xref: /dflybsd-src/sys/kern/uipc_socket.c (revision 0e1bdf492f13155bc45fb3fcd056ee51d3c763c0)
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 
102 #ifdef INET
103 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
104 #endif /* INET */
105 
106 static void 	filt_sordetach(struct knote *kn);
107 static int 	filt_soread(struct knote *kn, long hint);
108 static void 	filt_sowdetach(struct knote *kn);
109 static int	filt_sowrite(struct knote *kn, long hint);
110 static int	filt_solisten(struct knote *kn, long hint);
111 
112 static struct filterops solisten_filtops =
113 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
114 static struct filterops soread_filtops =
115 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
116 static struct filterops sowrite_filtops =
117 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
118 static struct filterops soexcept_filtops =
119 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
120 
121 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
122 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
123 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
124 
125 
126 static int somaxconn = SOMAXCONN;
127 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
128     &somaxconn, 0, "Maximum pending socket connection queue size");
129 
130 /*
131  * Socket operation routines.
132  * These routines are called by the routines in
133  * sys_socket.c or from a system process, and
134  * implement the semantics of socket operations by
135  * switching out to the protocol specific routines.
136  */
137 
138 /*
139  * Get a socket structure, and initialize it.
140  * Note that it would probably be better to allocate socket
141  * and PCB at the same time, but I'm not convinced that all
142  * the protocols can be easily modified to do this.
143  */
144 struct socket *
145 soalloc(int waitok)
146 {
147 	struct socket *so;
148 	unsigned waitmask;
149 
150 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
151 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
152 	if (so) {
153 		/* XXX race condition for reentrant kernel */
154 		TAILQ_INIT(&so->so_aiojobq);
155 		TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
156 		TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
157 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
158 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
159 		so->so_state = SS_NOFDREF;
160 		so->so_refs = 1;
161 	}
162 	return so;
163 }
164 
165 int
166 socreate(int dom, struct socket **aso, int type,
167 	int proto, struct thread *td)
168 {
169 	struct proc *p = td->td_proc;
170 	struct protosw *prp;
171 	struct socket *so;
172 	struct pru_attach_info ai;
173 	int error;
174 
175 	if (proto)
176 		prp = pffindproto(dom, proto, type);
177 	else
178 		prp = pffindtype(dom, type);
179 
180 	if (prp == 0 || prp->pr_usrreqs->pru_attach == 0)
181 		return (EPROTONOSUPPORT);
182 
183 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
184 	    prp->pr_domain->dom_family != PF_LOCAL &&
185 	    prp->pr_domain->dom_family != PF_INET &&
186 	    prp->pr_domain->dom_family != PF_INET6 &&
187 	    prp->pr_domain->dom_family != PF_ROUTE) {
188 		return (EPROTONOSUPPORT);
189 	}
190 
191 	if (prp->pr_type != type)
192 		return (EPROTOTYPE);
193 	so = soalloc(p != 0);
194 	if (so == NULL)
195 		return (ENOBUFS);
196 
197 	/*
198 	 * Callers of socreate() presumably will connect up a descriptor
199 	 * and call soclose() if they cannot.  This represents our so_refs
200 	 * (which should be 1) from soalloc().
201 	 */
202 	soclrstate(so, SS_NOFDREF);
203 
204 	/*
205 	 * Set a default port for protocol processing.  No action will occur
206 	 * on the socket on this port until an inpcb is attached to it and
207 	 * is able to match incoming packets, or until the socket becomes
208 	 * available to userland.
209 	 *
210 	 * We normally default the socket to the protocol thread on cpu 0.
211 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
212 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
213 	 */
214 	if (prp->pr_flags & PR_SYNC_PORT)
215 		so->so_port = &netisr_sync_port;
216 	else
217 		so->so_port = cpu_portfn(0);
218 
219 	TAILQ_INIT(&so->so_incomp);
220 	TAILQ_INIT(&so->so_comp);
221 	so->so_type = type;
222 	so->so_cred = crhold(p->p_ucred);
223 	so->so_proto = prp;
224 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
225 	ai.p_ucred = p->p_ucred;
226 	ai.fd_rdir = p->p_fd->fd_rdir;
227 
228 	/*
229 	 * Auto-sizing of socket buffers is managed by the protocols and
230 	 * the appropriate flags must be set in the pru_attach function.
231 	 */
232 	error = so_pru_attach(so, proto, &ai);
233 	if (error) {
234 		sosetstate(so, SS_NOFDREF);
235 		sofree(so);	/* from soalloc */
236 		return error;
237 	}
238 
239 	/*
240 	 * NOTE: Returns referenced socket.
241 	 */
242 	*aso = so;
243 	return (0);
244 }
245 
246 int
247 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
248 {
249 	int error;
250 
251 	error = so_pru_bind(so, nam, td);
252 	return (error);
253 }
254 
255 static void
256 sodealloc(struct socket *so)
257 {
258 	if (so->so_rcv.ssb_hiwat)
259 		(void)chgsbsize(so->so_cred->cr_uidinfo,
260 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
261 	if (so->so_snd.ssb_hiwat)
262 		(void)chgsbsize(so->so_cred->cr_uidinfo,
263 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
264 #ifdef INET
265 	/* remove accept filter if present */
266 	if (so->so_accf != NULL)
267 		do_setopt_accept_filter(so, NULL);
268 #endif /* INET */
269 	crfree(so->so_cred);
270 	kfree(so, M_SOCKET);
271 }
272 
273 int
274 solisten(struct socket *so, int backlog, struct thread *td)
275 {
276 	int error;
277 #ifdef SCTP
278 	short oldopt, oldqlimit;
279 #endif /* SCTP */
280 
281 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
282 		return (EINVAL);
283 
284 #ifdef SCTP
285 	oldopt = so->so_options;
286 	oldqlimit = so->so_qlimit;
287 #endif /* SCTP */
288 
289 	lwkt_gettoken(&so->so_rcv.ssb_token);
290 	if (TAILQ_EMPTY(&so->so_comp))
291 		so->so_options |= SO_ACCEPTCONN;
292 	lwkt_reltoken(&so->so_rcv.ssb_token);
293 	if (backlog < 0 || backlog > somaxconn)
294 		backlog = somaxconn;
295 	so->so_qlimit = backlog;
296 	/* SCTP needs to look at tweak both the inbound backlog parameter AND
297 	 * the so_options (UDP model both connect's and gets inbound
298 	 * connections .. implicitly).
299 	 */
300 	error = so_pru_listen(so, td);
301 	if (error) {
302 #ifdef SCTP
303 		/* Restore the params */
304 		so->so_options = oldopt;
305 		so->so_qlimit = oldqlimit;
306 #endif /* SCTP */
307 		return (error);
308 	}
309 	return (0);
310 }
311 
312 /*
313  * Destroy a disconnected socket.  This routine is a NOP if entities
314  * still have a reference on the socket:
315  *
316  *	so_pcb -	The protocol stack still has a reference
317  *	SS_NOFDREF -	There is no longer a file pointer reference
318  */
319 void
320 sofree(struct socket *so)
321 {
322 	struct socket *head;
323 
324 	/*
325 	 * This is a bit hackish at the moment.  We need to interlock
326 	 * any accept queue we are on before we potentially lose the
327 	 * last reference to avoid races against a re-reference from
328 	 * someone operating on the queue.
329 	 */
330 	while ((head = so->so_head) != NULL) {
331 		lwkt_getpooltoken(head);
332 		if (so->so_head == head)
333 			break;
334 		lwkt_relpooltoken(head);
335 	}
336 
337 	/*
338 	 * Arbitrage the last free.
339 	 */
340 	KKASSERT(so->so_refs > 0);
341 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
342 		if (head)
343 			lwkt_relpooltoken(head);
344 		return;
345 	}
346 
347 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
348 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
349 
350 	/*
351 	 * We're done, remove ourselves from the accept queue we are
352 	 * on, if we are on one.
353 	 */
354 	if (head != NULL) {
355 		if (so->so_state & SS_INCOMP) {
356 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
357 			head->so_incqlen--;
358 		} else if (so->so_state & SS_COMP) {
359 			/*
360 			 * We must not decommission a socket that's
361 			 * on the accept(2) queue.  If we do, then
362 			 * accept(2) may hang after select(2) indicated
363 			 * that the listening socket was ready.
364 			 */
365 			lwkt_relpooltoken(head);
366 			return;
367 		} else {
368 			panic("sofree: not queued");
369 		}
370 		soclrstate(so, SS_INCOMP);
371 		so->so_head = NULL;
372 		lwkt_relpooltoken(head);
373 	}
374 	ssb_release(&so->so_snd, so);
375 	sorflush(so);
376 	sodealloc(so);
377 }
378 
379 /*
380  * Close a socket on last file table reference removal.
381  * Initiate disconnect if connected.
382  * Free socket when disconnect complete.
383  */
384 int
385 soclose(struct socket *so, int fflag)
386 {
387 	int error = 0;
388 
389 	funsetown(&so->so_sigio);
390 	if (so->so_pcb == NULL)
391 		goto discard;
392 	if (so->so_state & SS_ISCONNECTED) {
393 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
394 			error = sodisconnect(so);
395 			if (error)
396 				goto drop;
397 		}
398 		if (so->so_options & SO_LINGER) {
399 			if ((so->so_state & SS_ISDISCONNECTING) &&
400 			    (fflag & FNONBLOCK))
401 				goto drop;
402 			while (so->so_state & SS_ISCONNECTED) {
403 				error = tsleep(&so->so_timeo, PCATCH,
404 					       "soclos", so->so_linger * hz);
405 				if (error)
406 					break;
407 			}
408 		}
409 	}
410 drop:
411 	if (so->so_pcb) {
412 		int error2;
413 
414 		error2 = so_pru_detach(so);
415 		if (error == 0)
416 			error = error2;
417 	}
418 discard:
419 	lwkt_getpooltoken(so);
420 	if (so->so_options & SO_ACCEPTCONN) {
421 		struct socket *sp;
422 
423 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
424 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
425 			soclrstate(sp, SS_INCOMP);
426 			sp->so_head = NULL;
427 			so->so_incqlen--;
428 			soaborta(sp);
429 		}
430 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
431 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
432 			soclrstate(sp, SS_COMP);
433 			sp->so_head = NULL;
434 			so->so_qlen--;
435 			soaborta(sp);
436 		}
437 	}
438 	lwkt_relpooltoken(so);
439 	if (so->so_state & SS_NOFDREF)
440 		panic("soclose: NOFDREF");
441 	sosetstate(so, SS_NOFDREF);	/* take ref */
442 	sofree(so);			/* dispose of ref */
443 	return (error);
444 }
445 
446 /*
447  * Abort and destroy a socket.  Only one abort can be in progress
448  * at any given moment.
449  */
450 void
451 soabort(struct socket *so)
452 {
453 	soreference(so);
454 	so_pru_abort(so);
455 }
456 
457 void
458 soaborta(struct socket *so)
459 {
460 	soreference(so);
461 	so_pru_aborta(so);
462 }
463 
464 void
465 soabort_oncpu(struct socket *so)
466 {
467 	soreference(so);
468 	so_pru_abort_oncpu(so);
469 }
470 
471 /*
472  * so is passed in ref'd, which becomes owned by
473  * the cleared SS_NOFDREF flag.
474  */
475 int
476 soaccept(struct socket *so, struct sockaddr **nam)
477 {
478 	int error;
479 
480 	if ((so->so_state & SS_NOFDREF) == 0)
481 		panic("soaccept: !NOFDREF");
482 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
483 	error = so_pru_accept_direct(so, nam);
484 	return (error);
485 }
486 
487 int
488 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
489 {
490 	int error;
491 
492 	if (so->so_options & SO_ACCEPTCONN)
493 		return (EOPNOTSUPP);
494 	/*
495 	 * If protocol is connection-based, can only connect once.
496 	 * Otherwise, if connected, try to disconnect first.
497 	 * This allows user to disconnect by connecting to, e.g.,
498 	 * a null address.
499 	 */
500 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
501 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
502 	    (error = sodisconnect(so)))) {
503 		error = EISCONN;
504 	} else {
505 		/*
506 		 * Prevent accumulated error from previous connection
507 		 * from biting us.
508 		 */
509 		so->so_error = 0;
510 		error = so_pru_connect(so, nam, td);
511 	}
512 	return (error);
513 }
514 
515 int
516 soconnect2(struct socket *so1, struct socket *so2)
517 {
518 	int error;
519 
520 	error = so_pru_connect2(so1, so2);
521 	return (error);
522 }
523 
524 int
525 sodisconnect(struct socket *so)
526 {
527 	int error;
528 
529 	if ((so->so_state & SS_ISCONNECTED) == 0) {
530 		error = ENOTCONN;
531 		goto bad;
532 	}
533 	if (so->so_state & SS_ISDISCONNECTING) {
534 		error = EALREADY;
535 		goto bad;
536 	}
537 	error = so_pru_disconnect(so);
538 bad:
539 	return (error);
540 }
541 
542 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
543 /*
544  * Send on a socket.
545  * If send must go all at once and message is larger than
546  * send buffering, then hard error.
547  * Lock against other senders.
548  * If must go all at once and not enough room now, then
549  * inform user that this would block and do nothing.
550  * Otherwise, if nonblocking, send as much as possible.
551  * The data to be sent is described by "uio" if nonzero,
552  * otherwise by the mbuf chain "top" (which must be null
553  * if uio is not).  Data provided in mbuf chain must be small
554  * enough to send all at once.
555  *
556  * Returns nonzero on error, timeout or signal; callers
557  * must check for short counts if EINTR/ERESTART are returned.
558  * Data and control buffers are freed on return.
559  */
560 int
561 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
562 	struct mbuf *top, struct mbuf *control, int flags,
563 	struct thread *td)
564 {
565 	struct mbuf **mp;
566 	struct mbuf *m;
567 	size_t resid;
568 	int space, len;
569 	int clen = 0, error, dontroute, mlen;
570 	int atomic = sosendallatonce(so) || top;
571 	int pru_flags;
572 
573 	if (uio) {
574 		resid = uio->uio_resid;
575 	} else {
576 		resid = (size_t)top->m_pkthdr.len;
577 #ifdef INVARIANTS
578 		len = 0;
579 		for (m = top; m; m = m->m_next)
580 			len += m->m_len;
581 		KKASSERT(top->m_pkthdr.len == len);
582 #endif
583 	}
584 
585 	/*
586 	 * WARNING!  resid is unsigned, space and len are signed.  space
587 	 * 	     can wind up negative if the sockbuf is overcommitted.
588 	 *
589 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
590 	 * type sockets since that's an error.
591 	 */
592 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
593 		error = EINVAL;
594 		goto out;
595 	}
596 
597 	dontroute =
598 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
599 	    (so->so_proto->pr_flags & PR_ATOMIC);
600 	if (td->td_lwp != NULL)
601 		td->td_lwp->lwp_ru.ru_msgsnd++;
602 	if (control)
603 		clen = control->m_len;
604 #define	gotoerr(errcode)	{ error = errcode; goto release; }
605 
606 restart:
607 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
608 	if (error)
609 		goto out;
610 
611 	do {
612 		if (so->so_state & SS_CANTSENDMORE)
613 			gotoerr(EPIPE);
614 		if (so->so_error) {
615 			error = so->so_error;
616 			so->so_error = 0;
617 			goto release;
618 		}
619 		if ((so->so_state & SS_ISCONNECTED) == 0) {
620 			/*
621 			 * `sendto' and `sendmsg' is allowed on a connection-
622 			 * based socket if it supports implied connect.
623 			 * Return ENOTCONN if not connected and no address is
624 			 * supplied.
625 			 */
626 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
627 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
628 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
629 				    !(resid == 0 && clen != 0))
630 					gotoerr(ENOTCONN);
631 			} else if (addr == 0)
632 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
633 				   ENOTCONN : EDESTADDRREQ);
634 		}
635 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
636 		    clen > so->so_snd.ssb_hiwat) {
637 			gotoerr(EMSGSIZE);
638 		}
639 		space = ssb_space(&so->so_snd);
640 		if (flags & MSG_OOB)
641 			space += 1024;
642 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
643 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
644 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
645 				gotoerr(EWOULDBLOCK);
646 			ssb_unlock(&so->so_snd);
647 			error = ssb_wait(&so->so_snd);
648 			if (error)
649 				goto out;
650 			goto restart;
651 		}
652 		mp = &top;
653 		space -= clen;
654 		do {
655 		    if (uio == NULL) {
656 			/*
657 			 * Data is prepackaged in "top".
658 			 */
659 			resid = 0;
660 			if (flags & MSG_EOR)
661 				top->m_flags |= M_EOR;
662 		    } else do {
663 			if (resid > INT_MAX)
664 				resid = INT_MAX;
665 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
666 				   top == NULL ? M_PKTHDR : 0, &mlen);
667 			if (top == NULL) {
668 				m->m_pkthdr.len = 0;
669 				m->m_pkthdr.rcvif = NULL;
670 			}
671 			len = imin((int)szmin(mlen, resid), space);
672 			if (resid < MINCLSIZE) {
673 				/*
674 				 * For datagram protocols, leave room
675 				 * for protocol headers in first mbuf.
676 				 */
677 				if (atomic && top == 0 && len < mlen)
678 					MH_ALIGN(m, len);
679 			}
680 			space -= len;
681 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
682 			resid = uio->uio_resid;
683 			m->m_len = len;
684 			*mp = m;
685 			top->m_pkthdr.len += len;
686 			if (error)
687 				goto release;
688 			mp = &m->m_next;
689 			if (resid == 0) {
690 				if (flags & MSG_EOR)
691 					top->m_flags |= M_EOR;
692 				break;
693 			}
694 		    } while (space > 0 && atomic);
695 		    if (dontroute)
696 			    so->so_options |= SO_DONTROUTE;
697 		    if (flags & MSG_OOB) {
698 		    	    pru_flags = PRUS_OOB;
699 		    } else if ((flags & MSG_EOF) &&
700 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
701 			       (resid == 0)) {
702 			    /*
703 			     * If the user set MSG_EOF, the protocol
704 			     * understands this flag and nothing left to
705 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
706 			     */
707 		    	    pru_flags = PRUS_EOF;
708 		    } else if (resid > 0 && space > 0) {
709 			    /* If there is more to send, set PRUS_MORETOCOME */
710 		    	    pru_flags = PRUS_MORETOCOME;
711 		    } else {
712 		    	    pru_flags = 0;
713 		    }
714 		    /*
715 		     * XXX all the SS_CANTSENDMORE checks previously
716 		     * done could be out of date.  We could have recieved
717 		     * a reset packet in an interrupt or maybe we slept
718 		     * while doing page faults in uiomove() etc. We could
719 		     * probably recheck again inside the splnet() protection
720 		     * here, but there are probably other places that this
721 		     * also happens.  We must rethink this.
722 		     */
723 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
724 		    if (dontroute)
725 			    so->so_options &= ~SO_DONTROUTE;
726 		    clen = 0;
727 		    control = 0;
728 		    top = NULL;
729 		    mp = &top;
730 		    if (error)
731 			    goto release;
732 		} while (resid && space > 0);
733 	} while (resid);
734 
735 release:
736 	ssb_unlock(&so->so_snd);
737 out:
738 	if (top)
739 		m_freem(top);
740 	if (control)
741 		m_freem(control);
742 	return (error);
743 }
744 
745 /*
746  * A specialization of sosend() for UDP based on protocol-specific knowledge:
747  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
748  *	sosendallatonce() returns true,
749  *	the "atomic" variable is true,
750  *	and sosendudp() blocks until space is available for the entire send.
751  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
752  *	PR_IMPLOPCL flags set.
753  *   UDP has no out-of-band data.
754  *   UDP has no control data.
755  *   UDP does not support MSG_EOR.
756  */
757 int
758 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
759 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
760 {
761 	boolean_t dontroute;		/* temporary SO_DONTROUTE setting */
762 	size_t resid;
763 	int error;
764 	int space;
765 
766 	if (td->td_lwp != NULL)
767 		td->td_lwp->lwp_ru.ru_msgsnd++;
768 	if (control)
769 		m_freem(control);
770 
771 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
772 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
773 
774 restart:
775 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
776 	if (error)
777 		goto out;
778 
779 	if (so->so_state & SS_CANTSENDMORE)
780 		gotoerr(EPIPE);
781 	if (so->so_error) {
782 		error = so->so_error;
783 		so->so_error = 0;
784 		goto release;
785 	}
786 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
787 		gotoerr(EDESTADDRREQ);
788 	if (resid > so->so_snd.ssb_hiwat)
789 		gotoerr(EMSGSIZE);
790 	space = ssb_space(&so->so_snd);
791 	if (uio && (space < 0 || (size_t)space < resid)) {
792 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
793 			gotoerr(EWOULDBLOCK);
794 		ssb_unlock(&so->so_snd);
795 		error = ssb_wait(&so->so_snd);
796 		if (error)
797 			goto out;
798 		goto restart;
799 	}
800 
801 	if (uio) {
802 		top = m_uiomove(uio);
803 		if (top == NULL)
804 			goto release;
805 	}
806 
807 	dontroute = (flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE);
808 	if (dontroute)
809 		so->so_options |= SO_DONTROUTE;
810 
811 	error = so_pru_send(so, 0, top, addr, NULL, td);
812 	top = NULL;		/* sent or freed in lower layer */
813 
814 	if (dontroute)
815 		so->so_options &= ~SO_DONTROUTE;
816 
817 release:
818 	ssb_unlock(&so->so_snd);
819 out:
820 	if (top)
821 		m_freem(top);
822 	return (error);
823 }
824 
825 int
826 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
827 	struct mbuf *top, struct mbuf *control, int flags,
828 	struct thread *td)
829 {
830 	struct mbuf **mp;
831 	struct mbuf *m;
832 	size_t resid;
833 	int space, len;
834 	int error, mlen;
835 	int allatonce;
836 	int pru_flags;
837 
838 	if (uio) {
839 		KKASSERT(top == NULL);
840 		allatonce = 0;
841 		resid = uio->uio_resid;
842 	} else {
843 		allatonce = 1;
844 		resid = (size_t)top->m_pkthdr.len;
845 #ifdef INVARIANTS
846 		len = 0;
847 		for (m = top; m; m = m->m_next)
848 			len += m->m_len;
849 		KKASSERT(top->m_pkthdr.len == len);
850 #endif
851 	}
852 
853 	/*
854 	 * WARNING!  resid is unsigned, space and len are signed.  space
855 	 * 	     can wind up negative if the sockbuf is overcommitted.
856 	 *
857 	 * Also check to make sure that MSG_EOR isn't used on TCP
858 	 */
859 	if (flags & MSG_EOR) {
860 		error = EINVAL;
861 		goto out;
862 	}
863 
864 	if (control) {
865 		/* TCP doesn't do control messages (rights, creds, etc) */
866 		if (control->m_len) {
867 			error = EINVAL;
868 			goto out;
869 		}
870 		m_freem(control);	/* empty control, just free it */
871 		control = NULL;
872 	}
873 
874 	if (td->td_lwp != NULL)
875 		td->td_lwp->lwp_ru.ru_msgsnd++;
876 
877 #define	gotoerr(errcode)	{ error = errcode; goto release; }
878 
879 restart:
880 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
881 	if (error)
882 		goto out;
883 
884 	do {
885 		if (so->so_state & SS_CANTSENDMORE)
886 			gotoerr(EPIPE);
887 		if (so->so_error) {
888 			error = so->so_error;
889 			so->so_error = 0;
890 			goto release;
891 		}
892 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
893 		    (so->so_state & SS_ISCONFIRMING) == 0)
894 			gotoerr(ENOTCONN);
895 		if (allatonce && resid > so->so_snd.ssb_hiwat)
896 			gotoerr(EMSGSIZE);
897 
898 		space = ssb_space(&so->so_snd);
899 		if (flags & MSG_OOB)
900 			space += 1024;
901 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
902 		    space < so->so_snd.ssb_lowat) {
903 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
904 				gotoerr(EWOULDBLOCK);
905 			ssb_unlock(&so->so_snd);
906 			error = ssb_wait(&so->so_snd);
907 			if (error)
908 				goto out;
909 			goto restart;
910 		}
911 		mp = &top;
912 		do {
913 		    int cnt = 0;
914 
915 		    if (uio == NULL) {
916 			/*
917 			 * Data is prepackaged in "top".
918 			 */
919 			resid = 0;
920 		    } else do {
921 			if (resid > INT_MAX)
922 				resid = INT_MAX;
923 			m = m_getl((int)resid, MB_WAIT, MT_DATA,
924 				   top == NULL ? M_PKTHDR : 0, &mlen);
925 			if (top == NULL) {
926 				m->m_pkthdr.len = 0;
927 				m->m_pkthdr.rcvif = NULL;
928 			}
929 			len = imin((int)szmin(mlen, resid), space);
930 			space -= len;
931 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
932 			resid = uio->uio_resid;
933 			m->m_len = len;
934 			*mp = m;
935 			top->m_pkthdr.len += len;
936 			if (error)
937 				goto release;
938 			mp = &m->m_next;
939 			if (resid == 0)
940 				break;
941 			++cnt;
942 		    } while (space > 0 && cnt < tcp_sosnd_agglim);
943 
944 		    if (flags & MSG_OOB) {
945 		    	    pru_flags = PRUS_OOB;
946 		    } else if (resid > 0 && space > 0) {
947 			    /* If there is more to send, set PRUS_MORETOCOME */
948 		    	    pru_flags = PRUS_MORETOCOME;
949 		    } else {
950 		    	    pru_flags = 0;
951 		    }
952 
953 		    /*
954 		     * XXX all the SS_CANTSENDMORE checks previously
955 		     * done could be out of date.  We could have recieved
956 		     * a reset packet in an interrupt or maybe we slept
957 		     * while doing page faults in uiomove() etc. We could
958 		     * probably recheck again inside the splnet() protection
959 		     * here, but there are probably other places that this
960 		     * also happens.  We must rethink this.
961 		     */
962 		    if ((pru_flags & PRUS_OOB) ||
963 		        (pru_flags & PRUS_MORETOCOME) == 0) {
964 			    error = so_pru_send(so, pru_flags, top,
965 			        NULL, NULL, td);
966 		    } else {
967 			    so_pru_send_async(so, pru_flags, top,
968 			        NULL, NULL, td);
969 			    error = 0;
970 		    }
971 
972 		    top = NULL;
973 		    mp = &top;
974 		    if (error)
975 			    goto release;
976 		} while (resid && space > 0);
977 	} while (resid);
978 
979 release:
980 	ssb_unlock(&so->so_snd);
981 out:
982 	if (top)
983 		m_freem(top);
984 	if (control)
985 		m_freem(control);
986 	return (error);
987 }
988 
989 /*
990  * Implement receive operations on a socket.
991  *
992  * We depend on the way that records are added to the signalsockbuf
993  * by sbappend*.  In particular, each record (mbufs linked through m_next)
994  * must begin with an address if the protocol so specifies,
995  * followed by an optional mbuf or mbufs containing ancillary data,
996  * and then zero or more mbufs of data.
997  *
998  * Although the signalsockbuf is locked, new data may still be appended.
999  * A token inside the ssb_lock deals with MP issues and still allows
1000  * the network to access the socket if we block in a uio.
1001  *
1002  * The caller may receive the data as a single mbuf chain by supplying
1003  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1004  * only for the count in uio_resid.
1005  */
1006 int
1007 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1008 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1009 {
1010 	struct mbuf *m, *n;
1011 	struct mbuf *free_chain = NULL;
1012 	int flags, len, error, offset;
1013 	struct protosw *pr = so->so_proto;
1014 	int moff, type = 0;
1015 	size_t resid, orig_resid;
1016 
1017 	if (uio)
1018 		resid = uio->uio_resid;
1019 	else
1020 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1021 	orig_resid = resid;
1022 
1023 	if (psa)
1024 		*psa = NULL;
1025 	if (controlp)
1026 		*controlp = NULL;
1027 	if (flagsp)
1028 		flags = *flagsp &~ MSG_EOR;
1029 	else
1030 		flags = 0;
1031 	if (flags & MSG_OOB) {
1032 		m = m_get(MB_WAIT, MT_DATA);
1033 		if (m == NULL)
1034 			return (ENOBUFS);
1035 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1036 		if (error)
1037 			goto bad;
1038 		if (sio) {
1039 			do {
1040 				sbappend(sio, m);
1041 				KKASSERT(resid >= (size_t)m->m_len);
1042 				resid -= (size_t)m->m_len;
1043 			} while (resid > 0 && m);
1044 		} else {
1045 			do {
1046 				uio->uio_resid = resid;
1047 				error = uiomove(mtod(m, caddr_t),
1048 						(int)szmin(resid, m->m_len),
1049 						uio);
1050 				resid = uio->uio_resid;
1051 				m = m_free(m);
1052 			} while (uio->uio_resid && error == 0 && m);
1053 		}
1054 bad:
1055 		if (m)
1056 			m_freem(m);
1057 		return (error);
1058 	}
1059 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1060 		so_pru_rcvd(so, 0);
1061 
1062 	/*
1063 	 * The token interlocks against the protocol thread while
1064 	 * ssb_lock is a blocking lock against other userland entities.
1065 	 */
1066 	lwkt_gettoken(&so->so_rcv.ssb_token);
1067 restart:
1068 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1069 	if (error)
1070 		goto done;
1071 
1072 	m = so->so_rcv.ssb_mb;
1073 	/*
1074 	 * If we have less data than requested, block awaiting more
1075 	 * (subject to any timeout) if:
1076 	 *   1. the current count is less than the low water mark, or
1077 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1078 	 *	receive operation at once if we block (resid <= hiwat).
1079 	 *   3. MSG_DONTWAIT is not set
1080 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1081 	 * we have to do the receive in sections, and thus risk returning
1082 	 * a short count if a timeout or signal occurs after we start.
1083 	 */
1084 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1085 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1086 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1087 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1088 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1089 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1090 		if (so->so_error) {
1091 			if (m)
1092 				goto dontblock;
1093 			error = so->so_error;
1094 			if ((flags & MSG_PEEK) == 0)
1095 				so->so_error = 0;
1096 			goto release;
1097 		}
1098 		if (so->so_state & SS_CANTRCVMORE) {
1099 			if (m)
1100 				goto dontblock;
1101 			else
1102 				goto release;
1103 		}
1104 		for (; m; m = m->m_next) {
1105 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1106 				m = so->so_rcv.ssb_mb;
1107 				goto dontblock;
1108 			}
1109 		}
1110 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1111 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1112 			error = ENOTCONN;
1113 			goto release;
1114 		}
1115 		if (resid == 0)
1116 			goto release;
1117 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1118 			error = EWOULDBLOCK;
1119 			goto release;
1120 		}
1121 		ssb_unlock(&so->so_rcv);
1122 		error = ssb_wait(&so->so_rcv);
1123 		if (error)
1124 			goto done;
1125 		goto restart;
1126 	}
1127 dontblock:
1128 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1129 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1130 
1131 	/*
1132 	 * note: m should be == sb_mb here.  Cache the next record while
1133 	 * cleaning up.  Note that calling m_free*() will break out critical
1134 	 * section.
1135 	 */
1136 	KKASSERT(m == so->so_rcv.ssb_mb);
1137 
1138 	/*
1139 	 * Skip any address mbufs prepending the record.
1140 	 */
1141 	if (pr->pr_flags & PR_ADDR) {
1142 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1143 		orig_resid = 0;
1144 		if (psa)
1145 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1146 		if (flags & MSG_PEEK)
1147 			m = m->m_next;
1148 		else
1149 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1150 	}
1151 
1152 	/*
1153 	 * Skip any control mbufs prepending the record.
1154 	 */
1155 #ifdef SCTP
1156 	if (pr->pr_flags & PR_ADDR_OPT) {
1157 		/*
1158 		 * For SCTP we may be getting a
1159 		 * whole message OR a partial delivery.
1160 		 */
1161 		if (m && m->m_type == MT_SONAME) {
1162 			orig_resid = 0;
1163 			if (psa)
1164 				*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1165 			if (flags & MSG_PEEK)
1166 				m = m->m_next;
1167 			else
1168 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1169 		}
1170 	}
1171 #endif /* SCTP */
1172 	while (m && m->m_type == MT_CONTROL && error == 0) {
1173 		if (flags & MSG_PEEK) {
1174 			if (controlp)
1175 				*controlp = m_copy(m, 0, m->m_len);
1176 			m = m->m_next;	/* XXX race */
1177 		} else {
1178 			if (controlp) {
1179 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1180 				if (pr->pr_domain->dom_externalize &&
1181 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1182 				    SCM_RIGHTS)
1183 				   error = (*pr->pr_domain->dom_externalize)(m);
1184 				*controlp = m;
1185 				m = n;
1186 			} else {
1187 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1188 			}
1189 		}
1190 		if (controlp && *controlp) {
1191 			orig_resid = 0;
1192 			controlp = &(*controlp)->m_next;
1193 		}
1194 	}
1195 
1196 	/*
1197 	 * flag OOB data.
1198 	 */
1199 	if (m) {
1200 		type = m->m_type;
1201 		if (type == MT_OOBDATA)
1202 			flags |= MSG_OOB;
1203 	}
1204 
1205 	/*
1206 	 * Copy to the UIO or mbuf return chain (*mp).
1207 	 */
1208 	moff = 0;
1209 	offset = 0;
1210 	while (m && resid > 0 && error == 0) {
1211 		if (m->m_type == MT_OOBDATA) {
1212 			if (type != MT_OOBDATA)
1213 				break;
1214 		} else if (type == MT_OOBDATA)
1215 			break;
1216 		else
1217 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1218 			("receive 3"));
1219 		soclrstate(so, SS_RCVATMARK);
1220 		len = (resid > INT_MAX) ? INT_MAX : resid;
1221 		if (so->so_oobmark && len > so->so_oobmark - offset)
1222 			len = so->so_oobmark - offset;
1223 		if (len > m->m_len - moff)
1224 			len = m->m_len - moff;
1225 
1226 		/*
1227 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1228 		 * The SIO is dealt with when we eat the mbuf, but deal
1229 		 * with the resid here either way.
1230 		 */
1231 		if (uio) {
1232 			uio->uio_resid = resid;
1233 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1234 			resid = uio->uio_resid;
1235 			if (error)
1236 				goto release;
1237 		} else {
1238 			resid -= (size_t)len;
1239 		}
1240 
1241 		/*
1242 		 * Eat the entire mbuf or just a piece of it
1243 		 */
1244 		if (len == m->m_len - moff) {
1245 			if (m->m_flags & M_EOR)
1246 				flags |= MSG_EOR;
1247 #ifdef SCTP
1248 			if (m->m_flags & M_NOTIFICATION)
1249 				flags |= MSG_NOTIFICATION;
1250 #endif /* SCTP */
1251 			if (flags & MSG_PEEK) {
1252 				m = m->m_next;
1253 				moff = 0;
1254 			} else {
1255 				if (sio) {
1256 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1257 					sbappend(sio, m);
1258 					m = n;
1259 				} else {
1260 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1261 				}
1262 			}
1263 		} else {
1264 			if (flags & MSG_PEEK) {
1265 				moff += len;
1266 			} else {
1267 				if (sio) {
1268 					n = m_copym(m, 0, len, MB_WAIT);
1269 					if (n)
1270 						sbappend(sio, n);
1271 				}
1272 				m->m_data += len;
1273 				m->m_len -= len;
1274 				so->so_rcv.ssb_cc -= len;
1275 			}
1276 		}
1277 		if (so->so_oobmark) {
1278 			if ((flags & MSG_PEEK) == 0) {
1279 				so->so_oobmark -= len;
1280 				if (so->so_oobmark == 0) {
1281 					sosetstate(so, SS_RCVATMARK);
1282 					break;
1283 				}
1284 			} else {
1285 				offset += len;
1286 				if (offset == so->so_oobmark)
1287 					break;
1288 			}
1289 		}
1290 		if (flags & MSG_EOR)
1291 			break;
1292 		/*
1293 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1294 		 * we must not quit until resid == 0 or an error
1295 		 * termination.  If a signal/timeout occurs, return
1296 		 * with a short count but without error.
1297 		 * Keep signalsockbuf locked against other readers.
1298 		 */
1299 		while ((flags & MSG_WAITALL) && m == NULL &&
1300 		       resid > 0 && !sosendallatonce(so) &&
1301 		       so->so_rcv.ssb_mb == NULL) {
1302 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1303 				break;
1304 			/*
1305 			 * The window might have closed to zero, make
1306 			 * sure we send an ack now that we've drained
1307 			 * the buffer or we might end up blocking until
1308 			 * the idle takes over (5 seconds).
1309 			 */
1310 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1311 				so_pru_rcvd(so, flags);
1312 			error = ssb_wait(&so->so_rcv);
1313 			if (error) {
1314 				ssb_unlock(&so->so_rcv);
1315 				error = 0;
1316 				goto done;
1317 			}
1318 			m = so->so_rcv.ssb_mb;
1319 		}
1320 	}
1321 
1322 	/*
1323 	 * If an atomic read was requested but unread data still remains
1324 	 * in the record, set MSG_TRUNC.
1325 	 */
1326 	if (m && pr->pr_flags & PR_ATOMIC)
1327 		flags |= MSG_TRUNC;
1328 
1329 	/*
1330 	 * Cleanup.  If an atomic read was requested drop any unread data.
1331 	 */
1332 	if ((flags & MSG_PEEK) == 0) {
1333 		if (m && (pr->pr_flags & PR_ATOMIC))
1334 			sbdroprecord(&so->so_rcv.sb);
1335 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1336 			so_pru_rcvd(so, flags);
1337 	}
1338 
1339 	if (orig_resid == resid && orig_resid &&
1340 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1341 		ssb_unlock(&so->so_rcv);
1342 		goto restart;
1343 	}
1344 
1345 	if (flagsp)
1346 		*flagsp |= flags;
1347 release:
1348 	ssb_unlock(&so->so_rcv);
1349 done:
1350 	lwkt_reltoken(&so->so_rcv.ssb_token);
1351 	if (free_chain)
1352 		m_freem(free_chain);
1353 	return (error);
1354 }
1355 
1356 /*
1357  * Shut a socket down.  Note that we do not get a frontend lock as we
1358  * want to be able to shut the socket down even if another thread is
1359  * blocked in a read(), thus waking it up.
1360  */
1361 int
1362 soshutdown(struct socket *so, int how)
1363 {
1364 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1365 		return (EINVAL);
1366 
1367 	if (how != SHUT_WR) {
1368 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1369 		sorflush(so);
1370 		/*ssb_unlock(&so->so_rcv);*/
1371 	}
1372 	if (how != SHUT_RD)
1373 		return (so_pru_shutdown(so));
1374 	return (0);
1375 }
1376 
1377 void
1378 sorflush(struct socket *so)
1379 {
1380 	struct signalsockbuf *ssb = &so->so_rcv;
1381 	struct protosw *pr = so->so_proto;
1382 	struct signalsockbuf asb;
1383 
1384 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1385 
1386 	lwkt_gettoken(&ssb->ssb_token);
1387 	socantrcvmore(so);
1388 	asb = *ssb;
1389 
1390 	/*
1391 	 * Can't just blow up the ssb structure here
1392 	 */
1393 	bzero(&ssb->sb, sizeof(ssb->sb));
1394 	ssb->ssb_timeo = 0;
1395 	ssb->ssb_lowat = 0;
1396 	ssb->ssb_hiwat = 0;
1397 	ssb->ssb_mbmax = 0;
1398 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1399 
1400 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1401 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1402 	ssb_release(&asb, so);
1403 
1404 	lwkt_reltoken(&ssb->ssb_token);
1405 }
1406 
1407 #ifdef INET
1408 static int
1409 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1410 {
1411 	struct accept_filter_arg	*afap = NULL;
1412 	struct accept_filter	*afp;
1413 	struct so_accf	*af = so->so_accf;
1414 	int	error = 0;
1415 
1416 	/* do not set/remove accept filters on non listen sockets */
1417 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1418 		error = EINVAL;
1419 		goto out;
1420 	}
1421 
1422 	/* removing the filter */
1423 	if (sopt == NULL) {
1424 		if (af != NULL) {
1425 			if (af->so_accept_filter != NULL &&
1426 				af->so_accept_filter->accf_destroy != NULL) {
1427 				af->so_accept_filter->accf_destroy(so);
1428 			}
1429 			if (af->so_accept_filter_str != NULL) {
1430 				FREE(af->so_accept_filter_str, M_ACCF);
1431 			}
1432 			FREE(af, M_ACCF);
1433 			so->so_accf = NULL;
1434 		}
1435 		so->so_options &= ~SO_ACCEPTFILTER;
1436 		return (0);
1437 	}
1438 	/* adding a filter */
1439 	/* must remove previous filter first */
1440 	if (af != NULL) {
1441 		error = EINVAL;
1442 		goto out;
1443 	}
1444 	/* don't put large objects on the kernel stack */
1445 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1446 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1447 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1448 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1449 	if (error)
1450 		goto out;
1451 	afp = accept_filt_get(afap->af_name);
1452 	if (afp == NULL) {
1453 		error = ENOENT;
1454 		goto out;
1455 	}
1456 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1457 	if (afp->accf_create != NULL) {
1458 		if (afap->af_name[0] != '\0') {
1459 			int len = strlen(afap->af_name) + 1;
1460 
1461 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1462 			strcpy(af->so_accept_filter_str, afap->af_name);
1463 		}
1464 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1465 		if (af->so_accept_filter_arg == NULL) {
1466 			FREE(af->so_accept_filter_str, M_ACCF);
1467 			FREE(af, M_ACCF);
1468 			so->so_accf = NULL;
1469 			error = EINVAL;
1470 			goto out;
1471 		}
1472 	}
1473 	af->so_accept_filter = afp;
1474 	so->so_accf = af;
1475 	so->so_options |= SO_ACCEPTFILTER;
1476 out:
1477 	if (afap != NULL)
1478 		FREE(afap, M_TEMP);
1479 	return (error);
1480 }
1481 #endif /* INET */
1482 
1483 /*
1484  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1485  * an additional variant to handle the case where the option value needs
1486  * to be some kind of integer, but not a specific size.
1487  * In addition to their use here, these functions are also called by the
1488  * protocol-level pr_ctloutput() routines.
1489  */
1490 int
1491 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1492 {
1493 	return soopt_to_kbuf(sopt, buf, len, minlen);
1494 }
1495 
1496 int
1497 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1498 {
1499 	size_t	valsize;
1500 
1501 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1502 	KKASSERT(kva_p(buf));
1503 
1504 	/*
1505 	 * If the user gives us more than we wanted, we ignore it,
1506 	 * but if we don't get the minimum length the caller
1507 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1508 	 * is set to however much we actually retrieved.
1509 	 */
1510 	if ((valsize = sopt->sopt_valsize) < minlen)
1511 		return EINVAL;
1512 	if (valsize > len)
1513 		sopt->sopt_valsize = valsize = len;
1514 
1515 	bcopy(sopt->sopt_val, buf, valsize);
1516 	return 0;
1517 }
1518 
1519 
1520 int
1521 sosetopt(struct socket *so, struct sockopt *sopt)
1522 {
1523 	int	error, optval;
1524 	struct	linger l;
1525 	struct	timeval tv;
1526 	u_long  val;
1527 	struct signalsockbuf *sotmp;
1528 
1529 	error = 0;
1530 	sopt->sopt_dir = SOPT_SET;
1531 	if (sopt->sopt_level != SOL_SOCKET) {
1532 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1533 			return (so_pr_ctloutput(so, sopt));
1534 		}
1535 		error = ENOPROTOOPT;
1536 	} else {
1537 		switch (sopt->sopt_name) {
1538 #ifdef INET
1539 		case SO_ACCEPTFILTER:
1540 			error = do_setopt_accept_filter(so, sopt);
1541 			if (error)
1542 				goto bad;
1543 			break;
1544 #endif /* INET */
1545 		case SO_LINGER:
1546 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1547 			if (error)
1548 				goto bad;
1549 
1550 			so->so_linger = l.l_linger;
1551 			if (l.l_onoff)
1552 				so->so_options |= SO_LINGER;
1553 			else
1554 				so->so_options &= ~SO_LINGER;
1555 			break;
1556 
1557 		case SO_DEBUG:
1558 		case SO_KEEPALIVE:
1559 		case SO_DONTROUTE:
1560 		case SO_USELOOPBACK:
1561 		case SO_BROADCAST:
1562 		case SO_REUSEADDR:
1563 		case SO_REUSEPORT:
1564 		case SO_OOBINLINE:
1565 		case SO_TIMESTAMP:
1566 			error = sooptcopyin(sopt, &optval, sizeof optval,
1567 					    sizeof optval);
1568 			if (error)
1569 				goto bad;
1570 			if (optval)
1571 				so->so_options |= sopt->sopt_name;
1572 			else
1573 				so->so_options &= ~sopt->sopt_name;
1574 			break;
1575 
1576 		case SO_SNDBUF:
1577 		case SO_RCVBUF:
1578 		case SO_SNDLOWAT:
1579 		case SO_RCVLOWAT:
1580 			error = sooptcopyin(sopt, &optval, sizeof optval,
1581 					    sizeof optval);
1582 			if (error)
1583 				goto bad;
1584 
1585 			/*
1586 			 * Values < 1 make no sense for any of these
1587 			 * options, so disallow them.
1588 			 */
1589 			if (optval < 1) {
1590 				error = EINVAL;
1591 				goto bad;
1592 			}
1593 
1594 			switch (sopt->sopt_name) {
1595 			case SO_SNDBUF:
1596 			case SO_RCVBUF:
1597 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
1598 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1599 				    so,
1600 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
1601 					error = ENOBUFS;
1602 					goto bad;
1603 				}
1604 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
1605 						&so->so_snd : &so->so_rcv;
1606 				atomic_clear_int(&sotmp->ssb_flags,
1607 						 SSB_AUTOSIZE);
1608 				break;
1609 
1610 			/*
1611 			 * Make sure the low-water is never greater than
1612 			 * the high-water.
1613 			 */
1614 			case SO_SNDLOWAT:
1615 				so->so_snd.ssb_lowat =
1616 				    (optval > so->so_snd.ssb_hiwat) ?
1617 				    so->so_snd.ssb_hiwat : optval;
1618 				atomic_clear_int(&so->so_snd.ssb_flags,
1619 						 SSB_AUTOLOWAT);
1620 				break;
1621 			case SO_RCVLOWAT:
1622 				so->so_rcv.ssb_lowat =
1623 				    (optval > so->so_rcv.ssb_hiwat) ?
1624 				    so->so_rcv.ssb_hiwat : optval;
1625 				atomic_clear_int(&so->so_rcv.ssb_flags,
1626 						 SSB_AUTOLOWAT);
1627 				break;
1628 			}
1629 			break;
1630 
1631 		case SO_SNDTIMEO:
1632 		case SO_RCVTIMEO:
1633 			error = sooptcopyin(sopt, &tv, sizeof tv,
1634 					    sizeof tv);
1635 			if (error)
1636 				goto bad;
1637 
1638 			/* assert(hz > 0); */
1639 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
1640 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1641 				error = EDOM;
1642 				goto bad;
1643 			}
1644 			/* assert(tick > 0); */
1645 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
1646 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
1647 			if (val > INT_MAX) {
1648 				error = EDOM;
1649 				goto bad;
1650 			}
1651 			if (val == 0 && tv.tv_usec != 0)
1652 				val = 1;
1653 
1654 			switch (sopt->sopt_name) {
1655 			case SO_SNDTIMEO:
1656 				so->so_snd.ssb_timeo = val;
1657 				break;
1658 			case SO_RCVTIMEO:
1659 				so->so_rcv.ssb_timeo = val;
1660 				break;
1661 			}
1662 			break;
1663 		default:
1664 			error = ENOPROTOOPT;
1665 			break;
1666 		}
1667 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
1668 			(void) so_pr_ctloutput(so, sopt);
1669 		}
1670 	}
1671 bad:
1672 	return (error);
1673 }
1674 
1675 /* Helper routine for getsockopt */
1676 int
1677 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1678 {
1679 	soopt_from_kbuf(sopt, buf, len);
1680 	return 0;
1681 }
1682 
1683 void
1684 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
1685 {
1686 	size_t	valsize;
1687 
1688 	if (len == 0) {
1689 		sopt->sopt_valsize = 0;
1690 		return;
1691 	}
1692 
1693 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1694 	KKASSERT(kva_p(buf));
1695 
1696 	/*
1697 	 * Documented get behavior is that we always return a value,
1698 	 * possibly truncated to fit in the user's buffer.
1699 	 * Traditional behavior is that we always tell the user
1700 	 * precisely how much we copied, rather than something useful
1701 	 * like the total amount we had available for her.
1702 	 * Note that this interface is not idempotent; the entire answer must
1703 	 * generated ahead of time.
1704 	 */
1705 	valsize = szmin(len, sopt->sopt_valsize);
1706 	sopt->sopt_valsize = valsize;
1707 	if (sopt->sopt_val != 0) {
1708 		bcopy(buf, sopt->sopt_val, valsize);
1709 	}
1710 }
1711 
1712 int
1713 sogetopt(struct socket *so, struct sockopt *sopt)
1714 {
1715 	int	error, optval;
1716 	long	optval_l;
1717 	struct	linger l;
1718 	struct	timeval tv;
1719 #ifdef INET
1720 	struct accept_filter_arg *afap;
1721 #endif
1722 
1723 	error = 0;
1724 	sopt->sopt_dir = SOPT_GET;
1725 	if (sopt->sopt_level != SOL_SOCKET) {
1726 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1727 			return (so_pr_ctloutput(so, sopt));
1728 		} else
1729 			return (ENOPROTOOPT);
1730 	} else {
1731 		switch (sopt->sopt_name) {
1732 #ifdef INET
1733 		case SO_ACCEPTFILTER:
1734 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1735 				return (EINVAL);
1736 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1737 				M_TEMP, M_WAITOK | M_ZERO);
1738 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1739 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1740 				if (so->so_accf->so_accept_filter_str != NULL)
1741 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1742 			}
1743 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1744 			FREE(afap, M_TEMP);
1745 			break;
1746 #endif /* INET */
1747 
1748 		case SO_LINGER:
1749 			l.l_onoff = so->so_options & SO_LINGER;
1750 			l.l_linger = so->so_linger;
1751 			error = sooptcopyout(sopt, &l, sizeof l);
1752 			break;
1753 
1754 		case SO_USELOOPBACK:
1755 		case SO_DONTROUTE:
1756 		case SO_DEBUG:
1757 		case SO_KEEPALIVE:
1758 		case SO_REUSEADDR:
1759 		case SO_REUSEPORT:
1760 		case SO_BROADCAST:
1761 		case SO_OOBINLINE:
1762 		case SO_TIMESTAMP:
1763 			optval = so->so_options & sopt->sopt_name;
1764 integer:
1765 			error = sooptcopyout(sopt, &optval, sizeof optval);
1766 			break;
1767 
1768 		case SO_TYPE:
1769 			optval = so->so_type;
1770 			goto integer;
1771 
1772 		case SO_ERROR:
1773 			optval = so->so_error;
1774 			so->so_error = 0;
1775 			goto integer;
1776 
1777 		case SO_SNDBUF:
1778 			optval = so->so_snd.ssb_hiwat;
1779 			goto integer;
1780 
1781 		case SO_RCVBUF:
1782 			optval = so->so_rcv.ssb_hiwat;
1783 			goto integer;
1784 
1785 		case SO_SNDLOWAT:
1786 			optval = so->so_snd.ssb_lowat;
1787 			goto integer;
1788 
1789 		case SO_RCVLOWAT:
1790 			optval = so->so_rcv.ssb_lowat;
1791 			goto integer;
1792 
1793 		case SO_SNDTIMEO:
1794 		case SO_RCVTIMEO:
1795 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1796 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
1797 
1798 			tv.tv_sec = optval / hz;
1799 			tv.tv_usec = (optval % hz) * ustick;
1800 			error = sooptcopyout(sopt, &tv, sizeof tv);
1801 			break;
1802 
1803 		case SO_SNDSPACE:
1804 			optval_l = ssb_space(&so->so_snd);
1805 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
1806 			break;
1807 
1808 		default:
1809 			error = ENOPROTOOPT;
1810 			break;
1811 		}
1812 		return (error);
1813 	}
1814 }
1815 
1816 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1817 int
1818 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1819 {
1820 	struct mbuf *m, *m_prev;
1821 	int sopt_size = sopt->sopt_valsize, msize;
1822 
1823 	m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT, MT_DATA,
1824 		   0, &msize);
1825 	if (m == NULL)
1826 		return (ENOBUFS);
1827 	m->m_len = min(msize, sopt_size);
1828 	sopt_size -= m->m_len;
1829 	*mp = m;
1830 	m_prev = m;
1831 
1832 	while (sopt_size > 0) {
1833 		m = m_getl(sopt_size, sopt->sopt_td ? MB_WAIT : MB_DONTWAIT,
1834 			   MT_DATA, 0, &msize);
1835 		if (m == NULL) {
1836 			m_freem(*mp);
1837 			return (ENOBUFS);
1838 		}
1839 		m->m_len = min(msize, sopt_size);
1840 		sopt_size -= m->m_len;
1841 		m_prev->m_next = m;
1842 		m_prev = m;
1843 	}
1844 	return (0);
1845 }
1846 
1847 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1848 int
1849 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1850 {
1851 	soopt_to_mbuf(sopt, m);
1852 	return 0;
1853 }
1854 
1855 void
1856 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
1857 {
1858 	size_t valsize;
1859 	void *val;
1860 
1861 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1862 	KKASSERT(kva_p(m));
1863 	if (sopt->sopt_val == NULL)
1864 		return;
1865 	val = sopt->sopt_val;
1866 	valsize = sopt->sopt_valsize;
1867 	while (m != NULL && valsize >= m->m_len) {
1868 		bcopy(val, mtod(m, char *), m->m_len);
1869 		valsize -= m->m_len;
1870 		val = (caddr_t)val + m->m_len;
1871 		m = m->m_next;
1872 	}
1873 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1874 		panic("ip6_sooptmcopyin");
1875 }
1876 
1877 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1878 int
1879 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1880 {
1881 	return soopt_from_mbuf(sopt, m);
1882 }
1883 
1884 int
1885 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
1886 {
1887 	struct mbuf *m0 = m;
1888 	size_t valsize = 0;
1889 	size_t maxsize;
1890 	void *val;
1891 
1892 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
1893 	KKASSERT(kva_p(m));
1894 	if (sopt->sopt_val == NULL)
1895 		return 0;
1896 	val = sopt->sopt_val;
1897 	maxsize = sopt->sopt_valsize;
1898 	while (m != NULL && maxsize >= m->m_len) {
1899 		bcopy(mtod(m, char *), val, m->m_len);
1900 	       maxsize -= m->m_len;
1901 	       val = (caddr_t)val + m->m_len;
1902 	       valsize += m->m_len;
1903 	       m = m->m_next;
1904 	}
1905 	if (m != NULL) {
1906 		/* enough soopt buffer should be given from user-land */
1907 		m_freem(m0);
1908 		return (EINVAL);
1909 	}
1910 	sopt->sopt_valsize = valsize;
1911 	return 0;
1912 }
1913 
1914 void
1915 sohasoutofband(struct socket *so)
1916 {
1917 	if (so->so_sigio != NULL)
1918 		pgsigio(so->so_sigio, SIGURG, 0);
1919 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
1920 }
1921 
1922 int
1923 sokqfilter(struct file *fp, struct knote *kn)
1924 {
1925 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1926 	struct signalsockbuf *ssb;
1927 
1928 	switch (kn->kn_filter) {
1929 	case EVFILT_READ:
1930 		if (so->so_options & SO_ACCEPTCONN)
1931 			kn->kn_fop = &solisten_filtops;
1932 		else
1933 			kn->kn_fop = &soread_filtops;
1934 		ssb = &so->so_rcv;
1935 		break;
1936 	case EVFILT_WRITE:
1937 		kn->kn_fop = &sowrite_filtops;
1938 		ssb = &so->so_snd;
1939 		break;
1940 	case EVFILT_EXCEPT:
1941 		kn->kn_fop = &soexcept_filtops;
1942 		ssb = &so->so_rcv;
1943 		break;
1944 	default:
1945 		return (EOPNOTSUPP);
1946 	}
1947 
1948 	knote_insert(&ssb->ssb_kq.ki_note, kn);
1949 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
1950 	return (0);
1951 }
1952 
1953 static void
1954 filt_sordetach(struct knote *kn)
1955 {
1956 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1957 
1958 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
1959 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
1960 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
1961 }
1962 
1963 /*ARGSUSED*/
1964 static int
1965 filt_soread(struct knote *kn, long hint)
1966 {
1967 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
1968 
1969 	if (kn->kn_sfflags & NOTE_OOB) {
1970 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
1971 			kn->kn_fflags |= NOTE_OOB;
1972 			return (1);
1973 		}
1974 		return (0);
1975 	}
1976 	kn->kn_data = so->so_rcv.ssb_cc;
1977 
1978 	if (so->so_state & SS_CANTRCVMORE) {
1979 		/*
1980 		 * Only set NODATA if all data has been exhausted.
1981 		 */
1982 		if (kn->kn_data == 0)
1983 			kn->kn_flags |= EV_NODATA;
1984 		kn->kn_flags |= EV_EOF;
1985 		kn->kn_fflags = so->so_error;
1986 		return (1);
1987 	}
1988 	if (so->so_error)	/* temporary udp error */
1989 		return (1);
1990 	if (kn->kn_sfflags & NOTE_LOWAT)
1991 		return (kn->kn_data >= kn->kn_sdata);
1992 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
1993 		!TAILQ_EMPTY(&so->so_comp));
1994 }
1995 
1996 static void
1997 filt_sowdetach(struct knote *kn)
1998 {
1999 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2000 
2001 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2002 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2003 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2004 }
2005 
2006 /*ARGSUSED*/
2007 static int
2008 filt_sowrite(struct knote *kn, long hint)
2009 {
2010 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2011 
2012 	kn->kn_data = ssb_space(&so->so_snd);
2013 	if (so->so_state & SS_CANTSENDMORE) {
2014 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2015 		kn->kn_fflags = so->so_error;
2016 		return (1);
2017 	}
2018 	if (so->so_error)	/* temporary udp error */
2019 		return (1);
2020 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2021 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2022 		return (0);
2023 	if (kn->kn_sfflags & NOTE_LOWAT)
2024 		return (kn->kn_data >= kn->kn_sdata);
2025 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2026 }
2027 
2028 /*ARGSUSED*/
2029 static int
2030 filt_solisten(struct knote *kn, long hint)
2031 {
2032 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2033 
2034 	kn->kn_data = so->so_qlen;
2035 	return (! TAILQ_EMPTY(&so->so_comp));
2036 }
2037