xref: /dflybsd-src/sys/kern/uipc_socket.c (revision a563ca70e68142ccf7f50a6f129665fd8cb66d98)
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  */
69 
70 #include "opt_inet.h"
71 #include "opt_sctp.h"
72 
73 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/fcntl.h>
76 #include <sys/malloc.h>
77 #include <sys/mbuf.h>
78 #include <sys/domain.h>
79 #include <sys/file.h>			/* for struct knote */
80 #include <sys/kernel.h>
81 #include <sys/event.h>
82 #include <sys/proc.h>
83 #include <sys/protosw.h>
84 #include <sys/socket.h>
85 #include <sys/socketvar.h>
86 #include <sys/socketops.h>
87 #include <sys/resourcevar.h>
88 #include <sys/signalvar.h>
89 #include <sys/sysctl.h>
90 #include <sys/uio.h>
91 #include <sys/jail.h>
92 #include <vm/vm_zone.h>
93 #include <vm/pmap.h>
94 
95 #include <sys/thread2.h>
96 #include <sys/socketvar2.h>
97 
98 #include <machine/limits.h>
99 
100 extern int tcp_sosnd_agglim;
101 extern int tcp_sosnd_async;
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 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
115 static struct filterops soread_filtops =
116 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
117 static struct filterops sowrite_filtops =
118 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
119 static struct filterops soexcept_filtops =
120 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, 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_kq.ki_mlist);
157 		TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
158 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
159 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
160 		so->so_state = SS_NOFDREF;
161 		so->so_refs = 1;
162 	}
163 	return so;
164 }
165 
166 int
167 socreate(int dom, struct socket **aso, int type,
168 	int proto, struct thread *td)
169 {
170 	struct proc *p = td->td_proc;
171 	struct protosw *prp;
172 	struct socket *so;
173 	struct pru_attach_info ai;
174 	int error;
175 
176 	if (proto)
177 		prp = pffindproto(dom, proto, type);
178 	else
179 		prp = pffindtype(dom, type);
180 
181 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
182 		return (EPROTONOSUPPORT);
183 
184 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
185 	    prp->pr_domain->dom_family != PF_LOCAL &&
186 	    prp->pr_domain->dom_family != PF_INET &&
187 	    prp->pr_domain->dom_family != PF_INET6 &&
188 	    prp->pr_domain->dom_family != PF_ROUTE) {
189 		return (EPROTONOSUPPORT);
190 	}
191 
192 	if (prp->pr_type != type)
193 		return (EPROTOTYPE);
194 	so = soalloc(p != 0);
195 	if (so == NULL)
196 		return (ENOBUFS);
197 
198 	/*
199 	 * Callers of socreate() presumably will connect up a descriptor
200 	 * and call soclose() if they cannot.  This represents our so_refs
201 	 * (which should be 1) from soalloc().
202 	 */
203 	soclrstate(so, SS_NOFDREF);
204 
205 	/*
206 	 * Set a default port for protocol processing.  No action will occur
207 	 * on the socket on this port until an inpcb is attached to it and
208 	 * is able to match incoming packets, or until the socket becomes
209 	 * available to userland.
210 	 *
211 	 * We normally default the socket to the protocol thread on cpu 0.
212 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
213 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
214 	 */
215 	if (prp->pr_flags & PR_SYNC_PORT)
216 		so->so_port = &netisr_sync_port;
217 	else
218 		so->so_port = cpu_portfn(0);
219 
220 	TAILQ_INIT(&so->so_incomp);
221 	TAILQ_INIT(&so->so_comp);
222 	so->so_type = type;
223 	so->so_cred = crhold(p->p_ucred);
224 	so->so_proto = prp;
225 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
226 	ai.p_ucred = p->p_ucred;
227 	ai.fd_rdir = p->p_fd->fd_rdir;
228 
229 	/*
230 	 * Auto-sizing of socket buffers is managed by the protocols and
231 	 * the appropriate flags must be set in the pru_attach function.
232 	 */
233 	error = so_pru_attach(so, proto, &ai);
234 	if (error) {
235 		sosetstate(so, SS_NOFDREF);
236 		sofree(so);	/* from soalloc */
237 		return error;
238 	}
239 
240 	/*
241 	 * NOTE: Returns referenced socket.
242 	 */
243 	*aso = so;
244 	return (0);
245 }
246 
247 int
248 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
249 {
250 	int error;
251 
252 	error = so_pru_bind(so, nam, td);
253 	return (error);
254 }
255 
256 static void
257 sodealloc(struct socket *so)
258 {
259 	if (so->so_rcv.ssb_hiwat)
260 		(void)chgsbsize(so->so_cred->cr_uidinfo,
261 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
262 	if (so->so_snd.ssb_hiwat)
263 		(void)chgsbsize(so->so_cred->cr_uidinfo,
264 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
265 #ifdef INET
266 	/* remove accept filter if present */
267 	if (so->so_accf != NULL)
268 		do_setopt_accept_filter(so, NULL);
269 #endif /* INET */
270 	crfree(so->so_cred);
271 	kfree(so, M_SOCKET);
272 }
273 
274 int
275 solisten(struct socket *so, int backlog, struct thread *td)
276 {
277 	int error;
278 #ifdef SCTP
279 	short oldopt, oldqlimit;
280 #endif /* SCTP */
281 
282 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
283 		return (EINVAL);
284 
285 #ifdef SCTP
286 	oldopt = so->so_options;
287 	oldqlimit = so->so_qlimit;
288 #endif /* SCTP */
289 
290 	lwkt_gettoken(&so->so_rcv.ssb_token);
291 	if (TAILQ_EMPTY(&so->so_comp))
292 		so->so_options |= SO_ACCEPTCONN;
293 	lwkt_reltoken(&so->so_rcv.ssb_token);
294 	if (backlog < 0 || backlog > somaxconn)
295 		backlog = somaxconn;
296 	so->so_qlimit = backlog;
297 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
298 	 * the so_options (UDP model both connect's and gets inbound
299 	 * connections .. implicitly).
300 	 */
301 	error = so_pru_listen(so, td);
302 	if (error) {
303 #ifdef SCTP
304 		/* Restore the params */
305 		so->so_options = oldopt;
306 		so->so_qlimit = oldqlimit;
307 #endif /* SCTP */
308 		return (error);
309 	}
310 	return (0);
311 }
312 
313 /*
314  * Destroy a disconnected socket.  This routine is a NOP if entities
315  * still have a reference on the socket:
316  *
317  *	so_pcb -	The protocol stack still has a reference
318  *	SS_NOFDREF -	There is no longer a file pointer reference
319  */
320 void
321 sofree(struct socket *so)
322 {
323 	struct socket *head;
324 
325 	/*
326 	 * This is a bit hackish at the moment.  We need to interlock
327 	 * any accept queue we are on before we potentially lose the
328 	 * last reference to avoid races against a re-reference from
329 	 * someone operating on the queue.
330 	 */
331 	while ((head = so->so_head) != NULL) {
332 		lwkt_getpooltoken(head);
333 		if (so->so_head == head)
334 			break;
335 		lwkt_relpooltoken(head);
336 	}
337 
338 	/*
339 	 * Arbitrage the last free.
340 	 */
341 	KKASSERT(so->so_refs > 0);
342 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
343 		if (head)
344 			lwkt_relpooltoken(head);
345 		return;
346 	}
347 
348 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
349 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
350 
351 	/*
352 	 * We're done, remove ourselves from the accept queue we are
353 	 * on, if we are on one.
354 	 */
355 	if (head != NULL) {
356 		if (so->so_state & SS_INCOMP) {
357 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
358 			head->so_incqlen--;
359 		} else if (so->so_state & SS_COMP) {
360 			/*
361 			 * We must not decommission a socket that's
362 			 * on the accept(2) queue.  If we do, then
363 			 * accept(2) may hang after select(2) indicated
364 			 * that the listening socket was ready.
365 			 */
366 			lwkt_relpooltoken(head);
367 			return;
368 		} else {
369 			panic("sofree: not queued");
370 		}
371 		soclrstate(so, SS_INCOMP);
372 		so->so_head = NULL;
373 		lwkt_relpooltoken(head);
374 	}
375 	ssb_release(&so->so_snd, so);
376 	sorflush(so);
377 	sodealloc(so);
378 }
379 
380 /*
381  * Close a socket on last file table reference removal.
382  * Initiate disconnect if connected.
383  * Free socket when disconnect complete.
384  */
385 int
386 soclose(struct socket *so, int fflag)
387 {
388 	int error = 0;
389 
390 	funsetown(&so->so_sigio);
391 	if (so->so_pcb == NULL)
392 		goto discard;
393 	if (so->so_state & SS_ISCONNECTED) {
394 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
395 			error = sodisconnect(so);
396 			if (error)
397 				goto drop;
398 		}
399 		if (so->so_options & SO_LINGER) {
400 			if ((so->so_state & SS_ISDISCONNECTING) &&
401 			    (fflag & FNONBLOCK))
402 				goto drop;
403 			while (so->so_state & SS_ISCONNECTED) {
404 				error = tsleep(&so->so_timeo, PCATCH,
405 					       "soclos", so->so_linger * hz);
406 				if (error)
407 					break;
408 			}
409 		}
410 	}
411 drop:
412 	if (so->so_pcb) {
413 		int error2;
414 
415 		error2 = so_pru_detach(so);
416 		if (error == 0)
417 			error = error2;
418 	}
419 discard:
420 	lwkt_getpooltoken(so);
421 	if (so->so_options & SO_ACCEPTCONN) {
422 		struct socket *sp;
423 
424 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
425 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
426 			soclrstate(sp, SS_INCOMP);
427 			sp->so_head = NULL;
428 			so->so_incqlen--;
429 			soaborta(sp);
430 		}
431 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
432 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
433 			soclrstate(sp, SS_COMP);
434 			sp->so_head = NULL;
435 			so->so_qlen--;
436 			soaborta(sp);
437 		}
438 	}
439 	lwkt_relpooltoken(so);
440 	if (so->so_state & SS_NOFDREF)
441 		panic("soclose: NOFDREF");
442 	sosetstate(so, SS_NOFDREF);	/* take ref */
443 
444 	/* Make sure all asychronous sending are done */
445 	so_pru_sync(so);
446 	sofree(so);			/* dispose of ref */
447 	return (error);
448 }
449 
450 /*
451  * Abort and destroy a socket.  Only one abort can be in progress
452  * at any given moment.
453  */
454 void
455 soabort(struct socket *so)
456 {
457 	soreference(so);
458 	so_pru_abort(so);
459 }
460 
461 void
462 soaborta(struct socket *so)
463 {
464 	soreference(so);
465 	so_pru_aborta(so);
466 }
467 
468 void
469 soabort_oncpu(struct socket *so)
470 {
471 	soreference(so);
472 	so_pru_abort_oncpu(so);
473 }
474 
475 /*
476  * so is passed in ref'd, which becomes owned by
477  * the cleared SS_NOFDREF flag.
478  */
479 int
480 soaccept(struct socket *so, struct sockaddr **nam)
481 {
482 	int error;
483 
484 	if ((so->so_state & SS_NOFDREF) == 0)
485 		panic("soaccept: !NOFDREF");
486 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
487 	error = so_pru_accept_direct(so, nam);
488 	return (error);
489 }
490 
491 int
492 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
493 {
494 	int error;
495 
496 	if (so->so_options & SO_ACCEPTCONN)
497 		return (EOPNOTSUPP);
498 	/*
499 	 * If protocol is connection-based, can only connect once.
500 	 * Otherwise, if connected, try to disconnect first.
501 	 * This allows user to disconnect by connecting to, e.g.,
502 	 * a null address.
503 	 */
504 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
505 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
506 	    (error = sodisconnect(so)))) {
507 		error = EISCONN;
508 	} else {
509 		/*
510 		 * Prevent accumulated error from previous connection
511 		 * from biting us.
512 		 */
513 		so->so_error = 0;
514 		error = so_pru_connect(so, nam, td);
515 	}
516 	return (error);
517 }
518 
519 int
520 soconnect2(struct socket *so1, struct socket *so2)
521 {
522 	int error;
523 
524 	error = so_pru_connect2(so1, so2);
525 	return (error);
526 }
527 
528 int
529 sodisconnect(struct socket *so)
530 {
531 	int error;
532 
533 	if ((so->so_state & SS_ISCONNECTED) == 0) {
534 		error = ENOTCONN;
535 		goto bad;
536 	}
537 	if (so->so_state & SS_ISDISCONNECTING) {
538 		error = EALREADY;
539 		goto bad;
540 	}
541 	error = so_pru_disconnect(so);
542 bad:
543 	return (error);
544 }
545 
546 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
547 /*
548  * Send on a socket.
549  * If send must go all at once and message is larger than
550  * send buffering, then hard error.
551  * Lock against other senders.
552  * If must go all at once and not enough room now, then
553  * inform user that this would block and do nothing.
554  * Otherwise, if nonblocking, send as much as possible.
555  * The data to be sent is described by "uio" if nonzero,
556  * otherwise by the mbuf chain "top" (which must be null
557  * if uio is not).  Data provided in mbuf chain must be small
558  * enough to send all at once.
559  *
560  * Returns nonzero on error, timeout or signal; callers
561  * must check for short counts if EINTR/ERESTART are returned.
562  * Data and control buffers are freed on return.
563  */
564 int
565 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
566 	struct mbuf *top, struct mbuf *control, int flags,
567 	struct thread *td)
568 {
569 	struct mbuf **mp;
570 	struct mbuf *m;
571 	size_t resid;
572 	int space, len;
573 	int clen = 0, error, dontroute, mlen;
574 	int atomic = sosendallatonce(so) || top;
575 	int pru_flags;
576 
577 	if (uio) {
578 		resid = uio->uio_resid;
579 	} else {
580 		resid = (size_t)top->m_pkthdr.len;
581 #ifdef INVARIANTS
582 		len = 0;
583 		for (m = top; m; m = m->m_next)
584 			len += m->m_len;
585 		KKASSERT(top->m_pkthdr.len == len);
586 #endif
587 	}
588 
589 	/*
590 	 * WARNING!  resid is unsigned, space and len are signed.  space
591 	 * 	     can wind up negative if the sockbuf is overcommitted.
592 	 *
593 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
594 	 * type sockets since that's an error.
595 	 */
596 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
597 		error = EINVAL;
598 		goto out;
599 	}
600 
601 	dontroute =
602 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
603 	    (so->so_proto->pr_flags & PR_ATOMIC);
604 	if (td->td_lwp != NULL)
605 		td->td_lwp->lwp_ru.ru_msgsnd++;
606 	if (control)
607 		clen = control->m_len;
608 #define	gotoerr(errcode)	{ error = errcode; goto release; }
609 
610 restart:
611 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
612 	if (error)
613 		goto out;
614 
615 	do {
616 		if (so->so_state & SS_CANTSENDMORE)
617 			gotoerr(EPIPE);
618 		if (so->so_error) {
619 			error = so->so_error;
620 			so->so_error = 0;
621 			goto release;
622 		}
623 		if ((so->so_state & SS_ISCONNECTED) == 0) {
624 			/*
625 			 * `sendto' and `sendmsg' is allowed on a connection-
626 			 * based socket if it supports implied connect.
627 			 * Return ENOTCONN if not connected and no address is
628 			 * supplied.
629 			 */
630 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
631 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
632 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
633 				    !(resid == 0 && clen != 0))
634 					gotoerr(ENOTCONN);
635 			} else if (addr == 0)
636 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
637 				   ENOTCONN : EDESTADDRREQ);
638 		}
639 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
640 		    clen > so->so_snd.ssb_hiwat) {
641 			gotoerr(EMSGSIZE);
642 		}
643 		space = ssb_space(&so->so_snd);
644 		if (flags & MSG_OOB)
645 			space += 1024;
646 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
647 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
648 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
649 				gotoerr(EWOULDBLOCK);
650 			ssb_unlock(&so->so_snd);
651 			error = ssb_wait(&so->so_snd);
652 			if (error)
653 				goto out;
654 			goto restart;
655 		}
656 		mp = &top;
657 		space -= clen;
658 		do {
659 		    if (uio == NULL) {
660 			/*
661 			 * Data is prepackaged in "top".
662 			 */
663 			resid = 0;
664 			if (flags & MSG_EOR)
665 				top->m_flags |= M_EOR;
666 		    } else do {
667 			if (resid > INT_MAX)
668 				resid = INT_MAX;
669 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
670 				   top == NULL ? M_PKTHDR : 0, &mlen);
671 			if (top == NULL) {
672 				m->m_pkthdr.len = 0;
673 				m->m_pkthdr.rcvif = NULL;
674 			}
675 			len = imin((int)szmin(mlen, resid), space);
676 			if (resid < MINCLSIZE) {
677 				/*
678 				 * For datagram protocols, leave room
679 				 * for protocol headers in first mbuf.
680 				 */
681 				if (atomic && top == 0 && len < mlen)
682 					MH_ALIGN(m, len);
683 			}
684 			space -= len;
685 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
686 			resid = uio->uio_resid;
687 			m->m_len = len;
688 			*mp = m;
689 			top->m_pkthdr.len += len;
690 			if (error)
691 				goto release;
692 			mp = &m->m_next;
693 			if (resid == 0) {
694 				if (flags & MSG_EOR)
695 					top->m_flags |= M_EOR;
696 				break;
697 			}
698 		    } while (space > 0 && atomic);
699 		    if (dontroute)
700 			    so->so_options |= SO_DONTROUTE;
701 		    if (flags & MSG_OOB) {
702 		    	    pru_flags = PRUS_OOB;
703 		    } else if ((flags & MSG_EOF) &&
704 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
705 			       (resid == 0)) {
706 			    /*
707 			     * If the user set MSG_EOF, the protocol
708 			     * understands this flag and nothing left to
709 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
710 			     */
711 		    	    pru_flags = PRUS_EOF;
712 		    } else if (resid > 0 && space > 0) {
713 			    /* If there is more to send, set PRUS_MORETOCOME */
714 		    	    pru_flags = PRUS_MORETOCOME;
715 		    } else {
716 		    	    pru_flags = 0;
717 		    }
718 		    /*
719 		     * XXX all the SS_CANTSENDMORE checks previously
720 		     * done could be out of date.  We could have recieved
721 		     * a reset packet in an interrupt or maybe we slept
722 		     * while doing page faults in uiomove() etc. We could
723 		     * probably recheck again inside the splnet() protection
724 		     * here, but there are probably other places that this
725 		     * also happens.  We must rethink this.
726 		     */
727 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
728 		    if (dontroute)
729 			    so->so_options &= ~SO_DONTROUTE;
730 		    clen = 0;
731 		    control = 0;
732 		    top = NULL;
733 		    mp = &top;
734 		    if (error)
735 			    goto release;
736 		} while (resid && space > 0);
737 	} while (resid);
738 
739 release:
740 	ssb_unlock(&so->so_snd);
741 out:
742 	if (top)
743 		m_freem(top);
744 	if (control)
745 		m_freem(control);
746 	return (error);
747 }
748 
749 /*
750  * A specialization of sosend() for UDP based on protocol-specific knowledge:
751  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
752  *	sosendallatonce() returns true,
753  *	the "atomic" variable is true,
754  *	and sosendudp() blocks until space is available for the entire send.
755  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
756  *	PR_IMPLOPCL flags set.
757  *   UDP has no out-of-band data.
758  *   UDP has no control data.
759  *   UDP does not support MSG_EOR.
760  */
761 int
762 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
763 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
764 {
765 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
766 	size_t resid;
767 	int error;
768 	int space;
769 
770 	if (td->td_lwp != NULL)
771 		td->td_lwp->lwp_ru.ru_msgsnd++;
772 	if (control)
773 		m_freem(control);
774 
775 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
776 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
777 
778 restart:
779 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
780 	if (error)
781 		goto out;
782 
783 	if (so->so_state & SS_CANTSENDMORE)
784 		gotoerr(EPIPE);
785 	if (so->so_error) {
786 		error = so->so_error;
787 		so->so_error = 0;
788 		goto release;
789 	}
790 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
791 		gotoerr(EDESTADDRREQ);
792 	if (resid > so->so_snd.ssb_hiwat)
793 		gotoerr(EMSGSIZE);
794 	space = ssb_space(&so->so_snd);
795 	if (uio && (space < 0 || (size_t)space < resid)) {
796 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
797 			gotoerr(EWOULDBLOCK);
798 		ssb_unlock(&so->so_snd);
799 		error = ssb_wait(&so->so_snd);
800 		if (error)
801 			goto out;
802 		goto restart;
803 	}
804 
805 	if (uio) {
806 		top = m_uiomove(uio);
807 		if (top == NULL)
808 			goto release;
809 	}
810 
811 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
812 	if (dontroute)
813 		so->so_options |= SO_DONTROUTE;
814 
815 	error = so_pru_send(so, 0, top, addr, NULL, td);
816 	top = NULL;		/* sent or freed in lower layer */
817 
818 	if (dontroute)
819 		so->so_options &= ~SO_DONTROUTE;
820 
821 release:
822 	ssb_unlock(&so->so_snd);
823 out:
824 	if (top)
825 		m_freem(top);
826 	return (error);
827 }
828 
829 int
830 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
831 	struct mbuf *top, struct mbuf *control, int flags,
832 	struct thread *td)
833 {
834 	struct mbuf **mp;
835 	struct mbuf *m;
836 	size_t resid;
837 	int space, len;
838 	int error, mlen;
839 	int allatonce;
840 	int pru_flags;
841 
842 	if (uio) {
843 		KKASSERT(top == NULL);
844 		allatonce = 0;
845 		resid = uio->uio_resid;
846 	} else {
847 		allatonce = 1;
848 		resid = (size_t)top->m_pkthdr.len;
849 #ifdef INVARIANTS
850 		len = 0;
851 		for (m = top; m; m = m->m_next)
852 			len += m->m_len;
853 		KKASSERT(top->m_pkthdr.len == len);
854 #endif
855 	}
856 
857 	/*
858 	 * WARNING!  resid is unsigned, space and len are signed.  space
859 	 * 	     can wind up negative if the sockbuf is overcommitted.
860 	 *
861 	 * Also check to make sure that MSG_EOR isn't used on TCP
862 	 */
863 	if (flags & MSG_EOR) {
864 		error = EINVAL;
865 		goto out;
866 	}
867 
868 	if (control) {
869 		/* TCP doesn't do control messages (rights, creds, etc) */
870 		if (control->m_len) {
871 			error = EINVAL;
872 			goto out;
873 		}
874 		m_freem(control);	/* empty control, just free it */
875 		control = NULL;
876 	}
877 
878 	if (td->td_lwp != NULL)
879 		td->td_lwp->lwp_ru.ru_msgsnd++;
880 
881 #define	gotoerr(errcode)	{ error = errcode; goto release; }
882 
883 restart:
884 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
885 	if (error)
886 		goto out;
887 
888 	do {
889 		if (so->so_state & SS_CANTSENDMORE)
890 			gotoerr(EPIPE);
891 		if (so->so_error) {
892 			error = so->so_error;
893 			so->so_error = 0;
894 			goto release;
895 		}
896 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
897 		    (so->so_state & SS_ISCONFIRMING) == 0)
898 			gotoerr(ENOTCONN);
899 		if (allatonce && resid > so->so_snd.ssb_hiwat)
900 			gotoerr(EMSGSIZE);
901 
902 		space = ssb_space(&so->so_snd);
903 		if (flags & MSG_OOB)
904 			space += 1024;
905 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
906 		    space < so->so_snd.ssb_lowat) {
907 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
908 				gotoerr(EWOULDBLOCK);
909 			ssb_unlock(&so->so_snd);
910 			error = ssb_wait(&so->so_snd);
911 			if (error)
912 				goto out;
913 			goto restart;
914 		}
915 		mp = &top;
916 		do {
917 		    int cnt = 0, async = 0;
918 
919 		    if (uio == NULL) {
920 			/*
921 			 * Data is prepackaged in "top".
922 			 */
923 			resid = 0;
924 		    } else do {
925 			if (resid > INT_MAX)
926 				resid = INT_MAX;
927 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
928 				   top == NULL ? M_PKTHDR : 0, &mlen);
929 			if (top == NULL) {
930 				m->m_pkthdr.len = 0;
931 				m->m_pkthdr.rcvif = NULL;
932 			}
933 			len = imin((int)szmin(mlen, resid), space);
934 			space -= len;
935 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
936 			resid = uio->uio_resid;
937 			m->m_len = len;
938 			*mp = m;
939 			top->m_pkthdr.len += len;
940 			if (error)
941 				goto release;
942 			mp = &m->m_next;
943 			if (resid == 0)
944 				break;
945 			++cnt;
946 		    } while (space > 0 && cnt < tcp_sosnd_agglim);
947 
948 		    if (tcp_sosnd_async)
949 			    async = 1;
950 
951 		    if (flags & MSG_OOB) {
952 		    	    pru_flags = PRUS_OOB;
953 			    async = 0;
954 		    } else if ((flags & MSG_EOF) && resid == 0) {
955 			    pru_flags = PRUS_EOF;
956 		    } else if (resid > 0 && space > 0) {
957 			    /* If there is more to send, set PRUS_MORETOCOME */
958 		    	    pru_flags = PRUS_MORETOCOME;
959 			    async = 1;
960 		    } else {
961 		    	    pru_flags = 0;
962 		    }
963 
964 		    if (flags & MSG_SYNC)
965 			    async = 0;
966 
967 		    /*
968 		     * XXX all the SS_CANTSENDMORE checks previously
969 		     * done could be out of date.  We could have recieved
970 		     * a reset packet in an interrupt or maybe we slept
971 		     * while doing page faults in uiomove() etc. We could
972 		     * probably recheck again inside the splnet() protection
973 		     * here, but there are probably other places that this
974 		     * also happens.  We must rethink this.
975 		     */
976 		    if (!async) {
977 			    error = so_pru_send(so, pru_flags, top,
978 			        NULL, NULL, td);
979 		    } else {
980 			    so_pru_send_async(so, pru_flags, top,
981 			        NULL, NULL, td);
982 			    error = 0;
983 		    }
984 
985 		    top = NULL;
986 		    mp = &top;
987 		    if (error)
988 			    goto release;
989 		} while (resid && space > 0);
990 	} while (resid);
991 
992 release:
993 	ssb_unlock(&so->so_snd);
994 out:
995 	if (top)
996 		m_freem(top);
997 	if (control)
998 		m_freem(control);
999 	return (error);
1000 }
1001 
1002 /*
1003  * Implement receive operations on a socket.
1004  *
1005  * We depend on the way that records are added to the signalsockbuf
1006  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1007  * must begin with an address if the protocol so specifies,
1008  * followed by an optional mbuf or mbufs containing ancillary data,
1009  * and then zero or more mbufs of data.
1010  *
1011  * Although the signalsockbuf is locked, new data may still be appended.
1012  * A token inside the ssb_lock deals with MP issues and still allows
1013  * the network to access the socket if we block in a uio.
1014  *
1015  * The caller may receive the data as a single mbuf chain by supplying
1016  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1017  * only for the count in uio_resid.
1018  */
1019 int
1020 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1021 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1022 {
1023 	struct mbuf *m, *n;
1024 	struct mbuf *free_chain = NULL;
1025 	int flags, len, error, offset;
1026 	struct protosw *pr = so->so_proto;
1027 	int moff, type = 0;
1028 	size_t resid, orig_resid;
1029 
1030 	if (uio)
1031 		resid = uio->uio_resid;
1032 	else
1033 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1034 	orig_resid = resid;
1035 
1036 	if (psa)
1037 		*psa = NULL;
1038 	if (controlp)
1039 		*controlp = NULL;
1040 	if (flagsp)
1041 		flags = *flagsp &~ MSG_EOR;
1042 	else
1043 		flags = 0;
1044 	if (flags & MSG_OOB) {
1045 		m = m_get(MB_WAIT, MT_DATA);
1046 		if (m == NULL)
1047 			return (ENOBUFS);
1048 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1049 		if (error)
1050 			goto bad;
1051 		if (sio) {
1052 			do {
1053 				sbappend(sio, m);
1054 				KKASSERT(resid >= (size_t)m->m_len);
1055 				resid -= (size_t)m->m_len;
1056 			} while (resid > 0 && m);
1057 		} else {
1058 			do {
1059 				uio->uio_resid = resid;
1060 				error = uiomove(mtod(m, caddr_t),
1061 						(int)szmin(resid, m->m_len),
1062 						uio);
1063 				resid = uio->uio_resid;
1064 				m = m_free(m);
1065 			} while (uio->uio_resid && error == 0 && m);
1066 		}
1067 bad:
1068 		if (m)
1069 			m_freem(m);
1070 		return (error);
1071 	}
1072 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1073 		so_pru_rcvd(so, 0);
1074 
1075 	/*
1076 	 * The token interlocks against the protocol thread while
1077 	 * ssb_lock is a blocking lock against other userland entities.
1078 	 */
1079 	lwkt_gettoken(&so->so_rcv.ssb_token);
1080 restart:
1081 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1082 	if (error)
1083 		goto done;
1084 
1085 	m = so->so_rcv.ssb_mb;
1086 	/*
1087 	 * If we have less data than requested, block awaiting more
1088 	 * (subject to any timeout) if:
1089 	 *   1. the current count is less than the low water mark, or
1090 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1091 	 *	receive operation at once if we block (resid <= hiwat).
1092 	 *   3. MSG_DONTWAIT is not set
1093 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1094 	 * we have to do the receive in sections, and thus risk returning
1095 	 * a short count if a timeout or signal occurs after we start.
1096 	 */
1097 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1098 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1099 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1100 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1101 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1102 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1103 		if (so->so_error) {
1104 			if (m)
1105 				goto dontblock;
1106 			error = so->so_error;
1107 			if ((flags & MSG_PEEK) == 0)
1108 				so->so_error = 0;
1109 			goto release;
1110 		}
1111 		if (so->so_state & SS_CANTRCVMORE) {
1112 			if (m)
1113 				goto dontblock;
1114 			else
1115 				goto release;
1116 		}
1117 		for (; m; m = m->m_next) {
1118 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1119 				m = so->so_rcv.ssb_mb;
1120 				goto dontblock;
1121 			}
1122 		}
1123 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1124 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1125 			error = ENOTCONN;
1126 			goto release;
1127 		}
1128 		if (resid == 0)
1129 			goto release;
1130 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1131 			error = EWOULDBLOCK;
1132 			goto release;
1133 		}
1134 		ssb_unlock(&so->so_rcv);
1135 		error = ssb_wait(&so->so_rcv);
1136 		if (error)
1137 			goto done;
1138 		goto restart;
1139 	}
1140 dontblock:
1141 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1142 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1143 
1144 	/*
1145 	 * note: m should be == sb_mb here.  Cache the next record while
1146 	 * cleaning up.  Note that calling m_free*() will break out critical
1147 	 * section.
1148 	 */
1149 	KKASSERT(m == so->so_rcv.ssb_mb);
1150 
1151 	/*
1152 	 * Skip any address mbufs prepending the record.
1153 	 */
1154 	if (pr->pr_flags & PR_ADDR) {
1155 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1156 		orig_resid = 0;
1157 		if (psa)
1158 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1159 		if (flags & MSG_PEEK)
1160 			m = m->m_next;
1161 		else
1162 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1163 	}
1164 
1165 	/*
1166 	 * Skip any control mbufs prepending the record.
1167 	 */
1168 #ifdef SCTP
1169 	if (pr->pr_flags & PR_ADDR_OPT) {
1170 		/*
1171 		 * For SCTP we may be getting a
1172 		 * whole message OR a partial delivery.
1173 		 */
1174 		if (m && m->m_type == MT_SONAME) {
1175 			orig_resid = 0;
1176 			if (psa)
1177 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1178 			if (flags & MSG_PEEK)
1179 				m = m->m_next;
1180 			else
1181 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1182 		}
1183 	}
1184 #endif /* SCTP */
1185 	while (m && m->m_type == MT_CONTROL && error == 0) {
1186 		if (flags & MSG_PEEK) {
1187 			if (controlp)
1188 				*controlp = m_copy(m, 0, m->m_len);
1189 			m = m->m_next;	/* XXX race */
1190 		} else {
1191 			if (controlp) {
1192 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1193 				if (pr->pr_domain->dom_externalize &&
1194 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1195 				    SCM_RIGHTS)
1196 				   error = (*pr->pr_domain->dom_externalize)(m);
1197 				*controlp = m;
1198 				m = n;
1199 			} else {
1200 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1201 			}
1202 		}
1203 		if (controlp && *controlp) {
1204 			orig_resid = 0;
1205 			controlp = &(*controlp)->m_next;
1206 		}
1207 	}
1208 
1209 	/*
1210 	 * flag OOB data.
1211 	 */
1212 	if (m) {
1213 		type = m->m_type;
1214 		if (type == MT_OOBDATA)
1215 			flags |= MSG_OOB;
1216 	}
1217 
1218 	/*
1219 	 * Copy to the UIO or mbuf return chain (*mp).
1220 	 */
1221 	moff = 0;
1222 	offset = 0;
1223 	while (m && resid > 0 && error == 0) {
1224 		if (m->m_type == MT_OOBDATA) {
1225 			if (type != MT_OOBDATA)
1226 				break;
1227 		} else if (type == MT_OOBDATA)
1228 			break;
1229 		else
1230 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1231 			("receive 3"));
1232 		soclrstate(so, SS_RCVATMARK);
1233 		len = (resid > INT_MAX) ? INT_MAX : resid;
1234 		if (so->so_oobmark && len > so->so_oobmark - offset)
1235 			len = so->so_oobmark - offset;
1236 		if (len > m->m_len - moff)
1237 			len = m->m_len - moff;
1238 
1239 		/*
1240 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1241 		 * The SIO is dealt with when we eat the mbuf, but deal
1242 		 * with the resid here either way.
1243 		 */
1244 		if (uio) {
1245 			uio->uio_resid = resid;
1246 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1247 			resid = uio->uio_resid;
1248 			if (error)
1249 				goto release;
1250 		} else {
1251 			resid -= (size_t)len;
1252 		}
1253 
1254 		/*
1255 		 * Eat the entire mbuf or just a piece of it
1256 		 */
1257 		if (len == m->m_len - moff) {
1258 			if (m->m_flags & M_EOR)
1259 				flags |= MSG_EOR;
1260 #ifdef SCTP
1261 			if (m->m_flags & M_NOTIFICATION)
1262 				flags |= MSG_NOTIFICATION;
1263 #endif /* SCTP */
1264 			if (flags & MSG_PEEK) {
1265 				m = m->m_next;
1266 				moff = 0;
1267 			} else {
1268 				if (sio) {
1269 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1270 					sbappend(sio, m);
1271 					m = n;
1272 				} else {
1273 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1274 				}
1275 			}
1276 		} else {
1277 			if (flags & MSG_PEEK) {
1278 				moff += len;
1279 			} else {
1280 				if (sio) {
1281 					n = m_copym(m, 0, len, MB_WAIT);
1282 					if (n)
1283 						sbappend(sio, n);
1284 				}
1285 				m->m_data += len;
1286 				m->m_len -= len;
1287 				so->so_rcv.ssb_cc -= len;
1288 			}
1289 		}
1290 		if (so->so_oobmark) {
1291 			if ((flags & MSG_PEEK) == 0) {
1292 				so->so_oobmark -= len;
1293 				if (so->so_oobmark == 0) {
1294 					sosetstate(so, SS_RCVATMARK);
1295 					break;
1296 				}
1297 			} else {
1298 				offset += len;
1299 				if (offset == so->so_oobmark)
1300 					break;
1301 			}
1302 		}
1303 		if (flags & MSG_EOR)
1304 			break;
1305 		/*
1306 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1307 		 * we must not quit until resid == 0 or an error
1308 		 * termination.  If a signal/timeout occurs, return
1309 		 * with a short count but without error.
1310 		 * Keep signalsockbuf locked against other readers.
1311 		 */
1312 		while ((flags & MSG_WAITALL) && m == NULL &&
1313 		       resid > 0 && !sosendallatonce(so) &&
1314 		       so->so_rcv.ssb_mb == NULL) {
1315 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1316 				break;
1317 			/*
1318 			 * The window might have closed to zero, make
1319 			 * sure we send an ack now that we've drained
1320 			 * the buffer or we might end up blocking until
1321 			 * the idle takes over (5 seconds).
1322 			 */
1323 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1324 				so_pru_rcvd(so, flags);
1325 			error = ssb_wait(&so->so_rcv);
1326 			if (error) {
1327 				ssb_unlock(&so->so_rcv);
1328 				error = 0;
1329 				goto done;
1330 			}
1331 			m = so->so_rcv.ssb_mb;
1332 		}
1333 	}
1334 
1335 	/*
1336 	 * If an atomic read was requested but unread data still remains
1337 	 * in the record, set MSG_TRUNC.
1338 	 */
1339 	if (m && pr->pr_flags & PR_ATOMIC)
1340 		flags |= MSG_TRUNC;
1341 
1342 	/*
1343 	 * Cleanup.  If an atomic read was requested drop any unread data.
1344 	 */
1345 	if ((flags & MSG_PEEK) == 0) {
1346 		if (m && (pr->pr_flags & PR_ATOMIC))
1347 			sbdroprecord(&so->so_rcv.sb);
1348 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1349 			so_pru_rcvd(so, flags);
1350 	}
1351 
1352 	if (orig_resid == resid && orig_resid &&
1353 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1354 		ssb_unlock(&so->so_rcv);
1355 		goto restart;
1356 	}
1357 
1358 	if (flagsp)
1359 		*flagsp |= flags;
1360 release:
1361 	ssb_unlock(&so->so_rcv);
1362 done:
1363 	lwkt_reltoken(&so->so_rcv.ssb_token);
1364 	if (free_chain)
1365 		m_freem(free_chain);
1366 	return (error);
1367 }
1368 
1369 /*
1370  * Shut a socket down.  Note that we do not get a frontend lock as we
1371  * want to be able to shut the socket down even if another thread is
1372  * blocked in a read(), thus waking it up.
1373  */
1374 int
1375 soshutdown(struct socket *so, int how)
1376 {
1377 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1378 		return (EINVAL);
1379 
1380 	if (how != SHUT_WR) {
1381 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1382 		sorflush(so);
1383 		/*ssb_unlock(&so->so_rcv);*/
1384 	}
1385 	if (how != SHUT_RD)
1386 		return (so_pru_shutdown(so));
1387 	return (0);
1388 }
1389 
1390 void
1391 sorflush(struct socket *so)
1392 {
1393 	struct signalsockbuf *ssb = &so->so_rcv;
1394 	struct protosw *pr = so->so_proto;
1395 	struct signalsockbuf asb;
1396 
1397 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1398 
1399 	lwkt_gettoken(&ssb->ssb_token);
1400 	socantrcvmore(so);
1401 	asb = *ssb;
1402 
1403 	/*
1404 	 * Can't just blow up the ssb structure here
1405 	 */
1406 	bzero(&ssb->sb, sizeof(ssb->sb));
1407 	ssb->ssb_timeo = 0;
1408 	ssb->ssb_lowat = 0;
1409 	ssb->ssb_hiwat = 0;
1410 	ssb->ssb_mbmax = 0;
1411 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1412 
1413 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1414 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1415 	ssb_release(&asb, so);
1416 
1417 	lwkt_reltoken(&ssb->ssb_token);
1418 }
1419 
1420 #ifdef INET
1421 static int
1422 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1423 {
1424 	struct accept_filter_arg	*afap = NULL;
1425 	struct accept_filter	*afp;
1426 	struct so_accf	*af = so->so_accf;
1427 	int	error = 0;
1428 
1429 	/* do not set/remove accept filters on non listen sockets */
1430 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1431 		error = EINVAL;
1432 		goto out;
1433 	}
1434 
1435 	/* removing the filter */
1436 	if (sopt == NULL) {
1437 		if (af != NULL) {
1438 			if (af->so_accept_filter != NULL &&
1439 				af->so_accept_filter->accf_destroy != NULL) {
1440 				af->so_accept_filter->accf_destroy(so);
1441 			}
1442 			if (af->so_accept_filter_str != NULL) {
1443 				FREE(af->so_accept_filter_str, M_ACCF);
1444 			}
1445 			FREE(af, M_ACCF);
1446 			so->so_accf = NULL;
1447 		}
1448 		so->so_options &= ~SO_ACCEPTFILTER;
1449 		return (0);
1450 	}
1451 	/* adding a filter */
1452 	/* must remove previous filter first */
1453 	if (af != NULL) {
1454 		error = EINVAL;
1455 		goto out;
1456 	}
1457 	/* don't put large objects on the kernel stack */
1458 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1459 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1460 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1461 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1462 	if (error)
1463 		goto out;
1464 	afp = accept_filt_get(afap->af_name);
1465 	if (afp == NULL) {
1466 		error = ENOENT;
1467 		goto out;
1468 	}
1469 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1470 	if (afp->accf_create != NULL) {
1471 		if (afap->af_name[0] != '\0') {
1472 			int len = strlen(afap->af_name) + 1;
1473 
1474 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1475 			strcpy(af->so_accept_filter_str, afap->af_name);
1476 		}
1477 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1478 		if (af->so_accept_filter_arg == NULL) {
1479 			FREE(af->so_accept_filter_str, M_ACCF);
1480 			FREE(af, M_ACCF);
1481 			so->so_accf = NULL;
1482 			error = EINVAL;
1483 			goto out;
1484 		}
1485 	}
1486 	af->so_accept_filter = afp;
1487 	so->so_accf = af;
1488 	so->so_options |= SO_ACCEPTFILTER;
1489 out:
1490 	if (afap != NULL)
1491 		FREE(afap, M_TEMP);
1492 	return (error);
1493 }
1494 #endif /* INET */
1495 
1496 /*
1497  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1498  * an additional variant to handle the case where the option value needs
1499  * to be some kind of integer, but not a specific size.
1500  * In addition to their use here, these functions are also called by the
1501  * protocol-level pr_ctloutput() routines.
1502  */
1503 int
1504 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1505 {
1506 	return soopt_to_kbuf(sopt, buf, len, minlen);
1507 }
1508 
1509 int
1510 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1511 {
1512 	size_t	valsize;
1513 
1514 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1515 	KKASSERT(kva_p(buf));
1516 
1517 	/*
1518 	 * If the user gives us more than we wanted, we ignore it,
1519 	 * but if we don't get the minimum length the caller
1520 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1521 	 * is set to however much we actually retrieved.
1522 	 */
1523 	if ((valsize = sopt->sopt_valsize) < minlen)
1524 		return EINVAL;
1525 	if (valsize > len)
1526 		sopt->sopt_valsize = valsize = len;
1527 
1528 	bcopy(sopt->sopt_val, buf, valsize);
1529 	return 0;
1530 }
1531 
1532 
1533 int
1534 sosetopt(struct socket *so, struct sockopt *sopt)
1535 {
1536 	int	error, optval;
1537 	struct	linger l;
1538 	struct	timeval tv;
1539 	u_long  val;
1540 	struct signalsockbuf *sotmp;
1541 
1542 	error = 0;
1543 	sopt->sopt_dir = SOPT_SET;
1544 	if (sopt->sopt_level != SOL_SOCKET) {
1545 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1546 			return (so_pr_ctloutput(so, sopt));
1547 		}
1548 		error = ENOPROTOOPT;
1549 	} else {
1550 		switch (sopt->sopt_name) {
1551 #ifdef INET
1552 		case SO_ACCEPTFILTER:
1553 			error = do_setopt_accept_filter(so, sopt);
1554 			if (error)
1555 				goto bad;
1556 			break;
1557 #endif /* INET */
1558 		case SO_LINGER:
1559 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1560 			if (error)
1561 				goto bad;
1562 
1563 			so->so_linger = l.l_linger;
1564 			if (l.l_onoff)
1565 				so->so_options |= SO_LINGER;
1566 			else
1567 				so->so_options &= ~SO_LINGER;
1568 			break;
1569 
1570 		case SO_DEBUG:
1571 		case SO_KEEPALIVE:
1572 		case SO_DONTROUTE:
1573 		case SO_USELOOPBACK:
1574 		case SO_BROADCAST:
1575 		case SO_REUSEADDR:
1576 		case SO_REUSEPORT:
1577 		case SO_OOBINLINE:
1578 		case SO_TIMESTAMP:
1579 			error = sooptcopyin(sopt, &optval, sizeof optval,
1580 					    sizeof optval);
1581 			if (error)
1582 				goto bad;
1583 			if (optval)
1584 				so->so_options |= sopt->sopt_name;
1585 			else
1586 				so->so_options &= ~sopt->sopt_name;
1587 			break;
1588 
1589 		case SO_SNDBUF:
1590 		case SO_RCVBUF:
1591 		case SO_SNDLOWAT:
1592 		case SO_RCVLOWAT:
1593 			error = sooptcopyin(sopt, &optval, sizeof optval,
1594 					    sizeof optval);
1595 			if (error)
1596 				goto bad;
1597 
1598 			/*
1599 			 * Values < 1 make no sense for any of these
1600 			 * options, so disallow them.
1601 			 */
1602 			if (optval < 1) {
1603 				error = EINVAL;
1604 				goto bad;
1605 			}
1606 
1607 			switch (sopt->sopt_name) {
1608 			case SO_SNDBUF:
1609 			case SO_RCVBUF:
1610 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1611 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1612 				    so,
1613 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1614 					error = ENOBUFS;
1615 					goto bad;
1616 				}
1617 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1618 						&so->so_snd : &so->so_rcv;
1619 				atomic_clear_int(&sotmp->ssb_flags,
1620 						 SSB_AUTOSIZE);
1621 				break;
1622 
1623 			/*
1624 			 * Make sure the low-water is never greater than
1625 			 * the high-water.
1626 			 */
1627 			case SO_SNDLOWAT:
1628 				so->so_snd.ssb_lowat =
1629 				    (optval > so->so_snd.ssb_hiwat) ?
1630 				    so->so_snd.ssb_hiwat : optval;
1631 				atomic_clear_int(&so->so_snd.ssb_flags,
1632 						 SSB_AUTOLOWAT);
1633 				break;
1634 			case SO_RCVLOWAT:
1635 				so->so_rcv.ssb_lowat =
1636 				    (optval > so->so_rcv.ssb_hiwat) ?
1637 				    so->so_rcv.ssb_hiwat : optval;
1638 				atomic_clear_int(&so->so_rcv.ssb_flags,
1639 						 SSB_AUTOLOWAT);
1640 				break;
1641 			}
1642 			break;
1643 
1644 		case SO_SNDTIMEO:
1645 		case SO_RCVTIMEO:
1646 			error = sooptcopyin(sopt, &tv, sizeof tv,
1647 					    sizeof tv);
1648 			if (error)
1649 				goto bad;
1650 
1651 			/* assert(hz > 0); */
1652 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1653 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1654 				error = EDOM;
1655 				goto bad;
1656 			}
1657 			/* assert(tick > 0); */
1658 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
1659 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1660 			if (val > INT_MAX) {
1661 				error = EDOM;
1662 				goto bad;
1663 			}
1664 			if (val == 0 && tv.tv_usec != 0)
1665 				val = 1;
1666 
1667 			switch (sopt->sopt_name) {
1668 			case SO_SNDTIMEO:
1669 				so->so_snd.ssb_timeo = val;
1670 				break;
1671 			case SO_RCVTIMEO:
1672 				so->so_rcv.ssb_timeo = val;
1673 				break;
1674 			}
1675 			break;
1676 		default:
1677 			error = ENOPROTOOPT;
1678 			break;
1679 		}
1680 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1681 			(void) so_pr_ctloutput(so, sopt);
1682 		}
1683 	}
1684 bad:
1685 	return (error);
1686 }
1687 
1688 /* Helper routine for getsockopt */
1689 int
1690 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1691 {
1692 	soopt_from_kbuf(sopt, buf, len);
1693 	return 0;
1694 }
1695 
1696 void
1697 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1698 {
1699 	size_t	valsize;
1700 
1701 	if (len == 0) {
1702 		sopt->sopt_valsize = 0;
1703 		return;
1704 	}
1705 
1706 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1707 	KKASSERT(kva_p(buf));
1708 
1709 	/*
1710 	 * Documented get behavior is that we always return a value,
1711 	 * possibly truncated to fit in the user's buffer.
1712 	 * Traditional behavior is that we always tell the user
1713 	 * precisely how much we copied, rather than something useful
1714 	 * like the total amount we had available for her.
1715 	 * Note that this interface is not idempotent; the entire answer must
1716 	 * generated ahead of time.
1717 	 */
1718 	valsize = szmin(len, sopt->sopt_valsize);
1719 	sopt->sopt_valsize = valsize;
1720 	if (sopt->sopt_val != 0) {
1721 		bcopy(buf, sopt->sopt_val, valsize);
1722 	}
1723 }
1724 
1725 int
1726 sogetopt(struct socket *so, struct sockopt *sopt)
1727 {
1728 	int	error, optval;
1729 	long	optval_l;
1730 	struct	linger l;
1731 	struct	timeval tv;
1732 #ifdef INET
1733 	struct accept_filter_arg *afap;
1734 #endif
1735 
1736 	error = 0;
1737 	sopt->sopt_dir = SOPT_GET;
1738 	if (sopt->sopt_level != SOL_SOCKET) {
1739 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1740 			return (so_pr_ctloutput(so, sopt));
1741 		} else
1742 			return (ENOPROTOOPT);
1743 	} else {
1744 		switch (sopt->sopt_name) {
1745 #ifdef INET
1746 		case SO_ACCEPTFILTER:
1747 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1748 				return (EINVAL);
1749 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1750 				M_TEMP, M_WAITOK | M_ZERO);
1751 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1752 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1753 				if (so->so_accf->so_accept_filter_str != NULL)
1754 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1755 			}
1756 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1757 			FREE(afap, M_TEMP);
1758 			break;
1759 #endif /* INET */
1760 
1761 		case SO_LINGER:
1762 			l.l_onoff = so->so_options & SO_LINGER;
1763 			l.l_linger = so->so_linger;
1764 			error = sooptcopyout(sopt, &l, sizeof l);
1765 			break;
1766 
1767 		case SO_USELOOPBACK:
1768 		case SO_DONTROUTE:
1769 		case SO_DEBUG:
1770 		case SO_KEEPALIVE:
1771 		case SO_REUSEADDR:
1772 		case SO_REUSEPORT:
1773 		case SO_BROADCAST:
1774 		case SO_OOBINLINE:
1775 		case SO_TIMESTAMP:
1776 			optval = so->so_options & sopt->sopt_name;
1777 integer:
1778 			error = sooptcopyout(sopt, &optval, sizeof optval);
1779 			break;
1780 
1781 		case SO_TYPE:
1782 			optval = so->so_type;
1783 			goto integer;
1784 
1785 		case SO_ERROR:
1786 			optval = so->so_error;
1787 			so->so_error = 0;
1788 			goto integer;
1789 
1790 		case SO_SNDBUF:
1791 			optval = so->so_snd.ssb_hiwat;
1792 			goto integer;
1793 
1794 		case SO_RCVBUF:
1795 			optval = so->so_rcv.ssb_hiwat;
1796 			goto integer;
1797 
1798 		case SO_SNDLOWAT:
1799 			optval = so->so_snd.ssb_lowat;
1800 			goto integer;
1801 
1802 		case SO_RCVLOWAT:
1803 			optval = so->so_rcv.ssb_lowat;
1804 			goto integer;
1805 
1806 		case SO_SNDTIMEO:
1807 		case SO_RCVTIMEO:
1808 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1809 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1810 
1811 			tv.tv_sec = optval / hz;
1812 			tv.tv_usec = (optval % hz) * ustick;
1813 			error = sooptcopyout(sopt, &tv, sizeof tv);
1814 			break;
1815 
1816 		case SO_SNDSPACE:
1817 			optval_l = ssb_space(&so->so_snd);
1818 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1819 			break;
1820 
1821 		default:
1822 			error = ENOPROTOOPT;
1823 			break;
1824 		}
1825 		return (error);
1826 	}
1827 }
1828 
1829 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1830 int
1831 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1832 {
1833 	struct mbuf *m, *m_prev;
1834 	int sopt_size = sopt->sopt_valsize, msize;
1835 
1836 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1837 		   0, &msize);
1838 	if (m == NULL)
1839 		return (ENOBUFS);
1840 	m->m_len = min(msize, sopt_size);
1841 	sopt_size -= m->m_len;
1842 	*mp = m;
1843 	m_prev = m;
1844 
1845 	while (sopt_size > 0) {
1846 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1847 			   MT_DATA, 0, &msize);
1848 		if (m == NULL) {
1849 			m_freem(*mp);
1850 			return (ENOBUFS);
1851 		}
1852 		m->m_len = min(msize, sopt_size);
1853 		sopt_size -= m->m_len;
1854 		m_prev->m_next = m;
1855 		m_prev = m;
1856 	}
1857 	return (0);
1858 }
1859 
1860 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1861 int
1862 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1863 {
1864 	soopt_to_mbuf(sopt, m);
1865 	return 0;
1866 }
1867 
1868 void
1869 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1870 {
1871 	size_t valsize;
1872 	void *val;
1873 
1874 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1875 	KKASSERT(kva_p(m));
1876 	if (sopt->sopt_val == NULL)
1877 		return;
1878 	val = sopt->sopt_val;
1879 	valsize = sopt->sopt_valsize;
1880 	while (m != NULL && valsize >= m->m_len) {
1881 		bcopy(val, mtod(m, char *), m->m_len);
1882 		valsize -= m->m_len;
1883 		val = (caddr_t)val + m->m_len;
1884 		m = m->m_next;
1885 	}
1886 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1887 		panic("ip6_sooptmcopyin");
1888 }
1889 
1890 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1891 int
1892 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1893 {
1894 	return soopt_from_mbuf(sopt, m);
1895 }
1896 
1897 int
1898 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1899 {
1900 	struct mbuf *m0 = m;
1901 	size_t valsize = 0;
1902 	size_t maxsize;
1903 	void *val;
1904 
1905 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1906 	KKASSERT(kva_p(m));
1907 	if (sopt->sopt_val == NULL)
1908 		return 0;
1909 	val = sopt->sopt_val;
1910 	maxsize = sopt->sopt_valsize;
1911 	while (m != NULL && maxsize >= m->m_len) {
1912 		bcopy(mtod(m, char *), val, m->m_len);
1913 	       maxsize -= m->m_len;
1914 	       val = (caddr_t)val + m->m_len;
1915 	       valsize += m->m_len;
1916 	       m = m->m_next;
1917 	}
1918 	if (m != NULL) {
1919 		/* enough soopt buffer should be given from user-land */
1920 		m_freem(m0);
1921 		return (EINVAL);
1922 	}
1923 	sopt->sopt_valsize = valsize;
1924 	return 0;
1925 }
1926 
1927 void
1928 sohasoutofband(struct socket *so)
1929 {
1930 	if (so->so_sigio != NULL)
1931 		pgsigio(so->so_sigio, SIGURG, 0);
1932 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1933 }
1934 
1935 int
1936 sokqfilter(struct file *fp, struct knote *kn)
1937 {
1938 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1939 	struct signalsockbuf *ssb;
1940 
1941 	switch (kn->kn_filter) {
1942 	case EVFILT_READ:
1943 		if (so->so_options & SO_ACCEPTCONN)
1944 			kn->kn_fop = &solisten_filtops;
1945 		else
1946 			kn->kn_fop = &soread_filtops;
1947 		ssb = &so->so_rcv;
1948 		break;
1949 	case EVFILT_WRITE:
1950 		kn->kn_fop = &sowrite_filtops;
1951 		ssb = &so->so_snd;
1952 		break;
1953 	case EVFILT_EXCEPT:
1954 		kn->kn_fop = &soexcept_filtops;
1955 		ssb = &so->so_rcv;
1956 		break;
1957 	default:
1958 		return (EOPNOTSUPP);
1959 	}
1960 
1961 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1962 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1963 	return (0);
1964 }
1965 
1966 static void
1967 filt_sordetach(struct knote *kn)
1968 {
1969 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1970 
1971 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1972 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1973 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1974 }
1975 
1976 /*ARGSUSED*/
1977 static int
1978 filt_soread(struct knote *kn, long hint)
1979 {
1980 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1981 
1982 	if (kn->kn_sfflags & NOTE_OOB) {
1983 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1984 			kn->kn_fflags |= NOTE_OOB;
1985 			return (1);
1986 		}
1987 		return (0);
1988 	}
1989 	kn->kn_data = so->so_rcv.ssb_cc;
1990 
1991 	if (so->so_state & SS_CANTRCVMORE) {
1992 		/*
1993 		 * Only set NODATA if all data has been exhausted.
1994 		 */
1995 		if (kn->kn_data == 0)
1996 			kn->kn_flags |= EV_NODATA;
1997 		kn->kn_flags |= EV_EOF;
1998 		kn->kn_fflags = so->so_error;
1999 		return (1);
2000 	}
2001 	if (so->so_error)	/* temporary udp error */
2002 		return (1);
2003 	if (kn->kn_sfflags & NOTE_LOWAT)
2004 		return (kn->kn_data >= kn->kn_sdata);
2005 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2006 		!TAILQ_EMPTY(&so->so_comp));
2007 }
2008 
2009 static void
2010 filt_sowdetach(struct knote *kn)
2011 {
2012 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2013 
2014 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2015 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2016 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2017 }
2018 
2019 /*ARGSUSED*/
2020 static int
2021 filt_sowrite(struct knote *kn, long hint)
2022 {
2023 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2024 
2025 	kn->kn_data = ssb_space(&so->so_snd);
2026 	if (so->so_state & SS_CANTSENDMORE) {
2027 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2028 		kn->kn_fflags = so->so_error;
2029 		return (1);
2030 	}
2031 	if (so->so_error)	/* temporary udp error */
2032 		return (1);
2033 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2034 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2035 		return (0);
2036 	if (kn->kn_sfflags & NOTE_LOWAT)
2037 		return (kn->kn_data >= kn->kn_sdata);
2038 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2039 }
2040 
2041 /*ARGSUSED*/
2042 static int
2043 filt_solisten(struct knote *kn, long hint)
2044 {
2045 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2046 
2047 	kn->kn_data = so->so_qlen;
2048 	return (! TAILQ_EMPTY(&so->so_comp));
2049 }
2050