xref: /dragonfly/sys/kern/uipc_socket.c (revision cae2835b)
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 static int use_socreate_fast = 1;
152 SYSCTL_INT(_kern_ipc, OID_AUTO, socreate_fast, CTLFLAG_RW,
153     &use_socreate_fast, 0, "Fast socket creation");
154 
155 /*
156  * Socket operation routines.
157  * These routines are called by the routines in
158  * sys_socket.c or from a system process, and
159  * implement the semantics of socket operations by
160  * switching out to the protocol specific routines.
161  */
162 
163 /*
164  * Get a socket structure, and initialize it.
165  * Note that it would probably be better to allocate socket
166  * and PCB at the same time, but I'm not convinced that all
167  * the protocols can be easily modified to do this.
168  */
169 struct socket *
170 soalloc(int waitok, struct protosw *pr)
171 {
172 	struct socket *so;
173 	unsigned waitmask;
174 
175 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
176 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
177 	if (so) {
178 		/* XXX race condition for reentrant kernel */
179 		so->so_proto = pr;
180 		TAILQ_INIT(&so->so_aiojobq);
181 		TAILQ_INIT(&so->so_rcv.ssb_mlist);
182 		TAILQ_INIT(&so->so_snd.ssb_mlist);
183 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
184 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
185 		spin_init(&so->so_rcvd_spin, "soalloc");
186 		netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
187 		    MSGF_DROPABLE | MSGF_PRIORITY,
188 		    so->so_proto->pr_usrreqs->pru_rcvd);
189 		so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
190 		so->so_state = SS_NOFDREF;
191 		so->so_refs = 1;
192 	}
193 	return so;
194 }
195 
196 int
197 socreate(int dom, struct socket **aso, int type,
198 	int proto, struct thread *td)
199 {
200 	struct proc *p = td->td_proc;
201 	struct protosw *prp;
202 	struct socket *so;
203 	struct pru_attach_info ai;
204 	int error;
205 
206 	if (proto)
207 		prp = pffindproto(dom, proto, type);
208 	else
209 		prp = pffindtype(dom, type);
210 
211 	if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
212 		return (EPROTONOSUPPORT);
213 
214 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
215 	    prp->pr_domain->dom_family != PF_LOCAL &&
216 	    prp->pr_domain->dom_family != PF_INET &&
217 	    prp->pr_domain->dom_family != PF_INET6 &&
218 	    prp->pr_domain->dom_family != PF_ROUTE) {
219 		return (EPROTONOSUPPORT);
220 	}
221 
222 	if (prp->pr_type != type)
223 		return (EPROTOTYPE);
224 	so = soalloc(p != NULL, prp);
225 	if (so == NULL)
226 		return (ENOBUFS);
227 
228 	/*
229 	 * Callers of socreate() presumably will connect up a descriptor
230 	 * and call soclose() if they cannot.  This represents our so_refs
231 	 * (which should be 1) from soalloc().
232 	 */
233 	soclrstate(so, SS_NOFDREF);
234 
235 	/*
236 	 * Set a default port for protocol processing.  No action will occur
237 	 * on the socket on this port until an inpcb is attached to it and
238 	 * is able to match incoming packets, or until the socket becomes
239 	 * available to userland.
240 	 *
241 	 * We normally default the socket to the protocol thread on cpu 0,
242 	 * if protocol does not provide its own method to initialize the
243 	 * default port.
244 	 *
245 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
246 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
247 	 */
248 	if (prp->pr_flags & PR_SYNC_PORT)
249 		so->so_port = &netisr_sync_port;
250 	else if (prp->pr_initport != NULL)
251 		so->so_port = prp->pr_initport();
252 	else
253 		so->so_port = netisr_cpuport(0);
254 
255 	TAILQ_INIT(&so->so_incomp);
256 	TAILQ_INIT(&so->so_comp);
257 	so->so_type = type;
258 	so->so_cred = crhold(p->p_ucred);
259 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
260 	ai.p_ucred = p->p_ucred;
261 	ai.fd_rdir = p->p_fd->fd_rdir;
262 
263 	/*
264 	 * Auto-sizing of socket buffers is managed by the protocols and
265 	 * the appropriate flags must be set in the pru_attach function.
266 	 */
267 	if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
268 		error = so_pru_attach_fast(so, proto, &ai);
269 	else
270 		error = so_pru_attach(so, proto, &ai);
271 	if (error) {
272 		sosetstate(so, SS_NOFDREF);
273 		sofree(so);	/* from soalloc */
274 		return error;
275 	}
276 
277 	/*
278 	 * NOTE: Returns referenced socket.
279 	 */
280 	*aso = so;
281 	return (0);
282 }
283 
284 int
285 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
286 {
287 	int error;
288 
289 	error = so_pru_bind(so, nam, td);
290 	return (error);
291 }
292 
293 static void
294 sodealloc(struct socket *so)
295 {
296 	KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
297 
298 #ifdef INVARIANTS
299 	if (so->so_options & SO_ACCEPTCONN) {
300 		KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
301 		KASSERT(TAILQ_EMPTY(&so->so_incomp),
302 		    ("so_incomp is not empty"));
303 	}
304 #endif
305 
306 	if (so->so_rcv.ssb_hiwat)
307 		(void)chgsbsize(so->so_cred->cr_uidinfo,
308 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
309 	if (so->so_snd.ssb_hiwat)
310 		(void)chgsbsize(so->so_cred->cr_uidinfo,
311 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
312 #ifdef INET
313 	/* remove accept filter if present */
314 	if (so->so_accf != NULL)
315 		do_setopt_accept_filter(so, NULL);
316 #endif /* INET */
317 	crfree(so->so_cred);
318 	if (so->so_faddr != NULL)
319 		kfree(so->so_faddr, M_SONAME);
320 	kfree(so, M_SOCKET);
321 }
322 
323 int
324 solisten(struct socket *so, int backlog, struct thread *td)
325 {
326 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
327 		return (EINVAL);
328 
329 	lwkt_gettoken(&so->so_rcv.ssb_token);
330 	if (TAILQ_EMPTY(&so->so_comp))
331 		so->so_options |= SO_ACCEPTCONN;
332 	lwkt_reltoken(&so->so_rcv.ssb_token);
333 	if (backlog < 0 || backlog > somaxconn)
334 		backlog = somaxconn;
335 	so->so_qlimit = backlog;
336 	return so_pru_listen(so, td);
337 }
338 
339 static void
340 soqflush(struct socket *so)
341 {
342 	lwkt_getpooltoken(so);
343 	if (so->so_options & SO_ACCEPTCONN) {
344 		struct socket *sp;
345 
346 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
347 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
348 			    SS_INCOMP);
349 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
350 			so->so_incqlen--;
351 			soclrstate(sp, SS_INCOMP);
352 			soabort_async(sp, TRUE);
353 		}
354 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
355 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
356 			    SS_COMP);
357 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
358 			so->so_qlen--;
359 			soclrstate(sp, SS_COMP);
360 			soabort_async(sp, TRUE);
361 		}
362 	}
363 	lwkt_relpooltoken(so);
364 }
365 
366 /*
367  * Destroy a disconnected socket.  This routine is a NOP if entities
368  * still have a reference on the socket:
369  *
370  *	so_pcb -	The protocol stack still has a reference
371  *	SS_NOFDREF -	There is no longer a file pointer reference
372  */
373 void
374 sofree(struct socket *so)
375 {
376 	struct socket *head;
377 
378 	/*
379 	 * This is a bit hackish at the moment.  We need to interlock
380 	 * any accept queue we are on before we potentially lose the
381 	 * last reference to avoid races against a re-reference from
382 	 * someone operating on the queue.
383 	 */
384 	while ((head = so->so_head) != NULL) {
385 		lwkt_getpooltoken(head);
386 		if (so->so_head == head)
387 			break;
388 		lwkt_relpooltoken(head);
389 	}
390 
391 	/*
392 	 * Arbitrage the last free.
393 	 */
394 	KKASSERT(so->so_refs > 0);
395 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
396 		if (head)
397 			lwkt_relpooltoken(head);
398 		return;
399 	}
400 
401 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
402 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
403 
404 	if (head != NULL) {
405 		/*
406 		 * We're done, remove ourselves from the accept queue we are
407 		 * on, if we are on one.
408 		 */
409 		if (so->so_state & SS_INCOMP) {
410 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
411 			    SS_INCOMP);
412 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
413 			head->so_incqlen--;
414 		} else if (so->so_state & SS_COMP) {
415 			/*
416 			 * We must not decommission a socket that's
417 			 * on the accept(2) queue.  If we do, then
418 			 * accept(2) may hang after select(2) indicated
419 			 * that the listening socket was ready.
420 			 */
421 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
422 			    SS_COMP);
423 			lwkt_relpooltoken(head);
424 			return;
425 		} else {
426 			panic("sofree: not queued");
427 		}
428 		soclrstate(so, SS_INCOMP);
429 		so->so_head = NULL;
430 		lwkt_relpooltoken(head);
431 	} else {
432 		/* Flush accept queues, if we are accepting. */
433 		soqflush(so);
434 	}
435 	ssb_release(&so->so_snd, so);
436 	sorflush(so);
437 	sodealloc(so);
438 }
439 
440 /*
441  * Close a socket on last file table reference removal.
442  * Initiate disconnect if connected.
443  * Free socket when disconnect complete.
444  */
445 int
446 soclose(struct socket *so, int fflag)
447 {
448 	int error;
449 
450 	funsetown(&so->so_sigio);
451 	sosetstate(so, SS_ISCLOSING);
452 	if (!use_soclose_fast ||
453 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
454 	    ((so->so_state & SS_ISCONNECTED) &&
455 	     (so->so_options & SO_LINGER))) {
456 		error = soclose_sync(so, fflag);
457 	} else {
458 		soclose_fast(so);
459 		error = 0;
460 	}
461 	return error;
462 }
463 
464 void
465 sodiscard(struct socket *so)
466 {
467 	if (so->so_state & SS_NOFDREF)
468 		panic("soclose: NOFDREF");
469 	sosetstate(so, SS_NOFDREF);	/* take ref */
470 }
471 
472 /*
473  * Append the completed queue of head to head_inh (inherting listen socket).
474  */
475 void
476 soinherit(struct socket *head, struct socket *head_inh)
477 {
478 	boolean_t do_wakeup = FALSE;
479 
480 	KASSERT(head->so_options & SO_ACCEPTCONN,
481 	    ("head does not accept connection"));
482 	KASSERT(head_inh->so_options & SO_ACCEPTCONN,
483 	    ("head_inh does not accept connection"));
484 
485 	lwkt_getpooltoken(head);
486 	lwkt_getpooltoken(head_inh);
487 
488 	if (head->so_qlen > 0)
489 		do_wakeup = TRUE;
490 
491 	while (!TAILQ_EMPTY(&head->so_comp)) {
492 		struct ucred *old_cr;
493 		struct socket *sp;
494 
495 		sp = TAILQ_FIRST(&head->so_comp);
496 		KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
497 
498 		/*
499 		 * Remove this socket from the current listen socket
500 		 * completed queue.
501 		 */
502 		TAILQ_REMOVE(&head->so_comp, sp, so_list);
503 		head->so_qlen--;
504 
505 		/* Save the old ucred for later free. */
506 		old_cr = sp->so_cred;
507 
508 		/*
509 		 * Install this socket to the inheriting listen socket
510 		 * completed queue.
511 		 */
512 		sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
513 		sp->so_head = head_inh;
514 
515 		TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
516 		head_inh->so_qlen++;
517 
518 		/*
519 		 * NOTE:
520 		 * crfree() may block and release the tokens temporarily.
521 		 * However, we are fine here, since the transition is done.
522 		 */
523 		crfree(old_cr);
524 	}
525 
526 	lwkt_relpooltoken(head_inh);
527 	lwkt_relpooltoken(head);
528 
529 	if (do_wakeup) {
530 		/*
531 		 * "New" connections have arrived
532 		 */
533 		sorwakeup(head_inh);
534 		wakeup(&head_inh->so_timeo);
535 	}
536 }
537 
538 static int
539 soclose_sync(struct socket *so, int fflag)
540 {
541 	int error = 0;
542 
543 	if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
544 		so_pru_sync(so); /* unpend async prus */
545 
546 	if (so->so_pcb == NULL)
547 		goto discard;
548 
549 	if (so->so_state & SS_ISCONNECTED) {
550 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
551 			error = sodisconnect(so);
552 			if (error)
553 				goto drop;
554 		}
555 		if (so->so_options & SO_LINGER) {
556 			if ((so->so_state & SS_ISDISCONNECTING) &&
557 			    (fflag & FNONBLOCK))
558 				goto drop;
559 			while (so->so_state & SS_ISCONNECTED) {
560 				error = tsleep(&so->so_timeo, PCATCH,
561 					       "soclos", so->so_linger * hz);
562 				if (error)
563 					break;
564 			}
565 		}
566 	}
567 drop:
568 	if (so->so_pcb) {
569 		int error2;
570 
571 		error2 = so_pru_detach(so);
572 		if (error2 == EJUSTRETURN) {
573 			/*
574 			 * Protocol will call sodiscard()
575 			 * and sofree() for us.
576 			 */
577 			return error;
578 		}
579 		if (error == 0)
580 			error = error2;
581 	}
582 discard:
583 	sodiscard(so);
584 	sofree(so);		/* dispose of ref */
585 
586 	return (error);
587 }
588 
589 static void
590 soclose_fast_handler(netmsg_t msg)
591 {
592 	struct socket *so = msg->base.nm_so;
593 
594 	if (so->so_pcb == NULL)
595 		goto discard;
596 
597 	if ((so->so_state & SS_ISCONNECTED) &&
598 	    (so->so_state & SS_ISDISCONNECTING) == 0)
599 		so_pru_disconnect_direct(so);
600 
601 	if (so->so_pcb) {
602 		int error;
603 
604 		error = so_pru_detach_direct(so);
605 		if (error == EJUSTRETURN) {
606 			/*
607 			 * Protocol will call sodiscard()
608 			 * and sofree() for us.
609 			 */
610 			return;
611 		}
612 	}
613 discard:
614 	sodiscard(so);
615 	sofree(so);
616 }
617 
618 static void
619 soclose_fast(struct socket *so)
620 {
621 	struct netmsg_base *base = &so->so_clomsg;
622 
623 	netmsg_init(base, so, &netisr_apanic_rport, 0,
624 	    soclose_fast_handler);
625 	lwkt_sendmsg(so->so_port, &base->lmsg);
626 }
627 
628 /*
629  * Abort and destroy a socket.  Only one abort can be in progress
630  * at any given moment.
631  */
632 void
633 soabort_async(struct socket *so, boolean_t clr_head)
634 {
635 	/*
636 	 * Keep a reference before clearing the so_head
637 	 * to avoid racing socket close in netisr.
638 	 */
639 	soreference(so);
640 	if (clr_head)
641 		so->so_head = NULL;
642 	so_pru_abort_async(so);
643 }
644 
645 void
646 soabort_direct(struct socket *so)
647 {
648 	soreference(so);
649 	so_pru_abort_direct(so);
650 }
651 
652 /*
653  * so is passed in ref'd, which becomes owned by
654  * the cleared SS_NOFDREF flag.
655  */
656 void
657 soaccept_generic(struct socket *so)
658 {
659 	if ((so->so_state & SS_NOFDREF) == 0)
660 		panic("soaccept: !NOFDREF");
661 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
662 }
663 
664 int
665 soaccept(struct socket *so, struct sockaddr **nam)
666 {
667 	int error;
668 
669 	soaccept_generic(so);
670 	error = so_pru_accept(so, nam);
671 	return (error);
672 }
673 
674 int
675 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
676     boolean_t sync)
677 {
678 	int error;
679 
680 	if (so->so_options & SO_ACCEPTCONN)
681 		return (EOPNOTSUPP);
682 	/*
683 	 * If protocol is connection-based, can only connect once.
684 	 * Otherwise, if connected, try to disconnect first.
685 	 * This allows user to disconnect by connecting to, e.g.,
686 	 * a null address.
687 	 */
688 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
689 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
690 	    (error = sodisconnect(so)))) {
691 		error = EISCONN;
692 	} else {
693 		/*
694 		 * Prevent accumulated error from previous connection
695 		 * from biting us.
696 		 */
697 		so->so_error = 0;
698 		if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
699 			error = so_pru_connect_async(so, nam, td);
700 		else
701 			error = so_pru_connect(so, nam, td);
702 	}
703 	return (error);
704 }
705 
706 int
707 soconnect2(struct socket *so1, struct socket *so2)
708 {
709 	int error;
710 
711 	error = so_pru_connect2(so1, so2);
712 	return (error);
713 }
714 
715 int
716 sodisconnect(struct socket *so)
717 {
718 	int error;
719 
720 	if ((so->so_state & SS_ISCONNECTED) == 0) {
721 		error = ENOTCONN;
722 		goto bad;
723 	}
724 	if (so->so_state & SS_ISDISCONNECTING) {
725 		error = EALREADY;
726 		goto bad;
727 	}
728 	error = so_pru_disconnect(so);
729 bad:
730 	return (error);
731 }
732 
733 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
734 /*
735  * Send on a socket.
736  * If send must go all at once and message is larger than
737  * send buffering, then hard error.
738  * Lock against other senders.
739  * If must go all at once and not enough room now, then
740  * inform user that this would block and do nothing.
741  * Otherwise, if nonblocking, send as much as possible.
742  * The data to be sent is described by "uio" if nonzero,
743  * otherwise by the mbuf chain "top" (which must be null
744  * if uio is not).  Data provided in mbuf chain must be small
745  * enough to send all at once.
746  *
747  * Returns nonzero on error, timeout or signal; callers
748  * must check for short counts if EINTR/ERESTART are returned.
749  * Data and control buffers are freed on return.
750  */
751 int
752 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
753 	struct mbuf *top, struct mbuf *control, int flags,
754 	struct thread *td)
755 {
756 	struct mbuf **mp;
757 	struct mbuf *m;
758 	size_t resid;
759 	int space, len;
760 	int clen = 0, error, dontroute, mlen;
761 	int atomic = sosendallatonce(so) || top;
762 	int pru_flags;
763 
764 	if (uio) {
765 		resid = uio->uio_resid;
766 	} else {
767 		resid = (size_t)top->m_pkthdr.len;
768 #ifdef INVARIANTS
769 		len = 0;
770 		for (m = top; m; m = m->m_next)
771 			len += m->m_len;
772 		KKASSERT(top->m_pkthdr.len == len);
773 #endif
774 	}
775 
776 	/*
777 	 * WARNING!  resid is unsigned, space and len are signed.  space
778 	 * 	     can wind up negative if the sockbuf is overcommitted.
779 	 *
780 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
781 	 * type sockets since that's an error.
782 	 */
783 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
784 		error = EINVAL;
785 		goto out;
786 	}
787 
788 	dontroute =
789 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
790 	    (so->so_proto->pr_flags & PR_ATOMIC);
791 	if (td->td_lwp != NULL)
792 		td->td_lwp->lwp_ru.ru_msgsnd++;
793 	if (control)
794 		clen = control->m_len;
795 #define	gotoerr(errcode)	{ error = errcode; goto release; }
796 
797 restart:
798 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
799 	if (error)
800 		goto out;
801 
802 	do {
803 		if (so->so_state & SS_CANTSENDMORE)
804 			gotoerr(EPIPE);
805 		if (so->so_error) {
806 			error = so->so_error;
807 			so->so_error = 0;
808 			goto release;
809 		}
810 		if ((so->so_state & SS_ISCONNECTED) == 0) {
811 			/*
812 			 * `sendto' and `sendmsg' is allowed on a connection-
813 			 * based socket if it supports implied connect.
814 			 * Return ENOTCONN if not connected and no address is
815 			 * supplied.
816 			 */
817 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
818 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
819 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
820 				    !(resid == 0 && clen != 0))
821 					gotoerr(ENOTCONN);
822 			} else if (addr == NULL)
823 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
824 				   ENOTCONN : EDESTADDRREQ);
825 		}
826 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
827 		    clen > so->so_snd.ssb_hiwat) {
828 			gotoerr(EMSGSIZE);
829 		}
830 		space = ssb_space(&so->so_snd);
831 		if (flags & MSG_OOB)
832 			space += 1024;
833 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
834 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
835 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
836 				gotoerr(EWOULDBLOCK);
837 			ssb_unlock(&so->so_snd);
838 			error = ssb_wait(&so->so_snd);
839 			if (error)
840 				goto out;
841 			goto restart;
842 		}
843 		mp = &top;
844 		space -= clen;
845 		do {
846 		    if (uio == NULL) {
847 			/*
848 			 * Data is prepackaged in "top".
849 			 */
850 			resid = 0;
851 			if (flags & MSG_EOR)
852 				top->m_flags |= M_EOR;
853 		    } else do {
854 			if (resid > INT_MAX)
855 				resid = INT_MAX;
856 			m = m_getl((int)resid, M_WAITOK, MT_DATA,
857 				   top == NULL ? M_PKTHDR : 0, &mlen);
858 			if (top == NULL) {
859 				m->m_pkthdr.len = 0;
860 				m->m_pkthdr.rcvif = NULL;
861 			}
862 			len = imin((int)szmin(mlen, resid), space);
863 			if (resid < MINCLSIZE) {
864 				/*
865 				 * For datagram protocols, leave room
866 				 * for protocol headers in first mbuf.
867 				 */
868 				if (atomic && top == NULL && len < mlen)
869 					MH_ALIGN(m, len);
870 			}
871 			space -= len;
872 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
873 			resid = uio->uio_resid;
874 			m->m_len = len;
875 			*mp = m;
876 			top->m_pkthdr.len += len;
877 			if (error)
878 				goto release;
879 			mp = &m->m_next;
880 			if (resid == 0) {
881 				if (flags & MSG_EOR)
882 					top->m_flags |= M_EOR;
883 				break;
884 			}
885 		    } while (space > 0 && atomic);
886 		    if (dontroute)
887 			    so->so_options |= SO_DONTROUTE;
888 		    if (flags & MSG_OOB) {
889 		    	    pru_flags = PRUS_OOB;
890 		    } else if ((flags & MSG_EOF) &&
891 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
892 			       (resid == 0)) {
893 			    /*
894 			     * If the user set MSG_EOF, the protocol
895 			     * understands this flag and nothing left to
896 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
897 			     */
898 		    	    pru_flags = PRUS_EOF;
899 		    } else if (resid > 0 && space > 0) {
900 			    /* If there is more to send, set PRUS_MORETOCOME */
901 		    	    pru_flags = PRUS_MORETOCOME;
902 		    } else {
903 		    	    pru_flags = 0;
904 		    }
905 		    /*
906 		     * XXX all the SS_CANTSENDMORE checks previously
907 		     * done could be out of date.  We could have recieved
908 		     * a reset packet in an interrupt or maybe we slept
909 		     * while doing page faults in uiomove() etc. We could
910 		     * probably recheck again inside the splnet() protection
911 		     * here, but there are probably other places that this
912 		     * also happens.  We must rethink this.
913 		     */
914 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
915 		    if (dontroute)
916 			    so->so_options &= ~SO_DONTROUTE;
917 		    clen = 0;
918 		    control = NULL;
919 		    top = NULL;
920 		    mp = &top;
921 		    if (error)
922 			    goto release;
923 		} while (resid && space > 0);
924 	} while (resid);
925 
926 release:
927 	ssb_unlock(&so->so_snd);
928 out:
929 	if (top)
930 		m_freem(top);
931 	if (control)
932 		m_freem(control);
933 	return (error);
934 }
935 
936 #ifdef INET
937 /*
938  * A specialization of sosend() for UDP based on protocol-specific knowledge:
939  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
940  *	sosendallatonce() returns true,
941  *	the "atomic" variable is true,
942  *	and sosendudp() blocks until space is available for the entire send.
943  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
944  *	PR_IMPLOPCL flags set.
945  *   UDP has no out-of-band data.
946  *   UDP has no control data.
947  *   UDP does not support MSG_EOR.
948  */
949 int
950 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
951 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
952 {
953 	size_t resid;
954 	int error, pru_flags = 0;
955 	int space;
956 
957 	if (td->td_lwp != NULL)
958 		td->td_lwp->lwp_ru.ru_msgsnd++;
959 	if (control)
960 		m_freem(control);
961 
962 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
963 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
964 
965 restart:
966 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
967 	if (error)
968 		goto out;
969 
970 	if (so->so_state & SS_CANTSENDMORE)
971 		gotoerr(EPIPE);
972 	if (so->so_error) {
973 		error = so->so_error;
974 		so->so_error = 0;
975 		goto release;
976 	}
977 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
978 		gotoerr(EDESTADDRREQ);
979 	if (resid > so->so_snd.ssb_hiwat)
980 		gotoerr(EMSGSIZE);
981 	space = ssb_space(&so->so_snd);
982 	if (uio && (space < 0 || (size_t)space < resid)) {
983 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
984 			gotoerr(EWOULDBLOCK);
985 		ssb_unlock(&so->so_snd);
986 		error = ssb_wait(&so->so_snd);
987 		if (error)
988 			goto out;
989 		goto restart;
990 	}
991 
992 	if (uio) {
993 		int hdrlen = max_hdr;
994 
995 		/*
996 		 * We try to optimize out the additional mbuf
997 		 * allocations in M_PREPEND() on output path, e.g.
998 		 * - udp_output(), when it tries to prepend protocol
999 		 *   headers.
1000 		 * - Link layer output function, when it tries to
1001 		 *   prepend link layer header.
1002 		 *
1003 		 * This probably will not benefit any data that will
1004 		 * be fragmented, so this optimization is only performed
1005 		 * when the size of data and max size of protocol+link
1006 		 * headers fit into one mbuf cluster.
1007 		 */
1008 		if (uio->uio_resid > MCLBYTES - hdrlen ||
1009 		    !udp_sosend_prepend) {
1010 			top = m_uiomove(uio);
1011 			if (top == NULL)
1012 				goto release;
1013 		} else {
1014 			int nsize;
1015 
1016 			top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1017 			    MT_DATA, M_PKTHDR, &nsize);
1018 			KASSERT(nsize >= uio->uio_resid + hdrlen,
1019 			    ("sosendudp invalid nsize %d, "
1020 			     "resid %zu, hdrlen %d",
1021 			     nsize, uio->uio_resid, hdrlen));
1022 
1023 			top->m_len = uio->uio_resid;
1024 			top->m_pkthdr.len = uio->uio_resid;
1025 			top->m_data += hdrlen;
1026 
1027 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1028 			if (error)
1029 				goto out;
1030 		}
1031 	}
1032 
1033 	if (flags & MSG_DONTROUTE)
1034 		pru_flags |= PRUS_DONTROUTE;
1035 
1036 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1037 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1038 		error = 0;
1039 	} else {
1040 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1041 	}
1042 	top = NULL;		/* sent or freed in lower layer */
1043 
1044 release:
1045 	ssb_unlock(&so->so_snd);
1046 out:
1047 	if (top)
1048 		m_freem(top);
1049 	return (error);
1050 }
1051 
1052 int
1053 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1054 	struct mbuf *top, struct mbuf *control, int flags,
1055 	struct thread *td)
1056 {
1057 	struct mbuf **mp;
1058 	struct mbuf *m;
1059 	size_t resid;
1060 	int space, len;
1061 	int error, mlen;
1062 	int allatonce;
1063 	int pru_flags;
1064 
1065 	if (uio) {
1066 		KKASSERT(top == NULL);
1067 		allatonce = 0;
1068 		resid = uio->uio_resid;
1069 	} else {
1070 		allatonce = 1;
1071 		resid = (size_t)top->m_pkthdr.len;
1072 #ifdef INVARIANTS
1073 		len = 0;
1074 		for (m = top; m; m = m->m_next)
1075 			len += m->m_len;
1076 		KKASSERT(top->m_pkthdr.len == len);
1077 #endif
1078 	}
1079 
1080 	/*
1081 	 * WARNING!  resid is unsigned, space and len are signed.  space
1082 	 * 	     can wind up negative if the sockbuf is overcommitted.
1083 	 *
1084 	 * Also check to make sure that MSG_EOR isn't used on TCP
1085 	 */
1086 	if (flags & MSG_EOR) {
1087 		error = EINVAL;
1088 		goto out;
1089 	}
1090 
1091 	if (control) {
1092 		/* TCP doesn't do control messages (rights, creds, etc) */
1093 		if (control->m_len) {
1094 			error = EINVAL;
1095 			goto out;
1096 		}
1097 		m_freem(control);	/* empty control, just free it */
1098 		control = NULL;
1099 	}
1100 
1101 	if (td->td_lwp != NULL)
1102 		td->td_lwp->lwp_ru.ru_msgsnd++;
1103 
1104 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1105 
1106 restart:
1107 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1108 	if (error)
1109 		goto out;
1110 
1111 	do {
1112 		if (so->so_state & SS_CANTSENDMORE)
1113 			gotoerr(EPIPE);
1114 		if (so->so_error) {
1115 			error = so->so_error;
1116 			so->so_error = 0;
1117 			goto release;
1118 		}
1119 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1120 		    (so->so_state & SS_ISCONFIRMING) == 0)
1121 			gotoerr(ENOTCONN);
1122 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1123 			gotoerr(EMSGSIZE);
1124 
1125 		space = ssb_space_prealloc(&so->so_snd);
1126 		if (flags & MSG_OOB)
1127 			space += 1024;
1128 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1129 		    space < so->so_snd.ssb_lowat) {
1130 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1131 				gotoerr(EWOULDBLOCK);
1132 			ssb_unlock(&so->so_snd);
1133 			error = ssb_wait(&so->so_snd);
1134 			if (error)
1135 				goto out;
1136 			goto restart;
1137 		}
1138 		mp = &top;
1139 		do {
1140 		    int cnt = 0, async = 0;
1141 
1142 		    if (uio == NULL) {
1143 			/*
1144 			 * Data is prepackaged in "top".
1145 			 */
1146 			resid = 0;
1147 		    } else do {
1148 			if (resid > INT_MAX)
1149 				resid = INT_MAX;
1150 			if (tcp_sosend_jcluster) {
1151 				m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1152 					   top == NULL ? M_PKTHDR : 0, &mlen);
1153 			} else {
1154 				m = m_getl((int)resid, M_WAITOK, MT_DATA,
1155 					   top == NULL ? M_PKTHDR : 0, &mlen);
1156 			}
1157 			if (top == NULL) {
1158 				m->m_pkthdr.len = 0;
1159 				m->m_pkthdr.rcvif = NULL;
1160 			}
1161 			len = imin((int)szmin(mlen, resid), space);
1162 			space -= len;
1163 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1164 			resid = uio->uio_resid;
1165 			m->m_len = len;
1166 			*mp = m;
1167 			top->m_pkthdr.len += len;
1168 			if (error)
1169 				goto release;
1170 			mp = &m->m_next;
1171 			if (resid == 0)
1172 				break;
1173 			++cnt;
1174 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1175 
1176 		    if (tcp_sosend_async)
1177 			    async = 1;
1178 
1179 		    if (flags & MSG_OOB) {
1180 		    	    pru_flags = PRUS_OOB;
1181 			    async = 0;
1182 		    } else if ((flags & MSG_EOF) && resid == 0) {
1183 			    pru_flags = PRUS_EOF;
1184 		    } else if (resid > 0 && space > 0) {
1185 			    /* If there is more to send, set PRUS_MORETOCOME */
1186 		    	    pru_flags = PRUS_MORETOCOME;
1187 			    async = 1;
1188 		    } else {
1189 		    	    pru_flags = 0;
1190 		    }
1191 
1192 		    if (flags & MSG_SYNC)
1193 			    async = 0;
1194 
1195 		    /*
1196 		     * XXX all the SS_CANTSENDMORE checks previously
1197 		     * done could be out of date.  We could have recieved
1198 		     * a reset packet in an interrupt or maybe we slept
1199 		     * while doing page faults in uiomove() etc. We could
1200 		     * probably recheck again inside the splnet() protection
1201 		     * here, but there are probably other places that this
1202 		     * also happens.  We must rethink this.
1203 		     */
1204 		    for (m = top; m; m = m->m_next)
1205 			    ssb_preallocstream(&so->so_snd, m);
1206 		    if (!async) {
1207 			    error = so_pru_send(so, pru_flags, top,
1208 			        NULL, NULL, td);
1209 		    } else {
1210 			    so_pru_send_async(so, pru_flags, top,
1211 			        NULL, NULL, td);
1212 			    error = 0;
1213 		    }
1214 
1215 		    top = NULL;
1216 		    mp = &top;
1217 		    if (error)
1218 			    goto release;
1219 		} while (resid && space > 0);
1220 	} while (resid);
1221 
1222 release:
1223 	ssb_unlock(&so->so_snd);
1224 out:
1225 	if (top)
1226 		m_freem(top);
1227 	if (control)
1228 		m_freem(control);
1229 	return (error);
1230 }
1231 #endif
1232 
1233 /*
1234  * Implement receive operations on a socket.
1235  *
1236  * We depend on the way that records are added to the signalsockbuf
1237  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1238  * must begin with an address if the protocol so specifies,
1239  * followed by an optional mbuf or mbufs containing ancillary data,
1240  * and then zero or more mbufs of data.
1241  *
1242  * Although the signalsockbuf is locked, new data may still be appended.
1243  * A token inside the ssb_lock deals with MP issues and still allows
1244  * the network to access the socket if we block in a uio.
1245  *
1246  * The caller may receive the data as a single mbuf chain by supplying
1247  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1248  * only for the count in uio_resid.
1249  */
1250 int
1251 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1252 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1253 {
1254 	struct mbuf *m, *n;
1255 	struct mbuf *free_chain = NULL;
1256 	int flags, len, error, offset;
1257 	struct protosw *pr = so->so_proto;
1258 	int moff, type = 0;
1259 	size_t resid, orig_resid;
1260 	boolean_t free_rights = FALSE;
1261 
1262 	if (uio)
1263 		resid = uio->uio_resid;
1264 	else
1265 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1266 	orig_resid = resid;
1267 
1268 	if (psa)
1269 		*psa = NULL;
1270 	if (controlp)
1271 		*controlp = NULL;
1272 	if (flagsp)
1273 		flags = *flagsp &~ MSG_EOR;
1274 	else
1275 		flags = 0;
1276 	if (flags & MSG_OOB) {
1277 		m = m_get(M_WAITOK, MT_DATA);
1278 		if (m == NULL)
1279 			return (ENOBUFS);
1280 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1281 		if (error)
1282 			goto bad;
1283 		if (sio) {
1284 			do {
1285 				sbappend(sio, m);
1286 				KKASSERT(resid >= (size_t)m->m_len);
1287 				resid -= (size_t)m->m_len;
1288 			} while (resid > 0 && m);
1289 		} else {
1290 			do {
1291 				uio->uio_resid = resid;
1292 				error = uiomove(mtod(m, caddr_t),
1293 						(int)szmin(resid, m->m_len),
1294 						uio);
1295 				resid = uio->uio_resid;
1296 				m = m_free(m);
1297 			} while (uio->uio_resid && error == 0 && m);
1298 		}
1299 bad:
1300 		if (m)
1301 			m_freem(m);
1302 		return (error);
1303 	}
1304 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1305 		so_pru_rcvd(so, 0);
1306 
1307 	/*
1308 	 * The token interlocks against the protocol thread while
1309 	 * ssb_lock is a blocking lock against other userland entities.
1310 	 */
1311 	lwkt_gettoken(&so->so_rcv.ssb_token);
1312 restart:
1313 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1314 	if (error)
1315 		goto done;
1316 
1317 	m = so->so_rcv.ssb_mb;
1318 	/*
1319 	 * If we have less data than requested, block awaiting more
1320 	 * (subject to any timeout) if:
1321 	 *   1. the current count is less than the low water mark, or
1322 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1323 	 *	receive operation at once if we block (resid <= hiwat).
1324 	 *   3. MSG_DONTWAIT is not set
1325 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1326 	 * we have to do the receive in sections, and thus risk returning
1327 	 * a short count if a timeout or signal occurs after we start.
1328 	 */
1329 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1330 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1331 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1332 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1333 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1334 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1335 		if (so->so_error) {
1336 			if (m)
1337 				goto dontblock;
1338 			error = so->so_error;
1339 			if ((flags & MSG_PEEK) == 0)
1340 				so->so_error = 0;
1341 			goto release;
1342 		}
1343 		if (so->so_state & SS_CANTRCVMORE) {
1344 			if (m)
1345 				goto dontblock;
1346 			else
1347 				goto release;
1348 		}
1349 		for (; m; m = m->m_next) {
1350 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1351 				m = so->so_rcv.ssb_mb;
1352 				goto dontblock;
1353 			}
1354 		}
1355 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1356 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1357 			error = ENOTCONN;
1358 			goto release;
1359 		}
1360 		if (resid == 0)
1361 			goto release;
1362 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1363 			error = EWOULDBLOCK;
1364 			goto release;
1365 		}
1366 		ssb_unlock(&so->so_rcv);
1367 		error = ssb_wait(&so->so_rcv);
1368 		if (error)
1369 			goto done;
1370 		goto restart;
1371 	}
1372 dontblock:
1373 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1374 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1375 
1376 	/*
1377 	 * note: m should be == sb_mb here.  Cache the next record while
1378 	 * cleaning up.  Note that calling m_free*() will break out critical
1379 	 * section.
1380 	 */
1381 	KKASSERT(m == so->so_rcv.ssb_mb);
1382 
1383 	/*
1384 	 * Skip any address mbufs prepending the record.
1385 	 */
1386 	if (pr->pr_flags & PR_ADDR) {
1387 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1388 		orig_resid = 0;
1389 		if (psa)
1390 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1391 		if (flags & MSG_PEEK)
1392 			m = m->m_next;
1393 		else
1394 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1395 	}
1396 
1397 	/*
1398 	 * Skip any control mbufs prepending the record.
1399 	 */
1400 	while (m && m->m_type == MT_CONTROL && error == 0) {
1401 		if (flags & MSG_PEEK) {
1402 			if (controlp)
1403 				*controlp = m_copy(m, 0, m->m_len);
1404 			m = m->m_next;	/* XXX race */
1405 		} else {
1406 			const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1407 
1408 			if (controlp) {
1409 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1410 				if (pr->pr_domain->dom_externalize &&
1411 				    cm->cmsg_level == SOL_SOCKET &&
1412 				    cm->cmsg_type == SCM_RIGHTS) {
1413 					error = pr->pr_domain->dom_externalize
1414 					    (m, flags);
1415 				}
1416 				*controlp = m;
1417 				m = n;
1418 			} else {
1419 				if (cm->cmsg_level == SOL_SOCKET &&
1420 				    cm->cmsg_type == SCM_RIGHTS)
1421 					free_rights = TRUE;
1422 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1423 			}
1424 		}
1425 		if (controlp && *controlp) {
1426 			orig_resid = 0;
1427 			controlp = &(*controlp)->m_next;
1428 		}
1429 	}
1430 
1431 	/*
1432 	 * flag OOB data.
1433 	 */
1434 	if (m) {
1435 		type = m->m_type;
1436 		if (type == MT_OOBDATA)
1437 			flags |= MSG_OOB;
1438 	}
1439 
1440 	/*
1441 	 * Copy to the UIO or mbuf return chain (*mp).
1442 	 */
1443 	moff = 0;
1444 	offset = 0;
1445 	while (m && resid > 0 && error == 0) {
1446 		if (m->m_type == MT_OOBDATA) {
1447 			if (type != MT_OOBDATA)
1448 				break;
1449 		} else if (type == MT_OOBDATA)
1450 			break;
1451 		else
1452 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1453 			("receive 3"));
1454 		soclrstate(so, SS_RCVATMARK);
1455 		len = (resid > INT_MAX) ? INT_MAX : resid;
1456 		if (so->so_oobmark && len > so->so_oobmark - offset)
1457 			len = so->so_oobmark - offset;
1458 		if (len > m->m_len - moff)
1459 			len = m->m_len - moff;
1460 
1461 		/*
1462 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1463 		 * The SIO is dealt with when we eat the mbuf, but deal
1464 		 * with the resid here either way.
1465 		 */
1466 		if (uio) {
1467 			uio->uio_resid = resid;
1468 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1469 			resid = uio->uio_resid;
1470 			if (error)
1471 				goto release;
1472 		} else {
1473 			resid -= (size_t)len;
1474 		}
1475 
1476 		/*
1477 		 * Eat the entire mbuf or just a piece of it
1478 		 */
1479 		if (len == m->m_len - moff) {
1480 			if (m->m_flags & M_EOR)
1481 				flags |= MSG_EOR;
1482 			if (flags & MSG_PEEK) {
1483 				m = m->m_next;
1484 				moff = 0;
1485 			} else {
1486 				if (sio) {
1487 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1488 					sbappend(sio, m);
1489 					m = n;
1490 				} else {
1491 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1492 				}
1493 			}
1494 		} else {
1495 			if (flags & MSG_PEEK) {
1496 				moff += len;
1497 			} else {
1498 				if (sio) {
1499 					n = m_copym(m, 0, len, M_WAITOK);
1500 					if (n)
1501 						sbappend(sio, n);
1502 				}
1503 				m->m_data += len;
1504 				m->m_len -= len;
1505 				so->so_rcv.ssb_cc -= len;
1506 			}
1507 		}
1508 		if (so->so_oobmark) {
1509 			if ((flags & MSG_PEEK) == 0) {
1510 				so->so_oobmark -= len;
1511 				if (so->so_oobmark == 0) {
1512 					sosetstate(so, SS_RCVATMARK);
1513 					break;
1514 				}
1515 			} else {
1516 				offset += len;
1517 				if (offset == so->so_oobmark)
1518 					break;
1519 			}
1520 		}
1521 		if (flags & MSG_EOR)
1522 			break;
1523 		/*
1524 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1525 		 * we must not quit until resid == 0 or an error
1526 		 * termination.  If a signal/timeout occurs, return
1527 		 * with a short count but without error.
1528 		 * Keep signalsockbuf locked against other readers.
1529 		 */
1530 		while ((flags & MSG_WAITALL) && m == NULL &&
1531 		       resid > 0 && !sosendallatonce(so) &&
1532 		       so->so_rcv.ssb_mb == NULL) {
1533 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1534 				break;
1535 			/*
1536 			 * The window might have closed to zero, make
1537 			 * sure we send an ack now that we've drained
1538 			 * the buffer or we might end up blocking until
1539 			 * the idle takes over (5 seconds).
1540 			 */
1541 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1542 				so_pru_rcvd(so, flags);
1543 			error = ssb_wait(&so->so_rcv);
1544 			if (error) {
1545 				ssb_unlock(&so->so_rcv);
1546 				error = 0;
1547 				goto done;
1548 			}
1549 			m = so->so_rcv.ssb_mb;
1550 		}
1551 	}
1552 
1553 	/*
1554 	 * If an atomic read was requested but unread data still remains
1555 	 * in the record, set MSG_TRUNC.
1556 	 */
1557 	if (m && pr->pr_flags & PR_ATOMIC)
1558 		flags |= MSG_TRUNC;
1559 
1560 	/*
1561 	 * Cleanup.  If an atomic read was requested drop any unread data.
1562 	 */
1563 	if ((flags & MSG_PEEK) == 0) {
1564 		if (m && (pr->pr_flags & PR_ATOMIC))
1565 			sbdroprecord(&so->so_rcv.sb);
1566 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1567 			so_pru_rcvd(so, flags);
1568 	}
1569 
1570 	if (orig_resid == resid && orig_resid &&
1571 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1572 		ssb_unlock(&so->so_rcv);
1573 		goto restart;
1574 	}
1575 
1576 	if (flagsp)
1577 		*flagsp |= flags;
1578 release:
1579 	ssb_unlock(&so->so_rcv);
1580 done:
1581 	lwkt_reltoken(&so->so_rcv.ssb_token);
1582 	if (free_chain) {
1583 		if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1584 		    pr->pr_domain->dom_dispose)
1585 			pr->pr_domain->dom_dispose(free_chain);
1586 		m_freem(free_chain);
1587 	}
1588 	return (error);
1589 }
1590 
1591 int
1592 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1593 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1594 {
1595 	struct mbuf *m, *n;
1596 	struct mbuf *free_chain = NULL;
1597 	int flags, len, error, offset;
1598 	struct protosw *pr = so->so_proto;
1599 	int moff;
1600 	int didoob;
1601 	size_t resid, orig_resid, restmp;
1602 
1603 	if (uio)
1604 		resid = uio->uio_resid;
1605 	else
1606 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1607 	orig_resid = resid;
1608 
1609 	if (psa)
1610 		*psa = NULL;
1611 	if (controlp)
1612 		*controlp = NULL;
1613 	if (flagsp)
1614 		flags = *flagsp &~ MSG_EOR;
1615 	else
1616 		flags = 0;
1617 	if (flags & MSG_OOB) {
1618 		m = m_get(M_WAITOK, MT_DATA);
1619 		if (m == NULL)
1620 			return (ENOBUFS);
1621 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1622 		if (error)
1623 			goto bad;
1624 		if (sio) {
1625 			do {
1626 				sbappend(sio, m);
1627 				KKASSERT(resid >= (size_t)m->m_len);
1628 				resid -= (size_t)m->m_len;
1629 			} while (resid > 0 && m);
1630 		} else {
1631 			do {
1632 				uio->uio_resid = resid;
1633 				error = uiomove(mtod(m, caddr_t),
1634 						(int)szmin(resid, m->m_len),
1635 						uio);
1636 				resid = uio->uio_resid;
1637 				m = m_free(m);
1638 			} while (uio->uio_resid && error == 0 && m);
1639 		}
1640 bad:
1641 		if (m)
1642 			m_freem(m);
1643 		return (error);
1644 	}
1645 
1646 	/*
1647 	 * The token interlocks against the protocol thread while
1648 	 * ssb_lock is a blocking lock against other userland entities.
1649 	 *
1650 	 * Lock a limited number of mbufs (not all, so sbcompress() still
1651 	 * works well).  The token is used as an interlock for sbwait() so
1652 	 * release it afterwords.
1653 	 */
1654 restart:
1655 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1656 	if (error)
1657 		goto done;
1658 
1659 	lwkt_gettoken(&so->so_rcv.ssb_token);
1660 	m = so->so_rcv.ssb_mb;
1661 
1662 	/*
1663 	 * If we have less data than requested, block awaiting more
1664 	 * (subject to any timeout) if:
1665 	 *   1. the current count is less than the low water mark, or
1666 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1667 	 *	receive operation at once if we block (resid <= hiwat).
1668 	 *   3. MSG_DONTWAIT is not set
1669 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1670 	 * we have to do the receive in sections, and thus risk returning
1671 	 * a short count if a timeout or signal occurs after we start.
1672 	 */
1673 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1674 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1675 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1676 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1677 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1678 		if (so->so_error) {
1679 			if (m)
1680 				goto dontblock;
1681 			lwkt_reltoken(&so->so_rcv.ssb_token);
1682 			error = so->so_error;
1683 			if ((flags & MSG_PEEK) == 0)
1684 				so->so_error = 0;
1685 			goto release;
1686 		}
1687 		if (so->so_state & SS_CANTRCVMORE) {
1688 			if (m)
1689 				goto dontblock;
1690 			lwkt_reltoken(&so->so_rcv.ssb_token);
1691 			goto release;
1692 		}
1693 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1694 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1695 			lwkt_reltoken(&so->so_rcv.ssb_token);
1696 			error = ENOTCONN;
1697 			goto release;
1698 		}
1699 		if (resid == 0) {
1700 			lwkt_reltoken(&so->so_rcv.ssb_token);
1701 			goto release;
1702 		}
1703 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1704 			lwkt_reltoken(&so->so_rcv.ssb_token);
1705 			error = EWOULDBLOCK;
1706 			goto release;
1707 		}
1708 		ssb_unlock(&so->so_rcv);
1709 		error = ssb_wait(&so->so_rcv);
1710 		lwkt_reltoken(&so->so_rcv.ssb_token);
1711 		if (error)
1712 			goto done;
1713 		goto restart;
1714 	}
1715 
1716 	/*
1717 	 * Token still held
1718 	 */
1719 dontblock:
1720 	n = m;
1721 	restmp = 0;
1722 	while (n && restmp < resid) {
1723 		n->m_flags |= M_SOLOCKED;
1724 		restmp += n->m_len;
1725 		if (n->m_next == NULL)
1726 			n = n->m_nextpkt;
1727 		else
1728 			n = n->m_next;
1729 	}
1730 
1731 	/*
1732 	 * Release token for loop
1733 	 */
1734 	lwkt_reltoken(&so->so_rcv.ssb_token);
1735 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1736 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1737 
1738 	/*
1739 	 * note: m should be == sb_mb here.  Cache the next record while
1740 	 * cleaning up.  Note that calling m_free*() will break out critical
1741 	 * section.
1742 	 */
1743 	KKASSERT(m == so->so_rcv.ssb_mb);
1744 
1745 	/*
1746 	 * Copy to the UIO or mbuf return chain (*mp).
1747 	 *
1748 	 * NOTE: Token is not held for loop
1749 	 */
1750 	moff = 0;
1751 	offset = 0;
1752 	didoob = 0;
1753 
1754 	while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1755 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1756 		    ("receive 3"));
1757 
1758 		soclrstate(so, SS_RCVATMARK);
1759 		len = (resid > INT_MAX) ? INT_MAX : resid;
1760 		if (so->so_oobmark && len > so->so_oobmark - offset)
1761 			len = so->so_oobmark - offset;
1762 		if (len > m->m_len - moff)
1763 			len = m->m_len - moff;
1764 
1765 		/*
1766 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1767 		 * The SIO is dealt with when we eat the mbuf, but deal
1768 		 * with the resid here either way.
1769 		 */
1770 		if (uio) {
1771 			uio->uio_resid = resid;
1772 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1773 			resid = uio->uio_resid;
1774 			if (error)
1775 				goto release;
1776 		} else {
1777 			resid -= (size_t)len;
1778 		}
1779 
1780 		/*
1781 		 * Eat the entire mbuf or just a piece of it
1782 		 */
1783 		offset += len;
1784 		if (len == m->m_len - moff) {
1785 			m = m->m_next;
1786 			moff = 0;
1787 		} else {
1788 			moff += len;
1789 		}
1790 
1791 		/*
1792 		 * Check oobmark
1793 		 */
1794 		if (so->so_oobmark && offset == so->so_oobmark) {
1795 			didoob = 1;
1796 			break;
1797 		}
1798 	}
1799 
1800 	/*
1801 	 * Synchronize sockbuf with data we read.
1802 	 *
1803 	 * NOTE: (m) is junk on entry (it could be left over from the
1804 	 *	 previous loop).
1805 	 */
1806 	if ((flags & MSG_PEEK) == 0) {
1807 		lwkt_gettoken(&so->so_rcv.ssb_token);
1808 		m = so->so_rcv.ssb_mb;
1809 		while (m && offset >= m->m_len) {
1810 			if (so->so_oobmark) {
1811 				so->so_oobmark -= m->m_len;
1812 				if (so->so_oobmark == 0) {
1813 					sosetstate(so, SS_RCVATMARK);
1814 					didoob = 1;
1815 				}
1816 			}
1817 			offset -= m->m_len;
1818 			if (sio) {
1819 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1820 				sbappend(sio, m);
1821 				m = n;
1822 			} else {
1823 				m = sbunlinkmbuf(&so->so_rcv.sb,
1824 						 m, &free_chain);
1825 			}
1826 		}
1827 		if (offset) {
1828 			KKASSERT(m);
1829 			if (sio) {
1830 				n = m_copym(m, 0, offset, M_WAITOK);
1831 				if (n)
1832 					sbappend(sio, n);
1833 			}
1834 			m->m_data += offset;
1835 			m->m_len -= offset;
1836 			so->so_rcv.ssb_cc -= offset;
1837 			if (so->so_oobmark) {
1838 				so->so_oobmark -= offset;
1839 				if (so->so_oobmark == 0) {
1840 					sosetstate(so, SS_RCVATMARK);
1841 					didoob = 1;
1842 				}
1843 			}
1844 			offset = 0;
1845 		}
1846 		lwkt_reltoken(&so->so_rcv.ssb_token);
1847 	}
1848 
1849 	/*
1850 	 * If the MSG_WAITALL flag is set (for non-atomic socket),
1851 	 * we must not quit until resid == 0 or an error termination.
1852 	 *
1853 	 * If a signal/timeout occurs, return with a short count but without
1854 	 * error.
1855 	 *
1856 	 * Keep signalsockbuf locked against other readers.
1857 	 *
1858 	 * XXX if MSG_PEEK we currently do quit.
1859 	 */
1860 	if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1861 	    didoob == 0 && resid > 0 &&
1862 	    !sosendallatonce(so)) {
1863 		lwkt_gettoken(&so->so_rcv.ssb_token);
1864 		error = 0;
1865 		while ((m = so->so_rcv.ssb_mb) == NULL) {
1866 			if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1867 				error = so->so_error;
1868 				break;
1869 			}
1870 			/*
1871 			 * The window might have closed to zero, make
1872 			 * sure we send an ack now that we've drained
1873 			 * the buffer or we might end up blocking until
1874 			 * the idle takes over (5 seconds).
1875 			 */
1876 			if (so->so_pcb)
1877 				so_pru_rcvd_async(so);
1878 			if (so->so_rcv.ssb_mb == NULL)
1879 				error = ssb_wait(&so->so_rcv);
1880 			if (error) {
1881 				lwkt_reltoken(&so->so_rcv.ssb_token);
1882 				ssb_unlock(&so->so_rcv);
1883 				error = 0;
1884 				goto done;
1885 			}
1886 		}
1887 		if (m && error == 0)
1888 			goto dontblock;
1889 		lwkt_reltoken(&so->so_rcv.ssb_token);
1890 	}
1891 
1892 	/*
1893 	 * Token not held here.
1894 	 *
1895 	 * Cleanup.  If an atomic read was requested drop any unread data XXX
1896 	 */
1897 	if ((flags & MSG_PEEK) == 0) {
1898 		if (so->so_pcb)
1899 			so_pru_rcvd_async(so);
1900 	}
1901 
1902 	if (orig_resid == resid && orig_resid &&
1903 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1904 		ssb_unlock(&so->so_rcv);
1905 		goto restart;
1906 	}
1907 
1908 	if (flagsp)
1909 		*flagsp |= flags;
1910 release:
1911 	ssb_unlock(&so->so_rcv);
1912 done:
1913 	if (free_chain)
1914 		m_freem(free_chain);
1915 	return (error);
1916 }
1917 
1918 /*
1919  * Shut a socket down.  Note that we do not get a frontend lock as we
1920  * want to be able to shut the socket down even if another thread is
1921  * blocked in a read(), thus waking it up.
1922  */
1923 int
1924 soshutdown(struct socket *so, int how)
1925 {
1926 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1927 		return (EINVAL);
1928 
1929 	if (how != SHUT_WR) {
1930 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1931 		sorflush(so);
1932 		/*ssb_unlock(&so->so_rcv);*/
1933 	}
1934 	if (how != SHUT_RD)
1935 		return (so_pru_shutdown(so));
1936 	return (0);
1937 }
1938 
1939 void
1940 sorflush(struct socket *so)
1941 {
1942 	struct signalsockbuf *ssb = &so->so_rcv;
1943 	struct protosw *pr = so->so_proto;
1944 	struct signalsockbuf asb;
1945 
1946 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1947 
1948 	lwkt_gettoken(&ssb->ssb_token);
1949 	socantrcvmore(so);
1950 	asb = *ssb;
1951 
1952 	/*
1953 	 * Can't just blow up the ssb structure here
1954 	 */
1955 	bzero(&ssb->sb, sizeof(ssb->sb));
1956 	ssb->ssb_timeo = 0;
1957 	ssb->ssb_lowat = 0;
1958 	ssb->ssb_hiwat = 0;
1959 	ssb->ssb_mbmax = 0;
1960 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1961 
1962 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1963 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1964 	ssb_release(&asb, so);
1965 
1966 	lwkt_reltoken(&ssb->ssb_token);
1967 }
1968 
1969 #ifdef INET
1970 static int
1971 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1972 {
1973 	struct accept_filter_arg	*afap = NULL;
1974 	struct accept_filter	*afp;
1975 	struct so_accf	*af = so->so_accf;
1976 	int	error = 0;
1977 
1978 	/* do not set/remove accept filters on non listen sockets */
1979 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1980 		error = EINVAL;
1981 		goto out;
1982 	}
1983 
1984 	/* removing the filter */
1985 	if (sopt == NULL) {
1986 		if (af != NULL) {
1987 			if (af->so_accept_filter != NULL &&
1988 				af->so_accept_filter->accf_destroy != NULL) {
1989 				af->so_accept_filter->accf_destroy(so);
1990 			}
1991 			if (af->so_accept_filter_str != NULL) {
1992 				kfree(af->so_accept_filter_str, M_ACCF);
1993 			}
1994 			kfree(af, M_ACCF);
1995 			so->so_accf = NULL;
1996 		}
1997 		so->so_options &= ~SO_ACCEPTFILTER;
1998 		return (0);
1999 	}
2000 	/* adding a filter */
2001 	/* must remove previous filter first */
2002 	if (af != NULL) {
2003 		error = EINVAL;
2004 		goto out;
2005 	}
2006 	/* don't put large objects on the kernel stack */
2007 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2008 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2009 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
2010 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2011 	if (error)
2012 		goto out;
2013 	afp = accept_filt_get(afap->af_name);
2014 	if (afp == NULL) {
2015 		error = ENOENT;
2016 		goto out;
2017 	}
2018 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2019 	if (afp->accf_create != NULL) {
2020 		if (afap->af_name[0] != '\0') {
2021 			int len = strlen(afap->af_name) + 1;
2022 
2023 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
2024 							   M_WAITOK);
2025 			strcpy(af->so_accept_filter_str, afap->af_name);
2026 		}
2027 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2028 		if (af->so_accept_filter_arg == NULL) {
2029 			kfree(af->so_accept_filter_str, M_ACCF);
2030 			kfree(af, M_ACCF);
2031 			so->so_accf = NULL;
2032 			error = EINVAL;
2033 			goto out;
2034 		}
2035 	}
2036 	af->so_accept_filter = afp;
2037 	so->so_accf = af;
2038 	so->so_options |= SO_ACCEPTFILTER;
2039 out:
2040 	if (afap != NULL)
2041 		kfree(afap, M_TEMP);
2042 	return (error);
2043 }
2044 #endif /* INET */
2045 
2046 /*
2047  * Perhaps this routine, and sooptcopyout(), below, ought to come in
2048  * an additional variant to handle the case where the option value needs
2049  * to be some kind of integer, but not a specific size.
2050  * In addition to their use here, these functions are also called by the
2051  * protocol-level pr_ctloutput() routines.
2052  */
2053 int
2054 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2055 {
2056 	return soopt_to_kbuf(sopt, buf, len, minlen);
2057 }
2058 
2059 int
2060 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2061 {
2062 	size_t	valsize;
2063 
2064 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2065 	KKASSERT(kva_p(buf));
2066 
2067 	/*
2068 	 * If the user gives us more than we wanted, we ignore it,
2069 	 * but if we don't get the minimum length the caller
2070 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
2071 	 * is set to however much we actually retrieved.
2072 	 */
2073 	if ((valsize = sopt->sopt_valsize) < minlen)
2074 		return EINVAL;
2075 	if (valsize > len)
2076 		sopt->sopt_valsize = valsize = len;
2077 
2078 	bcopy(sopt->sopt_val, buf, valsize);
2079 	return 0;
2080 }
2081 
2082 
2083 int
2084 sosetopt(struct socket *so, struct sockopt *sopt)
2085 {
2086 	int	error, optval;
2087 	struct	linger l;
2088 	struct	timeval tv;
2089 	u_long  val;
2090 	struct signalsockbuf *sotmp;
2091 
2092 	error = 0;
2093 	sopt->sopt_dir = SOPT_SET;
2094 	if (sopt->sopt_level != SOL_SOCKET) {
2095 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2096 			return (so_pr_ctloutput(so, sopt));
2097 		}
2098 		error = ENOPROTOOPT;
2099 	} else {
2100 		switch (sopt->sopt_name) {
2101 #ifdef INET
2102 		case SO_ACCEPTFILTER:
2103 			error = do_setopt_accept_filter(so, sopt);
2104 			if (error)
2105 				goto bad;
2106 			break;
2107 #endif /* INET */
2108 		case SO_LINGER:
2109 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2110 			if (error)
2111 				goto bad;
2112 
2113 			so->so_linger = l.l_linger;
2114 			if (l.l_onoff)
2115 				so->so_options |= SO_LINGER;
2116 			else
2117 				so->so_options &= ~SO_LINGER;
2118 			break;
2119 
2120 		case SO_DEBUG:
2121 		case SO_KEEPALIVE:
2122 		case SO_DONTROUTE:
2123 		case SO_USELOOPBACK:
2124 		case SO_BROADCAST:
2125 		case SO_REUSEADDR:
2126 		case SO_REUSEPORT:
2127 		case SO_OOBINLINE:
2128 		case SO_TIMESTAMP:
2129 		case SO_NOSIGPIPE:
2130 			error = sooptcopyin(sopt, &optval, sizeof optval,
2131 					    sizeof optval);
2132 			if (error)
2133 				goto bad;
2134 			if (optval)
2135 				so->so_options |= sopt->sopt_name;
2136 			else
2137 				so->so_options &= ~sopt->sopt_name;
2138 			break;
2139 
2140 		case SO_SNDBUF:
2141 		case SO_RCVBUF:
2142 		case SO_SNDLOWAT:
2143 		case SO_RCVLOWAT:
2144 			error = sooptcopyin(sopt, &optval, sizeof optval,
2145 					    sizeof optval);
2146 			if (error)
2147 				goto bad;
2148 
2149 			/*
2150 			 * Values < 1 make no sense for any of these
2151 			 * options, so disallow them.
2152 			 */
2153 			if (optval < 1) {
2154 				error = EINVAL;
2155 				goto bad;
2156 			}
2157 
2158 			switch (sopt->sopt_name) {
2159 			case SO_SNDBUF:
2160 			case SO_RCVBUF:
2161 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2162 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2163 				    so,
2164 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2165 					error = ENOBUFS;
2166 					goto bad;
2167 				}
2168 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2169 						&so->so_snd : &so->so_rcv;
2170 				atomic_clear_int(&sotmp->ssb_flags,
2171 						 SSB_AUTOSIZE);
2172 				break;
2173 
2174 			/*
2175 			 * Make sure the low-water is never greater than
2176 			 * the high-water.
2177 			 */
2178 			case SO_SNDLOWAT:
2179 				so->so_snd.ssb_lowat =
2180 				    (optval > so->so_snd.ssb_hiwat) ?
2181 				    so->so_snd.ssb_hiwat : optval;
2182 				atomic_clear_int(&so->so_snd.ssb_flags,
2183 						 SSB_AUTOLOWAT);
2184 				break;
2185 			case SO_RCVLOWAT:
2186 				so->so_rcv.ssb_lowat =
2187 				    (optval > so->so_rcv.ssb_hiwat) ?
2188 				    so->so_rcv.ssb_hiwat : optval;
2189 				atomic_clear_int(&so->so_rcv.ssb_flags,
2190 						 SSB_AUTOLOWAT);
2191 				break;
2192 			}
2193 			break;
2194 
2195 		case SO_SNDTIMEO:
2196 		case SO_RCVTIMEO:
2197 			error = sooptcopyin(sopt, &tv, sizeof tv,
2198 					    sizeof tv);
2199 			if (error)
2200 				goto bad;
2201 
2202 			/* assert(hz > 0); */
2203 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2204 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2205 				error = EDOM;
2206 				goto bad;
2207 			}
2208 			/* assert(tick > 0); */
2209 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2210 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2211 			if (val > INT_MAX) {
2212 				error = EDOM;
2213 				goto bad;
2214 			}
2215 			if (val == 0 && tv.tv_usec != 0)
2216 				val = 1;
2217 
2218 			switch (sopt->sopt_name) {
2219 			case SO_SNDTIMEO:
2220 				so->so_snd.ssb_timeo = val;
2221 				break;
2222 			case SO_RCVTIMEO:
2223 				so->so_rcv.ssb_timeo = val;
2224 				break;
2225 			}
2226 			break;
2227 		default:
2228 			error = ENOPROTOOPT;
2229 			break;
2230 		}
2231 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2232 			(void) so_pr_ctloutput(so, sopt);
2233 		}
2234 	}
2235 bad:
2236 	return (error);
2237 }
2238 
2239 /* Helper routine for getsockopt */
2240 int
2241 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2242 {
2243 	soopt_from_kbuf(sopt, buf, len);
2244 	return 0;
2245 }
2246 
2247 void
2248 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2249 {
2250 	size_t	valsize;
2251 
2252 	if (len == 0) {
2253 		sopt->sopt_valsize = 0;
2254 		return;
2255 	}
2256 
2257 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2258 	KKASSERT(kva_p(buf));
2259 
2260 	/*
2261 	 * Documented get behavior is that we always return a value,
2262 	 * possibly truncated to fit in the user's buffer.
2263 	 * Traditional behavior is that we always tell the user
2264 	 * precisely how much we copied, rather than something useful
2265 	 * like the total amount we had available for her.
2266 	 * Note that this interface is not idempotent; the entire answer must
2267 	 * generated ahead of time.
2268 	 */
2269 	valsize = szmin(len, sopt->sopt_valsize);
2270 	sopt->sopt_valsize = valsize;
2271 	if (sopt->sopt_val != 0) {
2272 		bcopy(buf, sopt->sopt_val, valsize);
2273 	}
2274 }
2275 
2276 int
2277 sogetopt(struct socket *so, struct sockopt *sopt)
2278 {
2279 	int	error, optval;
2280 	long	optval_l;
2281 	struct	linger l;
2282 	struct	timeval tv;
2283 #ifdef INET
2284 	struct accept_filter_arg *afap;
2285 #endif
2286 
2287 	error = 0;
2288 	sopt->sopt_dir = SOPT_GET;
2289 	if (sopt->sopt_level != SOL_SOCKET) {
2290 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2291 			return (so_pr_ctloutput(so, sopt));
2292 		} else
2293 			return (ENOPROTOOPT);
2294 	} else {
2295 		switch (sopt->sopt_name) {
2296 #ifdef INET
2297 		case SO_ACCEPTFILTER:
2298 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2299 				return (EINVAL);
2300 			afap = kmalloc(sizeof(*afap), M_TEMP,
2301 				       M_WAITOK | M_ZERO);
2302 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2303 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2304 				if (so->so_accf->so_accept_filter_str != NULL)
2305 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2306 			}
2307 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2308 			kfree(afap, M_TEMP);
2309 			break;
2310 #endif /* INET */
2311 
2312 		case SO_LINGER:
2313 			l.l_onoff = so->so_options & SO_LINGER;
2314 			l.l_linger = so->so_linger;
2315 			error = sooptcopyout(sopt, &l, sizeof l);
2316 			break;
2317 
2318 		case SO_USELOOPBACK:
2319 		case SO_DONTROUTE:
2320 		case SO_DEBUG:
2321 		case SO_KEEPALIVE:
2322 		case SO_REUSEADDR:
2323 		case SO_REUSEPORT:
2324 		case SO_BROADCAST:
2325 		case SO_OOBINLINE:
2326 		case SO_TIMESTAMP:
2327 		case SO_NOSIGPIPE:
2328 			optval = so->so_options & sopt->sopt_name;
2329 integer:
2330 			error = sooptcopyout(sopt, &optval, sizeof optval);
2331 			break;
2332 
2333 		case SO_TYPE:
2334 			optval = so->so_type;
2335 			goto integer;
2336 
2337 		case SO_ERROR:
2338 			optval = so->so_error;
2339 			so->so_error = 0;
2340 			goto integer;
2341 
2342 		case SO_SNDBUF:
2343 			optval = so->so_snd.ssb_hiwat;
2344 			goto integer;
2345 
2346 		case SO_RCVBUF:
2347 			optval = so->so_rcv.ssb_hiwat;
2348 			goto integer;
2349 
2350 		case SO_SNDLOWAT:
2351 			optval = so->so_snd.ssb_lowat;
2352 			goto integer;
2353 
2354 		case SO_RCVLOWAT:
2355 			optval = so->so_rcv.ssb_lowat;
2356 			goto integer;
2357 
2358 		case SO_SNDTIMEO:
2359 		case SO_RCVTIMEO:
2360 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2361 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2362 
2363 			tv.tv_sec = optval / hz;
2364 			tv.tv_usec = (optval % hz) * ustick;
2365 			error = sooptcopyout(sopt, &tv, sizeof tv);
2366 			break;
2367 
2368 		case SO_SNDSPACE:
2369 			optval_l = ssb_space(&so->so_snd);
2370 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2371 			break;
2372 
2373 		case SO_CPUHINT:
2374 			optval = -1; /* no hint */
2375 			goto integer;
2376 
2377 		default:
2378 			error = ENOPROTOOPT;
2379 			break;
2380 		}
2381 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2382 			so_pr_ctloutput(so, sopt);
2383 		return (error);
2384 	}
2385 }
2386 
2387 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2388 int
2389 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2390 {
2391 	struct mbuf *m, *m_prev;
2392 	int sopt_size = sopt->sopt_valsize, msize;
2393 
2394 	m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2395 		   0, &msize);
2396 	if (m == NULL)
2397 		return (ENOBUFS);
2398 	m->m_len = min(msize, sopt_size);
2399 	sopt_size -= m->m_len;
2400 	*mp = m;
2401 	m_prev = m;
2402 
2403 	while (sopt_size > 0) {
2404 		m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2405 			   MT_DATA, 0, &msize);
2406 		if (m == NULL) {
2407 			m_freem(*mp);
2408 			return (ENOBUFS);
2409 		}
2410 		m->m_len = min(msize, sopt_size);
2411 		sopt_size -= m->m_len;
2412 		m_prev->m_next = m;
2413 		m_prev = m;
2414 	}
2415 	return (0);
2416 }
2417 
2418 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2419 int
2420 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2421 {
2422 	soopt_to_mbuf(sopt, m);
2423 	return 0;
2424 }
2425 
2426 void
2427 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2428 {
2429 	size_t valsize;
2430 	void *val;
2431 
2432 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2433 	KKASSERT(kva_p(m));
2434 	if (sopt->sopt_val == NULL)
2435 		return;
2436 	val = sopt->sopt_val;
2437 	valsize = sopt->sopt_valsize;
2438 	while (m != NULL && valsize >= m->m_len) {
2439 		bcopy(val, mtod(m, char *), m->m_len);
2440 		valsize -= m->m_len;
2441 		val = (caddr_t)val + m->m_len;
2442 		m = m->m_next;
2443 	}
2444 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2445 		panic("ip6_sooptmcopyin");
2446 }
2447 
2448 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2449 int
2450 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2451 {
2452 	return soopt_from_mbuf(sopt, m);
2453 }
2454 
2455 int
2456 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2457 {
2458 	struct mbuf *m0 = m;
2459 	size_t valsize = 0;
2460 	size_t maxsize;
2461 	void *val;
2462 
2463 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2464 	KKASSERT(kva_p(m));
2465 	if (sopt->sopt_val == NULL)
2466 		return 0;
2467 	val = sopt->sopt_val;
2468 	maxsize = sopt->sopt_valsize;
2469 	while (m != NULL && maxsize >= m->m_len) {
2470 		bcopy(mtod(m, char *), val, m->m_len);
2471 	       maxsize -= m->m_len;
2472 	       val = (caddr_t)val + m->m_len;
2473 	       valsize += m->m_len;
2474 	       m = m->m_next;
2475 	}
2476 	if (m != NULL) {
2477 		/* enough soopt buffer should be given from user-land */
2478 		m_freem(m0);
2479 		return (EINVAL);
2480 	}
2481 	sopt->sopt_valsize = valsize;
2482 	return 0;
2483 }
2484 
2485 void
2486 sohasoutofband(struct socket *so)
2487 {
2488 	if (so->so_sigio != NULL)
2489 		pgsigio(so->so_sigio, SIGURG, 0);
2490 	/*
2491 	 * NOTE:
2492 	 * There is no need to use NOTE_OOB as KNOTE hint here:
2493 	 * soread filter depends on so_oobmark and SS_RCVATMARK
2494 	 * so_state.  NOTE_OOB would cause unnecessary penalty
2495 	 * in KNOTE, if there was knote processing contention.
2496 	 */
2497 	KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2498 }
2499 
2500 int
2501 sokqfilter(struct file *fp, struct knote *kn)
2502 {
2503 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2504 	struct signalsockbuf *ssb;
2505 
2506 	switch (kn->kn_filter) {
2507 	case EVFILT_READ:
2508 		if (so->so_options & SO_ACCEPTCONN)
2509 			kn->kn_fop = &solisten_filtops;
2510 		else
2511 			kn->kn_fop = &soread_filtops;
2512 		ssb = &so->so_rcv;
2513 		break;
2514 	case EVFILT_WRITE:
2515 		kn->kn_fop = &sowrite_filtops;
2516 		ssb = &so->so_snd;
2517 		break;
2518 	case EVFILT_EXCEPT:
2519 		kn->kn_fop = &soexcept_filtops;
2520 		ssb = &so->so_rcv;
2521 		break;
2522 	default:
2523 		return (EOPNOTSUPP);
2524 	}
2525 
2526 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2527 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2528 	return (0);
2529 }
2530 
2531 static void
2532 filt_sordetach(struct knote *kn)
2533 {
2534 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2535 
2536 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2537 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2538 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2539 }
2540 
2541 /*ARGSUSED*/
2542 static int
2543 filt_soread(struct knote *kn, long hint __unused)
2544 {
2545 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2546 
2547 	if (kn->kn_sfflags & NOTE_OOB) {
2548 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2549 			kn->kn_fflags |= NOTE_OOB;
2550 			return (1);
2551 		}
2552 		return (0);
2553 	}
2554 	kn->kn_data = so->so_rcv.ssb_cc;
2555 
2556 	if (so->so_state & SS_CANTRCVMORE) {
2557 		/*
2558 		 * Only set NODATA if all data has been exhausted.
2559 		 */
2560 		if (kn->kn_data == 0)
2561 			kn->kn_flags |= EV_NODATA;
2562 		kn->kn_flags |= EV_EOF;
2563 		kn->kn_fflags = so->so_error;
2564 		return (1);
2565 	}
2566 	if (so->so_error)	/* temporary udp error */
2567 		return (1);
2568 	if (kn->kn_sfflags & NOTE_LOWAT)
2569 		return (kn->kn_data >= kn->kn_sdata);
2570 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2571 		!TAILQ_EMPTY(&so->so_comp));
2572 }
2573 
2574 static void
2575 filt_sowdetach(struct knote *kn)
2576 {
2577 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2578 
2579 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2580 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2581 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2582 }
2583 
2584 /*ARGSUSED*/
2585 static int
2586 filt_sowrite(struct knote *kn, long hint __unused)
2587 {
2588 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2589 
2590 	if (so->so_snd.ssb_flags & SSB_PREALLOC)
2591 		kn->kn_data = ssb_space_prealloc(&so->so_snd);
2592 	else
2593 		kn->kn_data = ssb_space(&so->so_snd);
2594 
2595 	if (so->so_state & SS_CANTSENDMORE) {
2596 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2597 		kn->kn_fflags = so->so_error;
2598 		return (1);
2599 	}
2600 	if (so->so_error)	/* temporary udp error */
2601 		return (1);
2602 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2603 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2604 		return (0);
2605 	if (kn->kn_sfflags & NOTE_LOWAT)
2606 		return (kn->kn_data >= kn->kn_sdata);
2607 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2608 }
2609 
2610 /*ARGSUSED*/
2611 static int
2612 filt_solisten(struct knote *kn, long hint __unused)
2613 {
2614 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2615 
2616 	kn->kn_data = so->so_qlen;
2617 	return (! TAILQ_EMPTY(&so->so_comp));
2618 }
2619