xref: /dflybsd-src/sys/kern/uipc_socket.c (revision 70e491c09b1cb18f96facdbc91052080f2a455c3)
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 static struct filterops soexcept_filtops =
120 	{ 1, NULL, filt_sordetach, filt_soread };
121 
122 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
123 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
124 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
125 
126 
127 static int somaxconn = SOMAXCONN;
128 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
129     &somaxconn, 0, "Maximum pending socket connection queue size");
130 
131 /*
132  * Socket operation routines.
133  * These routines are called by the routines in
134  * sys_socket.c or from a system process, and
135  * implement the semantics of socket operations by
136  * switching out to the protocol specific routines.
137  */
138 
139 /*
140  * Get a socket structure, and initialize it.
141  * Note that it would probably be better to allocate socket
142  * and PCB at the same time, but I'm not convinced that all
143  * the protocols can be easily modified to do this.
144  */
145 struct socket *
146 soalloc(int waitok)
147 {
148 	struct socket *so;
149 	unsigned waitmask;
150 
151 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
152 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
153 	if (so) {
154 		/* XXX race condition for reentrant kernel */
155 		TAILQ_INIT(&so->so_aiojobq);
156 		TAILQ_INIT(&so->so_rcv.ssb_sel.si_mlist);
157 		TAILQ_INIT(&so->so_snd.ssb_sel.si_mlist);
158 	}
159 	return so;
160 }
161 
162 int
163 socreate(int dom, struct socket **aso, int type,
164 	int proto, struct thread *td)
165 {
166 	struct proc *p = td->td_proc;
167 	struct protosw *prp;
168 	struct socket *so;
169 	struct pru_attach_info ai;
170 	int error;
171 
172 	if (proto)
173 		prp = pffindproto(dom, proto, type);
174 	else
175 		prp = pffindtype(dom, type);
176 
177 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
178 		return (EPROTONOSUPPORT);
179 
180 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
181 	    prp->pr_domain->dom_family != PF_LOCAL &&
182 	    prp->pr_domain->dom_family != PF_INET &&
183 	    prp->pr_domain->dom_family != PF_INET6 &&
184 	    prp->pr_domain->dom_family != PF_ROUTE) {
185 		return (EPROTONOSUPPORT);
186 	}
187 
188 	if (prp->pr_type != type)
189 		return (EPROTOTYPE);
190 	so = soalloc(p != 0);
191 	if (so == 0)
192 		return (ENOBUFS);
193 
194 	/*
195 	 * Set a default port for protocol processing.  No action will occur
196 	 * on the socket on this port until an inpcb is attached to it and
197 	 * is able to match incoming packets, or until the socket becomes
198 	 * available to userland.
199 	 */
200 	so->so_port = cpu0_soport(so, NULL, NULL);
201 
202 	TAILQ_INIT(&so->so_incomp);
203 	TAILQ_INIT(&so->so_comp);
204 	so->so_type = type;
205 	so->so_cred = crhold(p->p_ucred);
206 	so->so_proto = prp;
207 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
208 	ai.p_ucred = p->p_ucred;
209 	ai.fd_rdir = p->p_fd->fd_rdir;
210 
211 	/*
212 	 * Auto-sizing of socket buffers is managed by the protocols and
213 	 * the appropriate flags must be set in the pru_attach function.
214 	 */
215 	error = so_pru_attach(so, proto, &ai);
216 	if (error) {
217 		so->so_state |= SS_NOFDREF;
218 		sofree(so);
219 		return (error);
220 	}
221 
222 	*aso = so;
223 	return (0);
224 }
225 
226 int
227 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
228 {
229 	int error;
230 
231 	crit_enter();
232 	error = so_pru_bind(so, nam, td);
233 	crit_exit();
234 	return (error);
235 }
236 
237 void
238 sodealloc(struct socket *so)
239 {
240 	if (so->so_rcv.ssb_hiwat)
241 		(void)chgsbsize(so->so_cred->cr_uidinfo,
242 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
243 	if (so->so_snd.ssb_hiwat)
244 		(void)chgsbsize(so->so_cred->cr_uidinfo,
245 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
246 #ifdef INET
247 	/* remove accept filter if present */
248 	if (so->so_accf != NULL)
249 		do_setopt_accept_filter(so, NULL);
250 #endif /* INET */
251 	crfree(so->so_cred);
252 	kfree(so, M_SOCKET);
253 }
254 
255 int
256 solisten(struct socket *so, int backlog, struct thread *td)
257 {
258 	int error;
259 #ifdef SCTP
260 	short oldopt, oldqlimit;
261 #endif /* SCTP */
262 
263 	crit_enter();
264 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING)) {
265 		crit_exit();
266 		return (EINVAL);
267 	}
268 
269 #ifdef SCTP
270 	oldopt = so->so_options;
271 	oldqlimit = so->so_qlimit;
272 #endif /* SCTP */
273 
274 	if (TAILQ_EMPTY(&so->so_comp))
275 		so->so_options |= SO_ACCEPTCONN;
276 	if (backlog < 0 || backlog > somaxconn)
277 		backlog = somaxconn;
278 	so->so_qlimit = backlog;
279 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
280 	 * the so_options (UDP model both connect's and gets inbound
281 	 * connections .. implicitly).
282 	 */
283 	error = so_pru_listen(so, td);
284 	if (error) {
285 #ifdef SCTP
286 		/* Restore the params */
287 		so->so_options = oldopt;
288 		so->so_qlimit = oldqlimit;
289 #endif /* SCTP */
290 		crit_exit();
291 		return (error);
292 	}
293 	crit_exit();
294 	return (0);
295 }
296 
297 /*
298  * Destroy a disconnected socket.  This routine is a NOP if entities
299  * still have a reference on the socket:
300  *
301  *	so_pcb -	The protocol stack still has a reference
302  *	SS_NOFDREF -	There is no longer a file pointer reference
303  *	SS_ABORTING -	An abort netmsg is in-flight
304  */
305 void
306 sofree(struct socket *so)
307 {
308 	struct socket *head = so->so_head;
309 
310 	if (so->so_pcb || (so->so_state & SS_NOFDREF) == 0)
311 		return;
312 	if (so->so_state & SS_ABORTING)
313 		return;
314 	if (head != NULL) {
315 		if (so->so_state & SS_INCOMP) {
316 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
317 			head->so_incqlen--;
318 		} else if (so->so_state & SS_COMP) {
319 			/*
320 			 * We must not decommission a socket that's
321 			 * on the accept(2) queue.  If we do, then
322 			 * accept(2) may hang after select(2) indicated
323 			 * that the listening socket was ready.
324 			 */
325 			return;
326 		} else {
327 			panic("sofree: not queued");
328 		}
329 		so->so_state &= ~SS_INCOMP;
330 		so->so_head = NULL;
331 	}
332 	ssb_release(&so->so_snd, so);
333 	sorflush(so);
334 	sodealloc(so);
335 }
336 
337 /*
338  * Close a socket on last file table reference removal.
339  * Initiate disconnect if connected.
340  * Free socket when disconnect complete.
341  */
342 int
343 soclose(struct socket *so, int fflag)
344 {
345 	int error = 0;
346 
347 	crit_enter();
348 	funsetown(so->so_sigio);
349 	if (so->so_pcb == NULL)
350 		goto discard;
351 	if (so->so_state & SS_ISCONNECTED) {
352 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
353 			error = sodisconnect(so);
354 			if (error)
355 				goto drop;
356 		}
357 		if (so->so_options & SO_LINGER) {
358 			if ((so->so_state & SS_ISDISCONNECTING) &&
359 			    (fflag & FNONBLOCK))
360 				goto drop;
361 			while (so->so_state & SS_ISCONNECTED) {
362 				error = tsleep((caddr_t)&so->so_timeo,
363 				    PCATCH, "soclos", so->so_linger * hz);
364 				if (error)
365 					break;
366 			}
367 		}
368 	}
369 drop:
370 	if (so->so_pcb) {
371 		int error2;
372 
373 		error2 = so_pru_detach(so);
374 		if (error == 0)
375 			error = error2;
376 	}
377 discard:
378 	if (so->so_options & SO_ACCEPTCONN) {
379 		struct socket *sp;
380 
381 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
382 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
383 			sp->so_state &= ~SS_INCOMP;
384 			sp->so_head = NULL;
385 			so->so_incqlen--;
386 			soaborta(sp);
387 		}
388 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
389 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
390 			sp->so_state &= ~SS_COMP;
391 			sp->so_head = NULL;
392 			so->so_qlen--;
393 			soaborta(sp);
394 		}
395 	}
396 	if (so->so_state & SS_NOFDREF)
397 		panic("soclose: NOFDREF");
398 	so->so_state |= SS_NOFDREF;
399 	sofree(so);
400 	crit_exit();
401 	return (error);
402 }
403 
404 /*
405  * Abort and destroy a socket.  Only one abort can be in progress
406  * at any given moment.
407  */
408 void
409 soabort(struct socket *so)
410 {
411 	if ((so->so_state & SS_ABORTING) == 0) {
412 		so->so_state |= SS_ABORTING;
413 		so_pru_abort(so);
414 	}
415 }
416 
417 void
418 soaborta(struct socket *so)
419 {
420 	if ((so->so_state & SS_ABORTING) == 0) {
421 		so->so_state |= SS_ABORTING;
422 		so_pru_aborta(so);
423 	}
424 }
425 
426 void
427 soabort_oncpu(struct socket *so)
428 {
429 	if ((so->so_state & SS_ABORTING) == 0) {
430 		so->so_state |= SS_ABORTING;
431 		so_pru_abort_oncpu(so);
432 	}
433 }
434 
435 int
436 soaccept(struct socket *so, struct sockaddr **nam)
437 {
438 	int error;
439 
440 	crit_enter();
441 	if ((so->so_state & SS_NOFDREF) == 0)
442 		panic("soaccept: !NOFDREF");
443 	so->so_state &= ~SS_NOFDREF;
444 	error = so_pru_accept(so, nam);
445 	crit_exit();
446 	return (error);
447 }
448 
449 int
450 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
451 {
452 	int error;
453 
454 	if (so->so_options & SO_ACCEPTCONN)
455 		return (EOPNOTSUPP);
456 	crit_enter();
457 	/*
458 	 * If protocol is connection-based, can only connect once.
459 	 * Otherwise, if connected, try to disconnect first.
460 	 * This allows user to disconnect by connecting to, e.g.,
461 	 * a null address.
462 	 */
463 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
464 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
465 	    (error = sodisconnect(so)))) {
466 		error = EISCONN;
467 	} else {
468 		/*
469 		 * Prevent accumulated error from previous connection
470 		 * from biting us.
471 		 */
472 		so->so_error = 0;
473 		error = so_pru_connect(so, nam, td);
474 	}
475 	crit_exit();
476 	return (error);
477 }
478 
479 int
480 soconnect2(struct socket *so1, struct socket *so2)
481 {
482 	int error;
483 
484 	crit_enter();
485 	error = so_pru_connect2(so1, so2);
486 	crit_exit();
487 	return (error);
488 }
489 
490 int
491 sodisconnect(struct socket *so)
492 {
493 	int error;
494 
495 	crit_enter();
496 	if ((so->so_state & SS_ISCONNECTED) == 0) {
497 		error = ENOTCONN;
498 		goto bad;
499 	}
500 	if (so->so_state & SS_ISDISCONNECTING) {
501 		error = EALREADY;
502 		goto bad;
503 	}
504 	error = so_pru_disconnect(so);
505 bad:
506 	crit_exit();
507 	return (error);
508 }
509 
510 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
511 /*
512  * Send on a socket.
513  * If send must go all at once and message is larger than
514  * send buffering, then hard error.
515  * Lock against other senders.
516  * If must go all at once and not enough room now, then
517  * inform user that this would block and do nothing.
518  * Otherwise, if nonblocking, send as much as possible.
519  * The data to be sent is described by "uio" if nonzero,
520  * otherwise by the mbuf chain "top" (which must be null
521  * if uio is not).  Data provided in mbuf chain must be small
522  * enough to send all at once.
523  *
524  * Returns nonzero on error, timeout or signal; callers
525  * must check for short counts if EINTR/ERESTART are returned.
526  * Data and control buffers are freed on return.
527  */
528 int
529 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
530 	struct mbuf *top, struct mbuf *control, int flags,
531 	struct thread *td)
532 {
533 	struct mbuf **mp;
534 	struct mbuf *m;
535 	size_t resid;
536 	int space, len;
537 	int clen = 0, error, dontroute, mlen;
538 	int atomic = sosendallatonce(so) || top;
539 	int pru_flags;
540 
541 	if (uio)
542 		resid = uio->uio_resid;
543 	else
544 		resid = (size_t)top->m_pkthdr.len;
545 
546 	/*
547 	 * WARNING!  resid is unsigned, space and len are signed.  space
548 	 * 	     can wind up negative if the sockbuf is overcommitted.
549 	 *
550 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
551 	 * type sockets since that's an error.
552 	 */
553 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
554 		error = EINVAL;
555 		goto out;
556 	}
557 
558 	dontroute =
559 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
560 	    (so->so_proto->pr_flags & PR_ATOMIC);
561 	if (td->td_lwp != NULL)
562 		td->td_lwp->lwp_ru.ru_msgsnd++;
563 	if (control)
564 		clen = control->m_len;
565 #define	gotoerr(errcode)	{ error = errcode; crit_exit(); goto release; }
566 
567 restart:
568 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
569 	if (error)
570 		goto out;
571 
572 	do {
573 		crit_enter();
574 		if (so->so_state & SS_CANTSENDMORE)
575 			gotoerr(EPIPE);
576 		if (so->so_error) {
577 			error = so->so_error;
578 			so->so_error = 0;
579 			crit_exit();
580 			goto release;
581 		}
582 		if ((so->so_state & SS_ISCONNECTED) == 0) {
583 			/*
584 			 * `sendto' and `sendmsg' is allowed on a connection-
585 			 * based socket if it supports implied connect.
586 			 * Return ENOTCONN if not connected and no address is
587 			 * supplied.
588 			 */
589 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
590 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
591 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
592 				    !(resid == 0 && clen != 0))
593 					gotoerr(ENOTCONN);
594 			} else if (addr == 0)
595 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
596 				   ENOTCONN : EDESTADDRREQ);
597 		}
598 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
599 		    clen > so->so_snd.ssb_hiwat) {
600 			gotoerr(EMSGSIZE);
601 		}
602 		space = ssb_space(&so->so_snd);
603 		if (flags & MSG_OOB)
604 			space += 1024;
605 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
606 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
607 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
608 				gotoerr(EWOULDBLOCK);
609 			ssb_unlock(&so->so_snd);
610 			error = ssb_wait(&so->so_snd);
611 			crit_exit();
612 			if (error)
613 				goto out;
614 			goto restart;
615 		}
616 		crit_exit();
617 		mp = &top;
618 		space -= clen;
619 		do {
620 		    if (uio == NULL) {
621 			/*
622 			 * Data is prepackaged in "top".
623 			 */
624 			resid = 0;
625 			if (flags & MSG_EOR)
626 				top->m_flags |= M_EOR;
627 		    } else do {
628 			if (resid > INT_MAX)
629 				resid = INT_MAX;
630 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
631 				   top == NULL ? M_PKTHDR : 0, &mlen);
632 			if (top == NULL) {
633 				m->m_pkthdr.len = 0;
634 				m->m_pkthdr.rcvif = NULL;
635 			}
636 			len = imin((int)szmin(mlen, resid), space);
637 			if (resid < MINCLSIZE) {
638 				/*
639 				 * For datagram protocols, leave room
640 				 * for protocol headers in first mbuf.
641 				 */
642 				if (atomic && top == 0 && len < mlen)
643 					MH_ALIGN(m, len);
644 			}
645 			space -= len;
646 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
647 			resid = uio->uio_resid;
648 			m->m_len = len;
649 			*mp = m;
650 			top->m_pkthdr.len += len;
651 			if (error)
652 				goto release;
653 			mp = &m->m_next;
654 			if (resid == 0) {
655 				if (flags & MSG_EOR)
656 					top->m_flags |= M_EOR;
657 				break;
658 			}
659 		    } while (space > 0 && atomic);
660 		    if (dontroute)
661 			    so->so_options |= SO_DONTROUTE;
662 		    if (flags & MSG_OOB) {
663 		    	    pru_flags = PRUS_OOB;
664 		    } else if ((flags & MSG_EOF) &&
665 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
666 			       (resid == 0)) {
667 			    /*
668 			     * If the user set MSG_EOF, the protocol
669 			     * understands this flag and nothing left to
670 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
671 			     */
672 		    	    pru_flags = PRUS_EOF;
673 		    } else if (resid > 0 && space > 0) {
674 			    /* If there is more to send, set PRUS_MORETOCOME */
675 		    	    pru_flags = PRUS_MORETOCOME;
676 		    } else {
677 		    	    pru_flags = 0;
678 		    }
679 		    crit_enter();
680 		    /*
681 		     * XXX all the SS_CANTSENDMORE checks previously
682 		     * done could be out of date.  We could have recieved
683 		     * a reset packet in an interrupt or maybe we slept
684 		     * while doing page faults in uiomove() etc. We could
685 		     * probably recheck again inside the splnet() protection
686 		     * here, but there are probably other places that this
687 		     * also happens.  We must rethink this.
688 		     */
689 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
690 		    crit_exit();
691 		    if (dontroute)
692 			    so->so_options &= ~SO_DONTROUTE;
693 		    clen = 0;
694 		    control = 0;
695 		    top = 0;
696 		    mp = &top;
697 		    if (error)
698 			    goto release;
699 		} while (resid && space > 0);
700 	} while (resid);
701 
702 release:
703 	ssb_unlock(&so->so_snd);
704 out:
705 	if (top)
706 		m_freem(top);
707 	if (control)
708 		m_freem(control);
709 	return (error);
710 }
711 
712 /*
713  * A specialization of sosend() for UDP based on protocol-specific knowledge:
714  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
715  *	sosendallatonce() returns true,
716  *	the "atomic" variable is true,
717  *	and sosendudp() blocks until space is available for the entire send.
718  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
719  *	PR_IMPLOPCL flags set.
720  *   UDP has no out-of-band data.
721  *   UDP has no control data.
722  *   UDP does not support MSG_EOR.
723  */
724 int
725 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
726 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
727 {
728 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
729 	size_t resid;
730 	int error;
731 	int space;
732 
733 	if (td->td_lwp != NULL)
734 		td->td_lwp->lwp_ru.ru_msgsnd++;
735 	if (control)
736 		m_freem(control);
737 
738 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
739 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
740 
741 restart:
742 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
743 	if (error)
744 		goto out;
745 
746 	crit_enter();
747 	if (so->so_state & SS_CANTSENDMORE)
748 		gotoerr(EPIPE);
749 	if (so->so_error) {
750 		error = so->so_error;
751 		so->so_error = 0;
752 		crit_exit();
753 		goto release;
754 	}
755 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
756 		gotoerr(EDESTADDRREQ);
757 	if (resid > so->so_snd.ssb_hiwat)
758 		gotoerr(EMSGSIZE);
759 	space = ssb_space(&so->so_snd);
760 	if (uio && (space < 0 || (size_t)space < resid)) {
761 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
762 			gotoerr(EWOULDBLOCK);
763 		ssb_unlock(&so->so_snd);
764 		error = ssb_wait(&so->so_snd);
765 		crit_exit();
766 		if (error)
767 			goto out;
768 		goto restart;
769 	}
770 	crit_exit();
771 
772 	if (uio) {
773 		top = m_uiomove(uio);
774 		if (top == NULL)
775 			goto release;
776 	}
777 
778 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
779 	if (dontroute)
780 		so->so_options |= SO_DONTROUTE;
781 
782 	error = so_pru_send(so, 0, top, addr, NULL, td);
783 	top = NULL;		/* sent or freed in lower layer */
784 
785 	if (dontroute)
786 		so->so_options &= ~SO_DONTROUTE;
787 
788 release:
789 	ssb_unlock(&so->so_snd);
790 out:
791 	if (top)
792 		m_freem(top);
793 	return (error);
794 }
795 
796 /*
797  * Implement receive operations on a socket.
798  * We depend on the way that records are added to the signalsockbuf
799  * by sbappend*.  In particular, each record (mbufs linked through m_next)
800  * must begin with an address if the protocol so specifies,
801  * followed by an optional mbuf or mbufs containing ancillary data,
802  * and then zero or more mbufs of data.
803  * In order to avoid blocking network interrupts for the entire time here,
804  * we exit the critical section while doing the actual copy to user space.
805  * Although the signalsockbuf is locked, new data may still be appended,
806  * and thus we must maintain consistency of the signalsockbuf during that time.
807  *
808  * The caller may receive the data as a single mbuf chain by supplying
809  * an mbuf **mp0 for use in returning the chain.  The uio is then used
810  * only for the count in uio_resid.
811  */
812 int
813 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
814 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
815 {
816 	struct mbuf *m, *n;
817 	struct mbuf *free_chain = NULL;
818 	int flags, len, error, offset;
819 	struct protosw *pr = so->so_proto;
820 	int moff, type = 0;
821 	size_t resid, orig_resid;
822 
823 	if (uio)
824 		resid = uio->uio_resid;
825 	else
826 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
827 	orig_resid = resid;
828 
829 	if (psa)
830 		*psa = NULL;
831 	if (controlp)
832 		*controlp = NULL;
833 	if (flagsp)
834 		flags = *flagsp &~ MSG_EOR;
835 	else
836 		flags = 0;
837 	if (flags & MSG_OOB) {
838 		m = m_get(MB_WAIT, MT_DATA);
839 		if (m == NULL)
840 			return (ENOBUFS);
841 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
842 		if (error)
843 			goto bad;
844 		if (sio) {
845 			do {
846 				sbappend(sio, m);
847 				KKASSERT(resid >= (size_t)m->m_len);
848 				resid -= (size_t)m->m_len;
849 			} while (resid > 0 && m);
850 		} else {
851 			do {
852 				uio->uio_resid = resid;
853 				error = uiomove(mtod(m, caddr_t),
854 						(int)szmin(resid, m->m_len),
855 						uio);
856 				resid = uio->uio_resid;
857 				m = m_free(m);
858 			} while (uio->uio_resid && error == 0 && m);
859 		}
860 bad:
861 		if (m)
862 			m_freem(m);
863 		return (error);
864 	}
865 	if ((so->so_state & SS_ISCONFIRMING) && resid)
866 		so_pru_rcvd(so, 0);
867 
868 restart:
869 	crit_enter();
870 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
871 	if (error)
872 		goto done;
873 
874 	m = so->so_rcv.ssb_mb;
875 	/*
876 	 * If we have less data than requested, block awaiting more
877 	 * (subject to any timeout) if:
878 	 *   1. the current count is less than the low water mark, or
879 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
880 	 *	receive operation at once if we block (resid <= hiwat).
881 	 *   3. MSG_DONTWAIT is not set
882 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
883 	 * we have to do the receive in sections, and thus risk returning
884 	 * a short count if a timeout or signal occurs after we start.
885 	 */
886 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
887 	    (size_t)so->so_rcv.ssb_cc < resid) &&
888 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
889 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
890 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
891 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
892 		if (so->so_error) {
893 			if (m)
894 				goto dontblock;
895 			error = so->so_error;
896 			if ((flags & MSG_PEEK) == 0)
897 				so->so_error = 0;
898 			goto release;
899 		}
900 		if (so->so_state & SS_CANTRCVMORE) {
901 			if (m)
902 				goto dontblock;
903 			else
904 				goto release;
905 		}
906 		for (; m; m = m->m_next) {
907 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
908 				m = so->so_rcv.ssb_mb;
909 				goto dontblock;
910 			}
911 		}
912 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
913 		    (pr->pr_flags & PR_CONNREQUIRED)) {
914 			error = ENOTCONN;
915 			goto release;
916 		}
917 		if (resid == 0)
918 			goto release;
919 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
920 			error = EWOULDBLOCK;
921 			goto release;
922 		}
923 		ssb_unlock(&so->so_rcv);
924 		error = ssb_wait(&so->so_rcv);
925 		if (error)
926 			goto done;
927 		crit_exit();
928 		goto restart;
929 	}
930 dontblock:
931 	if (uio && uio->uio_td && uio->uio_td->td_proc)
932 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
933 
934 	/*
935 	 * note: m should be == sb_mb here.  Cache the next record while
936 	 * cleaning up.  Note that calling m_free*() will break out critical
937 	 * section.
938 	 */
939 	KKASSERT(m == so->so_rcv.ssb_mb);
940 
941 	/*
942 	 * Skip any address mbufs prepending the record.
943 	 */
944 	if (pr->pr_flags & PR_ADDR) {
945 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
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 	/*
956 	 * Skip any control mbufs prepending the record.
957 	 */
958 #ifdef SCTP
959 	if (pr->pr_flags & PR_ADDR_OPT) {
960 		/*
961 		 * For SCTP we may be getting a
962 		 * whole message OR a partial delivery.
963 		 */
964 		if (m && m->m_type == MT_SONAME) {
965 			orig_resid = 0;
966 			if (psa)
967 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
968 			if (flags & MSG_PEEK)
969 				m = m->m_next;
970 			else
971 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
972 		}
973 	}
974 #endif /* SCTP */
975 	while (m && m->m_type == MT_CONTROL && error == 0) {
976 		if (flags & MSG_PEEK) {
977 			if (controlp)
978 				*controlp = m_copy(m, 0, m->m_len);
979 			m = m->m_next;	/* XXX race */
980 		} else {
981 			if (controlp) {
982 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
983 				if (pr->pr_domain->dom_externalize &&
984 				    mtod(m, struct cmsghdr *)->cmsg_type ==
985 				    SCM_RIGHTS)
986 				   error = (*pr->pr_domain->dom_externalize)(m);
987 				*controlp = m;
988 				m = n;
989 			} else {
990 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
991 			}
992 		}
993 		if (controlp && *controlp) {
994 			orig_resid = 0;
995 			controlp = &(*controlp)->m_next;
996 		}
997 	}
998 
999 	/*
1000 	 * flag OOB data.
1001 	 */
1002 	if (m) {
1003 		type = m->m_type;
1004 		if (type == MT_OOBDATA)
1005 			flags |= MSG_OOB;
1006 	}
1007 
1008 	/*
1009 	 * Copy to the UIO or mbuf return chain (*mp).
1010 	 */
1011 	moff = 0;
1012 	offset = 0;
1013 	while (m && resid > 0 && error == 0) {
1014 		if (m->m_type == MT_OOBDATA) {
1015 			if (type != MT_OOBDATA)
1016 				break;
1017 		} else if (type == MT_OOBDATA)
1018 			break;
1019 		else
1020 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1021 			("receive 3"));
1022 		so->so_state &= ~SS_RCVATMARK;
1023 		len = (resid > INT_MAX) ? INT_MAX : resid;
1024 		if (so->so_oobmark && len > so->so_oobmark - offset)
1025 			len = so->so_oobmark - offset;
1026 		if (len > m->m_len - moff)
1027 			len = m->m_len - moff;
1028 
1029 		/*
1030 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1031 		 * The SIO is dealt with when we eat the mbuf, but deal
1032 		 * with the resid here either way.
1033 		 */
1034 		if (uio) {
1035 			crit_exit();
1036 			uio->uio_resid = resid;
1037 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1038 			resid = uio->uio_resid;
1039 			crit_enter();
1040 			if (error)
1041 				goto release;
1042 		} else {
1043 			resid -= (size_t)len;
1044 		}
1045 
1046 		/*
1047 		 * Eat the entire mbuf or just a piece of it
1048 		 */
1049 		if (len == m->m_len - moff) {
1050 			if (m->m_flags & M_EOR)
1051 				flags |= MSG_EOR;
1052 #ifdef SCTP
1053 			if (m->m_flags & M_NOTIFICATION)
1054 				flags |= MSG_NOTIFICATION;
1055 #endif /* SCTP */
1056 			if (flags & MSG_PEEK) {
1057 				m = m->m_next;
1058 				moff = 0;
1059 			} else {
1060 				if (sio) {
1061 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1062 					sbappend(sio, m);
1063 					m = n;
1064 				} else {
1065 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1066 				}
1067 			}
1068 		} else {
1069 			if (flags & MSG_PEEK) {
1070 				moff += len;
1071 			} else {
1072 				if (sio) {
1073 					n = m_copym(m, 0, len, MB_WAIT);
1074 					if (n)
1075 						sbappend(sio, n);
1076 				}
1077 				m->m_data += len;
1078 				m->m_len -= len;
1079 				so->so_rcv.ssb_cc -= len;
1080 			}
1081 		}
1082 		if (so->so_oobmark) {
1083 			if ((flags & MSG_PEEK) == 0) {
1084 				so->so_oobmark -= len;
1085 				if (so->so_oobmark == 0) {
1086 					so->so_state |= SS_RCVATMARK;
1087 					break;
1088 				}
1089 			} else {
1090 				offset += len;
1091 				if (offset == so->so_oobmark)
1092 					break;
1093 			}
1094 		}
1095 		if (flags & MSG_EOR)
1096 			break;
1097 		/*
1098 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1099 		 * we must not quit until resid == 0 or an error
1100 		 * termination.  If a signal/timeout occurs, return
1101 		 * with a short count but without error.
1102 		 * Keep signalsockbuf locked against other readers.
1103 		 */
1104 		while ((flags & MSG_WAITALL) && m == NULL &&
1105 		       resid > 0 && !sosendallatonce(so) &&
1106 		       so->so_rcv.ssb_mb == NULL) {
1107 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1108 				break;
1109 			/*
1110 			 * The window might have closed to zero, make
1111 			 * sure we send an ack now that we've drained
1112 			 * the buffer or we might end up blocking until
1113 			 * the idle takes over (5 seconds).
1114 			 */
1115 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1116 				so_pru_rcvd(so, flags);
1117 			error = ssb_wait(&so->so_rcv);
1118 			if (error) {
1119 				ssb_unlock(&so->so_rcv);
1120 				error = 0;
1121 				goto done;
1122 			}
1123 			m = so->so_rcv.ssb_mb;
1124 		}
1125 	}
1126 
1127 	/*
1128 	 * If an atomic read was requested but unread data still remains
1129 	 * in the record, set MSG_TRUNC.
1130 	 */
1131 	if (m && pr->pr_flags & PR_ATOMIC)
1132 		flags |= MSG_TRUNC;
1133 
1134 	/*
1135 	 * Cleanup.  If an atomic read was requested drop any unread data.
1136 	 */
1137 	if ((flags & MSG_PEEK) == 0) {
1138 		if (m && (pr->pr_flags & PR_ATOMIC))
1139 			sbdroprecord(&so->so_rcv.sb);
1140 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1141 			so_pru_rcvd(so, flags);
1142 	}
1143 
1144 	if (orig_resid == resid && orig_resid &&
1145 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1146 		ssb_unlock(&so->so_rcv);
1147 		crit_exit();
1148 		goto restart;
1149 	}
1150 
1151 	if (flagsp)
1152 		*flagsp |= flags;
1153 release:
1154 	ssb_unlock(&so->so_rcv);
1155 done:
1156 	crit_exit();
1157 	if (free_chain)
1158 		m_freem(free_chain);
1159 	return (error);
1160 }
1161 
1162 int
1163 soshutdown(struct socket *so, int how)
1164 {
1165 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1166 		return (EINVAL);
1167 
1168 	if (how != SHUT_WR)
1169 		sorflush(so);
1170 	if (how != SHUT_RD)
1171 		return (so_pru_shutdown(so));
1172 	return (0);
1173 }
1174 
1175 void
1176 sorflush(struct socket *so)
1177 {
1178 	struct signalsockbuf *ssb = &so->so_rcv;
1179 	struct protosw *pr = so->so_proto;
1180 	struct signalsockbuf asb;
1181 
1182 	ssb->ssb_flags |= SSB_NOINTR;
1183 	(void) ssb_lock(ssb, M_WAITOK);
1184 
1185 	crit_enter();
1186 	socantrcvmore(so);
1187 	ssb_unlock(ssb);
1188 	asb = *ssb;
1189 	bzero((caddr_t)ssb, sizeof (*ssb));
1190 	if (asb.ssb_flags & SSB_KNOTE) {
1191 		ssb->ssb_sel.si_note = asb.ssb_sel.si_note;
1192 		ssb->ssb_flags = SSB_KNOTE;
1193 	}
1194 	crit_exit();
1195 
1196 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose)
1197 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1198 	ssb_release(&asb, so);
1199 }
1200 
1201 #ifdef INET
1202 static int
1203 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1204 {
1205 	struct accept_filter_arg	*afap = NULL;
1206 	struct accept_filter	*afp;
1207 	struct so_accf	*af = so->so_accf;
1208 	int	error = 0;
1209 
1210 	/* do not set/remove accept filters on non listen sockets */
1211 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1212 		error = EINVAL;
1213 		goto out;
1214 	}
1215 
1216 	/* removing the filter */
1217 	if (sopt == NULL) {
1218 		if (af != NULL) {
1219 			if (af->so_accept_filter != NULL &&
1220 				af->so_accept_filter->accf_destroy != NULL) {
1221 				af->so_accept_filter->accf_destroy(so);
1222 			}
1223 			if (af->so_accept_filter_str != NULL) {
1224 				FREE(af->so_accept_filter_str, M_ACCF);
1225 			}
1226 			FREE(af, M_ACCF);
1227 			so->so_accf = NULL;
1228 		}
1229 		so->so_options &= ~SO_ACCEPTFILTER;
1230 		return (0);
1231 	}
1232 	/* adding a filter */
1233 	/* must remove previous filter first */
1234 	if (af != NULL) {
1235 		error = EINVAL;
1236 		goto out;
1237 	}
1238 	/* don't put large objects on the kernel stack */
1239 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1240 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1241 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1242 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1243 	if (error)
1244 		goto out;
1245 	afp = accept_filt_get(afap->af_name);
1246 	if (afp == NULL) {
1247 		error = ENOENT;
1248 		goto out;
1249 	}
1250 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1251 	if (afp->accf_create != NULL) {
1252 		if (afap->af_name[0] != '\0') {
1253 			int len = strlen(afap->af_name) + 1;
1254 
1255 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1256 			strcpy(af->so_accept_filter_str, afap->af_name);
1257 		}
1258 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1259 		if (af->so_accept_filter_arg == NULL) {
1260 			FREE(af->so_accept_filter_str, M_ACCF);
1261 			FREE(af, M_ACCF);
1262 			so->so_accf = NULL;
1263 			error = EINVAL;
1264 			goto out;
1265 		}
1266 	}
1267 	af->so_accept_filter = afp;
1268 	so->so_accf = af;
1269 	so->so_options |= SO_ACCEPTFILTER;
1270 out:
1271 	if (afap != NULL)
1272 		FREE(afap, M_TEMP);
1273 	return (error);
1274 }
1275 #endif /* INET */
1276 
1277 /*
1278  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1279  * an additional variant to handle the case where the option value needs
1280  * to be some kind of integer, but not a specific size.
1281  * In addition to their use here, these functions are also called by the
1282  * protocol-level pr_ctloutput() routines.
1283  */
1284 int
1285 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1286 {
1287 	return soopt_to_kbuf(sopt, buf, len, minlen);
1288 }
1289 
1290 int
1291 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1292 {
1293 	size_t	valsize;
1294 
1295 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1296 	KKASSERT(kva_p(buf));
1297 
1298 	/*
1299 	 * If the user gives us more than we wanted, we ignore it,
1300 	 * but if we don't get the minimum length the caller
1301 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1302 	 * is set to however much we actually retrieved.
1303 	 */
1304 	if ((valsize = sopt->sopt_valsize) < minlen)
1305 		return EINVAL;
1306 	if (valsize > len)
1307 		sopt->sopt_valsize = valsize = len;
1308 
1309 	bcopy(sopt->sopt_val, buf, valsize);
1310 	return 0;
1311 }
1312 
1313 
1314 int
1315 sosetopt(struct socket *so, struct sockopt *sopt)
1316 {
1317 	int	error, optval;
1318 	struct	linger l;
1319 	struct	timeval tv;
1320 	u_long  val;
1321 
1322 	error = 0;
1323 	sopt->sopt_dir = SOPT_SET;
1324 	if (sopt->sopt_level != SOL_SOCKET) {
1325 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1326 			return (so_pru_ctloutput(so, sopt));
1327 		}
1328 		error = ENOPROTOOPT;
1329 	} else {
1330 		switch (sopt->sopt_name) {
1331 #ifdef INET
1332 		case SO_ACCEPTFILTER:
1333 			error = do_setopt_accept_filter(so, sopt);
1334 			if (error)
1335 				goto bad;
1336 			break;
1337 #endif /* INET */
1338 		case SO_LINGER:
1339 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1340 			if (error)
1341 				goto bad;
1342 
1343 			so->so_linger = l.l_linger;
1344 			if (l.l_onoff)
1345 				so->so_options |= SO_LINGER;
1346 			else
1347 				so->so_options &= ~SO_LINGER;
1348 			break;
1349 
1350 		case SO_DEBUG:
1351 		case SO_KEEPALIVE:
1352 		case SO_DONTROUTE:
1353 		case SO_USELOOPBACK:
1354 		case SO_BROADCAST:
1355 		case SO_REUSEADDR:
1356 		case SO_REUSEPORT:
1357 		case SO_OOBINLINE:
1358 		case SO_TIMESTAMP:
1359 			error = sooptcopyin(sopt, &optval, sizeof optval,
1360 					    sizeof optval);
1361 			if (error)
1362 				goto bad;
1363 			if (optval)
1364 				so->so_options |= sopt->sopt_name;
1365 			else
1366 				so->so_options &= ~sopt->sopt_name;
1367 			break;
1368 
1369 		case SO_SNDBUF:
1370 		case SO_RCVBUF:
1371 		case SO_SNDLOWAT:
1372 		case SO_RCVLOWAT:
1373 			error = sooptcopyin(sopt, &optval, sizeof optval,
1374 					    sizeof optval);
1375 			if (error)
1376 				goto bad;
1377 
1378 			/*
1379 			 * Values < 1 make no sense for any of these
1380 			 * options, so disallow them.
1381 			 */
1382 			if (optval < 1) {
1383 				error = EINVAL;
1384 				goto bad;
1385 			}
1386 
1387 			switch (sopt->sopt_name) {
1388 			case SO_SNDBUF:
1389 			case SO_RCVBUF:
1390 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1391 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1392 				    so,
1393 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1394 					error = ENOBUFS;
1395 					goto bad;
1396 				}
1397 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
1398 				    &so->so_rcv)->ssb_flags &= ~SSB_AUTOSIZE;
1399 				break;
1400 
1401 			/*
1402 			 * Make sure the low-water is never greater than
1403 			 * the high-water.
1404 			 */
1405 			case SO_SNDLOWAT:
1406 				so->so_snd.ssb_lowat =
1407 				    (optval > so->so_snd.ssb_hiwat) ?
1408 				    so->so_snd.ssb_hiwat : optval;
1409 				so->so_snd.ssb_flags &= ~SSB_AUTOLOWAT;
1410 				break;
1411 			case SO_RCVLOWAT:
1412 				so->so_rcv.ssb_lowat =
1413 				    (optval > so->so_rcv.ssb_hiwat) ?
1414 				    so->so_rcv.ssb_hiwat : optval;
1415 				so->so_rcv.ssb_flags &= ~SSB_AUTOLOWAT;
1416 				break;
1417 			}
1418 			break;
1419 
1420 		case SO_SNDTIMEO:
1421 		case SO_RCVTIMEO:
1422 			error = sooptcopyin(sopt, &tv, sizeof tv,
1423 					    sizeof tv);
1424 			if (error)
1425 				goto bad;
1426 
1427 			/* assert(hz > 0); */
1428 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1429 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1430 				error = EDOM;
1431 				goto bad;
1432 			}
1433 			/* assert(tick > 0); */
1434 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1435 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1436 			if (val > SHRT_MAX) {
1437 				error = EDOM;
1438 				goto bad;
1439 			}
1440 			if (val == 0 && tv.tv_usec != 0)
1441 				val = 1;
1442 
1443 			switch (sopt->sopt_name) {
1444 			case SO_SNDTIMEO:
1445 				so->so_snd.ssb_timeo = val;
1446 				break;
1447 			case SO_RCVTIMEO:
1448 				so->so_rcv.ssb_timeo = val;
1449 				break;
1450 			}
1451 			break;
1452 		default:
1453 			error = ENOPROTOOPT;
1454 			break;
1455 		}
1456 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1457 			(void) so_pru_ctloutput(so, sopt);
1458 		}
1459 	}
1460 bad:
1461 	return (error);
1462 }
1463 
1464 /* Helper routine for getsockopt */
1465 int
1466 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1467 {
1468 	soopt_from_kbuf(sopt, buf, len);
1469 	return 0;
1470 }
1471 
1472 void
1473 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1474 {
1475 	size_t	valsize;
1476 
1477 	if (len == 0) {
1478 		sopt->sopt_valsize = 0;
1479 		return;
1480 	}
1481 
1482 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1483 	KKASSERT(kva_p(buf));
1484 
1485 	/*
1486 	 * Documented get behavior is that we always return a value,
1487 	 * possibly truncated to fit in the user's buffer.
1488 	 * Traditional behavior is that we always tell the user
1489 	 * precisely how much we copied, rather than something useful
1490 	 * like the total amount we had available for her.
1491 	 * Note that this interface is not idempotent; the entire answer must
1492 	 * generated ahead of time.
1493 	 */
1494 	valsize = szmin(len, sopt->sopt_valsize);
1495 	sopt->sopt_valsize = valsize;
1496 	if (sopt->sopt_val != 0) {
1497 		bcopy(buf, sopt->sopt_val, valsize);
1498 	}
1499 }
1500 
1501 int
1502 sogetopt(struct socket *so, struct sockopt *sopt)
1503 {
1504 	int	error, optval;
1505 	struct	linger l;
1506 	struct	timeval tv;
1507 #ifdef INET
1508 	struct accept_filter_arg *afap;
1509 #endif
1510 
1511 	error = 0;
1512 	sopt->sopt_dir = SOPT_GET;
1513 	if (sopt->sopt_level != SOL_SOCKET) {
1514 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1515 			return (so_pru_ctloutput(so, sopt));
1516 		} else
1517 			return (ENOPROTOOPT);
1518 	} else {
1519 		switch (sopt->sopt_name) {
1520 #ifdef INET
1521 		case SO_ACCEPTFILTER:
1522 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1523 				return (EINVAL);
1524 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1525 				M_TEMP, M_WAITOK | M_ZERO);
1526 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1527 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1528 				if (so->so_accf->so_accept_filter_str != NULL)
1529 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1530 			}
1531 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1532 			FREE(afap, M_TEMP);
1533 			break;
1534 #endif /* INET */
1535 
1536 		case SO_LINGER:
1537 			l.l_onoff = so->so_options & SO_LINGER;
1538 			l.l_linger = so->so_linger;
1539 			error = sooptcopyout(sopt, &l, sizeof l);
1540 			break;
1541 
1542 		case SO_USELOOPBACK:
1543 		case SO_DONTROUTE:
1544 		case SO_DEBUG:
1545 		case SO_KEEPALIVE:
1546 		case SO_REUSEADDR:
1547 		case SO_REUSEPORT:
1548 		case SO_BROADCAST:
1549 		case SO_OOBINLINE:
1550 		case SO_TIMESTAMP:
1551 			optval = so->so_options & sopt->sopt_name;
1552 integer:
1553 			error = sooptcopyout(sopt, &optval, sizeof optval);
1554 			break;
1555 
1556 		case SO_TYPE:
1557 			optval = so->so_type;
1558 			goto integer;
1559 
1560 		case SO_ERROR:
1561 			optval = so->so_error;
1562 			so->so_error = 0;
1563 			goto integer;
1564 
1565 		case SO_SNDBUF:
1566 			optval = so->so_snd.ssb_hiwat;
1567 			goto integer;
1568 
1569 		case SO_RCVBUF:
1570 			optval = so->so_rcv.ssb_hiwat;
1571 			goto integer;
1572 
1573 		case SO_SNDLOWAT:
1574 			optval = so->so_snd.ssb_lowat;
1575 			goto integer;
1576 
1577 		case SO_RCVLOWAT:
1578 			optval = so->so_rcv.ssb_lowat;
1579 			goto integer;
1580 
1581 		case SO_SNDTIMEO:
1582 		case SO_RCVTIMEO:
1583 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1584 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1585 
1586 			tv.tv_sec = optval / hz;
1587 			tv.tv_usec = (optval % hz) * ustick;
1588 			error = sooptcopyout(sopt, &tv, sizeof tv);
1589 			break;
1590 
1591 		default:
1592 			error = ENOPROTOOPT;
1593 			break;
1594 		}
1595 		return (error);
1596 	}
1597 }
1598 
1599 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1600 int
1601 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1602 {
1603 	struct mbuf *m, *m_prev;
1604 	int sopt_size = sopt->sopt_valsize, msize;
1605 
1606 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1607 		   0, &msize);
1608 	if (m == NULL)
1609 		return (ENOBUFS);
1610 	m->m_len = min(msize, sopt_size);
1611 	sopt_size -= m->m_len;
1612 	*mp = m;
1613 	m_prev = m;
1614 
1615 	while (sopt_size > 0) {
1616 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1617 			   MT_DATA, 0, &msize);
1618 		if (m == NULL) {
1619 			m_freem(*mp);
1620 			return (ENOBUFS);
1621 		}
1622 		m->m_len = min(msize, sopt_size);
1623 		sopt_size -= m->m_len;
1624 		m_prev->m_next = m;
1625 		m_prev = m;
1626 	}
1627 	return (0);
1628 }
1629 
1630 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1631 int
1632 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1633 {
1634 	soopt_to_mbuf(sopt, m);
1635 	return 0;
1636 }
1637 
1638 void
1639 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1640 {
1641 	size_t valsize;
1642 	void *val;
1643 
1644 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1645 	KKASSERT(kva_p(m));
1646 	if (sopt->sopt_val == NULL)
1647 		return;
1648 	val = sopt->sopt_val;
1649 	valsize = sopt->sopt_valsize;
1650 	while (m != NULL && valsize >= m->m_len) {
1651 		bcopy(val, mtod(m, char *), m->m_len);
1652 		valsize -= m->m_len;
1653 		val = (caddr_t)val + m->m_len;
1654 		m = m->m_next;
1655 	}
1656 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1657 		panic("ip6_sooptmcopyin");
1658 }
1659 
1660 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1661 int
1662 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1663 {
1664 	return soopt_from_mbuf(sopt, m);
1665 }
1666 
1667 int
1668 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1669 {
1670 	struct mbuf *m0 = m;
1671 	size_t valsize = 0;
1672 	size_t maxsize;
1673 	void *val;
1674 
1675 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1676 	KKASSERT(kva_p(m));
1677 	if (sopt->sopt_val == NULL)
1678 		return 0;
1679 	val = sopt->sopt_val;
1680 	maxsize = sopt->sopt_valsize;
1681 	while (m != NULL && maxsize >= m->m_len) {
1682 		bcopy(mtod(m, char *), val, m->m_len);
1683 	       maxsize -= m->m_len;
1684 	       val = (caddr_t)val + m->m_len;
1685 	       valsize += m->m_len;
1686 	       m = m->m_next;
1687 	}
1688 	if (m != NULL) {
1689 		/* enough soopt buffer should be given from user-land */
1690 		m_freem(m0);
1691 		return (EINVAL);
1692 	}
1693 	sopt->sopt_valsize = valsize;
1694 	return 0;
1695 }
1696 
1697 void
1698 sohasoutofband(struct socket *so)
1699 {
1700 	if (so->so_sigio != NULL)
1701 		pgsigio(so->so_sigio, SIGURG, 0);
1702 	selwakeup(&so->so_rcv.ssb_sel);
1703 	KNOTE(&so->so_rcv.ssb_sel.si_note, NOTE_OOB);
1704 }
1705 
1706 int
1707 sopoll(struct socket *so, int events, struct ucred *cred, struct thread *td)
1708 {
1709 	int revents = 0;
1710 
1711 	crit_enter();
1712 
1713 	if (events & (POLLIN | POLLRDNORM))
1714 		if (soreadable(so))
1715 			revents |= events & (POLLIN | POLLRDNORM);
1716 
1717 	if (events & POLLINIGNEOF)
1718 		if (so->so_rcv.ssb_cc >= so->so_rcv.ssb_lowat ||
1719 			!TAILQ_EMPTY(&so->so_comp) || so->so_error)
1720 			revents |= POLLINIGNEOF;
1721 
1722 	if (events & (POLLOUT | POLLWRNORM))
1723 		if (sowriteable(so))
1724 			revents |= events & (POLLOUT | POLLWRNORM);
1725 
1726 	if (events & (POLLPRI | POLLRDBAND))
1727 		if (so->so_oobmark || (so->so_state & SS_RCVATMARK))
1728 			revents |= events & (POLLPRI | POLLRDBAND);
1729 
1730 	if (revents == 0) {
1731 		if (events &
1732 			(POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
1733 			 POLLRDBAND)) {
1734 			selrecord(td, &so->so_rcv.ssb_sel);
1735 			so->so_rcv.ssb_flags |= SSB_SEL;
1736 		}
1737 
1738 		if (events & (POLLOUT | POLLWRNORM)) {
1739 			selrecord(td, &so->so_snd.ssb_sel);
1740 			so->so_snd.ssb_flags |= SSB_SEL;
1741 		}
1742 	}
1743 
1744 	crit_exit();
1745 	return (revents);
1746 }
1747 
1748 int
1749 sokqfilter(struct file *fp, struct knote *kn)
1750 {
1751 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1752 	struct signalsockbuf *ssb;
1753 
1754 	switch (kn->kn_filter) {
1755 	case EVFILT_READ:
1756 		if (so->so_options & SO_ACCEPTCONN)
1757 			kn->kn_fop = &solisten_filtops;
1758 		else
1759 			kn->kn_fop = &soread_filtops;
1760 		ssb = &so->so_rcv;
1761 		break;
1762 	case EVFILT_WRITE:
1763 		kn->kn_fop = &sowrite_filtops;
1764 		ssb = &so->so_snd;
1765 		break;
1766 	case EVFILT_EXCEPT:
1767 		kn->kn_fop = &soexcept_filtops;
1768 		ssb = &so->so_rcv;
1769 		break;
1770 	default:
1771 		return (1);
1772 	}
1773 
1774 	crit_enter();
1775 	SLIST_INSERT_HEAD(&ssb->ssb_sel.si_note, kn, kn_selnext);
1776 	ssb->ssb_flags |= SSB_KNOTE;
1777 	crit_exit();
1778 	return (0);
1779 }
1780 
1781 static void
1782 filt_sordetach(struct knote *kn)
1783 {
1784 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1785 
1786 	crit_enter();
1787 	SLIST_REMOVE(&so->so_rcv.ssb_sel.si_note, kn, knote, kn_selnext);
1788 	if (SLIST_EMPTY(&so->so_rcv.ssb_sel.si_note))
1789 		so->so_rcv.ssb_flags &= ~SSB_KNOTE;
1790 	crit_exit();
1791 }
1792 
1793 /*ARGSUSED*/
1794 static int
1795 filt_soread(struct knote *kn, long hint)
1796 {
1797 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1798 
1799 	if (kn->kn_sfflags & NOTE_OOB) {
1800 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1801 			kn->kn_fflags |= NOTE_OOB;
1802 			return (1);
1803 		}
1804 		return (0);
1805 	}
1806 	kn->kn_data = so->so_rcv.ssb_cc;
1807 	if (so->so_state & SS_CANTRCVMORE) {
1808 		kn->kn_flags |= EV_EOF;
1809 		kn->kn_fflags = so->so_error;
1810 		return (1);
1811 	}
1812 	if (so->so_error)	/* temporary udp error */
1813 		return (1);
1814 	if (kn->kn_sfflags & NOTE_LOWAT)
1815 		return (kn->kn_data >= kn->kn_sdata);
1816 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1817 	    !TAILQ_EMPTY(&so->so_comp));
1818 }
1819 
1820 static void
1821 filt_sowdetach(struct knote *kn)
1822 {
1823 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1824 
1825 	crit_enter();
1826 	SLIST_REMOVE(&so->so_snd.ssb_sel.si_note, kn, knote, kn_selnext);
1827 	if (SLIST_EMPTY(&so->so_snd.ssb_sel.si_note))
1828 		so->so_snd.ssb_flags &= ~SSB_KNOTE;
1829 	crit_exit();
1830 }
1831 
1832 /*ARGSUSED*/
1833 static int
1834 filt_sowrite(struct knote *kn, long hint)
1835 {
1836 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1837 
1838 	kn->kn_data = ssb_space(&so->so_snd);
1839 	if (so->so_state & SS_CANTSENDMORE) {
1840 		kn->kn_flags |= EV_EOF;
1841 		kn->kn_fflags = so->so_error;
1842 		return (1);
1843 	}
1844 	if (so->so_error)	/* temporary udp error */
1845 		return (1);
1846 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
1847 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
1848 		return (0);
1849 	if (kn->kn_sfflags & NOTE_LOWAT)
1850 		return (kn->kn_data >= kn->kn_sdata);
1851 	return (kn->kn_data >= so->so_snd.ssb_lowat);
1852 }
1853 
1854 /*ARGSUSED*/
1855 static int
1856 filt_solisten(struct knote *kn, long hint)
1857 {
1858 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1859 
1860 	kn->kn_data = so->so_qlen;
1861 	return (! TAILQ_EMPTY(&so->so_comp));
1862 }
1863