xref: /dflybsd-src/sys/kern/uipc_socket.c (revision dae741e33c840b92a8a53bf9f01157ede145e256)
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 	sofree(so);			/* dispose of ref */
444 	return (error);
445 }
446 
447 /*
448  * Abort and destroy a socket.  Only one abort can be in progress
449  * at any given moment.
450  */
451 void
452 soabort(struct socket *so)
453 {
454 	soreference(so);
455 	so_pru_abort(so);
456 }
457 
458 void
459 soaborta(struct socket *so)
460 {
461 	soreference(so);
462 	so_pru_aborta(so);
463 }
464 
465 void
466 soabort_oncpu(struct socket *so)
467 {
468 	soreference(so);
469 	so_pru_abort_oncpu(so);
470 }
471 
472 /*
473  * so is passed in ref'd, which becomes owned by
474  * the cleared SS_NOFDREF flag.
475  */
476 int
477 soaccept(struct socket *so, struct sockaddr **nam)
478 {
479 	int error;
480 
481 	if ((so->so_state & SS_NOFDREF) == 0)
482 		panic("soaccept: !NOFDREF");
483 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
484 	error = so_pru_accept_direct(so, nam);
485 	return (error);
486 }
487 
488 int
489 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
490 {
491 	int error;
492 
493 	if (so->so_options & SO_ACCEPTCONN)
494 		return (EOPNOTSUPP);
495 	/*
496 	 * If protocol is connection-based, can only connect once.
497 	 * Otherwise, if connected, try to disconnect first.
498 	 * This allows user to disconnect by connecting to, e.g.,
499 	 * a null address.
500 	 */
501 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
502 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
503 	    (error = sodisconnect(so)))) {
504 		error = EISCONN;
505 	} else {
506 		/*
507 		 * Prevent accumulated error from previous connection
508 		 * from biting us.
509 		 */
510 		so->so_error = 0;
511 		error = so_pru_connect(so, nam, td);
512 	}
513 	return (error);
514 }
515 
516 int
517 soconnect2(struct socket *so1, struct socket *so2)
518 {
519 	int error;
520 
521 	error = so_pru_connect2(so1, so2);
522 	return (error);
523 }
524 
525 int
526 sodisconnect(struct socket *so)
527 {
528 	int error;
529 
530 	if ((so->so_state & SS_ISCONNECTED) == 0) {
531 		error = ENOTCONN;
532 		goto bad;
533 	}
534 	if (so->so_state & SS_ISDISCONNECTING) {
535 		error = EALREADY;
536 		goto bad;
537 	}
538 	error = so_pru_disconnect(so);
539 bad:
540 	return (error);
541 }
542 
543 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
544 /*
545  * Send on a socket.
546  * If send must go all at once and message is larger than
547  * send buffering, then hard error.
548  * Lock against other senders.
549  * If must go all at once and not enough room now, then
550  * inform user that this would block and do nothing.
551  * Otherwise, if nonblocking, send as much as possible.
552  * The data to be sent is described by "uio" if nonzero,
553  * otherwise by the mbuf chain "top" (which must be null
554  * if uio is not).  Data provided in mbuf chain must be small
555  * enough to send all at once.
556  *
557  * Returns nonzero on error, timeout or signal; callers
558  * must check for short counts if EINTR/ERESTART are returned.
559  * Data and control buffers are freed on return.
560  */
561 int
562 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
563 	struct mbuf *top, struct mbuf *control, int flags,
564 	struct thread *td)
565 {
566 	struct mbuf **mp;
567 	struct mbuf *m;
568 	size_t resid;
569 	int space, len;
570 	int clen = 0, error, dontroute, mlen;
571 	int atomic = sosendallatonce(so) || top;
572 	int pru_flags;
573 
574 	if (uio) {
575 		resid = uio->uio_resid;
576 	} else {
577 		resid = (size_t)top->m_pkthdr.len;
578 #ifdef INVARIANTS
579 		len = 0;
580 		for (m = top; m; m = m->m_next)
581 			len += m->m_len;
582 		KKASSERT(top->m_pkthdr.len == len);
583 #endif
584 	}
585 
586 	/*
587 	 * WARNING!  resid is unsigned, space and len are signed.  space
588 	 * 	     can wind up negative if the sockbuf is overcommitted.
589 	 *
590 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
591 	 * type sockets since that's an error.
592 	 */
593 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
594 		error = EINVAL;
595 		goto out;
596 	}
597 
598 	dontroute =
599 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
600 	    (so->so_proto->pr_flags & PR_ATOMIC);
601 	if (td->td_lwp != NULL)
602 		td->td_lwp->lwp_ru.ru_msgsnd++;
603 	if (control)
604 		clen = control->m_len;
605 #define	gotoerr(errcode)	{ error = errcode; goto release; }
606 
607 restart:
608 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
609 	if (error)
610 		goto out;
611 
612 	do {
613 		if (so->so_state & SS_CANTSENDMORE)
614 			gotoerr(EPIPE);
615 		if (so->so_error) {
616 			error = so->so_error;
617 			so->so_error = 0;
618 			goto release;
619 		}
620 		if ((so->so_state & SS_ISCONNECTED) == 0) {
621 			/*
622 			 * `sendto' and `sendmsg' is allowed on a connection-
623 			 * based socket if it supports implied connect.
624 			 * Return ENOTCONN if not connected and no address is
625 			 * supplied.
626 			 */
627 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
628 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
629 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
630 				    !(resid == 0 && clen != 0))
631 					gotoerr(ENOTCONN);
632 			} else if (addr == 0)
633 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
634 				   ENOTCONN : EDESTADDRREQ);
635 		}
636 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
637 		    clen > so->so_snd.ssb_hiwat) {
638 			gotoerr(EMSGSIZE);
639 		}
640 		space = ssb_space(&so->so_snd);
641 		if (flags & MSG_OOB)
642 			space += 1024;
643 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
644 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
645 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
646 				gotoerr(EWOULDBLOCK);
647 			ssb_unlock(&so->so_snd);
648 			error = ssb_wait(&so->so_snd);
649 			if (error)
650 				goto out;
651 			goto restart;
652 		}
653 		mp = &top;
654 		space -= clen;
655 		do {
656 		    if (uio == NULL) {
657 			/*
658 			 * Data is prepackaged in "top".
659 			 */
660 			resid = 0;
661 			if (flags & MSG_EOR)
662 				top->m_flags |= M_EOR;
663 		    } else do {
664 			if (resid > INT_MAX)
665 				resid = INT_MAX;
666 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
667 				   top == NULL ? M_PKTHDR : 0, &mlen);
668 			if (top == NULL) {
669 				m->m_pkthdr.len = 0;
670 				m->m_pkthdr.rcvif = NULL;
671 			}
672 			len = imin((int)szmin(mlen, resid), space);
673 			if (resid < MINCLSIZE) {
674 				/*
675 				 * For datagram protocols, leave room
676 				 * for protocol headers in first mbuf.
677 				 */
678 				if (atomic && top == 0 && len < mlen)
679 					MH_ALIGN(m, len);
680 			}
681 			space -= len;
682 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
683 			resid = uio->uio_resid;
684 			m->m_len = len;
685 			*mp = m;
686 			top->m_pkthdr.len += len;
687 			if (error)
688 				goto release;
689 			mp = &m->m_next;
690 			if (resid == 0) {
691 				if (flags & MSG_EOR)
692 					top->m_flags |= M_EOR;
693 				break;
694 			}
695 		    } while (space > 0 && atomic);
696 		    if (dontroute)
697 			    so->so_options |= SO_DONTROUTE;
698 		    if (flags & MSG_OOB) {
699 		    	    pru_flags = PRUS_OOB;
700 		    } else if ((flags & MSG_EOF) &&
701 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
702 			       (resid == 0)) {
703 			    /*
704 			     * If the user set MSG_EOF, the protocol
705 			     * understands this flag and nothing left to
706 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
707 			     */
708 		    	    pru_flags = PRUS_EOF;
709 		    } else if (resid > 0 && space > 0) {
710 			    /* If there is more to send, set PRUS_MORETOCOME */
711 		    	    pru_flags = PRUS_MORETOCOME;
712 		    } else {
713 		    	    pru_flags = 0;
714 		    }
715 		    /*
716 		     * XXX all the SS_CANTSENDMORE checks previously
717 		     * done could be out of date.  We could have recieved
718 		     * a reset packet in an interrupt or maybe we slept
719 		     * while doing page faults in uiomove() etc. We could
720 		     * probably recheck again inside the splnet() protection
721 		     * here, but there are probably other places that this
722 		     * also happens.  We must rethink this.
723 		     */
724 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
725 		    if (dontroute)
726 			    so->so_options &= ~SO_DONTROUTE;
727 		    clen = 0;
728 		    control = 0;
729 		    top = NULL;
730 		    mp = &top;
731 		    if (error)
732 			    goto release;
733 		} while (resid && space > 0);
734 	} while (resid);
735 
736 release:
737 	ssb_unlock(&so->so_snd);
738 out:
739 	if (top)
740 		m_freem(top);
741 	if (control)
742 		m_freem(control);
743 	return (error);
744 }
745 
746 /*
747  * A specialization of sosend() for UDP based on protocol-specific knowledge:
748  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
749  *	sosendallatonce() returns true,
750  *	the "atomic" variable is true,
751  *	and sosendudp() blocks until space is available for the entire send.
752  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
753  *	PR_IMPLOPCL flags set.
754  *   UDP has no out-of-band data.
755  *   UDP has no control data.
756  *   UDP does not support MSG_EOR.
757  */
758 int
759 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
760 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
761 {
762 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
763 	size_t resid;
764 	int error;
765 	int space;
766 
767 	if (td->td_lwp != NULL)
768 		td->td_lwp->lwp_ru.ru_msgsnd++;
769 	if (control)
770 		m_freem(control);
771 
772 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
773 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
774 
775 restart:
776 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
777 	if (error)
778 		goto out;
779 
780 	if (so->so_state & SS_CANTSENDMORE)
781 		gotoerr(EPIPE);
782 	if (so->so_error) {
783 		error = so->so_error;
784 		so->so_error = 0;
785 		goto release;
786 	}
787 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
788 		gotoerr(EDESTADDRREQ);
789 	if (resid > so->so_snd.ssb_hiwat)
790 		gotoerr(EMSGSIZE);
791 	space = ssb_space(&so->so_snd);
792 	if (uio && (space < 0 || (size_t)space < resid)) {
793 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
794 			gotoerr(EWOULDBLOCK);
795 		ssb_unlock(&so->so_snd);
796 		error = ssb_wait(&so->so_snd);
797 		if (error)
798 			goto out;
799 		goto restart;
800 	}
801 
802 	if (uio) {
803 		top = m_uiomove(uio);
804 		if (top == NULL)
805 			goto release;
806 	}
807 
808 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
809 	if (dontroute)
810 		so->so_options |= SO_DONTROUTE;
811 
812 	error = so_pru_send(so, 0, top, addr, NULL, td);
813 	top = NULL;		/* sent or freed in lower layer */
814 
815 	if (dontroute)
816 		so->so_options &= ~SO_DONTROUTE;
817 
818 release:
819 	ssb_unlock(&so->so_snd);
820 out:
821 	if (top)
822 		m_freem(top);
823 	return (error);
824 }
825 
826 int
827 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
828 	struct mbuf *top, struct mbuf *control, int flags,
829 	struct thread *td)
830 {
831 	struct mbuf **mp;
832 	struct mbuf *m;
833 	size_t resid;
834 	int space, len;
835 	int error, mlen;
836 	int allatonce;
837 	int pru_flags;
838 
839 	if (uio) {
840 		KKASSERT(top == NULL);
841 		allatonce = 0;
842 		resid = uio->uio_resid;
843 	} else {
844 		allatonce = 1;
845 		resid = (size_t)top->m_pkthdr.len;
846 #ifdef INVARIANTS
847 		len = 0;
848 		for (m = top; m; m = m->m_next)
849 			len += m->m_len;
850 		KKASSERT(top->m_pkthdr.len == len);
851 #endif
852 	}
853 
854 	/*
855 	 * WARNING!  resid is unsigned, space and len are signed.  space
856 	 * 	     can wind up negative if the sockbuf is overcommitted.
857 	 *
858 	 * Also check to make sure that MSG_EOR isn't used on TCP
859 	 */
860 	if (flags & MSG_EOR) {
861 		error = EINVAL;
862 		goto out;
863 	}
864 
865 	if (control) {
866 		/* TCP doesn't do control messages (rights, creds, etc) */
867 		if (control->m_len) {
868 			error = EINVAL;
869 			goto out;
870 		}
871 		m_freem(control);	/* empty control, just free it */
872 		control = NULL;
873 	}
874 
875 	if (td->td_lwp != NULL)
876 		td->td_lwp->lwp_ru.ru_msgsnd++;
877 
878 #define	gotoerr(errcode)	{ error = errcode; goto release; }
879 
880 restart:
881 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
882 	if (error)
883 		goto out;
884 
885 	do {
886 		if (so->so_state & SS_CANTSENDMORE)
887 			gotoerr(EPIPE);
888 		if (so->so_error) {
889 			error = so->so_error;
890 			so->so_error = 0;
891 			goto release;
892 		}
893 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
894 		    (so->so_state & SS_ISCONFIRMING) == 0)
895 			gotoerr(ENOTCONN);
896 		if (allatonce && resid > so->so_snd.ssb_hiwat)
897 			gotoerr(EMSGSIZE);
898 
899 		space = ssb_space(&so->so_snd);
900 		if (flags & MSG_OOB)
901 			space += 1024;
902 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
903 		    space < so->so_snd.ssb_lowat) {
904 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
905 				gotoerr(EWOULDBLOCK);
906 			ssb_unlock(&so->so_snd);
907 			error = ssb_wait(&so->so_snd);
908 			if (error)
909 				goto out;
910 			goto restart;
911 		}
912 		mp = &top;
913 		do {
914 		    int cnt = 0, async = 0;
915 
916 		    if (uio == NULL) {
917 			/*
918 			 * Data is prepackaged in "top".
919 			 */
920 			resid = 0;
921 		    } else do {
922 			if (resid > INT_MAX)
923 				resid = INT_MAX;
924 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
925 				   top == NULL ? M_PKTHDR : 0, &mlen);
926 			if (top == NULL) {
927 				m->m_pkthdr.len = 0;
928 				m->m_pkthdr.rcvif = NULL;
929 			}
930 			len = imin((int)szmin(mlen, resid), space);
931 			space -= len;
932 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
933 			resid = uio->uio_resid;
934 			m->m_len = len;
935 			*mp = m;
936 			top->m_pkthdr.len += len;
937 			if (error)
938 				goto release;
939 			mp = &m->m_next;
940 			if (resid == 0)
941 				break;
942 			++cnt;
943 		    } while (space > 0 && cnt < tcp_sosnd_agglim);
944 
945 		    if (tcp_sosnd_async)
946 			    async = 1;
947 
948 		    if (flags & MSG_OOB) {
949 		    	    pru_flags = PRUS_OOB;
950 			    async = 0;
951 		    } else if ((flags & MSG_EOF) && resid == 0) {
952 			    pru_flags = PRUS_EOF;
953 		    } else if (resid > 0 && space > 0) {
954 			    /* If there is more to send, set PRUS_MORETOCOME */
955 		    	    pru_flags = PRUS_MORETOCOME;
956 			    async = 1;
957 		    } else {
958 		    	    pru_flags = 0;
959 		    }
960 
961 		    if (flags & MSG_SYNC)
962 			    async = 0;
963 
964 		    /*
965 		     * XXX all the SS_CANTSENDMORE checks previously
966 		     * done could be out of date.  We could have recieved
967 		     * a reset packet in an interrupt or maybe we slept
968 		     * while doing page faults in uiomove() etc. We could
969 		     * probably recheck again inside the splnet() protection
970 		     * here, but there are probably other places that this
971 		     * also happens.  We must rethink this.
972 		     */
973 		    if (!async) {
974 			    error = so_pru_send(so, pru_flags, top,
975 			        NULL, NULL, td);
976 		    } else {
977 			    so_pru_send_async(so, pru_flags, top,
978 			        NULL, NULL, td);
979 			    error = 0;
980 		    }
981 
982 		    top = NULL;
983 		    mp = &top;
984 		    if (error)
985 			    goto release;
986 		} while (resid && space > 0);
987 	} while (resid);
988 
989 release:
990 	ssb_unlock(&so->so_snd);
991 out:
992 	if (top)
993 		m_freem(top);
994 	if (control)
995 		m_freem(control);
996 	return (error);
997 }
998 
999 /*
1000  * Implement receive operations on a socket.
1001  *
1002  * We depend on the way that records are added to the signalsockbuf
1003  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1004  * must begin with an address if the protocol so specifies,
1005  * followed by an optional mbuf or mbufs containing ancillary data,
1006  * and then zero or more mbufs of data.
1007  *
1008  * Although the signalsockbuf is locked, new data may still be appended.
1009  * A token inside the ssb_lock deals with MP issues and still allows
1010  * the network to access the socket if we block in a uio.
1011  *
1012  * The caller may receive the data as a single mbuf chain by supplying
1013  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1014  * only for the count in uio_resid.
1015  */
1016 int
1017 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1018 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1019 {
1020 	struct mbuf *m, *n;
1021 	struct mbuf *free_chain = NULL;
1022 	int flags, len, error, offset;
1023 	struct protosw *pr = so->so_proto;
1024 	int moff, type = 0;
1025 	size_t resid, orig_resid;
1026 
1027 	if (uio)
1028 		resid = uio->uio_resid;
1029 	else
1030 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1031 	orig_resid = resid;
1032 
1033 	if (psa)
1034 		*psa = NULL;
1035 	if (controlp)
1036 		*controlp = NULL;
1037 	if (flagsp)
1038 		flags = *flagsp &~ MSG_EOR;
1039 	else
1040 		flags = 0;
1041 	if (flags & MSG_OOB) {
1042 		m = m_get(MB_WAIT, MT_DATA);
1043 		if (m == NULL)
1044 			return (ENOBUFS);
1045 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1046 		if (error)
1047 			goto bad;
1048 		if (sio) {
1049 			do {
1050 				sbappend(sio, m);
1051 				KKASSERT(resid >= (size_t)m->m_len);
1052 				resid -= (size_t)m->m_len;
1053 			} while (resid > 0 && m);
1054 		} else {
1055 			do {
1056 				uio->uio_resid = resid;
1057 				error = uiomove(mtod(m, caddr_t),
1058 						(int)szmin(resid, m->m_len),
1059 						uio);
1060 				resid = uio->uio_resid;
1061 				m = m_free(m);
1062 			} while (uio->uio_resid && error == 0 && m);
1063 		}
1064 bad:
1065 		if (m)
1066 			m_freem(m);
1067 		return (error);
1068 	}
1069 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1070 		so_pru_rcvd(so, 0);
1071 
1072 	/*
1073 	 * The token interlocks against the protocol thread while
1074 	 * ssb_lock is a blocking lock against other userland entities.
1075 	 */
1076 	lwkt_gettoken(&so->so_rcv.ssb_token);
1077 restart:
1078 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1079 	if (error)
1080 		goto done;
1081 
1082 	m = so->so_rcv.ssb_mb;
1083 	/*
1084 	 * If we have less data than requested, block awaiting more
1085 	 * (subject to any timeout) if:
1086 	 *   1. the current count is less than the low water mark, or
1087 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1088 	 *	receive operation at once if we block (resid <= hiwat).
1089 	 *   3. MSG_DONTWAIT is not set
1090 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1091 	 * we have to do the receive in sections, and thus risk returning
1092 	 * a short count if a timeout or signal occurs after we start.
1093 	 */
1094 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1095 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1096 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1097 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1098 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1099 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1100 		if (so->so_error) {
1101 			if (m)
1102 				goto dontblock;
1103 			error = so->so_error;
1104 			if ((flags & MSG_PEEK) == 0)
1105 				so->so_error = 0;
1106 			goto release;
1107 		}
1108 		if (so->so_state & SS_CANTRCVMORE) {
1109 			if (m)
1110 				goto dontblock;
1111 			else
1112 				goto release;
1113 		}
1114 		for (; m; m = m->m_next) {
1115 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1116 				m = so->so_rcv.ssb_mb;
1117 				goto dontblock;
1118 			}
1119 		}
1120 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1121 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1122 			error = ENOTCONN;
1123 			goto release;
1124 		}
1125 		if (resid == 0)
1126 			goto release;
1127 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1128 			error = EWOULDBLOCK;
1129 			goto release;
1130 		}
1131 		ssb_unlock(&so->so_rcv);
1132 		error = ssb_wait(&so->so_rcv);
1133 		if (error)
1134 			goto done;
1135 		goto restart;
1136 	}
1137 dontblock:
1138 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1139 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1140 
1141 	/*
1142 	 * note: m should be == sb_mb here.  Cache the next record while
1143 	 * cleaning up.  Note that calling m_free*() will break out critical
1144 	 * section.
1145 	 */
1146 	KKASSERT(m == so->so_rcv.ssb_mb);
1147 
1148 	/*
1149 	 * Skip any address mbufs prepending the record.
1150 	 */
1151 	if (pr->pr_flags & PR_ADDR) {
1152 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1153 		orig_resid = 0;
1154 		if (psa)
1155 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1156 		if (flags & MSG_PEEK)
1157 			m = m->m_next;
1158 		else
1159 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1160 	}
1161 
1162 	/*
1163 	 * Skip any control mbufs prepending the record.
1164 	 */
1165 #ifdef SCTP
1166 	if (pr->pr_flags & PR_ADDR_OPT) {
1167 		/*
1168 		 * For SCTP we may be getting a
1169 		 * whole message OR a partial delivery.
1170 		 */
1171 		if (m && m->m_type == MT_SONAME) {
1172 			orig_resid = 0;
1173 			if (psa)
1174 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1175 			if (flags & MSG_PEEK)
1176 				m = m->m_next;
1177 			else
1178 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1179 		}
1180 	}
1181 #endif /* SCTP */
1182 	while (m && m->m_type == MT_CONTROL && error == 0) {
1183 		if (flags & MSG_PEEK) {
1184 			if (controlp)
1185 				*controlp = m_copy(m, 0, m->m_len);
1186 			m = m->m_next;	/* XXX race */
1187 		} else {
1188 			if (controlp) {
1189 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1190 				if (pr->pr_domain->dom_externalize &&
1191 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1192 				    SCM_RIGHTS)
1193 				   error = (*pr->pr_domain->dom_externalize)(m);
1194 				*controlp = m;
1195 				m = n;
1196 			} else {
1197 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1198 			}
1199 		}
1200 		if (controlp && *controlp) {
1201 			orig_resid = 0;
1202 			controlp = &(*controlp)->m_next;
1203 		}
1204 	}
1205 
1206 	/*
1207 	 * flag OOB data.
1208 	 */
1209 	if (m) {
1210 		type = m->m_type;
1211 		if (type == MT_OOBDATA)
1212 			flags |= MSG_OOB;
1213 	}
1214 
1215 	/*
1216 	 * Copy to the UIO or mbuf return chain (*mp).
1217 	 */
1218 	moff = 0;
1219 	offset = 0;
1220 	while (m && resid > 0 && error == 0) {
1221 		if (m->m_type == MT_OOBDATA) {
1222 			if (type != MT_OOBDATA)
1223 				break;
1224 		} else if (type == MT_OOBDATA)
1225 			break;
1226 		else
1227 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1228 			("receive 3"));
1229 		soclrstate(so, SS_RCVATMARK);
1230 		len = (resid > INT_MAX) ? INT_MAX : resid;
1231 		if (so->so_oobmark && len > so->so_oobmark - offset)
1232 			len = so->so_oobmark - offset;
1233 		if (len > m->m_len - moff)
1234 			len = m->m_len - moff;
1235 
1236 		/*
1237 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1238 		 * The SIO is dealt with when we eat the mbuf, but deal
1239 		 * with the resid here either way.
1240 		 */
1241 		if (uio) {
1242 			uio->uio_resid = resid;
1243 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1244 			resid = uio->uio_resid;
1245 			if (error)
1246 				goto release;
1247 		} else {
1248 			resid -= (size_t)len;
1249 		}
1250 
1251 		/*
1252 		 * Eat the entire mbuf or just a piece of it
1253 		 */
1254 		if (len == m->m_len - moff) {
1255 			if (m->m_flags & M_EOR)
1256 				flags |= MSG_EOR;
1257 #ifdef SCTP
1258 			if (m->m_flags & M_NOTIFICATION)
1259 				flags |= MSG_NOTIFICATION;
1260 #endif /* SCTP */
1261 			if (flags & MSG_PEEK) {
1262 				m = m->m_next;
1263 				moff = 0;
1264 			} else {
1265 				if (sio) {
1266 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1267 					sbappend(sio, m);
1268 					m = n;
1269 				} else {
1270 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1271 				}
1272 			}
1273 		} else {
1274 			if (flags & MSG_PEEK) {
1275 				moff += len;
1276 			} else {
1277 				if (sio) {
1278 					n = m_copym(m, 0, len, MB_WAIT);
1279 					if (n)
1280 						sbappend(sio, n);
1281 				}
1282 				m->m_data += len;
1283 				m->m_len -= len;
1284 				so->so_rcv.ssb_cc -= len;
1285 			}
1286 		}
1287 		if (so->so_oobmark) {
1288 			if ((flags & MSG_PEEK) == 0) {
1289 				so->so_oobmark -= len;
1290 				if (so->so_oobmark == 0) {
1291 					sosetstate(so, SS_RCVATMARK);
1292 					break;
1293 				}
1294 			} else {
1295 				offset += len;
1296 				if (offset == so->so_oobmark)
1297 					break;
1298 			}
1299 		}
1300 		if (flags & MSG_EOR)
1301 			break;
1302 		/*
1303 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1304 		 * we must not quit until resid == 0 or an error
1305 		 * termination.  If a signal/timeout occurs, return
1306 		 * with a short count but without error.
1307 		 * Keep signalsockbuf locked against other readers.
1308 		 */
1309 		while ((flags & MSG_WAITALL) && m == NULL &&
1310 		       resid > 0 && !sosendallatonce(so) &&
1311 		       so->so_rcv.ssb_mb == NULL) {
1312 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1313 				break;
1314 			/*
1315 			 * The window might have closed to zero, make
1316 			 * sure we send an ack now that we've drained
1317 			 * the buffer or we might end up blocking until
1318 			 * the idle takes over (5 seconds).
1319 			 */
1320 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1321 				so_pru_rcvd(so, flags);
1322 			error = ssb_wait(&so->so_rcv);
1323 			if (error) {
1324 				ssb_unlock(&so->so_rcv);
1325 				error = 0;
1326 				goto done;
1327 			}
1328 			m = so->so_rcv.ssb_mb;
1329 		}
1330 	}
1331 
1332 	/*
1333 	 * If an atomic read was requested but unread data still remains
1334 	 * in the record, set MSG_TRUNC.
1335 	 */
1336 	if (m && pr->pr_flags & PR_ATOMIC)
1337 		flags |= MSG_TRUNC;
1338 
1339 	/*
1340 	 * Cleanup.  If an atomic read was requested drop any unread data.
1341 	 */
1342 	if ((flags & MSG_PEEK) == 0) {
1343 		if (m && (pr->pr_flags & PR_ATOMIC))
1344 			sbdroprecord(&so->so_rcv.sb);
1345 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1346 			so_pru_rcvd(so, flags);
1347 	}
1348 
1349 	if (orig_resid == resid && orig_resid &&
1350 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1351 		ssb_unlock(&so->so_rcv);
1352 		goto restart;
1353 	}
1354 
1355 	if (flagsp)
1356 		*flagsp |= flags;
1357 release:
1358 	ssb_unlock(&so->so_rcv);
1359 done:
1360 	lwkt_reltoken(&so->so_rcv.ssb_token);
1361 	if (free_chain)
1362 		m_freem(free_chain);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * Shut a socket down.  Note that we do not get a frontend lock as we
1368  * want to be able to shut the socket down even if another thread is
1369  * blocked in a read(), thus waking it up.
1370  */
1371 int
1372 soshutdown(struct socket *so, int how)
1373 {
1374 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1375 		return (EINVAL);
1376 
1377 	if (how != SHUT_WR) {
1378 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1379 		sorflush(so);
1380 		/*ssb_unlock(&so->so_rcv);*/
1381 	}
1382 	if (how != SHUT_RD)
1383 		return (so_pru_shutdown(so));
1384 	return (0);
1385 }
1386 
1387 void
1388 sorflush(struct socket *so)
1389 {
1390 	struct signalsockbuf *ssb = &so->so_rcv;
1391 	struct protosw *pr = so->so_proto;
1392 	struct signalsockbuf asb;
1393 
1394 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1395 
1396 	lwkt_gettoken(&ssb->ssb_token);
1397 	socantrcvmore(so);
1398 	asb = *ssb;
1399 
1400 	/*
1401 	 * Can't just blow up the ssb structure here
1402 	 */
1403 	bzero(&ssb->sb, sizeof(ssb->sb));
1404 	ssb->ssb_timeo = 0;
1405 	ssb->ssb_lowat = 0;
1406 	ssb->ssb_hiwat = 0;
1407 	ssb->ssb_mbmax = 0;
1408 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1409 
1410 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1411 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1412 	ssb_release(&asb, so);
1413 
1414 	lwkt_reltoken(&ssb->ssb_token);
1415 }
1416 
1417 #ifdef INET
1418 static int
1419 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1420 {
1421 	struct accept_filter_arg	*afap = NULL;
1422 	struct accept_filter	*afp;
1423 	struct so_accf	*af = so->so_accf;
1424 	int	error = 0;
1425 
1426 	/* do not set/remove accept filters on non listen sockets */
1427 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1428 		error = EINVAL;
1429 		goto out;
1430 	}
1431 
1432 	/* removing the filter */
1433 	if (sopt == NULL) {
1434 		if (af != NULL) {
1435 			if (af->so_accept_filter != NULL &&
1436 				af->so_accept_filter->accf_destroy != NULL) {
1437 				af->so_accept_filter->accf_destroy(so);
1438 			}
1439 			if (af->so_accept_filter_str != NULL) {
1440 				FREE(af->so_accept_filter_str, M_ACCF);
1441 			}
1442 			FREE(af, M_ACCF);
1443 			so->so_accf = NULL;
1444 		}
1445 		so->so_options &= ~SO_ACCEPTFILTER;
1446 		return (0);
1447 	}
1448 	/* adding a filter */
1449 	/* must remove previous filter first */
1450 	if (af != NULL) {
1451 		error = EINVAL;
1452 		goto out;
1453 	}
1454 	/* don't put large objects on the kernel stack */
1455 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1456 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1457 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1458 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1459 	if (error)
1460 		goto out;
1461 	afp = accept_filt_get(afap->af_name);
1462 	if (afp == NULL) {
1463 		error = ENOENT;
1464 		goto out;
1465 	}
1466 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1467 	if (afp->accf_create != NULL) {
1468 		if (afap->af_name[0] != '\0') {
1469 			int len = strlen(afap->af_name) + 1;
1470 
1471 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1472 			strcpy(af->so_accept_filter_str, afap->af_name);
1473 		}
1474 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1475 		if (af->so_accept_filter_arg == NULL) {
1476 			FREE(af->so_accept_filter_str, M_ACCF);
1477 			FREE(af, M_ACCF);
1478 			so->so_accf = NULL;
1479 			error = EINVAL;
1480 			goto out;
1481 		}
1482 	}
1483 	af->so_accept_filter = afp;
1484 	so->so_accf = af;
1485 	so->so_options |= SO_ACCEPTFILTER;
1486 out:
1487 	if (afap != NULL)
1488 		FREE(afap, M_TEMP);
1489 	return (error);
1490 }
1491 #endif /* INET */
1492 
1493 /*
1494  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1495  * an additional variant to handle the case where the option value needs
1496  * to be some kind of integer, but not a specific size.
1497  * In addition to their use here, these functions are also called by the
1498  * protocol-level pr_ctloutput() routines.
1499  */
1500 int
1501 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1502 {
1503 	return soopt_to_kbuf(sopt, buf, len, minlen);
1504 }
1505 
1506 int
1507 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1508 {
1509 	size_t	valsize;
1510 
1511 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1512 	KKASSERT(kva_p(buf));
1513 
1514 	/*
1515 	 * If the user gives us more than we wanted, we ignore it,
1516 	 * but if we don't get the minimum length the caller
1517 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1518 	 * is set to however much we actually retrieved.
1519 	 */
1520 	if ((valsize = sopt->sopt_valsize) < minlen)
1521 		return EINVAL;
1522 	if (valsize > len)
1523 		sopt->sopt_valsize = valsize = len;
1524 
1525 	bcopy(sopt->sopt_val, buf, valsize);
1526 	return 0;
1527 }
1528 
1529 
1530 int
1531 sosetopt(struct socket *so, struct sockopt *sopt)
1532 {
1533 	int	error, optval;
1534 	struct	linger l;
1535 	struct	timeval tv;
1536 	u_long  val;
1537 	struct signalsockbuf *sotmp;
1538 
1539 	error = 0;
1540 	sopt->sopt_dir = SOPT_SET;
1541 	if (sopt->sopt_level != SOL_SOCKET) {
1542 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1543 			return (so_pr_ctloutput(so, sopt));
1544 		}
1545 		error = ENOPROTOOPT;
1546 	} else {
1547 		switch (sopt->sopt_name) {
1548 #ifdef INET
1549 		case SO_ACCEPTFILTER:
1550 			error = do_setopt_accept_filter(so, sopt);
1551 			if (error)
1552 				goto bad;
1553 			break;
1554 #endif /* INET */
1555 		case SO_LINGER:
1556 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1557 			if (error)
1558 				goto bad;
1559 
1560 			so->so_linger = l.l_linger;
1561 			if (l.l_onoff)
1562 				so->so_options |= SO_LINGER;
1563 			else
1564 				so->so_options &= ~SO_LINGER;
1565 			break;
1566 
1567 		case SO_DEBUG:
1568 		case SO_KEEPALIVE:
1569 		case SO_DONTROUTE:
1570 		case SO_USELOOPBACK:
1571 		case SO_BROADCAST:
1572 		case SO_REUSEADDR:
1573 		case SO_REUSEPORT:
1574 		case SO_OOBINLINE:
1575 		case SO_TIMESTAMP:
1576 			error = sooptcopyin(sopt, &optval, sizeof optval,
1577 					    sizeof optval);
1578 			if (error)
1579 				goto bad;
1580 			if (optval)
1581 				so->so_options |= sopt->sopt_name;
1582 			else
1583 				so->so_options &= ~sopt->sopt_name;
1584 			break;
1585 
1586 		case SO_SNDBUF:
1587 		case SO_RCVBUF:
1588 		case SO_SNDLOWAT:
1589 		case SO_RCVLOWAT:
1590 			error = sooptcopyin(sopt, &optval, sizeof optval,
1591 					    sizeof optval);
1592 			if (error)
1593 				goto bad;
1594 
1595 			/*
1596 			 * Values < 1 make no sense for any of these
1597 			 * options, so disallow them.
1598 			 */
1599 			if (optval < 1) {
1600 				error = EINVAL;
1601 				goto bad;
1602 			}
1603 
1604 			switch (sopt->sopt_name) {
1605 			case SO_SNDBUF:
1606 			case SO_RCVBUF:
1607 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1608 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1609 				    so,
1610 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1611 					error = ENOBUFS;
1612 					goto bad;
1613 				}
1614 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1615 						&so->so_snd : &so->so_rcv;
1616 				atomic_clear_int(&sotmp->ssb_flags,
1617 						 SSB_AUTOSIZE);
1618 				break;
1619 
1620 			/*
1621 			 * Make sure the low-water is never greater than
1622 			 * the high-water.
1623 			 */
1624 			case SO_SNDLOWAT:
1625 				so->so_snd.ssb_lowat =
1626 				    (optval > so->so_snd.ssb_hiwat) ?
1627 				    so->so_snd.ssb_hiwat : optval;
1628 				atomic_clear_int(&so->so_snd.ssb_flags,
1629 						 SSB_AUTOLOWAT);
1630 				break;
1631 			case SO_RCVLOWAT:
1632 				so->so_rcv.ssb_lowat =
1633 				    (optval > so->so_rcv.ssb_hiwat) ?
1634 				    so->so_rcv.ssb_hiwat : optval;
1635 				atomic_clear_int(&so->so_rcv.ssb_flags,
1636 						 SSB_AUTOLOWAT);
1637 				break;
1638 			}
1639 			break;
1640 
1641 		case SO_SNDTIMEO:
1642 		case SO_RCVTIMEO:
1643 			error = sooptcopyin(sopt, &tv, sizeof tv,
1644 					    sizeof tv);
1645 			if (error)
1646 				goto bad;
1647 
1648 			/* assert(hz > 0); */
1649 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1650 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1651 				error = EDOM;
1652 				goto bad;
1653 			}
1654 			/* assert(tick > 0); */
1655 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
1656 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1657 			if (val > INT_MAX) {
1658 				error = EDOM;
1659 				goto bad;
1660 			}
1661 			if (val == 0 && tv.tv_usec != 0)
1662 				val = 1;
1663 
1664 			switch (sopt->sopt_name) {
1665 			case SO_SNDTIMEO:
1666 				so->so_snd.ssb_timeo = val;
1667 				break;
1668 			case SO_RCVTIMEO:
1669 				so->so_rcv.ssb_timeo = val;
1670 				break;
1671 			}
1672 			break;
1673 		default:
1674 			error = ENOPROTOOPT;
1675 			break;
1676 		}
1677 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1678 			(void) so_pr_ctloutput(so, sopt);
1679 		}
1680 	}
1681 bad:
1682 	return (error);
1683 }
1684 
1685 /* Helper routine for getsockopt */
1686 int
1687 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1688 {
1689 	soopt_from_kbuf(sopt, buf, len);
1690 	return 0;
1691 }
1692 
1693 void
1694 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1695 {
1696 	size_t	valsize;
1697 
1698 	if (len == 0) {
1699 		sopt->sopt_valsize = 0;
1700 		return;
1701 	}
1702 
1703 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1704 	KKASSERT(kva_p(buf));
1705 
1706 	/*
1707 	 * Documented get behavior is that we always return a value,
1708 	 * possibly truncated to fit in the user's buffer.
1709 	 * Traditional behavior is that we always tell the user
1710 	 * precisely how much we copied, rather than something useful
1711 	 * like the total amount we had available for her.
1712 	 * Note that this interface is not idempotent; the entire answer must
1713 	 * generated ahead of time.
1714 	 */
1715 	valsize = szmin(len, sopt->sopt_valsize);
1716 	sopt->sopt_valsize = valsize;
1717 	if (sopt->sopt_val != 0) {
1718 		bcopy(buf, sopt->sopt_val, valsize);
1719 	}
1720 }
1721 
1722 int
1723 sogetopt(struct socket *so, struct sockopt *sopt)
1724 {
1725 	int	error, optval;
1726 	long	optval_l;
1727 	struct	linger l;
1728 	struct	timeval tv;
1729 #ifdef INET
1730 	struct accept_filter_arg *afap;
1731 #endif
1732 
1733 	error = 0;
1734 	sopt->sopt_dir = SOPT_GET;
1735 	if (sopt->sopt_level != SOL_SOCKET) {
1736 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1737 			return (so_pr_ctloutput(so, sopt));
1738 		} else
1739 			return (ENOPROTOOPT);
1740 	} else {
1741 		switch (sopt->sopt_name) {
1742 #ifdef INET
1743 		case SO_ACCEPTFILTER:
1744 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1745 				return (EINVAL);
1746 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1747 				M_TEMP, M_WAITOK | M_ZERO);
1748 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1749 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1750 				if (so->so_accf->so_accept_filter_str != NULL)
1751 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1752 			}
1753 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1754 			FREE(afap, M_TEMP);
1755 			break;
1756 #endif /* INET */
1757 
1758 		case SO_LINGER:
1759 			l.l_onoff = so->so_options & SO_LINGER;
1760 			l.l_linger = so->so_linger;
1761 			error = sooptcopyout(sopt, &l, sizeof l);
1762 			break;
1763 
1764 		case SO_USELOOPBACK:
1765 		case SO_DONTROUTE:
1766 		case SO_DEBUG:
1767 		case SO_KEEPALIVE:
1768 		case SO_REUSEADDR:
1769 		case SO_REUSEPORT:
1770 		case SO_BROADCAST:
1771 		case SO_OOBINLINE:
1772 		case SO_TIMESTAMP:
1773 			optval = so->so_options & sopt->sopt_name;
1774 integer:
1775 			error = sooptcopyout(sopt, &optval, sizeof optval);
1776 			break;
1777 
1778 		case SO_TYPE:
1779 			optval = so->so_type;
1780 			goto integer;
1781 
1782 		case SO_ERROR:
1783 			optval = so->so_error;
1784 			so->so_error = 0;
1785 			goto integer;
1786 
1787 		case SO_SNDBUF:
1788 			optval = so->so_snd.ssb_hiwat;
1789 			goto integer;
1790 
1791 		case SO_RCVBUF:
1792 			optval = so->so_rcv.ssb_hiwat;
1793 			goto integer;
1794 
1795 		case SO_SNDLOWAT:
1796 			optval = so->so_snd.ssb_lowat;
1797 			goto integer;
1798 
1799 		case SO_RCVLOWAT:
1800 			optval = so->so_rcv.ssb_lowat;
1801 			goto integer;
1802 
1803 		case SO_SNDTIMEO:
1804 		case SO_RCVTIMEO:
1805 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1806 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1807 
1808 			tv.tv_sec = optval / hz;
1809 			tv.tv_usec = (optval % hz) * ustick;
1810 			error = sooptcopyout(sopt, &tv, sizeof tv);
1811 			break;
1812 
1813 		case SO_SNDSPACE:
1814 			optval_l = ssb_space(&so->so_snd);
1815 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1816 			break;
1817 
1818 		default:
1819 			error = ENOPROTOOPT;
1820 			break;
1821 		}
1822 		return (error);
1823 	}
1824 }
1825 
1826 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1827 int
1828 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1829 {
1830 	struct mbuf *m, *m_prev;
1831 	int sopt_size = sopt->sopt_valsize, msize;
1832 
1833 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1834 		   0, &msize);
1835 	if (m == NULL)
1836 		return (ENOBUFS);
1837 	m->m_len = min(msize, sopt_size);
1838 	sopt_size -= m->m_len;
1839 	*mp = m;
1840 	m_prev = m;
1841 
1842 	while (sopt_size > 0) {
1843 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1844 			   MT_DATA, 0, &msize);
1845 		if (m == NULL) {
1846 			m_freem(*mp);
1847 			return (ENOBUFS);
1848 		}
1849 		m->m_len = min(msize, sopt_size);
1850 		sopt_size -= m->m_len;
1851 		m_prev->m_next = m;
1852 		m_prev = m;
1853 	}
1854 	return (0);
1855 }
1856 
1857 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1858 int
1859 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1860 {
1861 	soopt_to_mbuf(sopt, m);
1862 	return 0;
1863 }
1864 
1865 void
1866 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1867 {
1868 	size_t valsize;
1869 	void *val;
1870 
1871 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1872 	KKASSERT(kva_p(m));
1873 	if (sopt->sopt_val == NULL)
1874 		return;
1875 	val = sopt->sopt_val;
1876 	valsize = sopt->sopt_valsize;
1877 	while (m != NULL && valsize >= m->m_len) {
1878 		bcopy(val, mtod(m, char *), m->m_len);
1879 		valsize -= m->m_len;
1880 		val = (caddr_t)val + m->m_len;
1881 		m = m->m_next;
1882 	}
1883 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1884 		panic("ip6_sooptmcopyin");
1885 }
1886 
1887 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1888 int
1889 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1890 {
1891 	return soopt_from_mbuf(sopt, m);
1892 }
1893 
1894 int
1895 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1896 {
1897 	struct mbuf *m0 = m;
1898 	size_t valsize = 0;
1899 	size_t maxsize;
1900 	void *val;
1901 
1902 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1903 	KKASSERT(kva_p(m));
1904 	if (sopt->sopt_val == NULL)
1905 		return 0;
1906 	val = sopt->sopt_val;
1907 	maxsize = sopt->sopt_valsize;
1908 	while (m != NULL && maxsize >= m->m_len) {
1909 		bcopy(mtod(m, char *), val, m->m_len);
1910 	       maxsize -= m->m_len;
1911 	       val = (caddr_t)val + m->m_len;
1912 	       valsize += m->m_len;
1913 	       m = m->m_next;
1914 	}
1915 	if (m != NULL) {
1916 		/* enough soopt buffer should be given from user-land */
1917 		m_freem(m0);
1918 		return (EINVAL);
1919 	}
1920 	sopt->sopt_valsize = valsize;
1921 	return 0;
1922 }
1923 
1924 void
1925 sohasoutofband(struct socket *so)
1926 {
1927 	if (so->so_sigio != NULL)
1928 		pgsigio(so->so_sigio, SIGURG, 0);
1929 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1930 }
1931 
1932 int
1933 sokqfilter(struct file *fp, struct knote *kn)
1934 {
1935 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1936 	struct signalsockbuf *ssb;
1937 
1938 	switch (kn->kn_filter) {
1939 	case EVFILT_READ:
1940 		if (so->so_options & SO_ACCEPTCONN)
1941 			kn->kn_fop = &solisten_filtops;
1942 		else
1943 			kn->kn_fop = &soread_filtops;
1944 		ssb = &so->so_rcv;
1945 		break;
1946 	case EVFILT_WRITE:
1947 		kn->kn_fop = &sowrite_filtops;
1948 		ssb = &so->so_snd;
1949 		break;
1950 	case EVFILT_EXCEPT:
1951 		kn->kn_fop = &soexcept_filtops;
1952 		ssb = &so->so_rcv;
1953 		break;
1954 	default:
1955 		return (EOPNOTSUPP);
1956 	}
1957 
1958 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1959 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1960 	return (0);
1961 }
1962 
1963 static void
1964 filt_sordetach(struct knote *kn)
1965 {
1966 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1967 
1968 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1969 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1970 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1971 }
1972 
1973 /*ARGSUSED*/
1974 static int
1975 filt_soread(struct knote *kn, long hint)
1976 {
1977 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1978 
1979 	if (kn->kn_sfflags & NOTE_OOB) {
1980 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1981 			kn->kn_fflags |= NOTE_OOB;
1982 			return (1);
1983 		}
1984 		return (0);
1985 	}
1986 	kn->kn_data = so->so_rcv.ssb_cc;
1987 
1988 	if (so->so_state & SS_CANTRCVMORE) {
1989 		/*
1990 		 * Only set NODATA if all data has been exhausted.
1991 		 */
1992 		if (kn->kn_data == 0)
1993 			kn->kn_flags |= EV_NODATA;
1994 		kn->kn_flags |= EV_EOF;
1995 		kn->kn_fflags = so->so_error;
1996 		return (1);
1997 	}
1998 	if (so->so_error)	/* temporary udp error */
1999 		return (1);
2000 	if (kn->kn_sfflags & NOTE_LOWAT)
2001 		return (kn->kn_data >= kn->kn_sdata);
2002 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2003 		!TAILQ_EMPTY(&so->so_comp));
2004 }
2005 
2006 static void
2007 filt_sowdetach(struct knote *kn)
2008 {
2009 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2010 
2011 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2012 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2013 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2014 }
2015 
2016 /*ARGSUSED*/
2017 static int
2018 filt_sowrite(struct knote *kn, long hint)
2019 {
2020 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2021 
2022 	kn->kn_data = ssb_space(&so->so_snd);
2023 	if (so->so_state & SS_CANTSENDMORE) {
2024 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2025 		kn->kn_fflags = so->so_error;
2026 		return (1);
2027 	}
2028 	if (so->so_error)	/* temporary udp error */
2029 		return (1);
2030 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2031 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2032 		return (0);
2033 	if (kn->kn_sfflags & NOTE_LOWAT)
2034 		return (kn->kn_data >= kn->kn_sdata);
2035 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2036 }
2037 
2038 /*ARGSUSED*/
2039 static int
2040 filt_solisten(struct knote *kn, long hint)
2041 {
2042 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2043 
2044 	kn->kn_data = so->so_qlen;
2045 	return (! TAILQ_EMPTY(&so->so_comp));
2046 }
2047