xref: /dflybsd-src/sys/kern/uipc_socket.c (revision 3cc0afc616515a4d253a4a2bf5437858ae934c06)
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 	KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
290 	if (so->so_rcv.ssb_hiwat)
291 		(void)chgsbsize(so->so_cred->cr_uidinfo,
292 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
293 	if (so->so_snd.ssb_hiwat)
294 		(void)chgsbsize(so->so_cred->cr_uidinfo,
295 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
296 #ifdef INET
297 	/* remove accept filter if present */
298 	if (so->so_accf != NULL)
299 		do_setopt_accept_filter(so, NULL);
300 #endif /* INET */
301 	crfree(so->so_cred);
302 	if (so->so_faddr != NULL)
303 		kfree(so->so_faddr, M_SONAME);
304 	kfree(so, M_SOCKET);
305 }
306 
307 int
308 solisten(struct socket *so, int backlog, struct thread *td)
309 {
310 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
311 		return (EINVAL);
312 
313 	lwkt_gettoken(&so->so_rcv.ssb_token);
314 	if (TAILQ_EMPTY(&so->so_comp))
315 		so->so_options |= SO_ACCEPTCONN;
316 	lwkt_reltoken(&so->so_rcv.ssb_token);
317 	if (backlog < 0 || backlog > somaxconn)
318 		backlog = somaxconn;
319 	so->so_qlimit = backlog;
320 	return so_pru_listen(so, td);
321 }
322 
323 /*
324  * Destroy a disconnected socket.  This routine is a NOP if entities
325  * still have a reference on the socket:
326  *
327  *	so_pcb -	The protocol stack still has a reference
328  *	SS_NOFDREF -	There is no longer a file pointer reference
329  */
330 void
331 sofree(struct socket *so)
332 {
333 	struct socket *head;
334 
335 	/*
336 	 * This is a bit hackish at the moment.  We need to interlock
337 	 * any accept queue we are on before we potentially lose the
338 	 * last reference to avoid races against a re-reference from
339 	 * someone operating on the queue.
340 	 */
341 	while ((head = so->so_head) != NULL) {
342 		lwkt_getpooltoken(head);
343 		if (so->so_head == head)
344 			break;
345 		lwkt_relpooltoken(head);
346 	}
347 
348 	/*
349 	 * Arbitrage the last free.
350 	 */
351 	KKASSERT(so->so_refs > 0);
352 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
353 		if (head)
354 			lwkt_relpooltoken(head);
355 		return;
356 	}
357 
358 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
359 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
360 
361 	/*
362 	 * We're done, remove ourselves from the accept queue we are
363 	 * on, if we are on one.
364 	 */
365 	if (head != NULL) {
366 		if (so->so_state & SS_INCOMP) {
367 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
368 			    SS_INCOMP);
369 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
370 			head->so_incqlen--;
371 		} else if (so->so_state & SS_COMP) {
372 			/*
373 			 * We must not decommission a socket that's
374 			 * on the accept(2) queue.  If we do, then
375 			 * accept(2) may hang after select(2) indicated
376 			 * that the listening socket was ready.
377 			 */
378 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
379 			    SS_COMP);
380 			lwkt_relpooltoken(head);
381 			return;
382 		} else {
383 			panic("sofree: not queued");
384 		}
385 		soclrstate(so, SS_INCOMP);
386 		so->so_head = NULL;
387 		lwkt_relpooltoken(head);
388 	}
389 	ssb_release(&so->so_snd, so);
390 	sorflush(so);
391 	sodealloc(so);
392 }
393 
394 /*
395  * Close a socket on last file table reference removal.
396  * Initiate disconnect if connected.
397  * Free socket when disconnect complete.
398  */
399 int
400 soclose(struct socket *so, int fflag)
401 {
402 	int error;
403 
404 	funsetown(&so->so_sigio);
405 	sosetstate(so, SS_ISCLOSING);
406 	if (!use_soclose_fast ||
407 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
408 	    ((so->so_state & SS_ISCONNECTED) &&
409 	     (so->so_options & SO_LINGER))) {
410 		error = soclose_sync(so, fflag);
411 	} else {
412 		soclose_fast(so);
413 		error = 0;
414 	}
415 	return error;
416 }
417 
418 void
419 sodiscard(struct socket *so)
420 {
421 	lwkt_getpooltoken(so);
422 	if (so->so_options & SO_ACCEPTCONN) {
423 		struct socket *sp;
424 
425 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
426 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
427 			    SS_INCOMP);
428 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
429 			so->so_incqlen--;
430 			soclrstate(sp, SS_INCOMP);
431 			soabort_async(sp, TRUE);
432 		}
433 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
434 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
435 			    SS_COMP);
436 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
437 			so->so_qlen--;
438 			soclrstate(sp, SS_COMP);
439 			soabort_async(sp, TRUE);
440 		}
441 	}
442 	lwkt_relpooltoken(so);
443 
444 	if (so->so_state & SS_NOFDREF)
445 		panic("soclose: NOFDREF");
446 	sosetstate(so, SS_NOFDREF);	/* take ref */
447 }
448 
449 /*
450  * Append the completed queue of head to head_inh (inherting listen socket).
451  */
452 void
453 soinherit(struct socket *head, struct socket *head_inh)
454 {
455 	boolean_t do_wakeup = FALSE;
456 
457 	KASSERT(head->so_options & SO_ACCEPTCONN,
458 	    ("head does not accept connection"));
459 	KASSERT(head_inh->so_options & SO_ACCEPTCONN,
460 	    ("head_inh does not accept connection"));
461 
462 	lwkt_getpooltoken(head);
463 	lwkt_getpooltoken(head_inh);
464 
465 	if (head->so_qlen > 0)
466 		do_wakeup = TRUE;
467 
468 	while (!TAILQ_EMPTY(&head->so_comp)) {
469 		struct ucred *old_cr;
470 		struct socket *sp;
471 
472 		sp = TAILQ_FIRST(&head->so_comp);
473 		KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
474 
475 		/*
476 		 * Remove this socket from the current listen socket
477 		 * completed queue.
478 		 */
479 		TAILQ_REMOVE(&head->so_comp, sp, so_list);
480 		head->so_qlen--;
481 
482 		/* Save the old ucred for later free. */
483 		old_cr = sp->so_cred;
484 
485 		/*
486 		 * Install this socket to the inheriting listen socket
487 		 * completed queue.
488 		 */
489 		sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
490 		sp->so_head = head_inh;
491 
492 		TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
493 		head_inh->so_qlen++;
494 
495 		/*
496 		 * NOTE:
497 		 * crfree() may block and release the tokens temporarily.
498 		 * However, we are fine here, since the transition is done.
499 		 */
500 		crfree(old_cr);
501 	}
502 
503 	lwkt_relpooltoken(head_inh);
504 	lwkt_relpooltoken(head);
505 
506 	if (do_wakeup) {
507 		/*
508 		 * "New" connections have arrived
509 		 */
510 		sorwakeup(head_inh);
511 		wakeup(&head_inh->so_timeo);
512 	}
513 }
514 
515 static int
516 soclose_sync(struct socket *so, int fflag)
517 {
518 	int error = 0;
519 
520 	if (so->so_pcb == NULL)
521 		goto discard;
522 	if (so->so_state & SS_ISCONNECTED) {
523 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
524 			error = sodisconnect(so);
525 			if (error)
526 				goto drop;
527 		}
528 		if (so->so_options & SO_LINGER) {
529 			if ((so->so_state & SS_ISDISCONNECTING) &&
530 			    (fflag & FNONBLOCK))
531 				goto drop;
532 			while (so->so_state & SS_ISCONNECTED) {
533 				error = tsleep(&so->so_timeo, PCATCH,
534 					       "soclos", so->so_linger * hz);
535 				if (error)
536 					break;
537 			}
538 		}
539 	}
540 drop:
541 	if (so->so_pcb) {
542 		int error2;
543 
544 		error2 = so_pru_detach(so);
545 		if (error2 == EJUSTRETURN) {
546 			/*
547 			 * Protocol will call sodiscard()
548 			 * and sofree() for us.
549 			 */
550 			return error;
551 		}
552 		if (error == 0)
553 			error = error2;
554 	}
555 discard:
556 	sodiscard(so);
557 	so_pru_sync(so);	/* unpend async sending */
558 	sofree(so);		/* dispose of ref */
559 
560 	return (error);
561 }
562 
563 static void
564 soclose_sofree_async_handler(netmsg_t msg)
565 {
566 	sofree(msg->base.nm_so);
567 }
568 
569 static void
570 soclose_sofree_async(struct socket *so)
571 {
572 	struct netmsg_base *base = &so->so_clomsg;
573 
574 	netmsg_init(base, so, &netisr_apanic_rport, 0,
575 	    soclose_sofree_async_handler);
576 	lwkt_sendmsg(so->so_port, &base->lmsg);
577 }
578 
579 static void
580 soclose_disconn_async_handler(netmsg_t msg)
581 {
582 	struct socket *so = msg->base.nm_so;
583 
584 	if ((so->so_state & SS_ISCONNECTED) &&
585 	    (so->so_state & SS_ISDISCONNECTING) == 0)
586 		so_pru_disconnect_direct(so);
587 
588 	if (so->so_pcb) {
589 		int error;
590 
591 		error = so_pru_detach_direct(so);
592 		if (error == EJUSTRETURN) {
593 			/*
594 			 * Protocol will call sodiscard()
595 			 * and sofree() for us.
596 			 */
597 			return;
598 		}
599 	}
600 
601 	sodiscard(so);
602 	sofree(so);
603 }
604 
605 static void
606 soclose_disconn_async(struct socket *so)
607 {
608 	struct netmsg_base *base = &so->so_clomsg;
609 
610 	netmsg_init(base, so, &netisr_apanic_rport, 0,
611 	    soclose_disconn_async_handler);
612 	lwkt_sendmsg(so->so_port, &base->lmsg);
613 }
614 
615 static void
616 soclose_detach_async_handler(netmsg_t msg)
617 {
618 	struct socket *so = msg->base.nm_so;
619 
620 	if (so->so_pcb) {
621 		int error;
622 
623 		error = so_pru_detach_direct(so);
624 		if (error == EJUSTRETURN) {
625 			/*
626 			 * Protocol will call sodiscard()
627 			 * and sofree() for us.
628 			 */
629 			return;
630 		}
631 	}
632 
633 	sodiscard(so);
634 	sofree(so);
635 }
636 
637 static void
638 soclose_detach_async(struct socket *so)
639 {
640 	struct netmsg_base *base = &so->so_clomsg;
641 
642 	netmsg_init(base, so, &netisr_apanic_rport, 0,
643 	    soclose_detach_async_handler);
644 	lwkt_sendmsg(so->so_port, &base->lmsg);
645 }
646 
647 static void
648 soclose_fast(struct socket *so)
649 {
650 	if (so->so_pcb == NULL)
651 		goto discard;
652 
653 	if ((so->so_state & SS_ISCONNECTED) &&
654 	    (so->so_state & SS_ISDISCONNECTING) == 0) {
655 		soclose_disconn_async(so);
656 		return;
657 	}
658 
659 	if (so->so_pcb) {
660 		soclose_detach_async(so);
661 		return;
662 	}
663 
664 discard:
665 	sodiscard(so);
666 	soclose_sofree_async(so);
667 }
668 
669 /*
670  * Abort and destroy a socket.  Only one abort can be in progress
671  * at any given moment.
672  */
673 void
674 soabort_async(struct socket *so, boolean_t clr_head)
675 {
676 	/*
677 	 * Keep a reference before clearing the so_head
678 	 * to avoid racing socket close in netisr.
679 	 */
680 	soreference(so);
681 	if (clr_head)
682 		so->so_head = NULL;
683 	so_pru_abort_async(so);
684 }
685 
686 void
687 soabort_oncpu(struct socket *so)
688 {
689 	soreference(so);
690 	so_pru_abort_direct(so);
691 }
692 
693 /*
694  * so is passed in ref'd, which becomes owned by
695  * the cleared SS_NOFDREF flag.
696  */
697 void
698 soaccept_generic(struct socket *so)
699 {
700 	if ((so->so_state & SS_NOFDREF) == 0)
701 		panic("soaccept: !NOFDREF");
702 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
703 }
704 
705 int
706 soaccept(struct socket *so, struct sockaddr **nam)
707 {
708 	int error;
709 
710 	soaccept_generic(so);
711 	error = so_pru_accept(so, nam);
712 	return (error);
713 }
714 
715 int
716 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
717     boolean_t sync)
718 {
719 	int error;
720 
721 	if (so->so_options & SO_ACCEPTCONN)
722 		return (EOPNOTSUPP);
723 	/*
724 	 * If protocol is connection-based, can only connect once.
725 	 * Otherwise, if connected, try to disconnect first.
726 	 * This allows user to disconnect by connecting to, e.g.,
727 	 * a null address.
728 	 */
729 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
730 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
731 	    (error = sodisconnect(so)))) {
732 		error = EISCONN;
733 	} else {
734 		/*
735 		 * Prevent accumulated error from previous connection
736 		 * from biting us.
737 		 */
738 		so->so_error = 0;
739 		if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
740 			error = so_pru_connect_async(so, nam, td);
741 		else
742 			error = so_pru_connect(so, nam, td);
743 	}
744 	return (error);
745 }
746 
747 int
748 soconnect2(struct socket *so1, struct socket *so2)
749 {
750 	int error;
751 
752 	error = so_pru_connect2(so1, so2);
753 	return (error);
754 }
755 
756 int
757 sodisconnect(struct socket *so)
758 {
759 	int error;
760 
761 	if ((so->so_state & SS_ISCONNECTED) == 0) {
762 		error = ENOTCONN;
763 		goto bad;
764 	}
765 	if (so->so_state & SS_ISDISCONNECTING) {
766 		error = EALREADY;
767 		goto bad;
768 	}
769 	error = so_pru_disconnect(so);
770 bad:
771 	return (error);
772 }
773 
774 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
775 /*
776  * Send on a socket.
777  * If send must go all at once and message is larger than
778  * send buffering, then hard error.
779  * Lock against other senders.
780  * If must go all at once and not enough room now, then
781  * inform user that this would block and do nothing.
782  * Otherwise, if nonblocking, send as much as possible.
783  * The data to be sent is described by "uio" if nonzero,
784  * otherwise by the mbuf chain "top" (which must be null
785  * if uio is not).  Data provided in mbuf chain must be small
786  * enough to send all at once.
787  *
788  * Returns nonzero on error, timeout or signal; callers
789  * must check for short counts if EINTR/ERESTART are returned.
790  * Data and control buffers are freed on return.
791  */
792 int
793 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
794 	struct mbuf *top, struct mbuf *control, int flags,
795 	struct thread *td)
796 {
797 	struct mbuf **mp;
798 	struct mbuf *m;
799 	size_t resid;
800 	int space, len;
801 	int clen = 0, error, dontroute, mlen;
802 	int atomic = sosendallatonce(so) || top;
803 	int pru_flags;
804 
805 	if (uio) {
806 		resid = uio->uio_resid;
807 	} else {
808 		resid = (size_t)top->m_pkthdr.len;
809 #ifdef INVARIANTS
810 		len = 0;
811 		for (m = top; m; m = m->m_next)
812 			len += m->m_len;
813 		KKASSERT(top->m_pkthdr.len == len);
814 #endif
815 	}
816 
817 	/*
818 	 * WARNING!  resid is unsigned, space and len are signed.  space
819 	 * 	     can wind up negative if the sockbuf is overcommitted.
820 	 *
821 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
822 	 * type sockets since that's an error.
823 	 */
824 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
825 		error = EINVAL;
826 		goto out;
827 	}
828 
829 	dontroute =
830 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
831 	    (so->so_proto->pr_flags & PR_ATOMIC);
832 	if (td->td_lwp != NULL)
833 		td->td_lwp->lwp_ru.ru_msgsnd++;
834 	if (control)
835 		clen = control->m_len;
836 #define	gotoerr(errcode)	{ error = errcode; goto release; }
837 
838 restart:
839 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
840 	if (error)
841 		goto out;
842 
843 	do {
844 		if (so->so_state & SS_CANTSENDMORE)
845 			gotoerr(EPIPE);
846 		if (so->so_error) {
847 			error = so->so_error;
848 			so->so_error = 0;
849 			goto release;
850 		}
851 		if ((so->so_state & SS_ISCONNECTED) == 0) {
852 			/*
853 			 * `sendto' and `sendmsg' is allowed on a connection-
854 			 * based socket if it supports implied connect.
855 			 * Return ENOTCONN if not connected and no address is
856 			 * supplied.
857 			 */
858 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
859 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
860 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
861 				    !(resid == 0 && clen != 0))
862 					gotoerr(ENOTCONN);
863 			} else if (addr == NULL)
864 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
865 				   ENOTCONN : EDESTADDRREQ);
866 		}
867 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
868 		    clen > so->so_snd.ssb_hiwat) {
869 			gotoerr(EMSGSIZE);
870 		}
871 		space = ssb_space(&so->so_snd);
872 		if (flags & MSG_OOB)
873 			space += 1024;
874 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
875 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
876 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
877 				gotoerr(EWOULDBLOCK);
878 			ssb_unlock(&so->so_snd);
879 			error = ssb_wait(&so->so_snd);
880 			if (error)
881 				goto out;
882 			goto restart;
883 		}
884 		mp = &top;
885 		space -= clen;
886 		do {
887 		    if (uio == NULL) {
888 			/*
889 			 * Data is prepackaged in "top".
890 			 */
891 			resid = 0;
892 			if (flags & MSG_EOR)
893 				top->m_flags |= M_EOR;
894 		    } else do {
895 			if (resid > INT_MAX)
896 				resid = INT_MAX;
897 			m = m_getl((int)resid, M_WAITOK, MT_DATA,
898 				   top == NULL ? M_PKTHDR : 0, &mlen);
899 			if (top == NULL) {
900 				m->m_pkthdr.len = 0;
901 				m->m_pkthdr.rcvif = NULL;
902 			}
903 			len = imin((int)szmin(mlen, resid), space);
904 			if (resid < MINCLSIZE) {
905 				/*
906 				 * For datagram protocols, leave room
907 				 * for protocol headers in first mbuf.
908 				 */
909 				if (atomic && top == NULL && len < mlen)
910 					MH_ALIGN(m, len);
911 			}
912 			space -= len;
913 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
914 			resid = uio->uio_resid;
915 			m->m_len = len;
916 			*mp = m;
917 			top->m_pkthdr.len += len;
918 			if (error)
919 				goto release;
920 			mp = &m->m_next;
921 			if (resid == 0) {
922 				if (flags & MSG_EOR)
923 					top->m_flags |= M_EOR;
924 				break;
925 			}
926 		    } while (space > 0 && atomic);
927 		    if (dontroute)
928 			    so->so_options |= SO_DONTROUTE;
929 		    if (flags & MSG_OOB) {
930 		    	    pru_flags = PRUS_OOB;
931 		    } else if ((flags & MSG_EOF) &&
932 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
933 			       (resid == 0)) {
934 			    /*
935 			     * If the user set MSG_EOF, the protocol
936 			     * understands this flag and nothing left to
937 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
938 			     */
939 		    	    pru_flags = PRUS_EOF;
940 		    } else if (resid > 0 && space > 0) {
941 			    /* If there is more to send, set PRUS_MORETOCOME */
942 		    	    pru_flags = PRUS_MORETOCOME;
943 		    } else {
944 		    	    pru_flags = 0;
945 		    }
946 		    /*
947 		     * XXX all the SS_CANTSENDMORE checks previously
948 		     * done could be out of date.  We could have recieved
949 		     * a reset packet in an interrupt or maybe we slept
950 		     * while doing page faults in uiomove() etc. We could
951 		     * probably recheck again inside the splnet() protection
952 		     * here, but there are probably other places that this
953 		     * also happens.  We must rethink this.
954 		     */
955 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
956 		    if (dontroute)
957 			    so->so_options &= ~SO_DONTROUTE;
958 		    clen = 0;
959 		    control = NULL;
960 		    top = NULL;
961 		    mp = &top;
962 		    if (error)
963 			    goto release;
964 		} while (resid && space > 0);
965 	} while (resid);
966 
967 release:
968 	ssb_unlock(&so->so_snd);
969 out:
970 	if (top)
971 		m_freem(top);
972 	if (control)
973 		m_freem(control);
974 	return (error);
975 }
976 
977 #ifdef INET
978 /*
979  * A specialization of sosend() for UDP based on protocol-specific knowledge:
980  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
981  *	sosendallatonce() returns true,
982  *	the "atomic" variable is true,
983  *	and sosendudp() blocks until space is available for the entire send.
984  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
985  *	PR_IMPLOPCL flags set.
986  *   UDP has no out-of-band data.
987  *   UDP has no control data.
988  *   UDP does not support MSG_EOR.
989  */
990 int
991 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
992 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
993 {
994 	size_t resid;
995 	int error, pru_flags = 0;
996 	int space;
997 
998 	if (td->td_lwp != NULL)
999 		td->td_lwp->lwp_ru.ru_msgsnd++;
1000 	if (control)
1001 		m_freem(control);
1002 
1003 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
1004 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
1005 
1006 restart:
1007 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1008 	if (error)
1009 		goto out;
1010 
1011 	if (so->so_state & SS_CANTSENDMORE)
1012 		gotoerr(EPIPE);
1013 	if (so->so_error) {
1014 		error = so->so_error;
1015 		so->so_error = 0;
1016 		goto release;
1017 	}
1018 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
1019 		gotoerr(EDESTADDRREQ);
1020 	if (resid > so->so_snd.ssb_hiwat)
1021 		gotoerr(EMSGSIZE);
1022 	space = ssb_space(&so->so_snd);
1023 	if (uio && (space < 0 || (size_t)space < resid)) {
1024 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1025 			gotoerr(EWOULDBLOCK);
1026 		ssb_unlock(&so->so_snd);
1027 		error = ssb_wait(&so->so_snd);
1028 		if (error)
1029 			goto out;
1030 		goto restart;
1031 	}
1032 
1033 	if (uio) {
1034 		int hdrlen = max_hdr;
1035 
1036 		/*
1037 		 * We try to optimize out the additional mbuf
1038 		 * allocations in M_PREPEND() on output path, e.g.
1039 		 * - udp_output(), when it tries to prepend protocol
1040 		 *   headers.
1041 		 * - Link layer output function, when it tries to
1042 		 *   prepend link layer header.
1043 		 *
1044 		 * This probably will not benefit any data that will
1045 		 * be fragmented, so this optimization is only performed
1046 		 * when the size of data and max size of protocol+link
1047 		 * headers fit into one mbuf cluster.
1048 		 */
1049 		if (uio->uio_resid > MCLBYTES - hdrlen ||
1050 		    !udp_sosend_prepend) {
1051 			top = m_uiomove(uio);
1052 			if (top == NULL)
1053 				goto release;
1054 		} else {
1055 			int nsize;
1056 
1057 			top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1058 			    MT_DATA, M_PKTHDR, &nsize);
1059 			KASSERT(nsize >= uio->uio_resid + hdrlen,
1060 			    ("sosendudp invalid nsize %d, "
1061 			     "resid %zu, hdrlen %d",
1062 			     nsize, uio->uio_resid, hdrlen));
1063 
1064 			top->m_len = uio->uio_resid;
1065 			top->m_pkthdr.len = uio->uio_resid;
1066 			top->m_data += hdrlen;
1067 
1068 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1069 			if (error)
1070 				goto out;
1071 		}
1072 	}
1073 
1074 	if (flags & MSG_DONTROUTE)
1075 		pru_flags |= PRUS_DONTROUTE;
1076 
1077 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1078 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1079 		error = 0;
1080 	} else {
1081 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1082 	}
1083 	top = NULL;		/* sent or freed in lower layer */
1084 
1085 release:
1086 	ssb_unlock(&so->so_snd);
1087 out:
1088 	if (top)
1089 		m_freem(top);
1090 	return (error);
1091 }
1092 
1093 int
1094 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1095 	struct mbuf *top, struct mbuf *control, int flags,
1096 	struct thread *td)
1097 {
1098 	struct mbuf **mp;
1099 	struct mbuf *m;
1100 	size_t resid;
1101 	int space, len;
1102 	int error, mlen;
1103 	int allatonce;
1104 	int pru_flags;
1105 
1106 	if (uio) {
1107 		KKASSERT(top == NULL);
1108 		allatonce = 0;
1109 		resid = uio->uio_resid;
1110 	} else {
1111 		allatonce = 1;
1112 		resid = (size_t)top->m_pkthdr.len;
1113 #ifdef INVARIANTS
1114 		len = 0;
1115 		for (m = top; m; m = m->m_next)
1116 			len += m->m_len;
1117 		KKASSERT(top->m_pkthdr.len == len);
1118 #endif
1119 	}
1120 
1121 	/*
1122 	 * WARNING!  resid is unsigned, space and len are signed.  space
1123 	 * 	     can wind up negative if the sockbuf is overcommitted.
1124 	 *
1125 	 * Also check to make sure that MSG_EOR isn't used on TCP
1126 	 */
1127 	if (flags & MSG_EOR) {
1128 		error = EINVAL;
1129 		goto out;
1130 	}
1131 
1132 	if (control) {
1133 		/* TCP doesn't do control messages (rights, creds, etc) */
1134 		if (control->m_len) {
1135 			error = EINVAL;
1136 			goto out;
1137 		}
1138 		m_freem(control);	/* empty control, just free it */
1139 		control = NULL;
1140 	}
1141 
1142 	if (td->td_lwp != NULL)
1143 		td->td_lwp->lwp_ru.ru_msgsnd++;
1144 
1145 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1146 
1147 restart:
1148 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1149 	if (error)
1150 		goto out;
1151 
1152 	do {
1153 		if (so->so_state & SS_CANTSENDMORE)
1154 			gotoerr(EPIPE);
1155 		if (so->so_error) {
1156 			error = so->so_error;
1157 			so->so_error = 0;
1158 			goto release;
1159 		}
1160 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1161 		    (so->so_state & SS_ISCONFIRMING) == 0)
1162 			gotoerr(ENOTCONN);
1163 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1164 			gotoerr(EMSGSIZE);
1165 
1166 		space = ssb_space_prealloc(&so->so_snd);
1167 		if (flags & MSG_OOB)
1168 			space += 1024;
1169 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1170 		    space < so->so_snd.ssb_lowat) {
1171 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1172 				gotoerr(EWOULDBLOCK);
1173 			ssb_unlock(&so->so_snd);
1174 			error = ssb_wait(&so->so_snd);
1175 			if (error)
1176 				goto out;
1177 			goto restart;
1178 		}
1179 		mp = &top;
1180 		do {
1181 		    int cnt = 0, async = 0;
1182 
1183 		    if (uio == NULL) {
1184 			/*
1185 			 * Data is prepackaged in "top".
1186 			 */
1187 			resid = 0;
1188 		    } else do {
1189 			if (resid > INT_MAX)
1190 				resid = INT_MAX;
1191 			if (tcp_sosend_jcluster) {
1192 				m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1193 					   top == NULL ? M_PKTHDR : 0, &mlen);
1194 			} else {
1195 				m = m_getl((int)resid, M_WAITOK, MT_DATA,
1196 					   top == NULL ? M_PKTHDR : 0, &mlen);
1197 			}
1198 			if (top == NULL) {
1199 				m->m_pkthdr.len = 0;
1200 				m->m_pkthdr.rcvif = NULL;
1201 			}
1202 			len = imin((int)szmin(mlen, resid), space);
1203 			space -= len;
1204 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1205 			resid = uio->uio_resid;
1206 			m->m_len = len;
1207 			*mp = m;
1208 			top->m_pkthdr.len += len;
1209 			if (error)
1210 				goto release;
1211 			mp = &m->m_next;
1212 			if (resid == 0)
1213 				break;
1214 			++cnt;
1215 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1216 
1217 		    if (tcp_sosend_async)
1218 			    async = 1;
1219 
1220 		    if (flags & MSG_OOB) {
1221 		    	    pru_flags = PRUS_OOB;
1222 			    async = 0;
1223 		    } else if ((flags & MSG_EOF) && resid == 0) {
1224 			    pru_flags = PRUS_EOF;
1225 		    } else if (resid > 0 && space > 0) {
1226 			    /* If there is more to send, set PRUS_MORETOCOME */
1227 		    	    pru_flags = PRUS_MORETOCOME;
1228 			    async = 1;
1229 		    } else {
1230 		    	    pru_flags = 0;
1231 		    }
1232 
1233 		    if (flags & MSG_SYNC)
1234 			    async = 0;
1235 
1236 		    /*
1237 		     * XXX all the SS_CANTSENDMORE checks previously
1238 		     * done could be out of date.  We could have recieved
1239 		     * a reset packet in an interrupt or maybe we slept
1240 		     * while doing page faults in uiomove() etc. We could
1241 		     * probably recheck again inside the splnet() protection
1242 		     * here, but there are probably other places that this
1243 		     * also happens.  We must rethink this.
1244 		     */
1245 		    for (m = top; m; m = m->m_next)
1246 			    ssb_preallocstream(&so->so_snd, m);
1247 		    if (!async) {
1248 			    error = so_pru_send(so, pru_flags, top,
1249 			        NULL, NULL, td);
1250 		    } else {
1251 			    so_pru_send_async(so, pru_flags, top,
1252 			        NULL, NULL, td);
1253 			    error = 0;
1254 		    }
1255 
1256 		    top = NULL;
1257 		    mp = &top;
1258 		    if (error)
1259 			    goto release;
1260 		} while (resid && space > 0);
1261 	} while (resid);
1262 
1263 release:
1264 	ssb_unlock(&so->so_snd);
1265 out:
1266 	if (top)
1267 		m_freem(top);
1268 	if (control)
1269 		m_freem(control);
1270 	return (error);
1271 }
1272 #endif
1273 
1274 /*
1275  * Implement receive operations on a socket.
1276  *
1277  * We depend on the way that records are added to the signalsockbuf
1278  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1279  * must begin with an address if the protocol so specifies,
1280  * followed by an optional mbuf or mbufs containing ancillary data,
1281  * and then zero or more mbufs of data.
1282  *
1283  * Although the signalsockbuf is locked, new data may still be appended.
1284  * A token inside the ssb_lock deals with MP issues and still allows
1285  * the network to access the socket if we block in a uio.
1286  *
1287  * The caller may receive the data as a single mbuf chain by supplying
1288  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1289  * only for the count in uio_resid.
1290  */
1291 int
1292 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1293 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1294 {
1295 	struct mbuf *m, *n;
1296 	struct mbuf *free_chain = NULL;
1297 	int flags, len, error, offset;
1298 	struct protosw *pr = so->so_proto;
1299 	int moff, type = 0;
1300 	size_t resid, orig_resid;
1301 
1302 	if (uio)
1303 		resid = uio->uio_resid;
1304 	else
1305 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1306 	orig_resid = resid;
1307 
1308 	if (psa)
1309 		*psa = NULL;
1310 	if (controlp)
1311 		*controlp = NULL;
1312 	if (flagsp)
1313 		flags = *flagsp &~ MSG_EOR;
1314 	else
1315 		flags = 0;
1316 	if (flags & MSG_OOB) {
1317 		m = m_get(M_WAITOK, MT_DATA);
1318 		if (m == NULL)
1319 			return (ENOBUFS);
1320 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1321 		if (error)
1322 			goto bad;
1323 		if (sio) {
1324 			do {
1325 				sbappend(sio, m);
1326 				KKASSERT(resid >= (size_t)m->m_len);
1327 				resid -= (size_t)m->m_len;
1328 			} while (resid > 0 && m);
1329 		} else {
1330 			do {
1331 				uio->uio_resid = resid;
1332 				error = uiomove(mtod(m, caddr_t),
1333 						(int)szmin(resid, m->m_len),
1334 						uio);
1335 				resid = uio->uio_resid;
1336 				m = m_free(m);
1337 			} while (uio->uio_resid && error == 0 && m);
1338 		}
1339 bad:
1340 		if (m)
1341 			m_freem(m);
1342 		return (error);
1343 	}
1344 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1345 		so_pru_rcvd(so, 0);
1346 
1347 	/*
1348 	 * The token interlocks against the protocol thread while
1349 	 * ssb_lock is a blocking lock against other userland entities.
1350 	 */
1351 	lwkt_gettoken(&so->so_rcv.ssb_token);
1352 restart:
1353 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1354 	if (error)
1355 		goto done;
1356 
1357 	m = so->so_rcv.ssb_mb;
1358 	/*
1359 	 * If we have less data than requested, block awaiting more
1360 	 * (subject to any timeout) if:
1361 	 *   1. the current count is less than the low water mark, or
1362 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1363 	 *	receive operation at once if we block (resid <= hiwat).
1364 	 *   3. MSG_DONTWAIT is not set
1365 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1366 	 * we have to do the receive in sections, and thus risk returning
1367 	 * a short count if a timeout or signal occurs after we start.
1368 	 */
1369 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1370 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1371 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1372 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1373 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1374 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1375 		if (so->so_error) {
1376 			if (m)
1377 				goto dontblock;
1378 			error = so->so_error;
1379 			if ((flags & MSG_PEEK) == 0)
1380 				so->so_error = 0;
1381 			goto release;
1382 		}
1383 		if (so->so_state & SS_CANTRCVMORE) {
1384 			if (m)
1385 				goto dontblock;
1386 			else
1387 				goto release;
1388 		}
1389 		for (; m; m = m->m_next) {
1390 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1391 				m = so->so_rcv.ssb_mb;
1392 				goto dontblock;
1393 			}
1394 		}
1395 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1396 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1397 			error = ENOTCONN;
1398 			goto release;
1399 		}
1400 		if (resid == 0)
1401 			goto release;
1402 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1403 			error = EWOULDBLOCK;
1404 			goto release;
1405 		}
1406 		ssb_unlock(&so->so_rcv);
1407 		error = ssb_wait(&so->so_rcv);
1408 		if (error)
1409 			goto done;
1410 		goto restart;
1411 	}
1412 dontblock:
1413 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1414 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1415 
1416 	/*
1417 	 * note: m should be == sb_mb here.  Cache the next record while
1418 	 * cleaning up.  Note that calling m_free*() will break out critical
1419 	 * section.
1420 	 */
1421 	KKASSERT(m == so->so_rcv.ssb_mb);
1422 
1423 	/*
1424 	 * Skip any address mbufs prepending the record.
1425 	 */
1426 	if (pr->pr_flags & PR_ADDR) {
1427 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1428 		orig_resid = 0;
1429 		if (psa)
1430 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1431 		if (flags & MSG_PEEK)
1432 			m = m->m_next;
1433 		else
1434 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1435 	}
1436 
1437 	/*
1438 	 * Skip any control mbufs prepending the record.
1439 	 */
1440 	while (m && m->m_type == MT_CONTROL && error == 0) {
1441 		if (flags & MSG_PEEK) {
1442 			if (controlp)
1443 				*controlp = m_copy(m, 0, m->m_len);
1444 			m = m->m_next;	/* XXX race */
1445 		} else {
1446 			if (controlp) {
1447 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1448 				if (pr->pr_domain->dom_externalize &&
1449 				    mtod(m, struct cmsghdr *)->cmsg_type ==
1450 				    SCM_RIGHTS)
1451 				   error = (*pr->pr_domain->dom_externalize)(m);
1452 				*controlp = m;
1453 				m = n;
1454 			} else {
1455 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1456 			}
1457 		}
1458 		if (controlp && *controlp) {
1459 			orig_resid = 0;
1460 			controlp = &(*controlp)->m_next;
1461 		}
1462 	}
1463 
1464 	/*
1465 	 * flag OOB data.
1466 	 */
1467 	if (m) {
1468 		type = m->m_type;
1469 		if (type == MT_OOBDATA)
1470 			flags |= MSG_OOB;
1471 	}
1472 
1473 	/*
1474 	 * Copy to the UIO or mbuf return chain (*mp).
1475 	 */
1476 	moff = 0;
1477 	offset = 0;
1478 	while (m && resid > 0 && error == 0) {
1479 		if (m->m_type == MT_OOBDATA) {
1480 			if (type != MT_OOBDATA)
1481 				break;
1482 		} else if (type == MT_OOBDATA)
1483 			break;
1484 		else
1485 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1486 			("receive 3"));
1487 		soclrstate(so, SS_RCVATMARK);
1488 		len = (resid > INT_MAX) ? INT_MAX : resid;
1489 		if (so->so_oobmark && len > so->so_oobmark - offset)
1490 			len = so->so_oobmark - offset;
1491 		if (len > m->m_len - moff)
1492 			len = m->m_len - moff;
1493 
1494 		/*
1495 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1496 		 * The SIO is dealt with when we eat the mbuf, but deal
1497 		 * with the resid here either way.
1498 		 */
1499 		if (uio) {
1500 			uio->uio_resid = resid;
1501 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1502 			resid = uio->uio_resid;
1503 			if (error)
1504 				goto release;
1505 		} else {
1506 			resid -= (size_t)len;
1507 		}
1508 
1509 		/*
1510 		 * Eat the entire mbuf or just a piece of it
1511 		 */
1512 		if (len == m->m_len - moff) {
1513 			if (m->m_flags & M_EOR)
1514 				flags |= MSG_EOR;
1515 			if (flags & MSG_PEEK) {
1516 				m = m->m_next;
1517 				moff = 0;
1518 			} else {
1519 				if (sio) {
1520 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1521 					sbappend(sio, m);
1522 					m = n;
1523 				} else {
1524 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1525 				}
1526 			}
1527 		} else {
1528 			if (flags & MSG_PEEK) {
1529 				moff += len;
1530 			} else {
1531 				if (sio) {
1532 					n = m_copym(m, 0, len, M_WAITOK);
1533 					if (n)
1534 						sbappend(sio, n);
1535 				}
1536 				m->m_data += len;
1537 				m->m_len -= len;
1538 				so->so_rcv.ssb_cc -= len;
1539 			}
1540 		}
1541 		if (so->so_oobmark) {
1542 			if ((flags & MSG_PEEK) == 0) {
1543 				so->so_oobmark -= len;
1544 				if (so->so_oobmark == 0) {
1545 					sosetstate(so, SS_RCVATMARK);
1546 					break;
1547 				}
1548 			} else {
1549 				offset += len;
1550 				if (offset == so->so_oobmark)
1551 					break;
1552 			}
1553 		}
1554 		if (flags & MSG_EOR)
1555 			break;
1556 		/*
1557 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1558 		 * we must not quit until resid == 0 or an error
1559 		 * termination.  If a signal/timeout occurs, return
1560 		 * with a short count but without error.
1561 		 * Keep signalsockbuf locked against other readers.
1562 		 */
1563 		while ((flags & MSG_WAITALL) && m == NULL &&
1564 		       resid > 0 && !sosendallatonce(so) &&
1565 		       so->so_rcv.ssb_mb == NULL) {
1566 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1567 				break;
1568 			/*
1569 			 * The window might have closed to zero, make
1570 			 * sure we send an ack now that we've drained
1571 			 * the buffer or we might end up blocking until
1572 			 * the idle takes over (5 seconds).
1573 			 */
1574 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1575 				so_pru_rcvd(so, flags);
1576 			error = ssb_wait(&so->so_rcv);
1577 			if (error) {
1578 				ssb_unlock(&so->so_rcv);
1579 				error = 0;
1580 				goto done;
1581 			}
1582 			m = so->so_rcv.ssb_mb;
1583 		}
1584 	}
1585 
1586 	/*
1587 	 * If an atomic read was requested but unread data still remains
1588 	 * in the record, set MSG_TRUNC.
1589 	 */
1590 	if (m && pr->pr_flags & PR_ATOMIC)
1591 		flags |= MSG_TRUNC;
1592 
1593 	/*
1594 	 * Cleanup.  If an atomic read was requested drop any unread data.
1595 	 */
1596 	if ((flags & MSG_PEEK) == 0) {
1597 		if (m && (pr->pr_flags & PR_ATOMIC))
1598 			sbdroprecord(&so->so_rcv.sb);
1599 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1600 			so_pru_rcvd(so, flags);
1601 	}
1602 
1603 	if (orig_resid == resid && orig_resid &&
1604 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1605 		ssb_unlock(&so->so_rcv);
1606 		goto restart;
1607 	}
1608 
1609 	if (flagsp)
1610 		*flagsp |= flags;
1611 release:
1612 	ssb_unlock(&so->so_rcv);
1613 done:
1614 	lwkt_reltoken(&so->so_rcv.ssb_token);
1615 	if (free_chain)
1616 		m_freem(free_chain);
1617 	return (error);
1618 }
1619 
1620 int
1621 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1622 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1623 {
1624 	struct mbuf *m, *n;
1625 	struct mbuf *free_chain = NULL;
1626 	int flags, len, error, offset;
1627 	struct protosw *pr = so->so_proto;
1628 	int moff;
1629 	int didoob;
1630 	size_t resid, orig_resid, restmp;
1631 
1632 	if (uio)
1633 		resid = uio->uio_resid;
1634 	else
1635 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1636 	orig_resid = resid;
1637 
1638 	if (psa)
1639 		*psa = NULL;
1640 	if (controlp)
1641 		*controlp = NULL;
1642 	if (flagsp)
1643 		flags = *flagsp &~ MSG_EOR;
1644 	else
1645 		flags = 0;
1646 	if (flags & MSG_OOB) {
1647 		m = m_get(M_WAITOK, MT_DATA);
1648 		if (m == NULL)
1649 			return (ENOBUFS);
1650 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1651 		if (error)
1652 			goto bad;
1653 		if (sio) {
1654 			do {
1655 				sbappend(sio, m);
1656 				KKASSERT(resid >= (size_t)m->m_len);
1657 				resid -= (size_t)m->m_len;
1658 			} while (resid > 0 && m);
1659 		} else {
1660 			do {
1661 				uio->uio_resid = resid;
1662 				error = uiomove(mtod(m, caddr_t),
1663 						(int)szmin(resid, m->m_len),
1664 						uio);
1665 				resid = uio->uio_resid;
1666 				m = m_free(m);
1667 			} while (uio->uio_resid && error == 0 && m);
1668 		}
1669 bad:
1670 		if (m)
1671 			m_freem(m);
1672 		return (error);
1673 	}
1674 
1675 	/*
1676 	 * The token interlocks against the protocol thread while
1677 	 * ssb_lock is a blocking lock against other userland entities.
1678 	 *
1679 	 * Lock a limited number of mbufs (not all, so sbcompress() still
1680 	 * works well).  The token is used as an interlock for sbwait() so
1681 	 * release it afterwords.
1682 	 */
1683 restart:
1684 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1685 	if (error)
1686 		goto done;
1687 
1688 	lwkt_gettoken(&so->so_rcv.ssb_token);
1689 	m = so->so_rcv.ssb_mb;
1690 
1691 	/*
1692 	 * If we have less data than requested, block awaiting more
1693 	 * (subject to any timeout) if:
1694 	 *   1. the current count is less than the low water mark, or
1695 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1696 	 *	receive operation at once if we block (resid <= hiwat).
1697 	 *   3. MSG_DONTWAIT is not set
1698 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1699 	 * we have to do the receive in sections, and thus risk returning
1700 	 * a short count if a timeout or signal occurs after we start.
1701 	 */
1702 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1703 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1704 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1705 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1706 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1707 		if (so->so_error) {
1708 			if (m)
1709 				goto dontblock;
1710 			lwkt_reltoken(&so->so_rcv.ssb_token);
1711 			error = so->so_error;
1712 			if ((flags & MSG_PEEK) == 0)
1713 				so->so_error = 0;
1714 			goto release;
1715 		}
1716 		if (so->so_state & SS_CANTRCVMORE) {
1717 			if (m)
1718 				goto dontblock;
1719 			lwkt_reltoken(&so->so_rcv.ssb_token);
1720 			goto release;
1721 		}
1722 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1723 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1724 			lwkt_reltoken(&so->so_rcv.ssb_token);
1725 			error = ENOTCONN;
1726 			goto release;
1727 		}
1728 		if (resid == 0) {
1729 			lwkt_reltoken(&so->so_rcv.ssb_token);
1730 			goto release;
1731 		}
1732 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1733 			lwkt_reltoken(&so->so_rcv.ssb_token);
1734 			error = EWOULDBLOCK;
1735 			goto release;
1736 		}
1737 		ssb_unlock(&so->so_rcv);
1738 		error = ssb_wait(&so->so_rcv);
1739 		lwkt_reltoken(&so->so_rcv.ssb_token);
1740 		if (error)
1741 			goto done;
1742 		goto restart;
1743 	}
1744 
1745 	/*
1746 	 * Token still held
1747 	 */
1748 dontblock:
1749 	n = m;
1750 	restmp = 0;
1751 	while (n && restmp < resid) {
1752 		n->m_flags |= M_SOLOCKED;
1753 		restmp += n->m_len;
1754 		if (n->m_next == NULL)
1755 			n = n->m_nextpkt;
1756 		else
1757 			n = n->m_next;
1758 	}
1759 
1760 	/*
1761 	 * Release token for loop
1762 	 */
1763 	lwkt_reltoken(&so->so_rcv.ssb_token);
1764 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1765 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1766 
1767 	/*
1768 	 * note: m should be == sb_mb here.  Cache the next record while
1769 	 * cleaning up.  Note that calling m_free*() will break out critical
1770 	 * section.
1771 	 */
1772 	KKASSERT(m == so->so_rcv.ssb_mb);
1773 
1774 	/*
1775 	 * Copy to the UIO or mbuf return chain (*mp).
1776 	 *
1777 	 * NOTE: Token is not held for loop
1778 	 */
1779 	moff = 0;
1780 	offset = 0;
1781 	didoob = 0;
1782 
1783 	while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1784 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1785 		    ("receive 3"));
1786 
1787 		soclrstate(so, SS_RCVATMARK);
1788 		len = (resid > INT_MAX) ? INT_MAX : resid;
1789 		if (so->so_oobmark && len > so->so_oobmark - offset)
1790 			len = so->so_oobmark - offset;
1791 		if (len > m->m_len - moff)
1792 			len = m->m_len - moff;
1793 
1794 		/*
1795 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1796 		 * The SIO is dealt with when we eat the mbuf, but deal
1797 		 * with the resid here either way.
1798 		 */
1799 		if (uio) {
1800 			uio->uio_resid = resid;
1801 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1802 			resid = uio->uio_resid;
1803 			if (error)
1804 				goto release;
1805 		} else {
1806 			resid -= (size_t)len;
1807 		}
1808 
1809 		/*
1810 		 * Eat the entire mbuf or just a piece of it
1811 		 */
1812 		offset += len;
1813 		if (len == m->m_len - moff) {
1814 			m = m->m_next;
1815 			moff = 0;
1816 		} else {
1817 			moff += len;
1818 		}
1819 
1820 		/*
1821 		 * Check oobmark
1822 		 */
1823 		if (so->so_oobmark && offset == so->so_oobmark) {
1824 			didoob = 1;
1825 			break;
1826 		}
1827 	}
1828 
1829 	/*
1830 	 * Synchronize sockbuf with data we read.
1831 	 *
1832 	 * NOTE: (m) is junk on entry (it could be left over from the
1833 	 *	 previous loop).
1834 	 */
1835 	if ((flags & MSG_PEEK) == 0) {
1836 		lwkt_gettoken(&so->so_rcv.ssb_token);
1837 		m = so->so_rcv.ssb_mb;
1838 		while (m && offset >= m->m_len) {
1839 			if (so->so_oobmark) {
1840 				so->so_oobmark -= m->m_len;
1841 				if (so->so_oobmark == 0) {
1842 					sosetstate(so, SS_RCVATMARK);
1843 					didoob = 1;
1844 				}
1845 			}
1846 			offset -= m->m_len;
1847 			if (sio) {
1848 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1849 				sbappend(sio, m);
1850 				m = n;
1851 			} else {
1852 				m = sbunlinkmbuf(&so->so_rcv.sb,
1853 						 m, &free_chain);
1854 			}
1855 		}
1856 		if (offset) {
1857 			KKASSERT(m);
1858 			if (sio) {
1859 				n = m_copym(m, 0, offset, M_WAITOK);
1860 				if (n)
1861 					sbappend(sio, n);
1862 			}
1863 			m->m_data += offset;
1864 			m->m_len -= offset;
1865 			so->so_rcv.ssb_cc -= offset;
1866 			if (so->so_oobmark) {
1867 				so->so_oobmark -= offset;
1868 				if (so->so_oobmark == 0) {
1869 					sosetstate(so, SS_RCVATMARK);
1870 					didoob = 1;
1871 				}
1872 			}
1873 			offset = 0;
1874 		}
1875 		lwkt_reltoken(&so->so_rcv.ssb_token);
1876 	}
1877 
1878 	/*
1879 	 * If the MSG_WAITALL flag is set (for non-atomic socket),
1880 	 * we must not quit until resid == 0 or an error termination.
1881 	 *
1882 	 * If a signal/timeout occurs, return with a short count but without
1883 	 * error.
1884 	 *
1885 	 * Keep signalsockbuf locked against other readers.
1886 	 *
1887 	 * XXX if MSG_PEEK we currently do quit.
1888 	 */
1889 	if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1890 	    didoob == 0 && resid > 0 &&
1891 	    !sosendallatonce(so)) {
1892 		lwkt_gettoken(&so->so_rcv.ssb_token);
1893 		error = 0;
1894 		while ((m = so->so_rcv.ssb_mb) == NULL) {
1895 			if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1896 				error = so->so_error;
1897 				break;
1898 			}
1899 			/*
1900 			 * The window might have closed to zero, make
1901 			 * sure we send an ack now that we've drained
1902 			 * the buffer or we might end up blocking until
1903 			 * the idle takes over (5 seconds).
1904 			 */
1905 			if (so->so_pcb)
1906 				so_pru_rcvd_async(so);
1907 			if (so->so_rcv.ssb_mb == NULL)
1908 				error = ssb_wait(&so->so_rcv);
1909 			if (error) {
1910 				lwkt_reltoken(&so->so_rcv.ssb_token);
1911 				ssb_unlock(&so->so_rcv);
1912 				error = 0;
1913 				goto done;
1914 			}
1915 		}
1916 		if (m && error == 0)
1917 			goto dontblock;
1918 		lwkt_reltoken(&so->so_rcv.ssb_token);
1919 	}
1920 
1921 	/*
1922 	 * Token not held here.
1923 	 *
1924 	 * Cleanup.  If an atomic read was requested drop any unread data XXX
1925 	 */
1926 	if ((flags & MSG_PEEK) == 0) {
1927 		if (so->so_pcb)
1928 			so_pru_rcvd_async(so);
1929 	}
1930 
1931 	if (orig_resid == resid && orig_resid &&
1932 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1933 		ssb_unlock(&so->so_rcv);
1934 		goto restart;
1935 	}
1936 
1937 	if (flagsp)
1938 		*flagsp |= flags;
1939 release:
1940 	ssb_unlock(&so->so_rcv);
1941 done:
1942 	if (free_chain)
1943 		m_freem(free_chain);
1944 	return (error);
1945 }
1946 
1947 /*
1948  * Shut a socket down.  Note that we do not get a frontend lock as we
1949  * want to be able to shut the socket down even if another thread is
1950  * blocked in a read(), thus waking it up.
1951  */
1952 int
1953 soshutdown(struct socket *so, int how)
1954 {
1955 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1956 		return (EINVAL);
1957 
1958 	if (how != SHUT_WR) {
1959 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1960 		sorflush(so);
1961 		/*ssb_unlock(&so->so_rcv);*/
1962 	}
1963 	if (how != SHUT_RD)
1964 		return (so_pru_shutdown(so));
1965 	return (0);
1966 }
1967 
1968 void
1969 sorflush(struct socket *so)
1970 {
1971 	struct signalsockbuf *ssb = &so->so_rcv;
1972 	struct protosw *pr = so->so_proto;
1973 	struct signalsockbuf asb;
1974 
1975 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1976 
1977 	lwkt_gettoken(&ssb->ssb_token);
1978 	socantrcvmore(so);
1979 	asb = *ssb;
1980 
1981 	/*
1982 	 * Can't just blow up the ssb structure here
1983 	 */
1984 	bzero(&ssb->sb, sizeof(ssb->sb));
1985 	ssb->ssb_timeo = 0;
1986 	ssb->ssb_lowat = 0;
1987 	ssb->ssb_hiwat = 0;
1988 	ssb->ssb_mbmax = 0;
1989 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1990 
1991 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1992 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1993 	ssb_release(&asb, so);
1994 
1995 	lwkt_reltoken(&ssb->ssb_token);
1996 }
1997 
1998 #ifdef INET
1999 static int
2000 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
2001 {
2002 	struct accept_filter_arg	*afap = NULL;
2003 	struct accept_filter	*afp;
2004 	struct so_accf	*af = so->so_accf;
2005 	int	error = 0;
2006 
2007 	/* do not set/remove accept filters on non listen sockets */
2008 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
2009 		error = EINVAL;
2010 		goto out;
2011 	}
2012 
2013 	/* removing the filter */
2014 	if (sopt == NULL) {
2015 		if (af != NULL) {
2016 			if (af->so_accept_filter != NULL &&
2017 				af->so_accept_filter->accf_destroy != NULL) {
2018 				af->so_accept_filter->accf_destroy(so);
2019 			}
2020 			if (af->so_accept_filter_str != NULL) {
2021 				kfree(af->so_accept_filter_str, M_ACCF);
2022 			}
2023 			kfree(af, M_ACCF);
2024 			so->so_accf = NULL;
2025 		}
2026 		so->so_options &= ~SO_ACCEPTFILTER;
2027 		return (0);
2028 	}
2029 	/* adding a filter */
2030 	/* must remove previous filter first */
2031 	if (af != NULL) {
2032 		error = EINVAL;
2033 		goto out;
2034 	}
2035 	/* don't put large objects on the kernel stack */
2036 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2037 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2038 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
2039 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2040 	if (error)
2041 		goto out;
2042 	afp = accept_filt_get(afap->af_name);
2043 	if (afp == NULL) {
2044 		error = ENOENT;
2045 		goto out;
2046 	}
2047 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2048 	if (afp->accf_create != NULL) {
2049 		if (afap->af_name[0] != '\0') {
2050 			int len = strlen(afap->af_name) + 1;
2051 
2052 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
2053 							   M_WAITOK);
2054 			strcpy(af->so_accept_filter_str, afap->af_name);
2055 		}
2056 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2057 		if (af->so_accept_filter_arg == NULL) {
2058 			kfree(af->so_accept_filter_str, M_ACCF);
2059 			kfree(af, M_ACCF);
2060 			so->so_accf = NULL;
2061 			error = EINVAL;
2062 			goto out;
2063 		}
2064 	}
2065 	af->so_accept_filter = afp;
2066 	so->so_accf = af;
2067 	so->so_options |= SO_ACCEPTFILTER;
2068 out:
2069 	if (afap != NULL)
2070 		kfree(afap, M_TEMP);
2071 	return (error);
2072 }
2073 #endif /* INET */
2074 
2075 /*
2076  * Perhaps this routine, and sooptcopyout(), below, ought to come in
2077  * an additional variant to handle the case where the option value needs
2078  * to be some kind of integer, but not a specific size.
2079  * In addition to their use here, these functions are also called by the
2080  * protocol-level pr_ctloutput() routines.
2081  */
2082 int
2083 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2084 {
2085 	return soopt_to_kbuf(sopt, buf, len, minlen);
2086 }
2087 
2088 int
2089 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2090 {
2091 	size_t	valsize;
2092 
2093 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2094 	KKASSERT(kva_p(buf));
2095 
2096 	/*
2097 	 * If the user gives us more than we wanted, we ignore it,
2098 	 * but if we don't get the minimum length the caller
2099 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
2100 	 * is set to however much we actually retrieved.
2101 	 */
2102 	if ((valsize = sopt->sopt_valsize) < minlen)
2103 		return EINVAL;
2104 	if (valsize > len)
2105 		sopt->sopt_valsize = valsize = len;
2106 
2107 	bcopy(sopt->sopt_val, buf, valsize);
2108 	return 0;
2109 }
2110 
2111 
2112 int
2113 sosetopt(struct socket *so, struct sockopt *sopt)
2114 {
2115 	int	error, optval;
2116 	struct	linger l;
2117 	struct	timeval tv;
2118 	u_long  val;
2119 	struct signalsockbuf *sotmp;
2120 
2121 	error = 0;
2122 	sopt->sopt_dir = SOPT_SET;
2123 	if (sopt->sopt_level != SOL_SOCKET) {
2124 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2125 			return (so_pr_ctloutput(so, sopt));
2126 		}
2127 		error = ENOPROTOOPT;
2128 	} else {
2129 		switch (sopt->sopt_name) {
2130 #ifdef INET
2131 		case SO_ACCEPTFILTER:
2132 			error = do_setopt_accept_filter(so, sopt);
2133 			if (error)
2134 				goto bad;
2135 			break;
2136 #endif /* INET */
2137 		case SO_LINGER:
2138 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2139 			if (error)
2140 				goto bad;
2141 
2142 			so->so_linger = l.l_linger;
2143 			if (l.l_onoff)
2144 				so->so_options |= SO_LINGER;
2145 			else
2146 				so->so_options &= ~SO_LINGER;
2147 			break;
2148 
2149 		case SO_DEBUG:
2150 		case SO_KEEPALIVE:
2151 		case SO_DONTROUTE:
2152 		case SO_USELOOPBACK:
2153 		case SO_BROADCAST:
2154 		case SO_REUSEADDR:
2155 		case SO_REUSEPORT:
2156 		case SO_OOBINLINE:
2157 		case SO_TIMESTAMP:
2158 		case SO_NOSIGPIPE:
2159 			error = sooptcopyin(sopt, &optval, sizeof optval,
2160 					    sizeof optval);
2161 			if (error)
2162 				goto bad;
2163 			if (optval)
2164 				so->so_options |= sopt->sopt_name;
2165 			else
2166 				so->so_options &= ~sopt->sopt_name;
2167 			break;
2168 
2169 		case SO_SNDBUF:
2170 		case SO_RCVBUF:
2171 		case SO_SNDLOWAT:
2172 		case SO_RCVLOWAT:
2173 			error = sooptcopyin(sopt, &optval, sizeof optval,
2174 					    sizeof optval);
2175 			if (error)
2176 				goto bad;
2177 
2178 			/*
2179 			 * Values < 1 make no sense for any of these
2180 			 * options, so disallow them.
2181 			 */
2182 			if (optval < 1) {
2183 				error = EINVAL;
2184 				goto bad;
2185 			}
2186 
2187 			switch (sopt->sopt_name) {
2188 			case SO_SNDBUF:
2189 			case SO_RCVBUF:
2190 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2191 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2192 				    so,
2193 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2194 					error = ENOBUFS;
2195 					goto bad;
2196 				}
2197 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2198 						&so->so_snd : &so->so_rcv;
2199 				atomic_clear_int(&sotmp->ssb_flags,
2200 						 SSB_AUTOSIZE);
2201 				break;
2202 
2203 			/*
2204 			 * Make sure the low-water is never greater than
2205 			 * the high-water.
2206 			 */
2207 			case SO_SNDLOWAT:
2208 				so->so_snd.ssb_lowat =
2209 				    (optval > so->so_snd.ssb_hiwat) ?
2210 				    so->so_snd.ssb_hiwat : optval;
2211 				atomic_clear_int(&so->so_snd.ssb_flags,
2212 						 SSB_AUTOLOWAT);
2213 				break;
2214 			case SO_RCVLOWAT:
2215 				so->so_rcv.ssb_lowat =
2216 				    (optval > so->so_rcv.ssb_hiwat) ?
2217 				    so->so_rcv.ssb_hiwat : optval;
2218 				atomic_clear_int(&so->so_rcv.ssb_flags,
2219 						 SSB_AUTOLOWAT);
2220 				break;
2221 			}
2222 			break;
2223 
2224 		case SO_SNDTIMEO:
2225 		case SO_RCVTIMEO:
2226 			error = sooptcopyin(sopt, &tv, sizeof tv,
2227 					    sizeof tv);
2228 			if (error)
2229 				goto bad;
2230 
2231 			/* assert(hz > 0); */
2232 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2233 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2234 				error = EDOM;
2235 				goto bad;
2236 			}
2237 			/* assert(tick > 0); */
2238 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2239 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2240 			if (val > INT_MAX) {
2241 				error = EDOM;
2242 				goto bad;
2243 			}
2244 			if (val == 0 && tv.tv_usec != 0)
2245 				val = 1;
2246 
2247 			switch (sopt->sopt_name) {
2248 			case SO_SNDTIMEO:
2249 				so->so_snd.ssb_timeo = val;
2250 				break;
2251 			case SO_RCVTIMEO:
2252 				so->so_rcv.ssb_timeo = val;
2253 				break;
2254 			}
2255 			break;
2256 		default:
2257 			error = ENOPROTOOPT;
2258 			break;
2259 		}
2260 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2261 			(void) so_pr_ctloutput(so, sopt);
2262 		}
2263 	}
2264 bad:
2265 	return (error);
2266 }
2267 
2268 /* Helper routine for getsockopt */
2269 int
2270 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2271 {
2272 	soopt_from_kbuf(sopt, buf, len);
2273 	return 0;
2274 }
2275 
2276 void
2277 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2278 {
2279 	size_t	valsize;
2280 
2281 	if (len == 0) {
2282 		sopt->sopt_valsize = 0;
2283 		return;
2284 	}
2285 
2286 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2287 	KKASSERT(kva_p(buf));
2288 
2289 	/*
2290 	 * Documented get behavior is that we always return a value,
2291 	 * possibly truncated to fit in the user's buffer.
2292 	 * Traditional behavior is that we always tell the user
2293 	 * precisely how much we copied, rather than something useful
2294 	 * like the total amount we had available for her.
2295 	 * Note that this interface is not idempotent; the entire answer must
2296 	 * generated ahead of time.
2297 	 */
2298 	valsize = szmin(len, sopt->sopt_valsize);
2299 	sopt->sopt_valsize = valsize;
2300 	if (sopt->sopt_val != 0) {
2301 		bcopy(buf, sopt->sopt_val, valsize);
2302 	}
2303 }
2304 
2305 int
2306 sogetopt(struct socket *so, struct sockopt *sopt)
2307 {
2308 	int	error, optval;
2309 	long	optval_l;
2310 	struct	linger l;
2311 	struct	timeval tv;
2312 #ifdef INET
2313 	struct accept_filter_arg *afap;
2314 #endif
2315 
2316 	error = 0;
2317 	sopt->sopt_dir = SOPT_GET;
2318 	if (sopt->sopt_level != SOL_SOCKET) {
2319 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2320 			return (so_pr_ctloutput(so, sopt));
2321 		} else
2322 			return (ENOPROTOOPT);
2323 	} else {
2324 		switch (sopt->sopt_name) {
2325 #ifdef INET
2326 		case SO_ACCEPTFILTER:
2327 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2328 				return (EINVAL);
2329 			afap = kmalloc(sizeof(*afap), M_TEMP,
2330 				       M_WAITOK | M_ZERO);
2331 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2332 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2333 				if (so->so_accf->so_accept_filter_str != NULL)
2334 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2335 			}
2336 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2337 			kfree(afap, M_TEMP);
2338 			break;
2339 #endif /* INET */
2340 
2341 		case SO_LINGER:
2342 			l.l_onoff = so->so_options & SO_LINGER;
2343 			l.l_linger = so->so_linger;
2344 			error = sooptcopyout(sopt, &l, sizeof l);
2345 			break;
2346 
2347 		case SO_USELOOPBACK:
2348 		case SO_DONTROUTE:
2349 		case SO_DEBUG:
2350 		case SO_KEEPALIVE:
2351 		case SO_REUSEADDR:
2352 		case SO_REUSEPORT:
2353 		case SO_BROADCAST:
2354 		case SO_OOBINLINE:
2355 		case SO_TIMESTAMP:
2356 		case SO_NOSIGPIPE:
2357 			optval = so->so_options & sopt->sopt_name;
2358 integer:
2359 			error = sooptcopyout(sopt, &optval, sizeof optval);
2360 			break;
2361 
2362 		case SO_TYPE:
2363 			optval = so->so_type;
2364 			goto integer;
2365 
2366 		case SO_ERROR:
2367 			optval = so->so_error;
2368 			so->so_error = 0;
2369 			goto integer;
2370 
2371 		case SO_SNDBUF:
2372 			optval = so->so_snd.ssb_hiwat;
2373 			goto integer;
2374 
2375 		case SO_RCVBUF:
2376 			optval = so->so_rcv.ssb_hiwat;
2377 			goto integer;
2378 
2379 		case SO_SNDLOWAT:
2380 			optval = so->so_snd.ssb_lowat;
2381 			goto integer;
2382 
2383 		case SO_RCVLOWAT:
2384 			optval = so->so_rcv.ssb_lowat;
2385 			goto integer;
2386 
2387 		case SO_SNDTIMEO:
2388 		case SO_RCVTIMEO:
2389 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2390 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2391 
2392 			tv.tv_sec = optval / hz;
2393 			tv.tv_usec = (optval % hz) * ustick;
2394 			error = sooptcopyout(sopt, &tv, sizeof tv);
2395 			break;
2396 
2397 		case SO_SNDSPACE:
2398 			optval_l = ssb_space(&so->so_snd);
2399 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2400 			break;
2401 
2402 		case SO_CPUHINT:
2403 			optval = -1; /* no hint */
2404 			goto integer;
2405 
2406 		default:
2407 			error = ENOPROTOOPT;
2408 			break;
2409 		}
2410 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2411 			so_pr_ctloutput(so, sopt);
2412 		return (error);
2413 	}
2414 }
2415 
2416 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2417 int
2418 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2419 {
2420 	struct mbuf *m, *m_prev;
2421 	int sopt_size = sopt->sopt_valsize, msize;
2422 
2423 	m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2424 		   0, &msize);
2425 	if (m == NULL)
2426 		return (ENOBUFS);
2427 	m->m_len = min(msize, sopt_size);
2428 	sopt_size -= m->m_len;
2429 	*mp = m;
2430 	m_prev = m;
2431 
2432 	while (sopt_size > 0) {
2433 		m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2434 			   MT_DATA, 0, &msize);
2435 		if (m == NULL) {
2436 			m_freem(*mp);
2437 			return (ENOBUFS);
2438 		}
2439 		m->m_len = min(msize, sopt_size);
2440 		sopt_size -= m->m_len;
2441 		m_prev->m_next = m;
2442 		m_prev = m;
2443 	}
2444 	return (0);
2445 }
2446 
2447 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2448 int
2449 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2450 {
2451 	soopt_to_mbuf(sopt, m);
2452 	return 0;
2453 }
2454 
2455 void
2456 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2457 {
2458 	size_t valsize;
2459 	void *val;
2460 
2461 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2462 	KKASSERT(kva_p(m));
2463 	if (sopt->sopt_val == NULL)
2464 		return;
2465 	val = sopt->sopt_val;
2466 	valsize = sopt->sopt_valsize;
2467 	while (m != NULL && valsize >= m->m_len) {
2468 		bcopy(val, mtod(m, char *), m->m_len);
2469 		valsize -= m->m_len;
2470 		val = (caddr_t)val + m->m_len;
2471 		m = m->m_next;
2472 	}
2473 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2474 		panic("ip6_sooptmcopyin");
2475 }
2476 
2477 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2478 int
2479 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2480 {
2481 	return soopt_from_mbuf(sopt, m);
2482 }
2483 
2484 int
2485 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2486 {
2487 	struct mbuf *m0 = m;
2488 	size_t valsize = 0;
2489 	size_t maxsize;
2490 	void *val;
2491 
2492 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2493 	KKASSERT(kva_p(m));
2494 	if (sopt->sopt_val == NULL)
2495 		return 0;
2496 	val = sopt->sopt_val;
2497 	maxsize = sopt->sopt_valsize;
2498 	while (m != NULL && maxsize >= m->m_len) {
2499 		bcopy(mtod(m, char *), val, m->m_len);
2500 	       maxsize -= m->m_len;
2501 	       val = (caddr_t)val + m->m_len;
2502 	       valsize += m->m_len;
2503 	       m = m->m_next;
2504 	}
2505 	if (m != NULL) {
2506 		/* enough soopt buffer should be given from user-land */
2507 		m_freem(m0);
2508 		return (EINVAL);
2509 	}
2510 	sopt->sopt_valsize = valsize;
2511 	return 0;
2512 }
2513 
2514 void
2515 sohasoutofband(struct socket *so)
2516 {
2517 	if (so->so_sigio != NULL)
2518 		pgsigio(so->so_sigio, SIGURG, 0);
2519 	KNOTE(&so->so_rcv.ssb_kq.ki_note, NOTE_OOB);
2520 }
2521 
2522 int
2523 sokqfilter(struct file *fp, struct knote *kn)
2524 {
2525 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2526 	struct signalsockbuf *ssb;
2527 
2528 	switch (kn->kn_filter) {
2529 	case EVFILT_READ:
2530 		if (so->so_options & SO_ACCEPTCONN)
2531 			kn->kn_fop = &solisten_filtops;
2532 		else
2533 			kn->kn_fop = &soread_filtops;
2534 		ssb = &so->so_rcv;
2535 		break;
2536 	case EVFILT_WRITE:
2537 		kn->kn_fop = &sowrite_filtops;
2538 		ssb = &so->so_snd;
2539 		break;
2540 	case EVFILT_EXCEPT:
2541 		kn->kn_fop = &soexcept_filtops;
2542 		ssb = &so->so_rcv;
2543 		break;
2544 	default:
2545 		return (EOPNOTSUPP);
2546 	}
2547 
2548 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2549 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2550 	return (0);
2551 }
2552 
2553 static void
2554 filt_sordetach(struct knote *kn)
2555 {
2556 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2557 
2558 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2559 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2560 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2561 }
2562 
2563 /*ARGSUSED*/
2564 static int
2565 filt_soread(struct knote *kn, long hint)
2566 {
2567 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2568 
2569 	if (kn->kn_sfflags & NOTE_OOB) {
2570 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2571 			kn->kn_fflags |= NOTE_OOB;
2572 			return (1);
2573 		}
2574 		return (0);
2575 	}
2576 	kn->kn_data = so->so_rcv.ssb_cc;
2577 
2578 	if (so->so_state & SS_CANTRCVMORE) {
2579 		/*
2580 		 * Only set NODATA if all data has been exhausted.
2581 		 */
2582 		if (kn->kn_data == 0)
2583 			kn->kn_flags |= EV_NODATA;
2584 		kn->kn_flags |= EV_EOF;
2585 		kn->kn_fflags = so->so_error;
2586 		return (1);
2587 	}
2588 	if (so->so_error)	/* temporary udp error */
2589 		return (1);
2590 	if (kn->kn_sfflags & NOTE_LOWAT)
2591 		return (kn->kn_data >= kn->kn_sdata);
2592 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2593 		!TAILQ_EMPTY(&so->so_comp));
2594 }
2595 
2596 static void
2597 filt_sowdetach(struct knote *kn)
2598 {
2599 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2600 
2601 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2602 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2603 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2604 }
2605 
2606 /*ARGSUSED*/
2607 static int
2608 filt_sowrite(struct knote *kn, long hint)
2609 {
2610 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2611 
2612 	kn->kn_data = ssb_space(&so->so_snd);
2613 	if (so->so_state & SS_CANTSENDMORE) {
2614 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2615 		kn->kn_fflags = so->so_error;
2616 		return (1);
2617 	}
2618 	if (so->so_error)	/* temporary udp error */
2619 		return (1);
2620 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2621 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2622 		return (0);
2623 	if (kn->kn_sfflags & NOTE_LOWAT)
2624 		return (kn->kn_data >= kn->kn_sdata);
2625 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2626 }
2627 
2628 /*ARGSUSED*/
2629 static int
2630 filt_solisten(struct knote *kn, long hint)
2631 {
2632 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2633 
2634 	kn->kn_data = so->so_qlen;
2635 	return (! TAILQ_EMPTY(&so->so_comp));
2636 }
2637