xref: /dflybsd-src/sys/kern/uipc_socket.c (revision e4bdac6bd0bece3ae6b3233ad260e8e82d21ba76)
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. Neither the name of the University nor the names of its contributors
47  *    may be used to endorse or promote products derived from this software
48  *    without specific prior written permission.
49  *
50  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
51  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
52  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
53  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
54  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
55  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
56  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
57  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
58  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
59  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
60  * SUCH DAMAGE.
61  *
62  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
63  * $FreeBSD: src/sys/kern/uipc_socket.c,v 1.68.2.24 2003/11/11 17:18:18 silby Exp $
64  */
65 
66 #include "opt_inet.h"
67 
68 #include <sys/param.h>
69 #include <sys/systm.h>
70 #include <sys/fcntl.h>
71 #include <sys/malloc.h>
72 #include <sys/mbuf.h>
73 #include <sys/domain.h>
74 #include <sys/file.h>			/* for struct knote */
75 #include <sys/kernel.h>
76 #include <sys/event.h>
77 #include <sys/proc.h>
78 #include <sys/protosw.h>
79 #include <sys/socket.h>
80 #include <sys/socketvar.h>
81 #include <sys/socketops.h>
82 #include <sys/resourcevar.h>
83 #include <sys/signalvar.h>
84 #include <sys/sysctl.h>
85 #include <sys/uio.h>
86 #include <sys/jail.h>
87 #include <vm/vm_zone.h>
88 #include <vm/pmap.h>
89 #include <net/netmsg2.h>
90 #include <net/netisr2.h>
91 
92 #include <sys/thread2.h>
93 #include <sys/socketvar2.h>
94 #include <sys/spinlock2.h>
95 
96 #include <machine/limits.h>
97 
98 #ifdef INET
99 extern int tcp_sosend_agglim;
100 extern int tcp_sosend_async;
101 extern int tcp_sosend_jcluster;
102 extern int udp_sosend_async;
103 extern int udp_sosend_prepend;
104 
105 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
106 #endif /* INET */
107 
108 static void 	filt_sordetach(struct knote *kn);
109 static int 	filt_soread(struct knote *kn, long hint);
110 static void 	filt_sowdetach(struct knote *kn);
111 static int	filt_sowrite(struct knote *kn, long hint);
112 static int	filt_solisten(struct knote *kn, long hint);
113 
114 static int	soclose_sync(struct socket *so, int fflag);
115 static void	soclose_fast(struct socket *so);
116 
117 static struct filterops solisten_filtops =
118 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
119 static struct filterops soread_filtops =
120 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
121 static struct filterops sowrite_filtops =
122 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
123 static struct filterops soexcept_filtops =
124 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
125 
126 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
127 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
128 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
129 
130 
131 static int somaxconn = SOMAXCONN;
132 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
133     &somaxconn, 0, "Maximum pending socket connection queue size");
134 
135 static int use_soclose_fast = 1;
136 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
137     &use_soclose_fast, 0, "Fast socket close");
138 
139 int use_soaccept_pred_fast = 1;
140 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
141     &use_soaccept_pred_fast, 0, "Fast socket accept predication");
142 
143 int use_sendfile_async = 1;
144 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
145     &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
146 
147 int use_soconnect_async = 1;
148 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW,
149     &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect");
150 
151 /*
152  * Socket operation routines.
153  * These routines are called by the routines in
154  * sys_socket.c or from a system process, and
155  * implement the semantics of socket operations by
156  * switching out to the protocol specific routines.
157  */
158 
159 /*
160  * Get a socket structure, and initialize it.
161  * Note that it would probably be better to allocate socket
162  * and PCB at the same time, but I'm not convinced that all
163  * the protocols can be easily modified to do this.
164  */
165 struct socket *
166 soalloc(int waitok, struct protosw *pr)
167 {
168 	struct socket *so;
169 	unsigned waitmask;
170 
171 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
172 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
173 	if (so) {
174 		/* XXX race condition for reentrant kernel */
175 		so->so_proto = pr;
176 		TAILQ_INIT(&so->so_aiojobq);
177 		TAILQ_INIT(&so->so_rcv.ssb_kq.ki_mlist);
178 		TAILQ_INIT(&so->so_snd.ssb_kq.ki_mlist);
179 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
180 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
181 		spin_init(&so->so_rcvd_spin, "soalloc");
182 		netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
183 		    MSGF_DROPABLE | MSGF_PRIORITY,
184 		    so->so_proto->pr_usrreqs->pru_rcvd);
185 		so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
186 		so->so_state = SS_NOFDREF;
187 		so->so_refs = 1;
188 	}
189 	return so;
190 }
191 
192 int
193 socreate(int dom, struct socket **aso, int type,
194 	int proto, struct thread *td)
195 {
196 	struct proc *p = td->td_proc;
197 	struct protosw *prp;
198 	struct socket *so;
199 	struct pru_attach_info ai;
200 	int error;
201 
202 	if (proto)
203 		prp = pffindproto(dom, proto, type);
204 	else
205 		prp = pffindtype(dom, type);
206 
207 	if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
208 		return (EPROTONOSUPPORT);
209 
210 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
211 	    prp->pr_domain->dom_family != PF_LOCAL &&
212 	    prp->pr_domain->dom_family != PF_INET &&
213 	    prp->pr_domain->dom_family != PF_INET6 &&
214 	    prp->pr_domain->dom_family != PF_ROUTE) {
215 		return (EPROTONOSUPPORT);
216 	}
217 
218 	if (prp->pr_type != type)
219 		return (EPROTOTYPE);
220 	so = soalloc(p != NULL, prp);
221 	if (so == NULL)
222 		return (ENOBUFS);
223 
224 	/*
225 	 * Callers of socreate() presumably will connect up a descriptor
226 	 * and call soclose() if they cannot.  This represents our so_refs
227 	 * (which should be 1) from soalloc().
228 	 */
229 	soclrstate(so, SS_NOFDREF);
230 
231 	/*
232 	 * Set a default port for protocol processing.  No action will occur
233 	 * on the socket on this port until an inpcb is attached to it and
234 	 * is able to match incoming packets, or until the socket becomes
235 	 * available to userland.
236 	 *
237 	 * We normally default the socket to the protocol thread on cpu 0,
238 	 * if protocol does not provide its own method to initialize the
239 	 * default port.
240 	 *
241 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
242 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
243 	 */
244 	if (prp->pr_flags & PR_SYNC_PORT)
245 		so->so_port = &netisr_sync_port;
246 	else if (prp->pr_initport != NULL)
247 		so->so_port = prp->pr_initport();
248 	else
249 		so->so_port = netisr_cpuport(0);
250 
251 	TAILQ_INIT(&so->so_incomp);
252 	TAILQ_INIT(&so->so_comp);
253 	so->so_type = type;
254 	so->so_cred = crhold(p->p_ucred);
255 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
256 	ai.p_ucred = p->p_ucred;
257 	ai.fd_rdir = p->p_fd->fd_rdir;
258 
259 	/*
260 	 * Auto-sizing of socket buffers is managed by the protocols and
261 	 * the appropriate flags must be set in the pru_attach function.
262 	 */
263 	error = so_pru_attach(so, proto, &ai);
264 	if (error) {
265 		sosetstate(so, SS_NOFDREF);
266 		sofree(so);	/* from soalloc */
267 		return error;
268 	}
269 
270 	/*
271 	 * NOTE: Returns referenced socket.
272 	 */
273 	*aso = so;
274 	return (0);
275 }
276 
277 int
278 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
279 {
280 	int error;
281 
282 	error = so_pru_bind(so, nam, td);
283 	return (error);
284 }
285 
286 static void
287 sodealloc(struct socket *so)
288 {
289 	if (so->so_rcv.ssb_hiwat)
290 		(void)chgsbsize(so->so_cred->cr_uidinfo,
291 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
292 	if (so->so_snd.ssb_hiwat)
293 		(void)chgsbsize(so->so_cred->cr_uidinfo,
294 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
295 #ifdef INET
296 	/* remove accept filter if present */
297 	if (so->so_accf != NULL)
298 		do_setopt_accept_filter(so, NULL);
299 #endif /* INET */
300 	crfree(so->so_cred);
301 	if (so->so_faddr != NULL)
302 		kfree(so->so_faddr, M_SONAME);
303 	kfree(so, M_SOCKET);
304 }
305 
306 int
307 solisten(struct socket *so, int backlog, struct thread *td)
308 {
309 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
310 		return (EINVAL);
311 
312 	lwkt_gettoken(&so->so_rcv.ssb_token);
313 	if (TAILQ_EMPTY(&so->so_comp))
314 		so->so_options |= SO_ACCEPTCONN;
315 	lwkt_reltoken(&so->so_rcv.ssb_token);
316 	if (backlog < 0 || backlog > somaxconn)
317 		backlog = somaxconn;
318 	so->so_qlimit = backlog;
319 	return so_pru_listen(so, td);
320 }
321 
322 /*
323  * Destroy a disconnected socket.  This routine is a NOP if entities
324  * still have a reference on the socket:
325  *
326  *	so_pcb -	The protocol stack still has a reference
327  *	SS_NOFDREF -	There is no longer a file pointer reference
328  */
329 void
330 sofree(struct socket *so)
331 {
332 	struct socket *head;
333 
334 	/*
335 	 * This is a bit hackish at the moment.  We need to interlock
336 	 * any accept queue we are on before we potentially lose the
337 	 * last reference to avoid races against a re-reference from
338 	 * someone operating on the queue.
339 	 */
340 	while ((head = so->so_head) != NULL) {
341 		lwkt_getpooltoken(head);
342 		if (so->so_head == head)
343 			break;
344 		lwkt_relpooltoken(head);
345 	}
346 
347 	/*
348 	 * Arbitrage the last free.
349 	 */
350 	KKASSERT(so->so_refs > 0);
351 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
352 		if (head)
353 			lwkt_relpooltoken(head);
354 		return;
355 	}
356 
357 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
358 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
359 
360 	/*
361 	 * We're done, remove ourselves from the accept queue we are
362 	 * on, if we are on one.
363 	 */
364 	if (head != NULL) {
365 		if (so->so_state & SS_INCOMP) {
366 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
367 			head->so_incqlen--;
368 		} else if (so->so_state & SS_COMP) {
369 			/*
370 			 * We must not decommission a socket that's
371 			 * on the accept(2) queue.  If we do, then
372 			 * accept(2) may hang after select(2) indicated
373 			 * that the listening socket was ready.
374 			 */
375 			lwkt_relpooltoken(head);
376 			return;
377 		} else {
378 			panic("sofree: not queued");
379 		}
380 		soclrstate(so, SS_INCOMP);
381 		so->so_head = NULL;
382 		lwkt_relpooltoken(head);
383 	}
384 	ssb_release(&so->so_snd, so);
385 	sorflush(so);
386 	sodealloc(so);
387 }
388 
389 /*
390  * Close a socket on last file table reference removal.
391  * Initiate disconnect if connected.
392  * Free socket when disconnect complete.
393  */
394 int
395 soclose(struct socket *so, int fflag)
396 {
397 	int error;
398 
399 	funsetown(&so->so_sigio);
400 	sosetstate(so, SS_ISCLOSING);
401 	if (!use_soclose_fast ||
402 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
403 	    ((so->so_state & SS_ISCONNECTED) &&
404 	     (so->so_options & SO_LINGER))) {
405 		error = soclose_sync(so, fflag);
406 	} else {
407 		soclose_fast(so);
408 		error = 0;
409 	}
410 	return error;
411 }
412 
413 void
414 sodiscard(struct socket *so)
415 {
416 	lwkt_getpooltoken(so);
417 	if (so->so_options & SO_ACCEPTCONN) {
418 		struct socket *sp;
419 
420 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
421 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
422 			so->so_incqlen--;
423 			soclrstate(sp, SS_INCOMP);
424 			soabort_async(sp, TRUE);
425 		}
426 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
427 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
428 			so->so_qlen--;
429 			soclrstate(sp, SS_COMP);
430 			soabort_async(sp, TRUE);
431 		}
432 	}
433 	lwkt_relpooltoken(so);
434 
435 	if (so->so_state & SS_NOFDREF)
436 		panic("soclose: NOFDREF");
437 	sosetstate(so, SS_NOFDREF);	/* take ref */
438 }
439 
440 /*
441  * Append the completed queue of head to head_inh (inherting listen socket).
442  */
443 void
444 soinherit(struct socket *head, struct socket *head_inh)
445 {
446 	boolean_t do_wakeup = FALSE;
447 
448 	KASSERT(head->so_options & SO_ACCEPTCONN,
449 	    ("head does not accept connection"));
450 	KASSERT(head_inh->so_options & SO_ACCEPTCONN,
451 	    ("head_inh does not accept connection"));
452 
453 	lwkt_getpooltoken(head);
454 	lwkt_getpooltoken(head_inh);
455 
456 	if (head->so_qlen > 0)
457 		do_wakeup = TRUE;
458 
459 	while (!TAILQ_EMPTY(&head->so_comp)) {
460 		struct ucred *old_cr;
461 		struct socket *sp;
462 
463 		sp = TAILQ_FIRST(&head->so_comp);
464 
465 		/*
466 		 * Remove this socket from the current listen socket
467 		 * completed queue.
468 		 */
469 		TAILQ_REMOVE(&head->so_comp, sp, so_list);
470 		head->so_qlen--;
471 
472 		/* Save the old ucred for later free. */
473 		old_cr = sp->so_cred;
474 
475 		/*
476 		 * Install this socket to the inheriting listen socket
477 		 * completed queue.
478 		 */
479 		sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
480 		sp->so_head = head_inh;
481 
482 		TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
483 		head_inh->so_qlen++;
484 
485 		/*
486 		 * NOTE:
487 		 * crfree() may block and release the tokens temporarily.
488 		 * However, we are fine here, since the transition is done.
489 		 */
490 		crfree(old_cr);
491 	}
492 
493 	lwkt_relpooltoken(head_inh);
494 	lwkt_relpooltoken(head);
495 
496 	if (do_wakeup) {
497 		/*
498 		 * "New" connections have arrived
499 		 */
500 		sorwakeup(head_inh);
501 		wakeup(&head_inh->so_timeo);
502 	}
503 }
504 
505 static int
506 soclose_sync(struct socket *so, int fflag)
507 {
508 	int error = 0;
509 
510 	if (so->so_pcb == NULL)
511 		goto discard;
512 	if (so->so_state & SS_ISCONNECTED) {
513 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
514 			error = sodisconnect(so);
515 			if (error)
516 				goto drop;
517 		}
518 		if (so->so_options & SO_LINGER) {
519 			if ((so->so_state & SS_ISDISCONNECTING) &&
520 			    (fflag & FNONBLOCK))
521 				goto drop;
522 			while (so->so_state & SS_ISCONNECTED) {
523 				error = tsleep(&so->so_timeo, PCATCH,
524 					       "soclos", so->so_linger * hz);
525 				if (error)
526 					break;
527 			}
528 		}
529 	}
530 drop:
531 	if (so->so_pcb) {
532 		int error2;
533 
534 		error2 = so_pru_detach(so);
535 		if (error2 == EJUSTRETURN) {
536 			/*
537 			 * Protocol will call sodiscard()
538 			 * and sofree() for us.
539 			 */
540 			return error;
541 		}
542 		if (error == 0)
543 			error = error2;
544 	}
545 discard:
546 	sodiscard(so);
547 	so_pru_sync(so);	/* unpend async sending */
548 	sofree(so);		/* dispose of ref */
549 
550 	return (error);
551 }
552 
553 static void
554 soclose_sofree_async_handler(netmsg_t msg)
555 {
556 	sofree(msg->base.nm_so);
557 }
558 
559 static void
560 soclose_sofree_async(struct socket *so)
561 {
562 	struct netmsg_base *base = &so->so_clomsg;
563 
564 	netmsg_init(base, so, &netisr_apanic_rport, 0,
565 	    soclose_sofree_async_handler);
566 	lwkt_sendmsg(so->so_port, &base->lmsg);
567 }
568 
569 static void
570 soclose_disconn_async_handler(netmsg_t msg)
571 {
572 	struct socket *so = msg->base.nm_so;
573 
574 	if ((so->so_state & SS_ISCONNECTED) &&
575 	    (so->so_state & SS_ISDISCONNECTING) == 0)
576 		so_pru_disconnect_direct(so);
577 
578 	if (so->so_pcb) {
579 		int error;
580 
581 		error = so_pru_detach_direct(so);
582 		if (error == EJUSTRETURN) {
583 			/*
584 			 * Protocol will call sodiscard()
585 			 * and sofree() for us.
586 			 */
587 			return;
588 		}
589 	}
590 
591 	sodiscard(so);
592 	sofree(so);
593 }
594 
595 static void
596 soclose_disconn_async(struct socket *so)
597 {
598 	struct netmsg_base *base = &so->so_clomsg;
599 
600 	netmsg_init(base, so, &netisr_apanic_rport, 0,
601 	    soclose_disconn_async_handler);
602 	lwkt_sendmsg(so->so_port, &base->lmsg);
603 }
604 
605 static void
606 soclose_detach_async_handler(netmsg_t msg)
607 {
608 	struct socket *so = msg->base.nm_so;
609 
610 	if (so->so_pcb) {
611 		int error;
612 
613 		error = so_pru_detach_direct(so);
614 		if (error == EJUSTRETURN) {
615 			/*
616 			 * Protocol will call sodiscard()
617 			 * and sofree() for us.
618 			 */
619 			return;
620 		}
621 	}
622 
623 	sodiscard(so);
624 	sofree(so);
625 }
626 
627 static void
628 soclose_detach_async(struct socket *so)
629 {
630 	struct netmsg_base *base = &so->so_clomsg;
631 
632 	netmsg_init(base, so, &netisr_apanic_rport, 0,
633 	    soclose_detach_async_handler);
634 	lwkt_sendmsg(so->so_port, &base->lmsg);
635 }
636 
637 static void
638 soclose_fast(struct socket *so)
639 {
640 	if (so->so_pcb == NULL)
641 		goto discard;
642 
643 	if ((so->so_state & SS_ISCONNECTED) &&
644 	    (so->so_state & SS_ISDISCONNECTING) == 0) {
645 		soclose_disconn_async(so);
646 		return;
647 	}
648 
649 	if (so->so_pcb) {
650 		soclose_detach_async(so);
651 		return;
652 	}
653 
654 discard:
655 	sodiscard(so);
656 	soclose_sofree_async(so);
657 }
658 
659 /*
660  * Abort and destroy a socket.  Only one abort can be in progress
661  * at any given moment.
662  */
663 void
664 soabort_async(struct socket *so, boolean_t clr_head)
665 {
666 	/*
667 	 * Keep a reference before clearing the so_head
668 	 * to avoid racing socket close in netisr.
669 	 */
670 	soreference(so);
671 	if (clr_head)
672 		so->so_head = NULL;
673 	so_pru_abort_async(so);
674 }
675 
676 void
677 soabort_oncpu(struct socket *so)
678 {
679 	soreference(so);
680 	so_pru_abort_direct(so);
681 }
682 
683 /*
684  * so is passed in ref'd, which becomes owned by
685  * the cleared SS_NOFDREF flag.
686  */
687 void
688 soaccept_generic(struct socket *so)
689 {
690 	if ((so->so_state & SS_NOFDREF) == 0)
691 		panic("soaccept: !NOFDREF");
692 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
693 }
694 
695 int
696 soaccept(struct socket *so, struct sockaddr **nam)
697 {
698 	int error;
699 
700 	soaccept_generic(so);
701 	error = so_pru_accept(so, nam);
702 	return (error);
703 }
704 
705 int
706 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
707     boolean_t sync)
708 {
709 	int error;
710 
711 	if (so->so_options & SO_ACCEPTCONN)
712 		return (EOPNOTSUPP);
713 	/*
714 	 * If protocol is connection-based, can only connect once.
715 	 * Otherwise, if connected, try to disconnect first.
716 	 * This allows user to disconnect by connecting to, e.g.,
717 	 * a null address.
718 	 */
719 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
720 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
721 	    (error = sodisconnect(so)))) {
722 		error = EISCONN;
723 	} else {
724 		/*
725 		 * Prevent accumulated error from previous connection
726 		 * from biting us.
727 		 */
728 		so->so_error = 0;
729 		if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
730 			error = so_pru_connect_async(so, nam, td);
731 		else
732 			error = so_pru_connect(so, nam, td);
733 	}
734 	return (error);
735 }
736 
737 int
738 soconnect2(struct socket *so1, struct socket *so2)
739 {
740 	int error;
741 
742 	error = so_pru_connect2(so1, so2);
743 	return (error);
744 }
745 
746 int
747 sodisconnect(struct socket *so)
748 {
749 	int error;
750 
751 	if ((so->so_state & SS_ISCONNECTED) == 0) {
752 		error = ENOTCONN;
753 		goto bad;
754 	}
755 	if (so->so_state & SS_ISDISCONNECTING) {
756 		error = EALREADY;
757 		goto bad;
758 	}
759 	error = so_pru_disconnect(so);
760 bad:
761 	return (error);
762 }
763 
764 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
765 /*
766  * Send on a socket.
767  * If send must go all at once and message is larger than
768  * send buffering, then hard error.
769  * Lock against other senders.
770  * If must go all at once and not enough room now, then
771  * inform user that this would block and do nothing.
772  * Otherwise, if nonblocking, send as much as possible.
773  * The data to be sent is described by "uio" if nonzero,
774  * otherwise by the mbuf chain "top" (which must be null
775  * if uio is not).  Data provided in mbuf chain must be small
776  * enough to send all at once.
777  *
778  * Returns nonzero on error, timeout or signal; callers
779  * must check for short counts if EINTR/ERESTART are returned.
780  * Data and control buffers are freed on return.
781  */
782 int
783 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
784 	struct mbuf *top, struct mbuf *control, int flags,
785 	struct thread *td)
786 {
787 	struct mbuf **mp;
788 	struct mbuf *m;
789 	size_t resid;
790 	int space, len;
791 	int clen = 0, error, dontroute, mlen;
792 	int atomic = sosendallatonce(so) || top;
793 	int pru_flags;
794 
795 	if (uio) {
796 		resid = uio->uio_resid;
797 	} else {
798 		resid = (size_t)top->m_pkthdr.len;
799 #ifdef INVARIANTS
800 		len = 0;
801 		for (m = top; m; m = m->m_next)
802 			len += m->m_len;
803 		KKASSERT(top->m_pkthdr.len == len);
804 #endif
805 	}
806 
807 	/*
808 	 * WARNING!  resid is unsigned, space and len are signed.  space
809 	 * 	     can wind up negative if the sockbuf is overcommitted.
810 	 *
811 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
812 	 * type sockets since that's an error.
813 	 */
814 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
815 		error = EINVAL;
816 		goto out;
817 	}
818 
819 	dontroute =
820 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
821 	    (so->so_proto->pr_flags & PR_ATOMIC);
822 	if (td->td_lwp != NULL)
823 		td->td_lwp->lwp_ru.ru_msgsnd++;
824 	if (control)
825 		clen = control->m_len;
826 #define	gotoerr(errcode)	{ error = errcode; goto release; }
827 
828 restart:
829 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
830 	if (error)
831 		goto out;
832 
833 	do {
834 		if (so->so_state & SS_CANTSENDMORE)
835 			gotoerr(EPIPE);
836 		if (so->so_error) {
837 			error = so->so_error;
838 			so->so_error = 0;
839 			goto release;
840 		}
841 		if ((so->so_state & SS_ISCONNECTED) == 0) {
842 			/*
843 			 * `sendto' and `sendmsg' is allowed on a connection-
844 			 * based socket if it supports implied connect.
845 			 * Return ENOTCONN if not connected and no address is
846 			 * supplied.
847 			 */
848 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
849 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
850 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
851 				    !(resid == 0 && clen != 0))
852 					gotoerr(ENOTCONN);
853 			} else if (addr == NULL)
854 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
855 				   ENOTCONN : EDESTADDRREQ);
856 		}
857 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
858 		    clen > so->so_snd.ssb_hiwat) {
859 			gotoerr(EMSGSIZE);
860 		}
861 		space = ssb_space(&so->so_snd);
862 		if (flags & MSG_OOB)
863 			space += 1024;
864 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
865 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
866 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
867 				gotoerr(EWOULDBLOCK);
868 			ssb_unlock(&so->so_snd);
869 			error = ssb_wait(&so->so_snd);
870 			if (error)
871 				goto out;
872 			goto restart;
873 		}
874 		mp = &top;
875 		space -= clen;
876 		do {
877 		    if (uio == NULL) {
878 			/*
879 			 * Data is prepackaged in "top".
880 			 */
881 			resid = 0;
882 			if (flags & MSG_EOR)
883 				top->m_flags |= M_EOR;
884 		    } else do {
885 			if (resid > INT_MAX)
886 				resid = INT_MAX;
887 			m = m_getl((int)resid, M_WAITOK, MT_DATA,
888 				   top == NULL ? M_PKTHDR : 0, &mlen);
889 			if (top == NULL) {
890 				m->m_pkthdr.len = 0;
891 				m->m_pkthdr.rcvif = NULL;
892 			}
893 			len = imin((int)szmin(mlen, resid), space);
894 			if (resid < MINCLSIZE) {
895 				/*
896 				 * For datagram protocols, leave room
897 				 * for protocol headers in first mbuf.
898 				 */
899 				if (atomic && top == NULL && len < mlen)
900 					MH_ALIGN(m, len);
901 			}
902 			space -= len;
903 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
904 			resid = uio->uio_resid;
905 			m->m_len = len;
906 			*mp = m;
907 			top->m_pkthdr.len += len;
908 			if (error)
909 				goto release;
910 			mp = &m->m_next;
911 			if (resid == 0) {
912 				if (flags & MSG_EOR)
913 					top->m_flags |= M_EOR;
914 				break;
915 			}
916 		    } while (space > 0 && atomic);
917 		    if (dontroute)
918 			    so->so_options |= SO_DONTROUTE;
919 		    if (flags & MSG_OOB) {
920 		    	    pru_flags = PRUS_OOB;
921 		    } else if ((flags & MSG_EOF) &&
922 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
923 			       (resid == 0)) {
924 			    /*
925 			     * If the user set MSG_EOF, the protocol
926 			     * understands this flag and nothing left to
927 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
928 			     */
929 		    	    pru_flags = PRUS_EOF;
930 		    } else if (resid > 0 && space > 0) {
931 			    /* If there is more to send, set PRUS_MORETOCOME */
932 		    	    pru_flags = PRUS_MORETOCOME;
933 		    } else {
934 		    	    pru_flags = 0;
935 		    }
936 		    /*
937 		     * XXX all the SS_CANTSENDMORE checks previously
938 		     * done could be out of date.  We could have recieved
939 		     * a reset packet in an interrupt or maybe we slept
940 		     * while doing page faults in uiomove() etc. We could
941 		     * probably recheck again inside the splnet() protection
942 		     * here, but there are probably other places that this
943 		     * also happens.  We must rethink this.
944 		     */
945 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
946 		    if (dontroute)
947 			    so->so_options &= ~SO_DONTROUTE;
948 		    clen = 0;
949 		    control = NULL;
950 		    top = NULL;
951 		    mp = &top;
952 		    if (error)
953 			    goto release;
954 		} while (resid && space > 0);
955 	} while (resid);
956 
957 release:
958 	ssb_unlock(&so->so_snd);
959 out:
960 	if (top)
961 		m_freem(top);
962 	if (control)
963 		m_freem(control);
964 	return (error);
965 }
966 
967 #ifdef INET
968 /*
969  * A specialization of sosend() for UDP based on protocol-specific knowledge:
970  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
971  *	sosendallatonce() returns true,
972  *	the "atomic" variable is true,
973  *	and sosendudp() blocks until space is available for the entire send.
974  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
975  *	PR_IMPLOPCL flags set.
976  *   UDP has no out-of-band data.
977  *   UDP has no control data.
978  *   UDP does not support MSG_EOR.
979  */
980 int
981 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
982 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
983 {
984 	size_t resid;
985 	int error, pru_flags = 0;
986 	int space;
987 
988 	if (td->td_lwp != NULL)
989 		td->td_lwp->lwp_ru.ru_msgsnd++;
990 	if (control)
991 		m_freem(control);
992 
993 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
994 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
995 
996 restart:
997 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
998 	if (error)
999 		goto out;
1000 
1001 	if (so->so_state & SS_CANTSENDMORE)
1002 		gotoerr(EPIPE);
1003 	if (so->so_error) {
1004 		error = so->so_error;
1005 		so->so_error = 0;
1006 		goto release;
1007 	}
1008 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
1009 		gotoerr(EDESTADDRREQ);
1010 	if (resid > so->so_snd.ssb_hiwat)
1011 		gotoerr(EMSGSIZE);
1012 	space = ssb_space(&so->so_snd);
1013 	if (uio && (space < 0 || (size_t)space < resid)) {
1014 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1015 			gotoerr(EWOULDBLOCK);
1016 		ssb_unlock(&so->so_snd);
1017 		error = ssb_wait(&so->so_snd);
1018 		if (error)
1019 			goto out;
1020 		goto restart;
1021 	}
1022 
1023 	if (uio) {
1024 		int hdrlen = max_hdr;
1025 
1026 		/*
1027 		 * We try to optimize out the additional mbuf
1028 		 * allocations in M_PREPEND() on output path, e.g.
1029 		 * - udp_output(), when it tries to prepend protocol
1030 		 *   headers.
1031 		 * - Link layer output function, when it tries to
1032 		 *   prepend link layer header.
1033 		 *
1034 		 * This probably will not benefit any data that will
1035 		 * be fragmented, so this optimization is only performed
1036 		 * when the size of data and max size of protocol+link
1037 		 * headers fit into one mbuf cluster.
1038 		 */
1039 		if (uio->uio_resid > MCLBYTES - hdrlen ||
1040 		    !udp_sosend_prepend) {
1041 			top = m_uiomove(uio);
1042 			if (top == NULL)
1043 				goto release;
1044 		} else {
1045 			int nsize;
1046 
1047 			top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1048 			    MT_DATA, M_PKTHDR, &nsize);
1049 			KASSERT(nsize >= uio->uio_resid + hdrlen,
1050 			    ("sosendudp invalid nsize %d, "
1051 			     "resid %zu, hdrlen %d",
1052 			     nsize, uio->uio_resid, hdrlen));
1053 
1054 			top->m_len = uio->uio_resid;
1055 			top->m_pkthdr.len = uio->uio_resid;
1056 			top->m_data += hdrlen;
1057 
1058 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1059 			if (error)
1060 				goto out;
1061 		}
1062 	}
1063 
1064 	if (flags & MSG_DONTROUTE)
1065 		pru_flags |= PRUS_DONTROUTE;
1066 
1067 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1068 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1069 		error = 0;
1070 	} else {
1071 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1072 	}
1073 	top = NULL;		/* sent or freed in lower layer */
1074 
1075 release:
1076 	ssb_unlock(&so->so_snd);
1077 out:
1078 	if (top)
1079 		m_freem(top);
1080 	return (error);
1081 }
1082 
1083 int
1084 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1085 	struct mbuf *top, struct mbuf *control, int flags,
1086 	struct thread *td)
1087 {
1088 	struct mbuf **mp;
1089 	struct mbuf *m;
1090 	size_t resid;
1091 	int space, len;
1092 	int error, mlen;
1093 	int allatonce;
1094 	int pru_flags;
1095 
1096 	if (uio) {
1097 		KKASSERT(top == NULL);
1098 		allatonce = 0;
1099 		resid = uio->uio_resid;
1100 	} else {
1101 		allatonce = 1;
1102 		resid = (size_t)top->m_pkthdr.len;
1103 #ifdef INVARIANTS
1104 		len = 0;
1105 		for (m = top; m; m = m->m_next)
1106 			len += m->m_len;
1107 		KKASSERT(top->m_pkthdr.len == len);
1108 #endif
1109 	}
1110 
1111 	/*
1112 	 * WARNING!  resid is unsigned, space and len are signed.  space
1113 	 * 	     can wind up negative if the sockbuf is overcommitted.
1114 	 *
1115 	 * Also check to make sure that MSG_EOR isn't used on TCP
1116 	 */
1117 	if (flags & MSG_EOR) {
1118 		error = EINVAL;
1119 		goto out;
1120 	}
1121 
1122 	if (control) {
1123 		/* TCP doesn't do control messages (rights, creds, etc) */
1124 		if (control->m_len) {
1125 			error = EINVAL;
1126 			goto out;
1127 		}
1128 		m_freem(control);	/* empty control, just free it */
1129 		control = NULL;
1130 	}
1131 
1132 	if (td->td_lwp != NULL)
1133 		td->td_lwp->lwp_ru.ru_msgsnd++;
1134 
1135 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1136 
1137 restart:
1138 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1139 	if (error)
1140 		goto out;
1141 
1142 	do {
1143 		if (so->so_state & SS_CANTSENDMORE)
1144 			gotoerr(EPIPE);
1145 		if (so->so_error) {
1146 			error = so->so_error;
1147 			so->so_error = 0;
1148 			goto release;
1149 		}
1150 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1151 		    (so->so_state & SS_ISCONFIRMING) == 0)
1152 			gotoerr(ENOTCONN);
1153 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1154 			gotoerr(EMSGSIZE);
1155 
1156 		space = ssb_space_prealloc(&so->so_snd);
1157 		if (flags & MSG_OOB)
1158 			space += 1024;
1159 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1160 		    space < so->so_snd.ssb_lowat) {
1161 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1162 				gotoerr(EWOULDBLOCK);
1163 			ssb_unlock(&so->so_snd);
1164 			error = ssb_wait(&so->so_snd);
1165 			if (error)
1166 				goto out;
1167 			goto restart;
1168 		}
1169 		mp = &top;
1170 		do {
1171 		    int cnt = 0, async = 0;
1172 
1173 		    if (uio == NULL) {
1174 			/*
1175 			 * Data is prepackaged in "top".
1176 			 */
1177 			resid = 0;
1178 		    } else do {
1179 			if (resid > INT_MAX)
1180 				resid = INT_MAX;
1181 			if (tcp_sosend_jcluster) {
1182 				m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1183 					   top == NULL ? M_PKTHDR : 0, &mlen);
1184 			} else {
1185 				m = m_getl((int)resid, M_WAITOK, MT_DATA,
1186 					   top == NULL ? M_PKTHDR : 0, &mlen);
1187 			}
1188 			if (top == NULL) {
1189 				m->m_pkthdr.len = 0;
1190 				m->m_pkthdr.rcvif = NULL;
1191 			}
1192 			len = imin((int)szmin(mlen, resid), space);
1193 			space -= len;
1194 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1195 			resid = uio->uio_resid;
1196 			m->m_len = len;
1197 			*mp = m;
1198 			top->m_pkthdr.len += len;
1199 			if (error)
1200 				goto release;
1201 			mp = &m->m_next;
1202 			if (resid == 0)
1203 				break;
1204 			++cnt;
1205 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1206 
1207 		    if (tcp_sosend_async)
1208 			    async = 1;
1209 
1210 		    if (flags & MSG_OOB) {
1211 		    	    pru_flags = PRUS_OOB;
1212 			    async = 0;
1213 		    } else if ((flags & MSG_EOF) && resid == 0) {
1214 			    pru_flags = PRUS_EOF;
1215 		    } else if (resid > 0 && space > 0) {
1216 			    /* If there is more to send, set PRUS_MORETOCOME */
1217 		    	    pru_flags = PRUS_MORETOCOME;
1218 			    async = 1;
1219 		    } else {
1220 		    	    pru_flags = 0;
1221 		    }
1222 
1223 		    if (flags & MSG_SYNC)
1224 			    async = 0;
1225 
1226 		    /*
1227 		     * XXX all the SS_CANTSENDMORE checks previously
1228 		     * done could be out of date.  We could have recieved
1229 		     * a reset packet in an interrupt or maybe we slept
1230 		     * while doing page faults in uiomove() etc. We could
1231 		     * probably recheck again inside the splnet() protection
1232 		     * here, but there are probably other places that this
1233 		     * also happens.  We must rethink this.
1234 		     */
1235 		    for (m = top; m; m = m->m_next)
1236 			    ssb_preallocstream(&so->so_snd, m);
1237 		    if (!async) {
1238 			    error = so_pru_send(so, pru_flags, top,
1239 			        NULL, NULL, td);
1240 		    } else {
1241 			    so_pru_send_async(so, pru_flags, top,
1242 			        NULL, NULL, td);
1243 			    error = 0;
1244 		    }
1245 
1246 		    top = NULL;
1247 		    mp = &top;
1248 		    if (error)
1249 			    goto release;
1250 		} while (resid && space > 0);
1251 	} while (resid);
1252 
1253 release:
1254 	ssb_unlock(&so->so_snd);
1255 out:
1256 	if (top)
1257 		m_freem(top);
1258 	if (control)
1259 		m_freem(control);
1260 	return (error);
1261 }
1262 #endif
1263 
1264 /*
1265  * Implement receive operations on a socket.
1266  *
1267  * We depend on the way that records are added to the signalsockbuf
1268  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1269  * must begin with an address if the protocol so specifies,
1270  * followed by an optional mbuf or mbufs containing ancillary data,
1271  * and then zero or more mbufs of data.
1272  *
1273  * Although the signalsockbuf is locked, new data may still be appended.
1274  * A token inside the ssb_lock deals with MP issues and still allows
1275  * the network to access the socket if we block in a uio.
1276  *
1277  * The caller may receive the data as a single mbuf chain by supplying
1278  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1279  * only for the count in uio_resid.
1280  */
1281 int
1282 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1283 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1284 {
1285 	struct mbuf *m, *n;
1286 	struct mbuf *free_chain = NULL;
1287 	int flags, len, error, offset;
1288 	struct protosw *pr = so->so_proto;
1289 	int moff, type = 0;
1290 	size_t resid, orig_resid;
1291 
1292 	if (uio)
1293 		resid = uio->uio_resid;
1294 	else
1295 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1296 	orig_resid = resid;
1297 
1298 	if (psa)
1299 		*psa = NULL;
1300 	if (controlp)
1301 		*controlp = NULL;
1302 	if (flagsp)
1303 		flags = *flagsp &~ MSG_EOR;
1304 	else
1305 		flags = 0;
1306 	if (flags & MSG_OOB) {
1307 		m = m_get(M_WAITOK, MT_DATA);
1308 		if (m == NULL)
1309 			return (ENOBUFS);
1310 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1311 		if (error)
1312 			goto bad;
1313 		if (sio) {
1314 			do {
1315 				sbappend(sio, m);
1316 				KKASSERT(resid >= (size_t)m->m_len);
1317 				resid -= (size_t)m->m_len;
1318 			} while (resid > 0 && m);
1319 		} else {
1320 			do {
1321 				uio->uio_resid = resid;
1322 				error = uiomove(mtod(m, caddr_t),
1323 						(int)szmin(resid, m->m_len),
1324 						uio);
1325 				resid = uio->uio_resid;
1326 				m = m_free(m);
1327 			} while (uio->uio_resid && error == 0 && m);
1328 		}
1329 bad:
1330 		if (m)
1331 			m_freem(m);
1332 		return (error);
1333 	}
1334 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1335 		so_pru_rcvd(so, 0);
1336 
1337 	/*
1338 	 * The token interlocks against the protocol thread while
1339 	 * ssb_lock is a blocking lock against other userland entities.
1340 	 */
1341 	lwkt_gettoken(&so->so_rcv.ssb_token);
1342 restart:
1343 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1344 	if (error)
1345 		goto done;
1346 
1347 	m = so->so_rcv.ssb_mb;
1348 	/*
1349 	 * If we have less data than requested, block awaiting more
1350 	 * (subject to any timeout) if:
1351 	 *   1. the current count is less than the low water mark, or
1352 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1353 	 *	receive operation at once if we block (resid <= hiwat).
1354 	 *   3. MSG_DONTWAIT is not set
1355 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1356 	 * we have to do the receive in sections, and thus risk returning
1357 	 * a short count if a timeout or signal occurs after we start.
1358 	 */
1359 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1360 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1361 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1362 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1363 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1364 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1365 		if (so->so_error) {
1366 			if (m)
1367 				goto dontblock;
1368 			error = so->so_error;
1369 			if ((flags & MSG_PEEK) == 0)
1370 				so->so_error = 0;
1371 			goto release;
1372 		}
1373 		if (so->so_state & SS_CANTRCVMORE) {
1374 			if (m)
1375 				goto dontblock;
1376 			else
1377 				goto release;
1378 		}
1379 		for (; m; m = m->m_next) {
1380 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1381 				m = so->so_rcv.ssb_mb;
1382 				goto dontblock;
1383 			}
1384 		}
1385 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1386 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1387 			error = ENOTCONN;
1388 			goto release;
1389 		}
1390 		if (resid == 0)
1391 			goto release;
1392 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1393 			error = EWOULDBLOCK;
1394 			goto release;
1395 		}
1396 		ssb_unlock(&so->so_rcv);
1397 		error = ssb_wait(&so->so_rcv);
1398 		if (error)
1399 			goto done;
1400 		goto restart;
1401 	}
1402 dontblock:
1403 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1404 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1405 
1406 	/*
1407 	 * note: m should be == sb_mb here.  Cache the next record while
1408 	 * cleaning up.  Note that calling m_free*() will break out critical
1409 	 * section.
1410 	 */
1411 	KKASSERT(m == so->so_rcv.ssb_mb);
1412 
1413 	/*
1414 	 * Skip any address mbufs prepending the record.
1415 	 */
1416 	if (pr->pr_flags & PR_ADDR) {
1417 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1418 		orig_resid = 0;
1419 		if (psa)
1420 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1421 		if (flags & MSG_PEEK)
1422 			m = m->m_next;
1423 		else
1424 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1425 	}
1426 
1427 	/*
1428 	 * Skip any control mbufs prepending the record.
1429 	 */
1430 	while (m && m->m_type == MT_CONTROL && error == 0) {
1431 		if (flags & MSG_PEEK) {
1432 			if (controlp)
1433 				*controlp = m_copy(m, 0, m->m_len);
1434 			m = m->m_next;	/* XXX race */
1435 		} else {
1436 			if (controlp) {
1437 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1438 				if (pr->pr_domain->dom_externalize &&
1439 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1440 				    SCM_RIGHTS)
1441 				   error = (*pr->pr_domain->dom_externalize)(m);
1442 				*controlp = m;
1443 				m = n;
1444 			} else {
1445 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1446 			}
1447 		}
1448 		if (controlp && *controlp) {
1449 			orig_resid = 0;
1450 			controlp = &(*controlp)->m_next;
1451 		}
1452 	}
1453 
1454 	/*
1455 	 * flag OOB data.
1456 	 */
1457 	if (m) {
1458 		type = m->m_type;
1459 		if (type == MT_OOBDATA)
1460 			flags |= MSG_OOB;
1461 	}
1462 
1463 	/*
1464 	 * Copy to the UIO or mbuf return chain (*mp).
1465 	 */
1466 	moff = 0;
1467 	offset = 0;
1468 	while (m && resid > 0 && error == 0) {
1469 		if (m->m_type == MT_OOBDATA) {
1470 			if (type != MT_OOBDATA)
1471 				break;
1472 		} else if (type == MT_OOBDATA)
1473 			break;
1474 		else
1475 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1476 			("receive 3"));
1477 		soclrstate(so, SS_RCVATMARK);
1478 		len = (resid > INT_MAX) ? INT_MAX : resid;
1479 		if (so->so_oobmark && len > so->so_oobmark - offset)
1480 			len = so->so_oobmark - offset;
1481 		if (len > m->m_len - moff)
1482 			len = m->m_len - moff;
1483 
1484 		/*
1485 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1486 		 * The SIO is dealt with when we eat the mbuf, but deal
1487 		 * with the resid here either way.
1488 		 */
1489 		if (uio) {
1490 			uio->uio_resid = resid;
1491 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1492 			resid = uio->uio_resid;
1493 			if (error)
1494 				goto release;
1495 		} else {
1496 			resid -= (size_t)len;
1497 		}
1498 
1499 		/*
1500 		 * Eat the entire mbuf or just a piece of it
1501 		 */
1502 		if (len == m->m_len - moff) {
1503 			if (m->m_flags & M_EOR)
1504 				flags |= MSG_EOR;
1505 			if (flags & MSG_PEEK) {
1506 				m = m->m_next;
1507 				moff = 0;
1508 			} else {
1509 				if (sio) {
1510 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1511 					sbappend(sio, m);
1512 					m = n;
1513 				} else {
1514 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1515 				}
1516 			}
1517 		} else {
1518 			if (flags & MSG_PEEK) {
1519 				moff += len;
1520 			} else {
1521 				if (sio) {
1522 					n = m_copym(m, 0, len, M_WAITOK);
1523 					if (n)
1524 						sbappend(sio, n);
1525 				}
1526 				m->m_data += len;
1527 				m->m_len -= len;
1528 				so->so_rcv.ssb_cc -= len;
1529 			}
1530 		}
1531 		if (so->so_oobmark) {
1532 			if ((flags & MSG_PEEK) == 0) {
1533 				so->so_oobmark -= len;
1534 				if (so->so_oobmark == 0) {
1535 					sosetstate(so, SS_RCVATMARK);
1536 					break;
1537 				}
1538 			} else {
1539 				offset += len;
1540 				if (offset == so->so_oobmark)
1541 					break;
1542 			}
1543 		}
1544 		if (flags & MSG_EOR)
1545 			break;
1546 		/*
1547 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1548 		 * we must not quit until resid == 0 or an error
1549 		 * termination.  If a signal/timeout occurs, return
1550 		 * with a short count but without error.
1551 		 * Keep signalsockbuf locked against other readers.
1552 		 */
1553 		while ((flags & MSG_WAITALL) && m == NULL &&
1554 		       resid > 0 && !sosendallatonce(so) &&
1555 		       so->so_rcv.ssb_mb == NULL) {
1556 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1557 				break;
1558 			/*
1559 			 * The window might have closed to zero, make
1560 			 * sure we send an ack now that we've drained
1561 			 * the buffer or we might end up blocking until
1562 			 * the idle takes over (5 seconds).
1563 			 */
1564 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1565 				so_pru_rcvd(so, flags);
1566 			error = ssb_wait(&so->so_rcv);
1567 			if (error) {
1568 				ssb_unlock(&so->so_rcv);
1569 				error = 0;
1570 				goto done;
1571 			}
1572 			m = so->so_rcv.ssb_mb;
1573 		}
1574 	}
1575 
1576 	/*
1577 	 * If an atomic read was requested but unread data still remains
1578 	 * in the record, set MSG_TRUNC.
1579 	 */
1580 	if (m && pr->pr_flags & PR_ATOMIC)
1581 		flags |= MSG_TRUNC;
1582 
1583 	/*
1584 	 * Cleanup.  If an atomic read was requested drop any unread data.
1585 	 */
1586 	if ((flags & MSG_PEEK) == 0) {
1587 		if (m && (pr->pr_flags & PR_ATOMIC))
1588 			sbdroprecord(&so->so_rcv.sb);
1589 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1590 			so_pru_rcvd(so, flags);
1591 	}
1592 
1593 	if (orig_resid == resid && orig_resid &&
1594 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1595 		ssb_unlock(&so->so_rcv);
1596 		goto restart;
1597 	}
1598 
1599 	if (flagsp)
1600 		*flagsp |= flags;
1601 release:
1602 	ssb_unlock(&so->so_rcv);
1603 done:
1604 	lwkt_reltoken(&so->so_rcv.ssb_token);
1605 	if (free_chain)
1606 		m_freem(free_chain);
1607 	return (error);
1608 }
1609 
1610 int
1611 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1612 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1613 {
1614 	struct mbuf *m, *n;
1615 	struct mbuf *free_chain = NULL;
1616 	int flags, len, error, offset;
1617 	struct protosw *pr = so->so_proto;
1618 	int moff;
1619 	int didoob;
1620 	size_t resid, orig_resid, restmp;
1621 
1622 	if (uio)
1623 		resid = uio->uio_resid;
1624 	else
1625 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1626 	orig_resid = resid;
1627 
1628 	if (psa)
1629 		*psa = NULL;
1630 	if (controlp)
1631 		*controlp = NULL;
1632 	if (flagsp)
1633 		flags = *flagsp &~ MSG_EOR;
1634 	else
1635 		flags = 0;
1636 	if (flags & MSG_OOB) {
1637 		m = m_get(M_WAITOK, MT_DATA);
1638 		if (m == NULL)
1639 			return (ENOBUFS);
1640 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1641 		if (error)
1642 			goto bad;
1643 		if (sio) {
1644 			do {
1645 				sbappend(sio, m);
1646 				KKASSERT(resid >= (size_t)m->m_len);
1647 				resid -= (size_t)m->m_len;
1648 			} while (resid > 0 && m);
1649 		} else {
1650 			do {
1651 				uio->uio_resid = resid;
1652 				error = uiomove(mtod(m, caddr_t),
1653 						(int)szmin(resid, m->m_len),
1654 						uio);
1655 				resid = uio->uio_resid;
1656 				m = m_free(m);
1657 			} while (uio->uio_resid && error == 0 && m);
1658 		}
1659 bad:
1660 		if (m)
1661 			m_freem(m);
1662 		return (error);
1663 	}
1664 
1665 	/*
1666 	 * The token interlocks against the protocol thread while
1667 	 * ssb_lock is a blocking lock against other userland entities.
1668 	 *
1669 	 * Lock a limited number of mbufs (not all, so sbcompress() still
1670 	 * works well).  The token is used as an interlock for sbwait() so
1671 	 * release it afterwords.
1672 	 */
1673 restart:
1674 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1675 	if (error)
1676 		goto done;
1677 
1678 	lwkt_gettoken(&so->so_rcv.ssb_token);
1679 	m = so->so_rcv.ssb_mb;
1680 
1681 	/*
1682 	 * If we have less data than requested, block awaiting more
1683 	 * (subject to any timeout) if:
1684 	 *   1. the current count is less than the low water mark, or
1685 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1686 	 *	receive operation at once if we block (resid <= hiwat).
1687 	 *   3. MSG_DONTWAIT is not set
1688 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1689 	 * we have to do the receive in sections, and thus risk returning
1690 	 * a short count if a timeout or signal occurs after we start.
1691 	 */
1692 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1693 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1694 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1695 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1696 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1697 		if (so->so_error) {
1698 			if (m)
1699 				goto dontblock;
1700 			lwkt_reltoken(&so->so_rcv.ssb_token);
1701 			error = so->so_error;
1702 			if ((flags & MSG_PEEK) == 0)
1703 				so->so_error = 0;
1704 			goto release;
1705 		}
1706 		if (so->so_state & SS_CANTRCVMORE) {
1707 			if (m)
1708 				goto dontblock;
1709 			lwkt_reltoken(&so->so_rcv.ssb_token);
1710 			goto release;
1711 		}
1712 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1713 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1714 			lwkt_reltoken(&so->so_rcv.ssb_token);
1715 			error = ENOTCONN;
1716 			goto release;
1717 		}
1718 		if (resid == 0) {
1719 			lwkt_reltoken(&so->so_rcv.ssb_token);
1720 			goto release;
1721 		}
1722 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1723 			lwkt_reltoken(&so->so_rcv.ssb_token);
1724 			error = EWOULDBLOCK;
1725 			goto release;
1726 		}
1727 		ssb_unlock(&so->so_rcv);
1728 		error = ssb_wait(&so->so_rcv);
1729 		lwkt_reltoken(&so->so_rcv.ssb_token);
1730 		if (error)
1731 			goto done;
1732 		goto restart;
1733 	}
1734 
1735 	/*
1736 	 * Token still held
1737 	 */
1738 dontblock:
1739 	n = m;
1740 	restmp = 0;
1741 	while (n && restmp < resid) {
1742 		n->m_flags |= M_SOLOCKED;
1743 		restmp += n->m_len;
1744 		if (n->m_next == NULL)
1745 			n = n->m_nextpkt;
1746 		else
1747 			n = n->m_next;
1748 	}
1749 
1750 	/*
1751 	 * Release token for loop
1752 	 */
1753 	lwkt_reltoken(&so->so_rcv.ssb_token);
1754 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1755 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1756 
1757 	/*
1758 	 * note: m should be == sb_mb here.  Cache the next record while
1759 	 * cleaning up.  Note that calling m_free*() will break out critical
1760 	 * section.
1761 	 */
1762 	KKASSERT(m == so->so_rcv.ssb_mb);
1763 
1764 	/*
1765 	 * Copy to the UIO or mbuf return chain (*mp).
1766 	 *
1767 	 * NOTE: Token is not held for loop
1768 	 */
1769 	moff = 0;
1770 	offset = 0;
1771 	didoob = 0;
1772 
1773 	while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1774 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1775 		    ("receive 3"));
1776 
1777 		soclrstate(so, SS_RCVATMARK);
1778 		len = (resid > INT_MAX) ? INT_MAX : resid;
1779 		if (so->so_oobmark && len > so->so_oobmark - offset)
1780 			len = so->so_oobmark - offset;
1781 		if (len > m->m_len - moff)
1782 			len = m->m_len - moff;
1783 
1784 		/*
1785 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1786 		 * The SIO is dealt with when we eat the mbuf, but deal
1787 		 * with the resid here either way.
1788 		 */
1789 		if (uio) {
1790 			uio->uio_resid = resid;
1791 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1792 			resid = uio->uio_resid;
1793 			if (error)
1794 				goto release;
1795 		} else {
1796 			resid -= (size_t)len;
1797 		}
1798 
1799 		/*
1800 		 * Eat the entire mbuf or just a piece of it
1801 		 */
1802 		offset += len;
1803 		if (len == m->m_len - moff) {
1804 			m = m->m_next;
1805 			moff = 0;
1806 		} else {
1807 			moff += len;
1808 		}
1809 
1810 		/*
1811 		 * Check oobmark
1812 		 */
1813 		if (so->so_oobmark && offset == so->so_oobmark) {
1814 			didoob = 1;
1815 			break;
1816 		}
1817 	}
1818 
1819 	/*
1820 	 * Synchronize sockbuf with data we read.
1821 	 *
1822 	 * NOTE: (m) is junk on entry (it could be left over from the
1823 	 *	 previous loop).
1824 	 */
1825 	if ((flags & MSG_PEEK) == 0) {
1826 		lwkt_gettoken(&so->so_rcv.ssb_token);
1827 		m = so->so_rcv.ssb_mb;
1828 		while (m && offset >= m->m_len) {
1829 			if (so->so_oobmark) {
1830 				so->so_oobmark -= m->m_len;
1831 				if (so->so_oobmark == 0) {
1832 					sosetstate(so, SS_RCVATMARK);
1833 					didoob = 1;
1834 				}
1835 			}
1836 			offset -= m->m_len;
1837 			if (sio) {
1838 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1839 				sbappend(sio, m);
1840 				m = n;
1841 			} else {
1842 				m = sbunlinkmbuf(&so->so_rcv.sb,
1843 						 m, &free_chain);
1844 			}
1845 		}
1846 		if (offset) {
1847 			KKASSERT(m);
1848 			if (sio) {
1849 				n = m_copym(m, 0, offset, M_WAITOK);
1850 				if (n)
1851 					sbappend(sio, n);
1852 			}
1853 			m->m_data += offset;
1854 			m->m_len -= offset;
1855 			so->so_rcv.ssb_cc -= offset;
1856 			if (so->so_oobmark) {
1857 				so->so_oobmark -= offset;
1858 				if (so->so_oobmark == 0) {
1859 					sosetstate(so, SS_RCVATMARK);
1860 					didoob = 1;
1861 				}
1862 			}
1863 			offset = 0;
1864 		}
1865 		lwkt_reltoken(&so->so_rcv.ssb_token);
1866 	}
1867 
1868 	/*
1869 	 * If the MSG_WAITALL flag is set (for non-atomic socket),
1870 	 * we must not quit until resid == 0 or an error termination.
1871 	 *
1872 	 * If a signal/timeout occurs, return with a short count but without
1873 	 * error.
1874 	 *
1875 	 * Keep signalsockbuf locked against other readers.
1876 	 *
1877 	 * XXX if MSG_PEEK we currently do quit.
1878 	 */
1879 	if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1880 	    didoob == 0 && resid > 0 &&
1881 	    !sosendallatonce(so)) {
1882 		lwkt_gettoken(&so->so_rcv.ssb_token);
1883 		error = 0;
1884 		while ((m = so->so_rcv.ssb_mb) == NULL) {
1885 			if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1886 				error = so->so_error;
1887 				break;
1888 			}
1889 			/*
1890 			 * The window might have closed to zero, make
1891 			 * sure we send an ack now that we've drained
1892 			 * the buffer or we might end up blocking until
1893 			 * the idle takes over (5 seconds).
1894 			 */
1895 			if (so->so_pcb)
1896 				so_pru_rcvd_async(so);
1897 			if (so->so_rcv.ssb_mb == NULL)
1898 				error = ssb_wait(&so->so_rcv);
1899 			if (error) {
1900 				lwkt_reltoken(&so->so_rcv.ssb_token);
1901 				ssb_unlock(&so->so_rcv);
1902 				error = 0;
1903 				goto done;
1904 			}
1905 		}
1906 		if (m && error == 0)
1907 			goto dontblock;
1908 		lwkt_reltoken(&so->so_rcv.ssb_token);
1909 	}
1910 
1911 	/*
1912 	 * Token not held here.
1913 	 *
1914 	 * Cleanup.  If an atomic read was requested drop any unread data XXX
1915 	 */
1916 	if ((flags & MSG_PEEK) == 0) {
1917 		if (so->so_pcb)
1918 			so_pru_rcvd_async(so);
1919 	}
1920 
1921 	if (orig_resid == resid && orig_resid &&
1922 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1923 		ssb_unlock(&so->so_rcv);
1924 		goto restart;
1925 	}
1926 
1927 	if (flagsp)
1928 		*flagsp |= flags;
1929 release:
1930 	ssb_unlock(&so->so_rcv);
1931 done:
1932 	if (free_chain)
1933 		m_freem(free_chain);
1934 	return (error);
1935 }
1936 
1937 /*
1938  * Shut a socket down.  Note that we do not get a frontend lock as we
1939  * want to be able to shut the socket down even if another thread is
1940  * blocked in a read(), thus waking it up.
1941  */
1942 int
1943 soshutdown(struct socket *so, int how)
1944 {
1945 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1946 		return (EINVAL);
1947 
1948 	if (how != SHUT_WR) {
1949 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1950 		sorflush(so);
1951 		/*ssb_unlock(&so->so_rcv);*/
1952 	}
1953 	if (how != SHUT_RD)
1954 		return (so_pru_shutdown(so));
1955 	return (0);
1956 }
1957 
1958 void
1959 sorflush(struct socket *so)
1960 {
1961 	struct signalsockbuf *ssb = &so->so_rcv;
1962 	struct protosw *pr = so->so_proto;
1963 	struct signalsockbuf asb;
1964 
1965 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1966 
1967 	lwkt_gettoken(&ssb->ssb_token);
1968 	socantrcvmore(so);
1969 	asb = *ssb;
1970 
1971 	/*
1972 	 * Can't just blow up the ssb structure here
1973 	 */
1974 	bzero(&ssb->sb, sizeof(ssb->sb));
1975 	ssb->ssb_timeo = 0;
1976 	ssb->ssb_lowat = 0;
1977 	ssb->ssb_hiwat = 0;
1978 	ssb->ssb_mbmax = 0;
1979 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1980 
1981 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1982 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1983 	ssb_release(&asb, so);
1984 
1985 	lwkt_reltoken(&ssb->ssb_token);
1986 }
1987 
1988 #ifdef INET
1989 static int
1990 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1991 {
1992 	struct accept_filter_arg	*afap = NULL;
1993 	struct accept_filter	*afp;
1994 	struct so_accf	*af = so->so_accf;
1995 	int	error = 0;
1996 
1997 	/* do not set/remove accept filters on non listen sockets */
1998 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1999 		error = EINVAL;
2000 		goto out;
2001 	}
2002 
2003 	/* removing the filter */
2004 	if (sopt == NULL) {
2005 		if (af != NULL) {
2006 			if (af->so_accept_filter != NULL &&
2007 				af->so_accept_filter->accf_destroy != NULL) {
2008 				af->so_accept_filter->accf_destroy(so);
2009 			}
2010 			if (af->so_accept_filter_str != NULL) {
2011 				kfree(af->so_accept_filter_str, M_ACCF);
2012 			}
2013 			kfree(af, M_ACCF);
2014 			so->so_accf = NULL;
2015 		}
2016 		so->so_options &= ~SO_ACCEPTFILTER;
2017 		return (0);
2018 	}
2019 	/* adding a filter */
2020 	/* must remove previous filter first */
2021 	if (af != NULL) {
2022 		error = EINVAL;
2023 		goto out;
2024 	}
2025 	/* don't put large objects on the kernel stack */
2026 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2027 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2028 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
2029 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2030 	if (error)
2031 		goto out;
2032 	afp = accept_filt_get(afap->af_name);
2033 	if (afp == NULL) {
2034 		error = ENOENT;
2035 		goto out;
2036 	}
2037 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2038 	if (afp->accf_create != NULL) {
2039 		if (afap->af_name[0] != '\0') {
2040 			int len = strlen(afap->af_name) + 1;
2041 
2042 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
2043 							   M_WAITOK);
2044 			strcpy(af->so_accept_filter_str, afap->af_name);
2045 		}
2046 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2047 		if (af->so_accept_filter_arg == NULL) {
2048 			kfree(af->so_accept_filter_str, M_ACCF);
2049 			kfree(af, M_ACCF);
2050 			so->so_accf = NULL;
2051 			error = EINVAL;
2052 			goto out;
2053 		}
2054 	}
2055 	af->so_accept_filter = afp;
2056 	so->so_accf = af;
2057 	so->so_options |= SO_ACCEPTFILTER;
2058 out:
2059 	if (afap != NULL)
2060 		kfree(afap, M_TEMP);
2061 	return (error);
2062 }
2063 #endif /* INET */
2064 
2065 /*
2066  * Perhaps this routine, and sooptcopyout(), below, ought to come in
2067  * an additional variant to handle the case where the option value needs
2068  * to be some kind of integer, but not a specific size.
2069  * In addition to their use here, these functions are also called by the
2070  * protocol-level pr_ctloutput() routines.
2071  */
2072 int
2073 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2074 {
2075 	return soopt_to_kbuf(sopt, buf, len, minlen);
2076 }
2077 
2078 int
2079 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2080 {
2081 	size_t	valsize;
2082 
2083 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2084 	KKASSERT(kva_p(buf));
2085 
2086 	/*
2087 	 * If the user gives us more than we wanted, we ignore it,
2088 	 * but if we don't get the minimum length the caller
2089 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
2090 	 * is set to however much we actually retrieved.
2091 	 */
2092 	if ((valsize = sopt->sopt_valsize) < minlen)
2093 		return EINVAL;
2094 	if (valsize > len)
2095 		sopt->sopt_valsize = valsize = len;
2096 
2097 	bcopy(sopt->sopt_val, buf, valsize);
2098 	return 0;
2099 }
2100 
2101 
2102 int
2103 sosetopt(struct socket *so, struct sockopt *sopt)
2104 {
2105 	int	error, optval;
2106 	struct	linger l;
2107 	struct	timeval tv;
2108 	u_long  val;
2109 	struct signalsockbuf *sotmp;
2110 
2111 	error = 0;
2112 	sopt->sopt_dir = SOPT_SET;
2113 	if (sopt->sopt_level != SOL_SOCKET) {
2114 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2115 			return (so_pr_ctloutput(so, sopt));
2116 		}
2117 		error = ENOPROTOOPT;
2118 	} else {
2119 		switch (sopt->sopt_name) {
2120 #ifdef INET
2121 		case SO_ACCEPTFILTER:
2122 			error = do_setopt_accept_filter(so, sopt);
2123 			if (error)
2124 				goto bad;
2125 			break;
2126 #endif /* INET */
2127 		case SO_LINGER:
2128 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2129 			if (error)
2130 				goto bad;
2131 
2132 			so->so_linger = l.l_linger;
2133 			if (l.l_onoff)
2134 				so->so_options |= SO_LINGER;
2135 			else
2136 				so->so_options &= ~SO_LINGER;
2137 			break;
2138 
2139 		case SO_DEBUG:
2140 		case SO_KEEPALIVE:
2141 		case SO_DONTROUTE:
2142 		case SO_USELOOPBACK:
2143 		case SO_BROADCAST:
2144 		case SO_REUSEADDR:
2145 		case SO_REUSEPORT:
2146 		case SO_OOBINLINE:
2147 		case SO_TIMESTAMP:
2148 		case SO_NOSIGPIPE:
2149 			error = sooptcopyin(sopt, &optval, sizeof optval,
2150 					    sizeof optval);
2151 			if (error)
2152 				goto bad;
2153 			if (optval)
2154 				so->so_options |= sopt->sopt_name;
2155 			else
2156 				so->so_options &= ~sopt->sopt_name;
2157 			break;
2158 
2159 		case SO_SNDBUF:
2160 		case SO_RCVBUF:
2161 		case SO_SNDLOWAT:
2162 		case SO_RCVLOWAT:
2163 			error = sooptcopyin(sopt, &optval, sizeof optval,
2164 					    sizeof optval);
2165 			if (error)
2166 				goto bad;
2167 
2168 			/*
2169 			 * Values < 1 make no sense for any of these
2170 			 * options, so disallow them.
2171 			 */
2172 			if (optval < 1) {
2173 				error = EINVAL;
2174 				goto bad;
2175 			}
2176 
2177 			switch (sopt->sopt_name) {
2178 			case SO_SNDBUF:
2179 			case SO_RCVBUF:
2180 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2181 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2182 				    so,
2183 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2184 					error = ENOBUFS;
2185 					goto bad;
2186 				}
2187 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2188 						&so->so_snd : &so->so_rcv;
2189 				atomic_clear_int(&sotmp->ssb_flags,
2190 						 SSB_AUTOSIZE);
2191 				break;
2192 
2193 			/*
2194 			 * Make sure the low-water is never greater than
2195 			 * the high-water.
2196 			 */
2197 			case SO_SNDLOWAT:
2198 				so->so_snd.ssb_lowat =
2199 				    (optval > so->so_snd.ssb_hiwat) ?
2200 				    so->so_snd.ssb_hiwat : optval;
2201 				atomic_clear_int(&so->so_snd.ssb_flags,
2202 						 SSB_AUTOLOWAT);
2203 				break;
2204 			case SO_RCVLOWAT:
2205 				so->so_rcv.ssb_lowat =
2206 				    (optval > so->so_rcv.ssb_hiwat) ?
2207 				    so->so_rcv.ssb_hiwat : optval;
2208 				atomic_clear_int(&so->so_rcv.ssb_flags,
2209 						 SSB_AUTOLOWAT);
2210 				break;
2211 			}
2212 			break;
2213 
2214 		case SO_SNDTIMEO:
2215 		case SO_RCVTIMEO:
2216 			error = sooptcopyin(sopt, &tv, sizeof tv,
2217 					    sizeof tv);
2218 			if (error)
2219 				goto bad;
2220 
2221 			/* assert(hz > 0); */
2222 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2223 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2224 				error = EDOM;
2225 				goto bad;
2226 			}
2227 			/* assert(tick > 0); */
2228 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2229 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2230 			if (val > INT_MAX) {
2231 				error = EDOM;
2232 				goto bad;
2233 			}
2234 			if (val == 0 && tv.tv_usec != 0)
2235 				val = 1;
2236 
2237 			switch (sopt->sopt_name) {
2238 			case SO_SNDTIMEO:
2239 				so->so_snd.ssb_timeo = val;
2240 				break;
2241 			case SO_RCVTIMEO:
2242 				so->so_rcv.ssb_timeo = val;
2243 				break;
2244 			}
2245 			break;
2246 		default:
2247 			error = ENOPROTOOPT;
2248 			break;
2249 		}
2250 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2251 			(void) so_pr_ctloutput(so, sopt);
2252 		}
2253 	}
2254 bad:
2255 	return (error);
2256 }
2257 
2258 /* Helper routine for getsockopt */
2259 int
2260 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2261 {
2262 	soopt_from_kbuf(sopt, buf, len);
2263 	return 0;
2264 }
2265 
2266 void
2267 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2268 {
2269 	size_t	valsize;
2270 
2271 	if (len == 0) {
2272 		sopt->sopt_valsize = 0;
2273 		return;
2274 	}
2275 
2276 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2277 	KKASSERT(kva_p(buf));
2278 
2279 	/*
2280 	 * Documented get behavior is that we always return a value,
2281 	 * possibly truncated to fit in the user's buffer.
2282 	 * Traditional behavior is that we always tell the user
2283 	 * precisely how much we copied, rather than something useful
2284 	 * like the total amount we had available for her.
2285 	 * Note that this interface is not idempotent; the entire answer must
2286 	 * generated ahead of time.
2287 	 */
2288 	valsize = szmin(len, sopt->sopt_valsize);
2289 	sopt->sopt_valsize = valsize;
2290 	if (sopt->sopt_val != 0) {
2291 		bcopy(buf, sopt->sopt_val, valsize);
2292 	}
2293 }
2294 
2295 int
2296 sogetopt(struct socket *so, struct sockopt *sopt)
2297 {
2298 	int	error, optval;
2299 	long	optval_l;
2300 	struct	linger l;
2301 	struct	timeval tv;
2302 #ifdef INET
2303 	struct accept_filter_arg *afap;
2304 #endif
2305 
2306 	error = 0;
2307 	sopt->sopt_dir = SOPT_GET;
2308 	if (sopt->sopt_level != SOL_SOCKET) {
2309 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2310 			return (so_pr_ctloutput(so, sopt));
2311 		} else
2312 			return (ENOPROTOOPT);
2313 	} else {
2314 		switch (sopt->sopt_name) {
2315 #ifdef INET
2316 		case SO_ACCEPTFILTER:
2317 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2318 				return (EINVAL);
2319 			afap = kmalloc(sizeof(*afap), M_TEMP,
2320 				       M_WAITOK | M_ZERO);
2321 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2322 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2323 				if (so->so_accf->so_accept_filter_str != NULL)
2324 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2325 			}
2326 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2327 			kfree(afap, M_TEMP);
2328 			break;
2329 #endif /* INET */
2330 
2331 		case SO_LINGER:
2332 			l.l_onoff = so->so_options & SO_LINGER;
2333 			l.l_linger = so->so_linger;
2334 			error = sooptcopyout(sopt, &l, sizeof l);
2335 			break;
2336 
2337 		case SO_USELOOPBACK:
2338 		case SO_DONTROUTE:
2339 		case SO_DEBUG:
2340 		case SO_KEEPALIVE:
2341 		case SO_REUSEADDR:
2342 		case SO_REUSEPORT:
2343 		case SO_BROADCAST:
2344 		case SO_OOBINLINE:
2345 		case SO_TIMESTAMP:
2346 		case SO_NOSIGPIPE:
2347 			optval = so->so_options & sopt->sopt_name;
2348 integer:
2349 			error = sooptcopyout(sopt, &optval, sizeof optval);
2350 			break;
2351 
2352 		case SO_TYPE:
2353 			optval = so->so_type;
2354 			goto integer;
2355 
2356 		case SO_ERROR:
2357 			optval = so->so_error;
2358 			so->so_error = 0;
2359 			goto integer;
2360 
2361 		case SO_SNDBUF:
2362 			optval = so->so_snd.ssb_hiwat;
2363 			goto integer;
2364 
2365 		case SO_RCVBUF:
2366 			optval = so->so_rcv.ssb_hiwat;
2367 			goto integer;
2368 
2369 		case SO_SNDLOWAT:
2370 			optval = so->so_snd.ssb_lowat;
2371 			goto integer;
2372 
2373 		case SO_RCVLOWAT:
2374 			optval = so->so_rcv.ssb_lowat;
2375 			goto integer;
2376 
2377 		case SO_SNDTIMEO:
2378 		case SO_RCVTIMEO:
2379 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2380 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2381 
2382 			tv.tv_sec = optval / hz;
2383 			tv.tv_usec = (optval % hz) * ustick;
2384 			error = sooptcopyout(sopt, &tv, sizeof tv);
2385 			break;
2386 
2387 		case SO_SNDSPACE:
2388 			optval_l = ssb_space(&so->so_snd);
2389 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2390 			break;
2391 
2392 		case SO_CPUHINT:
2393 			optval = -1; /* no hint */
2394 			goto integer;
2395 
2396 		default:
2397 			error = ENOPROTOOPT;
2398 			break;
2399 		}
2400 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2401 			so_pr_ctloutput(so, sopt);
2402 		return (error);
2403 	}
2404 }
2405 
2406 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2407 int
2408 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2409 {
2410 	struct mbuf *m, *m_prev;
2411 	int sopt_size = sopt->sopt_valsize, msize;
2412 
2413 	m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2414 		   0, &msize);
2415 	if (m == NULL)
2416 		return (ENOBUFS);
2417 	m->m_len = min(msize, sopt_size);
2418 	sopt_size -= m->m_len;
2419 	*mp = m;
2420 	m_prev = m;
2421 
2422 	while (sopt_size > 0) {
2423 		m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2424 			   MT_DATA, 0, &msize);
2425 		if (m == NULL) {
2426 			m_freem(*mp);
2427 			return (ENOBUFS);
2428 		}
2429 		m->m_len = min(msize, sopt_size);
2430 		sopt_size -= m->m_len;
2431 		m_prev->m_next = m;
2432 		m_prev = m;
2433 	}
2434 	return (0);
2435 }
2436 
2437 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2438 int
2439 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2440 {
2441 	soopt_to_mbuf(sopt, m);
2442 	return 0;
2443 }
2444 
2445 void
2446 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2447 {
2448 	size_t valsize;
2449 	void *val;
2450 
2451 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2452 	KKASSERT(kva_p(m));
2453 	if (sopt->sopt_val == NULL)
2454 		return;
2455 	val = sopt->sopt_val;
2456 	valsize = sopt->sopt_valsize;
2457 	while (m != NULL && valsize >= m->m_len) {
2458 		bcopy(val, mtod(m, char *), m->m_len);
2459 		valsize -= m->m_len;
2460 		val = (caddr_t)val + m->m_len;
2461 		m = m->m_next;
2462 	}
2463 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2464 		panic("ip6_sooptmcopyin");
2465 }
2466 
2467 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2468 int
2469 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2470 {
2471 	return soopt_from_mbuf(sopt, m);
2472 }
2473 
2474 int
2475 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2476 {
2477 	struct mbuf *m0 = m;
2478 	size_t valsize = 0;
2479 	size_t maxsize;
2480 	void *val;
2481 
2482 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2483 	KKASSERT(kva_p(m));
2484 	if (sopt->sopt_val == NULL)
2485 		return 0;
2486 	val = sopt->sopt_val;
2487 	maxsize = sopt->sopt_valsize;
2488 	while (m != NULL && maxsize >= m->m_len) {
2489 		bcopy(mtod(m, char *), val, m->m_len);
2490 	       maxsize -= m->m_len;
2491 	       val = (caddr_t)val + m->m_len;
2492 	       valsize += m->m_len;
2493 	       m = m->m_next;
2494 	}
2495 	if (m != NULL) {
2496 		/* enough soopt buffer should be given from user-land */
2497 		m_freem(m0);
2498 		return (EINVAL);
2499 	}
2500 	sopt->sopt_valsize = valsize;
2501 	return 0;
2502 }
2503 
2504 void
2505 sohasoutofband(struct socket *so)
2506 {
2507 	if (so->so_sigio != NULL)
2508 		pgsigio(so->so_sigio, SIGURG, 0);
2509 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2510 }
2511 
2512 int
2513 sokqfilter(struct file *fp, struct knote *kn)
2514 {
2515 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2516 	struct signalsockbuf *ssb;
2517 
2518 	switch (kn->kn_filter) {
2519 	case EVFILT_READ:
2520 		if (so->so_options & SO_ACCEPTCONN)
2521 			kn->kn_fop = &solisten_filtops;
2522 		else
2523 			kn->kn_fop = &soread_filtops;
2524 		ssb = &so->so_rcv;
2525 		break;
2526 	case EVFILT_WRITE:
2527 		kn->kn_fop = &sowrite_filtops;
2528 		ssb = &so->so_snd;
2529 		break;
2530 	case EVFILT_EXCEPT:
2531 		kn->kn_fop = &soexcept_filtops;
2532 		ssb = &so->so_rcv;
2533 		break;
2534 	default:
2535 		return (EOPNOTSUPP);
2536 	}
2537 
2538 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2539 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2540 	return (0);
2541 }
2542 
2543 static void
2544 filt_sordetach(struct knote *kn)
2545 {
2546 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2547 
2548 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2549 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2550 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2551 }
2552 
2553 /*ARGSUSED*/
2554 static int
2555 filt_soread(struct knote *kn, long hint)
2556 {
2557 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2558 
2559 	if (kn->kn_sfflags & NOTE_OOB) {
2560 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2561 			kn->kn_fflags |= NOTE_OOB;
2562 			return (1);
2563 		}
2564 		return (0);
2565 	}
2566 	kn->kn_data = so->so_rcv.ssb_cc;
2567 
2568 	if (so->so_state & SS_CANTRCVMORE) {
2569 		/*
2570 		 * Only set NODATA if all data has been exhausted.
2571 		 */
2572 		if (kn->kn_data == 0)
2573 			kn->kn_flags |= EV_NODATA;
2574 		kn->kn_flags |= EV_EOF;
2575 		kn->kn_fflags = so->so_error;
2576 		return (1);
2577 	}
2578 	if (so->so_error)	/* temporary udp error */
2579 		return (1);
2580 	if (kn->kn_sfflags & NOTE_LOWAT)
2581 		return (kn->kn_data >= kn->kn_sdata);
2582 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2583 		!TAILQ_EMPTY(&so->so_comp));
2584 }
2585 
2586 static void
2587 filt_sowdetach(struct knote *kn)
2588 {
2589 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2590 
2591 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2592 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2593 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2594 }
2595 
2596 /*ARGSUSED*/
2597 static int
2598 filt_sowrite(struct knote *kn, long hint)
2599 {
2600 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2601 
2602 	kn->kn_data = ssb_space(&so->so_snd);
2603 	if (so->so_state & SS_CANTSENDMORE) {
2604 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2605 		kn->kn_fflags = so->so_error;
2606 		return (1);
2607 	}
2608 	if (so->so_error)	/* temporary udp error */
2609 		return (1);
2610 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2611 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2612 		return (0);
2613 	if (kn->kn_sfflags & NOTE_LOWAT)
2614 		return (kn->kn_data >= kn->kn_sdata);
2615 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2616 }
2617 
2618 /*ARGSUSED*/
2619 static int
2620 filt_solisten(struct knote *kn, long hint)
2621 {
2622 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2623 
2624 	kn->kn_data = so->so_qlen;
2625 	return (! TAILQ_EMPTY(&so->so_comp));
2626 }
2627