_snd) < resid) {
744 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
745 			gotoerr(EWOULDBLOCK);
746 		ssb_unlock(&so->so_snd);
747 		error = ssb_wait(&so->so_snd);
748 		crit_exit();
749 		if (error)
750 			goto out;
751 		goto restart;
752 	}
753 	crit_exit();
754 
755 	if (uio) {
756 		top = m_uiomove(uio);
757 		if (top == NULL)
758 			goto release;
759 	}
760 
761 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
762 	if (dontroute)
763 		so->so_options |= SO_DONTROUTE;
764 
765 	error = so_pru_send(so, 0, top, addr, NULL, td);
766 	top = NULL;		/* sent or freed in lower layer */
767 
768 	if (dontroute)
769 		so->so_options &= ~SO_DONTROUTE;
770 
771 release:
772 	ssb_unlock(&so->so_snd);
773 out:
774 	if (top)
775 		m_freem(top);
776 	return (error);
777 }
778 
779 /*
780  * Implement receive operations on a socket.
781  * We depend on the way that records are added to the signalsockbuf
782  * by sbappend*.  In particular, each record (mbufs linked through m_next)
783  * must begin with an address if the protocol so specifies,
784  * followed by an optional mbuf or mbufs containing ancillary data,
785  * and then zero or more mbufs of data.
786  * In order to avoid blocking network interrupts for the entire time here,
787  * we exit the critical section while doing the actual copy to user space.
788  * Although the signalsockbuf is locked, new data may still be appended,
789  * and thus we must maintain consistency of the signalsockbuf during that time.
790  *
791  * The caller may receive the data as a single mbuf chain by supplying
792  * an mbuf **mp0 for use in returning the chain.  The uio is then used
793  * only for the count in uio_resid.
794  */
795 int
796 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
797 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
798 {
799 	struct mbuf *m, *n;
800 	struct mbuf *free_chain = NULL;
801 	int flags, len, error, offset;
802 	struct protosw *pr = so->so_proto;
803 	int moff, type = 0;
804 	int resid, orig_resid;
805 
806 	if (uio)
807 		resid = uio->uio_resid;
808 	else
809 		resid = (int)(sio->sb_climit - sio->sb_cc);
810 	orig_resid = resid;
811 
812 	if (psa)
813 		*psa = NULL;
814 	if (controlp)
815 		*controlp = NULL;
816 	if (flagsp)
817 		flags = *flagsp &~ MSG_EOR;
818 	else
819 		flags = 0;
820 	if (flags & MSG_OOB) {
821 		m = m_get(MB_WAIT, MT_DATA);
822 		if (m == NULL)
823 			return (ENOBUFS);
824 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
825 		if (error)
826 			goto bad;
827 		if (sio) {
828 			do {
829 				sbappend(sio, m);
830 				resid -= m->m_len;
831 			} while (resid > 0 && m);
832 		} else {
833 			do {
834 				uio->uio_resid = resid;
835 				error = uiomove(mtod(m, caddr_t),
836 						(int)min(resid, m->m_len), uio);
837 				resid = uio->uio_resid;
838 				m = m_free(m);
839 			} while (uio->uio_resid && error == 0 && m);
840 		}
841 bad:
842 		if (m)
843 			m_freem(m);
844 		return (error);
845 	}
846 	if (so->so_state & SS_ISCONFIRMING && resid)
847 		so_pru_rcvd(so, 0);
848 
849 restart:
850 	crit_enter();
851 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
852 	if (error)
853 		goto done;
854 
855 	m = so->so_rcv.ssb_mb;
856 	/*
857 	 * If we have less data than requested, block awaiting more
858 	 * (subject to any timeout) if:
859 	 *   1. the current count is less than the low water mark, or
860 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
861 	 *	receive operation at once if we block (resid <= hiwat).
862 	 *   3. MSG_DONTWAIT is not set
863 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
864 	 * we have to do the receive in sections, and thus risk returning
865 	 * a short count if a timeout or signal occurs after we start.
866 	 */
867 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
868 	    so->so_rcv.ssb_cc < resid) &&
869 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
870 	    ((flags & MSG_WAITALL) && resid <= so->so_rcv.ssb_hiwat)) &&
871 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
872 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
873 		if (so->so_error) {
874 			if (m)
875 				goto dontblock;
876 			error = so->so_error;
877 			if ((flags & MSG_PEEK) == 0)
878 				so->so_error = 0;
879 			goto release;
880 		}
881 		if (so->so_state & SS_CANTRCVMORE) {
882 			if (m)
883 				goto dontblock;
884 			else
885 				goto release;
886 		}
887 		for (; m; m = m->m_next) {
888 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
889 				m = so->so_rcv.ssb_mb;
890 				goto dontblock;
891 			}
892 		}
893 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
894 		    (pr->pr_flags & PR_CONNREQUIRED)) {
895 			error = ENOTCONN;
896 			goto release;
897 		}
898 		if (resid == 0)
899 			goto release;
900 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
901 			error = EWOULDBLOCK;
902 			goto release;
903 		}
904 		ssb_unlock(&so->so_rcv);
905 		error = ssb_wait(&so->so_rcv);
906 		if (error)
907 			goto done;
908 		crit_exit();
909 		goto restart;
910 	}
911 dontblock:
912 	if (uio && uio->uio_td && uio->uio_td->td_proc)
913 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
914 
915 	/*
916 	 * note: m should be == sb_mb here.  Cache the next record while
917 	 * cleaning up.  Note that calling m_free*() will break out critical
918 	 * section.
919 	 */
920 	KKASSERT(m == so->so_rcv.ssb_mb);
921 
922 	/*
923 	 * Skip any address mbufs prepending the record.
924 	 */
925 	if (pr->pr_flags & PR_ADDR) {
926 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
927 		orig_resid = 0;
928 		if (psa)
929 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
930 		if (flags & MSG_PEEK)
931 			m = m->m_next;
932 		else
933 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
934 	}
935 
936 	/*
937 	 * Skip any control mbufs prepending the record.
938 	 */
939 #ifdef SCTP
940 	if (pr->pr_flags & PR_ADDR_OPT) {
941 		/*
942 		 * For SCTP we may be getting a
943 		 * whole message OR a partial delivery.
944 		 */
945 		if (m && m->m_type == MT_SONAME) {
946 			orig_resid = 0;
947 			if (psa)
948 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
949 			if (flags & MSG_PEEK)
950 				m = m->m_next;
951 			else
952 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
953 		}
954 	}
955 #endif /* SCTP */
956 	while (m && m->m_type == MT_CONTROL && error == 0) {
957 		if (flags & MSG_PEEK) {
958 			if (controlp)
959 				*controlp = m_copy(m, 0, m->m_len);
960 			m = m->m_next;	/* XXX race */
961 		} else {
962 			if (controlp) {
963 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
964 				if (pr->pr_domain->dom_externalize &&
965 				    mtod(m, struct cmsghdr *)->cmsg_type ==
966 				    SCM_RIGHTS)
967 				   error = (*pr->pr_domain->dom_externalize)(m);
968 				*controlp = m;
969 				m = n;
970 			} else {
971 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
972 			}
973 		}
974 		if (controlp && *controlp) {
975 			orig_resid = 0;
976 			controlp = &(*controlp)->m_next;
977 		}
978 	}
979 
980 	/*
981 	 * flag OOB data.
982 	 */
983 	if (m) {
984 		type = m->m_type;
985 		if (type == MT_OOBDATA)
986 			flags |= MSG_OOB;
987 	}
988 
989 	/*
990 	 * Copy to the UIO or mbuf return chain (*mp).
991 	 */
992 	moff = 0;
993 	offset = 0;
994 	while (m && resid > 0 && error == 0) {
995 		if (m->m_type == MT_OOBDATA) {
996 			if (type != MT_OOBDATA)
997 				break;
998 		} else if (type == MT_OOBDATA)
999 			break;
1000 		else
1001 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1002 			("receive 3"));
1003 		so->so_state &= ~SS_RCVATMARK;
1004 		len = resid;
1005 		if (so->so_oobmark && len > so->so_oobmark - offset)
1006 			len = so->so_oobmark - offset;
1007 		if (len > m->m_len - moff)
1008 			len = m->m_len - moff;
1009 
1010 		/*
1011 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1012 		 * The SIO is dealt with when we eat the mbuf, but deal
1013 		 * with the resid here either way.
1014 		 */
1015 		if (uio) {
1016 			crit_exit();
1017 			uio->uio_resid = resid;
1018 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1019 			resid = uio->uio_resid;
1020 			crit_enter();
1021 			if (error)
1022 				goto release;
1023 		} else {
1024 			resid -= len;
1025 		}
1026 
1027 		/*
1028 		 * Eat the entire mbuf or just a piece of it
1029 		 */
1030 		if (len == m->m_len - moff) {
1031 			if (m->m_flags & M_EOR)
1032 				flags |= MSG_EOR;
1033 #ifdef SCTP
1034 			if (m->m_flags & M_NOTIFICATION)
1035 				flags |= MSG_NOTIFICATION;
1036 #endif /* SCTP */
1037 			if (flags & MSG_PEEK) {
1038 				m = m->m_next;
1039 				moff = 0;
1040 			} else {
1041 				if (sio) {
1042 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1043 					sbappend(sio, m);
1044 					m = n;
1045 				} else {
1046 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1047 				}
1048 			}
1049 		} else {
1050 			if (flags & MSG_PEEK) {
1051 				moff += len;
1052 			} else {
1053 				if (sio) {
1054 					n = m_copym(m, 0, len, MB_WAIT);
1055 					if (n)
1056 						sbappend(sio, n);
1057 				}
1058 				m->m_data += len;
1059 				m->m_len -= len;
1060 				so->so_rcv.ssb_cc -= len;
1061 			}
1062 		}
1063 		if (so->so_oobmark) {
1064 			if ((flags & MSG_PEEK) == 0) {
1065 				so->so_oobmark -= len;
1066 				if (so->so_oobmark == 0) {
1067 					so->so_state |= SS_RCVATMARK;
1068 					break;
1069 				}
1070 			} else {
1071 				offset += len;
1072 				if (offset == so->so_oobmark)
1073 					break;
1074 			}
1075 		}
1076 		if (flags & MSG_EOR)
1077 			break;
1078 		/*
1079 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1080 		 * we must not quit until resid == 0 or an error
1081 		 * termination.  If a signal/timeout occurs, return
1082 		 * with a short count but without error.
1083 		 * Keep signalsockbuf locked against other readers.
1084 		 */
1085 		while ((flags & MSG_WAITALL) && m == NULL &&
1086 		       resid > 0 && !sosendallatonce(so) &&
1087 		       so->so_rcv.ssb_mb == NULL) {
1088 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1089 				break;
1090 			/*
1091 			 * The window might have closed to zero, make
1092 			 * sure we send an ack now that we've drained
1093 			 * the buffer or we might end up blocking until
1094 			 * the idle takes over (5 seconds).
1095 			 */
1096 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1097 				so_pru_rcvd(so, flags);
1098 			error = ssb_wait(&so->so_rcv);
1099 			if (error) {
1100 				ssb_unlock(&so->so_rcv);
1101 				error = 0;
1102 				goto done;
1103 			}
1104 			m = so->so_rcv.ssb_mb;
1105 		}
1106 	}
1107 
1108 	/*
1109 	 * If an atomic read was requested but unread data still remains
1110 	 * in the record, set MSG_TRUNC.
1111 	 */
1112 	if (m && pr->pr_flags & PR_ATOMIC)
1113 		flags |= MSG_TRUNC;
1114 
1115 	/*
1116 	 * Cleanup.  If an atomic read was requested drop any unread data.
1117 	 */
1118 	if ((flags & MSG_PEEK) == 0) {
1119 		if (m && (pr->pr_flags & PR_ATOMIC))
1120 			sbdroprecord(&so->so_rcv.sb);
1121 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1122 			so_pru_rcvd(so, flags);
1123 	}
1124 
1125 	if (orig_resid == resid && orig_resid &&
1126 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1127 		ssb_unlock(&so->so_rcv);
1128 		crit_exit();
1129 		goto restart;
1130 	}
1131 
1132 	if (flagsp)
1133 		*flagsp |= flags;
1134 release:
1135 	ssb_unlock(&so->so_rcv);
1136 done:
1137 	crit_exit();
1138 	if (free_chain)
1139 		m_freem(free_chain);
1140 	return (error);
1141 }
1142 
1143 int
1144 soshutdown(struct socket *so, int how)
1145 {
1146 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1147 		return (EINVAL);
1148 
1149 	if (how != SHUT_WR)
1150 		sorflush(so);
1151 	if (how != SHUT_RD)
1152 		return (so_pru_shutdown(so));
1153 	return (0);
1154 }
1155 
1156 void
1157 sorflush(struct socket *so)
1158 {
1159 	struct signalsockbuf *ssb = &so->so_rcv;
1160 	struct protosw *pr = so->so_proto;
1161 	struct signalsockbuf asb;
1162 
1163 	ssb->ssb_flags |= SSB_NOINTR;
1164 	(void) ssb_lock(ssb, M_WAITOK);
1165 
1166 	crit_enter();
1167 	socantrcvmore(so);
1168 	ssb_unlock(ssb);
1169 	asb = *ssb;
1170 	bzero((caddr_t)ssb, sizeof (*ssb));
1171 	if (asb.ssb_flags & SSB_KNOTE) {
1172 		ssb->ssb_sel.si_note = asb.ssb_sel.si_note;
1173 		ssb->ssb_flags = SSB_KNOTE;
1174 	}
1175 	crit_exit();
1176 
1177 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1178 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1179 	ssb_release(&asb, so);
1180 }
1181 
1182 #ifdef INET
1183 static int
1184 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1185 {
1186 	struct accept_filter_arg	*afap = NULL;
1187 	struct accept_filter	*afp;
1188 	struct so_accf	*af = so->so_accf;
1189 	int	error = 0;
1190 
1191 	/* do not set/remove accept filters on non listen sockets */
1192 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1193 		error = EINVAL;
1194 		goto out;
1195 	}
1196 
1197 	/* removing the filter */
1198 	if (sopt == NULL) {
1199 		if (af != NULL) {
1200 			if (af->so_accept_filter != NULL &&
1201 				af->so_accept_filter->accf_destroy != NULL) {
1202 				af->so_accept_filter->accf_destroy(so);
1203 			}
1204 			if (af->so_accept_filter_str != NULL) {
1205 				FREE(af->so_accept_filter_str, M_ACCF);
1206 			}
1207 			FREE(af, M_ACCF);
1208 			so->so_accf = NULL;
1209 		}
1210 		so->so_options &= ~SO_ACCEPTFILTER;
1211 		return (0);
1212 	}
1213 	/* adding a filter */
1214 	/* must remove previous filter first */
1215 	if (af != NULL) {
1216 		error = EINVAL;
1217 		goto out;
1218 	}
1219 	/* don't put large objects on the kernel stack */
1220 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1221 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1222 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1223 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1224 	if (error)
1225 		goto out;
1226 	afp = accept_filt_get(afap->af_name);
1227 	if (afp == NULL) {
1228 		error = ENOENT;
1229 		goto out;
1230 	}
1231 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1232 	if (afp->accf_create != NULL) {
1233 		if (afap->af_name[0] != '\0') {
1234 			int len = strlen(afap->af_name) + 1;
1235 
1236 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1237 			strcpy(af->so_accept_filter_str, afap->af_name);
1238 		}
1239 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1240 		if (af->so_accept_filter_arg == NULL) {
1241 			FREE(af->so_accept_filter_str, M_ACCF);
1242 			FREE(af, M_ACCF);
1243 			so->so_accf = NULL;
1244 			error = EINVAL;
1245 			goto out;
1246 		}
1247 	}
1248 	af->so_accept_filter = afp;
1249 	so->so_accf = af;
1250 	so->so_options |= SO_ACCEPTFILTER;
1251 out:
1252 	if (afap != NULL)
1253 		FREE(afap, M_TEMP);
1254 	return (error);
1255 }
1256 #endif /* INET */
1257 
1258 /*
1259  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1260  * an additional variant to handle the case where the option value needs
1261  * to be some kind of integer, but not a specific size.
1262  * In addition to their use here, these functions are also called by the
1263  * protocol-level pr_ctloutput() routines.
1264  */
1265 int
1266 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1267 {
1268 	return soopt_to_kbuf(sopt, buf, len, minlen);
1269 }
1270 
1271 int
1272 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1273 {
1274 	size_t	valsize;
1275 
1276 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1277 	KKASSERT(kva_p(buf));
1278 
1279 	/*
1280 	 * If the user gives us more than we wanted, we ignore it,
1281 	 * but if we don't get the minimum length the caller
1282 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1283 	 * is set to however much we actually retrieved.
1284 	 */
1285 	if ((valsize = sopt->sopt_valsize) < minlen)
1286 		return EINVAL;
1287 	if (valsize > len)
1288 		sopt->sopt_valsize = valsize = len;
1289 
1290 	bcopy(sopt->sopt_val, buf, valsize);
1291 	return 0;
1292 }
1293 
1294 
1295 int
1296 sosetopt(struct socket *so, struct sockopt *sopt)
1297 {
1298 	int	error, optval;
1299 	struct	linger l;
1300 	struct	timeval tv;
1301 	u_long  val;
1302 
1303 	error = 0;
1304 	sopt->sopt_dir = SOPT_SET;
1305 	if (sopt->sopt_level != SOL_SOCKET) {
1306 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1307 			return (so_pru_ctloutput(so, sopt));
1308 		}
1309 		error = ENOPROTOOPT;
1310 	} else {
1311 		switch (sopt->sopt_name) {
1312 #ifdef INET
1313 		case SO_ACCEPTFILTER:
1314 			error = do_setopt_accept_filter(so, sopt);
1315 			if (error)
1316 				goto bad;
1317 			break;
1318 #endif /* INET */
1319 		case SO_LINGER:
1320 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1321 			if (error)
1322 				goto bad;
1323 
1324 			so->so_linger = l.l_linger;
1325 			if (l.l_onoff)
1326 				so->so_options |= SO_LINGER;
1327 			else
1328 				so->so_options &= ~SO_LINGER;
1329 			break;
1330 
1331 		case SO_DEBUG:
1332 		case SO_KEEPALIVE:
1333 		case SO_DONTROUTE:
1334 		case SO_USELOOPBACK:
1335 		case SO_BROADCAST:
1336 		case SO_REUSEADDR:
1337 		case SO_REUSEPORT:
1338 		case SO_OOBINLINE:
1339 		case SO_TIMESTAMP:
1340 			error = sooptcopyin(sopt, &optval, sizeof optval,
1341 					    sizeof optval);
1342 			if (error)
1343 				goto bad;
1344 			if (optval)
1345 				so->so_options |= sopt->sopt_name;
1346 			else
1347 				so->so_options &= ~sopt->sopt_name;
1348 			break;
1349 
1350 		case SO_SNDBUF:
1351 		case SO_RCVBUF:
1352 		case SO_SNDLOWAT:
1353 		case SO_RCVLOWAT:
1354 			error = sooptcopyin(sopt, &optval, sizeof optval,
1355 					    sizeof optval);
1356 			if (error)
1357 				goto bad;
1358 
1359 			/*
1360 			 * Values < 1 make no sense for any of these
1361 			 * options, so disallow them.
1362 			 */
1363 			if (optval < 1) {
1364 				error = EINVAL;
1365 				goto bad;
1366 			}
1367 
1368 			switch (sopt->sopt_name) {
1369 			case SO_SNDBUF:
1370 			case SO_RCVBUF:
1371 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1372 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1373 				    so,
1374 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1375 					error = ENOBUFS;
1376 					goto bad;
1377 				}
1378 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
1379 				    &so->so_rcv)->ssb_flags &= ~SSB_AUTOSIZE;
1380 				break;
1381 
1382 			/*
1383 			 * Make sure the low-water is never greater than
1384 			 * the high-water.
1385 			 */
1386 			case SO_SNDLOWAT:
1387 				so->so_snd.ssb_lowat =
1388 				    (optval > so->so_snd.ssb_hiwat) ?
1389 				    so->so_snd.ssb_hiwat : optval;
1390 				break;
1391 			case SO_RCVLOWAT:
1392 				so->so_rcv.ssb_lowat =
1393 				    (optval > so->so_rcv.ssb_hiwat) ?
1394 				    so->so_rcv.ssb_hiwat : optval;
1395 				break;
1396 			}
1397 			break;
1398 
1399 		case SO_SNDTIMEO:
1400 		case SO_RCVTIMEO:
1401 			error = sooptcopyin(sopt, &tv, sizeof tv,
1402 					    sizeof tv);
1403 			if (error)
1404 				goto bad;
1405 
1406 			/* assert(hz > 0); */
1407 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1408 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1409 				error = EDOM;
1410 				goto bad;
1411 			}
1412 			/* assert(tick > 0); */
1413 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1414 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1415 			if (val > SHRT_MAX) {
1416 				error = EDOM;
1417 				goto bad;
1418 			}
1419 			if (val == 0 && tv.tv_usec != 0)
1420 				val = 1;
1421 
1422 			switch (sopt->sopt_name) {
1423 			case SO_SNDTIMEO:
1424 				so->so_snd.ssb_timeo = val;
1425 				break;
1426 			case SO_RCVTIMEO:
1427 				so->so_rcv.ssb_timeo = val;
1428 				break;
1429 			}
1430 			break;
1431 		default:
1432 			error = ENOPROTOOPT;
1433 			break;
1434 		}
1435 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1436 			(void) so_pru_ctloutput(so, sopt);
1437 		}
1438 	}
1439 bad:
1440 	return (error);
1441 }
1442 
1443 /* Helper routine for getsockopt */
1444 int
1445 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1446 {
1447 	soopt_from_kbuf(sopt, buf, len);
1448 	return 0;
1449 }
1450 
1451 void
1452 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1453 {
1454 	size_t	valsize;
1455 
1456 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1457 	KKASSERT(kva_p(buf));
1458 
1459 	/*
1460 	 * Documented get behavior is that we always return a value,
1461 	 * possibly truncated to fit in the user's buffer.
1462 	 * Traditional behavior is that we always tell the user
1463 	 * precisely how much we copied, rather than something useful
1464 	 * like the total amount we had available for her.
1465 	 * Note that this interface is not idempotent; the entire answer must
1466 	 * generated ahead of time.
1467 	 */
1468 	valsize = min(len, sopt->sopt_valsize);
1469 	sopt->sopt_valsize = valsize;
1470 	if (sopt->sopt_val != 0) {
1471 		bcopy(buf, sopt->sopt_val, valsize);
1472 	}
1473 }
1474 
1475 int
1476 sogetopt(struct socket *so, struct sockopt *sopt)
1477 {
1478 	int	error, optval;
1479 	struct	linger l;
1480 	struct	timeval tv;
1481 #ifdef INET
1482 	struct accept_filter_arg *afap;
1483 #endif
1484 
1485 	error = 0;
1486 	sopt->sopt_dir = SOPT_GET;
1487 	if (sopt->sopt_level != SOL_SOCKET) {
1488 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1489 			return (so_pru_ctloutput(so, sopt));
1490 		} else
1491 			return (ENOPROTOOPT);
1492 	} else {
1493 		switch (sopt->sopt_name) {
1494 #ifdef INET
1495 		case SO_ACCEPTFILTER:
1496 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1497 				return (EINVAL);
1498 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1499 				M_TEMP, M_WAITOK | M_ZERO);
1500 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1501 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1502 				if (so->so_accf->so_accept_filter_str != NULL)
1503 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1504 			}
1505 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1506 			FREE(afap, M_TEMP);
1507 			break;
1508 #endif /* INET */
1509 
1510 		case SO_LINGER:
1511 			l.l_onoff = so->so_options & SO_LINGER;
1512 			l.l_linger = so->so_linger;
1513 			error = sooptcopyout(sopt, &l, sizeof l);
1514 			break;
1515 
1516 		case SO_USELOOPBACK:
1517 		case SO_DONTROUTE:
1518 		case SO_DEBUG:
1519 		case SO_KEEPALIVE:
1520 		case SO_REUSEADDR:
1521 		case SO_REUSEPORT:
1522 		case SO_BROADCAST:
1523 		case SO_OOBINLINE:
1524 		case SO_TIMESTAMP:
1525 			optval = so->so_options & sopt->sopt_name;
1526 integer:
1527 			error = sooptcopyout(sopt, &optval, sizeof optval);
1528 			break;
1529 
1530 		case SO_TYPE:
1531 			optval = so->so_type;
1532 			goto integer;
1533 
1534 		case SO_ERROR:
1535 			optval = so->so_error;
1536 			so->so_error = 0;
1537 			goto integer;
1538 
1539 		case SO_SNDBUF:
1540 			optval = so->so_snd.ssb_hiwat;
1541 			goto integer;
1542 
1543 		case SO_RCVBUF:
1544 			optval = so->so_rcv.ssb_hiwat;
1545 			goto integer;
1546 
1547 		case SO_SNDLOWAT:
1548 			optval = so->so_snd.ssb_lowat;
1549 			goto integer;
1550 
1551 		case SO_RCVLOWAT:
1552 			optval = so->so_rcv.ssb_lowat;
1553 			goto integer;
1554 
1555 		case SO_SNDTIMEO:
1556 		case SO_RCVTIMEO:
1557 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1558 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1559 
1560 			tv.tv_sec = optval / hz;
1561 			tv.tv_usec = (optval % hz) * tick;
1562 			error = sooptcopyout(sopt, &tv, sizeof tv);
1563 			break;
1564 
1565 		default:
1566 			error = ENOPROTOOPT;
1567 			break;
1568 		}
1569 		return (error);
1570 	}
1571 }
1572 
1573 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1574 int
1575 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1576 {
1577 	struct mbuf *m, *m_prev;
1578 	int sopt_size = sopt->sopt_valsize, msize;
1579 
1580 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1581 		   0, &msize);
1582 	if (m == NULL)
1583 		return (ENOBUFS);
1584 	m->m_len = min(msize, sopt_size);
1585 	sopt_size -= m->m_len;
1586 	*mp = m;
1587 	m_prev = m;
1588 
1589 	while (sopt_size > 0) {
1590 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1591 			   MT_DATA, 0, &msize);
1592 		if (m == NULL) {
1593 			m_freem(*mp);
1594 			return (ENOBUFS);
1595 		}
1596 		m->m_len = min(msize, sopt_size);
1597 		sopt_size -= m->m_len;
1598 		m_prev->m_next = m;
1599 		m_prev = m;
1600 	}
1601 	return (0);
1602 }
1603 
1604 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1605 int
1606 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1607 {
1608 	soopt_to_mbuf(sopt, m);
1609 	return 0;
1610 }
1611 
1612 void
1613 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1614 {
1615 	size_t valsize;
1616 	void *val;
1617 
1618 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1619 	KKASSERT(kva_p(m));
1620 	if (sopt->sopt_val == NULL)
1621 		return;
1622 	val = sopt->sopt_val;
1623 	valsize = sopt->sopt_valsize;
1624 	while (m != NULL && valsize >= m->m_len) {
1625 		bcopy(val, mtod(m, char *), m->m_len);
1626 		valsize -= m->m_len;
1627 		val = (caddr_t)val + m->m_len;
1628 		m = m->m_next;
1629 	}
1630 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1631 		panic("ip6_sooptmcopyin");
1632 }
1633 
1634 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1635 int
1636 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1637 {
1638 	return soopt_from_mbuf(sopt, m);
1639 }
1640 
1641 int
1642 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1643 {
1644 	struct mbuf *m0 = m;
1645 	size_t valsize = 0;
1646 	size_t maxsize;
1647 	void *val;
1648 
1649 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1650 	KKASSERT(kva_p(m));
1651 	if (sopt->sopt_val == NULL)
1652 		return 0;
1653 	val = sopt->sopt_val;
1654 	maxsize = sopt->sopt_valsize;
1655 	while (m != NULL && maxsize >= m->m_len) {
1656 		bcopy(mtod(m, char *), val, m->m_len);
1657 	       maxsize -= m->m_len;
1658 	       val = (caddr_t)val + m->m_len;
1659 	       valsize += m->m_len;
1660 	       m = m->m_next;
1661 	}
1662 	if (m != NULL) {
1663 		/* enough soopt buffer should be given from user-land */
1664 		m_freem(m0);
1665 		return (EINVAL);
1666 	}
1667 	sopt->sopt_valsize = valsize;
1668 	return 0;
1669 }
1670 
1671 void
1672 sohasoutofband(struct socket *so)
1673 {
1674 	if (so->so_sigio != NULL)
1675 		pgsigio(so->so_sigio, SIGURG, 0);
1676 	selwakeup(&so->so_rcv.ssb_sel);
1677 }
1678 
1679 int
1680 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1681 {
1682 	int revents = 0;
1683 
1684 	crit_enter();
1685 
1686 	if (events & (POLLIN | POLLRDNORM))
1687 		if (soreadable(so))
1688 			revents |= events & (POLLIN | POLLRDNORM);
1689 
1690 	if (events & POLLINIGNEOF)
1691 		if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat ||
1692 			!TAILQ_EMPTY(&so->so_comp) || so->so_error)
1693 			revents |= POLLINIGNEOF;
1694 
1695 	if (events & (POLLOUT | POLLWRNORM))
1696 		if (sowriteable(so))
1697 			revents |= events & (POLLOUT | POLLWRNORM);
1698 
1699 	if (events & (POLLPRI | POLLRDBAND))
1700 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1701 			revents |= events & (POLLPRI | POLLRDBAND);
1702 
1703 	if (revents == 0) {
1704 		if (events &
1705 			(POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1706 			 POLLRDBAND)) {
1707 			selrecord(td, &so->so_rcv.ssb_sel);
1708 			so->so_rcv.ssb_flags |= SSB_SEL;
1709 		}
1710 
1711 		if (events & (POLLOUT | POLLWRNORM)) {
1712 			selrecord(td, &so->so_snd.ssb_sel);
1713 			so->so_snd.ssb_flags |= SSB_SEL;
1714 		}
1715 	}
1716 
1717 	crit_exit();
1718 	return (revents);
1719 }
1720 
1721 int
1722 sokqfilter(struct file *fp, struct knote *kn)
1723 {
1724 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1725 	struct signalsockbuf *ssb;
1726 
1727 	switch (kn->kn_filter) {
1728 	case EVFILT_READ:
1729 		if (so->so_options & SO_ACCEPTCONN)
1730 			kn->kn_fop = &solisten_filtops;
1731 		else
1732 			kn->kn_fop = &soread_filtops;
1733 		ssb = &so->so_rcv;
1734 		break;
1735 	case EVFILT_WRITE:
1736 		kn->kn_fop = &sowrite_filtops;
1737 		ssb = &so->so_snd;
1738 		break;
1739 	default:
1740 		return (1);
1741 	}
1742 
1743 	crit_enter();
1744 	SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext);
1745 	ssb->ssb_flags |= SSB_KNOTE;
1746 	crit_exit();
1747 	return (0);
1748 }
1749 
1750 static void
1751 filt_sordetach(struct knote *kn)
1752 {
1753 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1754 
1755 	crit_enter();
1756 	SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext);
1757 	if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note))
1758 		so->so_rcv.ssb_flags &= ~SSB_KNOTE;
1759 	crit_exit();
1760 }
1761 
1762 /*ARGSUSED*/
1763 static int
1764 filt_soread(struct knote *kn, long hint)
1765 {
1766 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1767 
1768 	kn->kn_data = so->so_rcv.ssb_cc;
1769 	if (so->so_state & SS_CANTRCVMORE) {
1770 		kn->kn_flags |= EV_EOF;
1771 		kn->kn_fflags = so->so_error;
1772 		return (1);
1773 	}
1774 	if (so->so_error)	/* temporary udp error */
1775 		return (1);
1776 	if (kn->kn_sfflags & NOTE_LOWAT)
1777 		return (kn->kn_data >= kn->kn_sdata);
1778 	return (kn->kn_data >= so->so_rcv.ssb_lowat);
1779 }
1780 
1781 static void
1782 filt_sowdetach(struct knote *kn)
1783 {
1784 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1785 
1786 	crit_enter();
1787 	SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext);
1788 	if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note))
1789 		so->so_snd.ssb_flags &= ~SSB_KNOTE;
1790 	crit_exit();
1791 }
1792 
1793 /*ARGSUSED*/
1794 static int
1795 filt_sowrite(struct knote *kn, long hint)
1796 {
1797 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1798 
1799 	kn->kn_data = ssb_space(&so->so_snd);
1800 	if (so->so_state & SS_CANTSENDMORE) {
1801 		kn->kn_flags |= EV_EOF;
1802 		kn->kn_fflags = so->so_error;
1803 		return (1);
1804 	}
1805 	if (so->so_error)	/* temporary udp error */
1806 		return (1);
1807 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1808 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1809 		return (0);
1810 	if (kn->kn_sfflags & NOTE_LOWAT)
1811 		return (kn->kn_data >= kn->kn_sdata);
1812 	return (kn->kn_data >= so->so_snd.ssb_lowat);
1813 }
1814 
1815 /*ARGSUSED*/
1816 static int
1817 filt_solisten(struct knote *kn, long hint)
1818 {
1819 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1820 
1821 	kn->kn_data = so->so_qlen;
1822 	return (! TAILQ_EMPTY(&so->so_comp));
1823 }
1824