xref: /dflybsd-src/sys/kern/uipc_socket.c (revision ef94481424b6bf8cbda60b2a9633e25f1513329c)
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 static int soavailconn = 32;
156 SYSCTL_INT(_kern_ipc, OID_AUTO, soavailconn, CTLFLAG_RW,
157     &soavailconn, 0, "Maximum available socket connection queue size");
158 
159 /*
160  * Socket operation routines.
161  * These routines are called by the routines in
162  * sys_socket.c or from a system process, and
163  * implement the semantics of socket operations by
164  * switching out to the protocol specific routines.
165  */
166 
167 /*
168  * Get a socket structure, and initialize it.
169  * Note that it would probably be better to allocate socket
170  * and PCB at the same time, but I'm not convinced that all
171  * the protocols can be easily modified to do this.
172  */
173 struct socket *
174 soalloc(int waitok, struct protosw *pr)
175 {
176 	globaldata_t gd = mycpu;
177 	struct socket *so;
178 	unsigned waitmask;
179 
180 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
181 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
182 	if (so) {
183 		/* XXX race condition for reentrant kernel */
184 		so->so_proto = pr;
185 		TAILQ_INIT(&so->so_aiojobq);
186 		TAILQ_INIT(&so->so_rcv.ssb_mlist);
187 		TAILQ_INIT(&so->so_snd.ssb_mlist);
188 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
189 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
190 		spin_init(&so->so_rcvd_spin, "soalloc");
191 		netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
192 		    MSGF_DROPABLE | MSGF_PRIORITY,
193 		    so->so_proto->pr_usrreqs->pru_rcvd);
194 		so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
195 		so->so_state = SS_NOFDREF;
196 		so->so_refs = 1;
197 		so->so_inum = gd->gd_anoninum++ * ncpus + gd->gd_cpuid + 2;
198 	}
199 	return so;
200 }
201 
202 int
203 socreate(int dom, struct socket **aso, int type,
204 	int proto, struct thread *td)
205 {
206 	struct proc *p = td->td_proc;
207 	struct protosw *prp;
208 	struct socket *so;
209 	struct pru_attach_info ai;
210 	int error;
211 
212 	if (proto)
213 		prp = pffindproto(dom, proto, type);
214 	else
215 		prp = pffindtype(dom, type);
216 
217 	if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
218 		return (EPROTONOSUPPORT);
219 
220 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
221 	    prp->pr_domain->dom_family != PF_LOCAL &&
222 	    prp->pr_domain->dom_family != PF_INET &&
223 	    prp->pr_domain->dom_family != PF_INET6 &&
224 	    prp->pr_domain->dom_family != PF_ROUTE) {
225 		return (EPROTONOSUPPORT);
226 	}
227 
228 	if (prp->pr_type != type)
229 		return (EPROTOTYPE);
230 	so = soalloc(p != NULL, prp);
231 	if (so == NULL)
232 		return (ENOBUFS);
233 
234 	/*
235 	 * Callers of socreate() presumably will connect up a descriptor
236 	 * and call soclose() if they cannot.  This represents our so_refs
237 	 * (which should be 1) from soalloc().
238 	 */
239 	soclrstate(so, SS_NOFDREF);
240 
241 	/*
242 	 * Set a default port for protocol processing.  No action will occur
243 	 * on the socket on this port until an inpcb is attached to it and
244 	 * is able to match incoming packets, or until the socket becomes
245 	 * available to userland.
246 	 *
247 	 * We normally default the socket to the protocol thread on cpu 0,
248 	 * if protocol does not provide its own method to initialize the
249 	 * default port.
250 	 *
251 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
252 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
253 	 */
254 	if (prp->pr_flags & PR_SYNC_PORT)
255 		so->so_port = &netisr_sync_port;
256 	else if (prp->pr_initport != NULL)
257 		so->so_port = prp->pr_initport();
258 	else
259 		so->so_port = netisr_cpuport(0);
260 
261 	TAILQ_INIT(&so->so_incomp);
262 	TAILQ_INIT(&so->so_comp);
263 	so->so_type = type;
264 	so->so_cred = crhold(p->p_ucred);
265 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
266 	ai.p_ucred = p->p_ucred;
267 	ai.fd_rdir = p->p_fd->fd_rdir;
268 
269 	/*
270 	 * Auto-sizing of socket buffers is managed by the protocols and
271 	 * the appropriate flags must be set in the pru_attach function.
272 	 */
273 	if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
274 		error = so_pru_attach_fast(so, proto, &ai);
275 	else
276 		error = so_pru_attach(so, proto, &ai);
277 	if (error) {
278 		sosetstate(so, SS_NOFDREF);
279 		sofree(so);	/* from soalloc */
280 		return error;
281 	}
282 
283 	/*
284 	 * NOTE: Returns referenced socket.
285 	 */
286 	*aso = so;
287 	return (0);
288 }
289 
290 int
291 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
292 {
293 	int error;
294 
295 	error = so_pru_bind(so, nam, td);
296 	return (error);
297 }
298 
299 static void
300 sodealloc(struct socket *so)
301 {
302 	KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
303 
304 #ifdef INVARIANTS
305 	if (so->so_options & SO_ACCEPTCONN) {
306 		KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
307 		KASSERT(TAILQ_EMPTY(&so->so_incomp),
308 		    ("so_incomp is not empty"));
309 	}
310 #endif
311 
312 	if (so->so_rcv.ssb_hiwat)
313 		(void)chgsbsize(so->so_cred->cr_uidinfo,
314 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
315 	if (so->so_snd.ssb_hiwat)
316 		(void)chgsbsize(so->so_cred->cr_uidinfo,
317 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
318 #ifdef INET
319 	/* remove accept filter if present */
320 	if (so->so_accf != NULL)
321 		do_setopt_accept_filter(so, NULL);
322 #endif /* INET */
323 	crfree(so->so_cred);
324 	if (so->so_faddr != NULL)
325 		kfree(so->so_faddr, M_SONAME);
326 	kfree(so, M_SOCKET);
327 }
328 
329 int
330 solisten(struct socket *so, int backlog, struct thread *td)
331 {
332 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
333 		return (EINVAL);
334 
335 	lwkt_gettoken(&so->so_rcv.ssb_token);
336 	if (TAILQ_EMPTY(&so->so_comp))
337 		so->so_options |= SO_ACCEPTCONN;
338 	lwkt_reltoken(&so->so_rcv.ssb_token);
339 	if (backlog < 0 || backlog > somaxconn)
340 		backlog = somaxconn;
341 	so->so_qlimit = backlog;
342 	return so_pru_listen(so, td);
343 }
344 
345 static void
346 soqflush(struct socket *so)
347 {
348 	lwkt_getpooltoken(so);
349 	if (so->so_options & SO_ACCEPTCONN) {
350 		struct socket *sp;
351 
352 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
353 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
354 			    SS_INCOMP);
355 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
356 			so->so_incqlen--;
357 			soclrstate(sp, SS_INCOMP);
358 			soabort_async(sp, TRUE);
359 		}
360 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
361 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
362 			    SS_COMP);
363 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
364 			so->so_qlen--;
365 			soclrstate(sp, SS_COMP);
366 			soabort_async(sp, TRUE);
367 		}
368 	}
369 	lwkt_relpooltoken(so);
370 }
371 
372 /*
373  * Destroy a disconnected socket.  This routine is a NOP if entities
374  * still have a reference on the socket:
375  *
376  *	so_pcb -	The protocol stack still has a reference
377  *	SS_NOFDREF -	There is no longer a file pointer reference
378  */
379 void
380 sofree(struct socket *so)
381 {
382 	struct socket *head;
383 
384 	/*
385 	 * This is a bit hackish at the moment.  We need to interlock
386 	 * any accept queue we are on before we potentially lose the
387 	 * last reference to avoid races against a re-reference from
388 	 * someone operating on the queue.
389 	 */
390 	while ((head = so->so_head) != NULL) {
391 		lwkt_getpooltoken(head);
392 		if (so->so_head == head)
393 			break;
394 		lwkt_relpooltoken(head);
395 	}
396 
397 	/*
398 	 * Arbitrage the last free.
399 	 */
400 	KKASSERT(so->so_refs > 0);
401 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
402 		if (head)
403 			lwkt_relpooltoken(head);
404 		return;
405 	}
406 
407 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
408 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
409 
410 	if (head != NULL) {
411 		/*
412 		 * We're done, remove ourselves from the accept queue we are
413 		 * on, if we are on one.
414 		 */
415 		if (so->so_state & SS_INCOMP) {
416 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
417 			    SS_INCOMP);
418 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
419 			head->so_incqlen--;
420 		} else if (so->so_state & SS_COMP) {
421 			/*
422 			 * We must not decommission a socket that's
423 			 * on the accept(2) queue.  If we do, then
424 			 * accept(2) may hang after select(2) indicated
425 			 * that the listening socket was ready.
426 			 */
427 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
428 			    SS_COMP);
429 			lwkt_relpooltoken(head);
430 			return;
431 		} else {
432 			panic("sofree: not queued");
433 		}
434 		soclrstate(so, SS_INCOMP);
435 		so->so_head = NULL;
436 		lwkt_relpooltoken(head);
437 	} else {
438 		/* Flush accept queues, if we are accepting. */
439 		soqflush(so);
440 	}
441 	ssb_release(&so->so_snd, so);
442 	sorflush(so);
443 	sodealloc(so);
444 }
445 
446 /*
447  * Close a socket on last file table reference removal.
448  * Initiate disconnect if connected.
449  * Free socket when disconnect complete.
450  */
451 int
452 soclose(struct socket *so, int fflag)
453 {
454 	int error;
455 
456 	funsetown(&so->so_sigio);
457 	sosetstate(so, SS_ISCLOSING);
458 	if (!use_soclose_fast ||
459 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
460 	    ((so->so_state & SS_ISCONNECTED) &&
461 	     (so->so_options & SO_LINGER) &&
462 	     so->so_linger != 0)) {
463 		error = soclose_sync(so, fflag);
464 	} else {
465 		soclose_fast(so);
466 		error = 0;
467 	}
468 	return error;
469 }
470 
471 void
472 sodiscard(struct socket *so)
473 {
474 	if (so->so_state & SS_NOFDREF)
475 		panic("soclose: NOFDREF");
476 	sosetstate(so, SS_NOFDREF);	/* take ref */
477 }
478 
479 /*
480  * Append the completed queue of head to head_inh (inherting listen socket).
481  */
482 void
483 soinherit(struct socket *head, struct socket *head_inh)
484 {
485 	boolean_t do_wakeup = FALSE;
486 
487 	KASSERT(head->so_options & SO_ACCEPTCONN,
488 	    ("head does not accept connection"));
489 	KASSERT(head_inh->so_options & SO_ACCEPTCONN,
490 	    ("head_inh does not accept connection"));
491 
492 	lwkt_getpooltoken(head);
493 	lwkt_getpooltoken(head_inh);
494 
495 	if (head->so_qlen > 0)
496 		do_wakeup = TRUE;
497 
498 	while (!TAILQ_EMPTY(&head->so_comp)) {
499 		struct ucred *old_cr;
500 		struct socket *sp;
501 
502 		sp = TAILQ_FIRST(&head->so_comp);
503 		KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
504 
505 		/*
506 		 * Remove this socket from the current listen socket
507 		 * completed queue.
508 		 */
509 		TAILQ_REMOVE(&head->so_comp, sp, so_list);
510 		head->so_qlen--;
511 
512 		/* Save the old ucred for later free. */
513 		old_cr = sp->so_cred;
514 
515 		/*
516 		 * Install this socket to the inheriting listen socket
517 		 * completed queue.
518 		 */
519 		sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
520 		sp->so_head = head_inh;
521 
522 		TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
523 		head_inh->so_qlen++;
524 
525 		/*
526 		 * NOTE:
527 		 * crfree() may block and release the tokens temporarily.
528 		 * However, we are fine here, since the transition is done.
529 		 */
530 		crfree(old_cr);
531 	}
532 
533 	lwkt_relpooltoken(head_inh);
534 	lwkt_relpooltoken(head);
535 
536 	if (do_wakeup) {
537 		/*
538 		 * "New" connections have arrived
539 		 */
540 		sorwakeup(head_inh);
541 		wakeup(&head_inh->so_timeo);
542 	}
543 }
544 
545 static int
546 soclose_sync(struct socket *so, int fflag)
547 {
548 	int error = 0;
549 
550 	if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
551 		so_pru_sync(so); /* unpend async prus */
552 
553 	if (so->so_pcb == NULL)
554 		goto discard;
555 
556 	if (so->so_state & SS_ISCONNECTED) {
557 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
558 			error = sodisconnect(so);
559 			if (error)
560 				goto drop;
561 		}
562 		if (so->so_options & SO_LINGER) {
563 			if ((so->so_state & SS_ISDISCONNECTING) &&
564 			    (fflag & FNONBLOCK))
565 				goto drop;
566 			while (so->so_state & SS_ISCONNECTED) {
567 				error = tsleep(&so->so_timeo, PCATCH,
568 					       "soclos", so->so_linger * hz);
569 				if (error)
570 					break;
571 			}
572 		}
573 	}
574 drop:
575 	if (so->so_pcb) {
576 		int error2;
577 
578 		error2 = so_pru_detach(so);
579 		if (error2 == EJUSTRETURN) {
580 			/*
581 			 * Protocol will call sodiscard()
582 			 * and sofree() for us.
583 			 */
584 			return error;
585 		}
586 		if (error == 0)
587 			error = error2;
588 	}
589 discard:
590 	sodiscard(so);
591 	sofree(so);		/* dispose of ref */
592 
593 	return (error);
594 }
595 
596 static void
597 soclose_fast_handler(netmsg_t msg)
598 {
599 	struct socket *so = msg->base.nm_so;
600 
601 	if (so->so_pcb == NULL)
602 		goto discard;
603 
604 	if ((so->so_state & SS_ISCONNECTED) &&
605 	    (so->so_state & SS_ISDISCONNECTING) == 0)
606 		so_pru_disconnect_direct(so);
607 
608 	if (so->so_pcb) {
609 		int error;
610 
611 		error = so_pru_detach_direct(so);
612 		if (error == EJUSTRETURN) {
613 			/*
614 			 * Protocol will call sodiscard()
615 			 * and sofree() for us.
616 			 */
617 			return;
618 		}
619 	}
620 discard:
621 	sodiscard(so);
622 	sofree(so);
623 }
624 
625 static void
626 soclose_fast(struct socket *so)
627 {
628 	struct netmsg_base *base = &so->so_clomsg;
629 
630 	netmsg_init(base, so, &netisr_apanic_rport, 0,
631 	    soclose_fast_handler);
632 	if (so->so_port == netisr_curport())
633 		lwkt_sendmsg_oncpu(so->so_port, &base->lmsg);
634 	else
635 		lwkt_sendmsg(so->so_port, &base->lmsg);
636 }
637 
638 /*
639  * Abort and destroy a socket.  Only one abort can be in progress
640  * at any given moment.
641  */
642 void
643 soabort_async(struct socket *so, boolean_t clr_head)
644 {
645 	/*
646 	 * Keep a reference before clearing the so_head
647 	 * to avoid racing socket close in netisr.
648 	 */
649 	soreference(so);
650 	if (clr_head)
651 		so->so_head = NULL;
652 	so_pru_abort_async(so);
653 }
654 
655 void
656 soabort_direct(struct socket *so)
657 {
658 	soreference(so);
659 	so_pru_abort_direct(so);
660 }
661 
662 /*
663  * so is passed in ref'd, which becomes owned by
664  * the cleared SS_NOFDREF flag.
665  */
666 void
667 soaccept_generic(struct socket *so)
668 {
669 	if ((so->so_state & SS_NOFDREF) == 0)
670 		panic("soaccept: !NOFDREF");
671 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
672 }
673 
674 int
675 soaccept(struct socket *so, struct sockaddr **nam)
676 {
677 	int error;
678 
679 	soaccept_generic(so);
680 	error = so_pru_accept(so, nam);
681 	return (error);
682 }
683 
684 int
685 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
686     boolean_t sync)
687 {
688 	int error;
689 
690 	if (so->so_options & SO_ACCEPTCONN)
691 		return (EOPNOTSUPP);
692 	/*
693 	 * If protocol is connection-based, can only connect once.
694 	 * Otherwise, if connected, try to disconnect first.
695 	 * This allows user to disconnect by connecting to, e.g.,
696 	 * a null address.
697 	 */
698 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
699 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
700 	    (error = sodisconnect(so)))) {
701 		error = EISCONN;
702 	} else {
703 		/*
704 		 * Prevent accumulated error from previous connection
705 		 * from biting us.
706 		 */
707 		so->so_error = 0;
708 		if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
709 			error = so_pru_connect_async(so, nam, td);
710 		else
711 			error = so_pru_connect(so, nam, td);
712 	}
713 	return (error);
714 }
715 
716 int
717 soconnect2(struct socket *so1, struct socket *so2)
718 {
719 	int error;
720 
721 	error = so_pru_connect2(so1, so2);
722 	return (error);
723 }
724 
725 int
726 sodisconnect(struct socket *so)
727 {
728 	int error;
729 
730 	if ((so->so_state & SS_ISCONNECTED) == 0) {
731 		error = ENOTCONN;
732 		goto bad;
733 	}
734 	if (so->so_state & SS_ISDISCONNECTING) {
735 		error = EALREADY;
736 		goto bad;
737 	}
738 	error = so_pru_disconnect(so);
739 bad:
740 	return (error);
741 }
742 
743 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
744 /*
745  * Send on a socket.
746  * If send must go all at once and message is larger than
747  * send buffering, then hard error.
748  * Lock against other senders.
749  * If must go all at once and not enough room now, then
750  * inform user that this would block and do nothing.
751  * Otherwise, if nonblocking, send as much as possible.
752  * The data to be sent is described by "uio" if nonzero,
753  * otherwise by the mbuf chain "top" (which must be null
754  * if uio is not).  Data provided in mbuf chain must be small
755  * enough to send all at once.
756  *
757  * Returns nonzero on error, timeout or signal; callers
758  * must check for short counts if EINTR/ERESTART are returned.
759  * Data and control buffers are freed on return.
760  */
761 int
762 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
763 	struct mbuf *top, struct mbuf *control, int flags,
764 	struct thread *td)
765 {
766 	struct mbuf **mp;
767 	struct mbuf *m;
768 	size_t resid;
769 	int space, len;
770 	int clen = 0, error, dontroute, mlen;
771 	int atomic = sosendallatonce(so) || top;
772 	int pru_flags;
773 
774 	if (uio) {
775 		resid = uio->uio_resid;
776 	} else {
777 		resid = (size_t)top->m_pkthdr.len;
778 #ifdef INVARIANTS
779 		len = 0;
780 		for (m = top; m; m = m->m_next)
781 			len += m->m_len;
782 		KKASSERT(top->m_pkthdr.len == len);
783 #endif
784 	}
785 
786 	/*
787 	 * WARNING!  resid is unsigned, space and len are signed.  space
788 	 * 	     can wind up negative if the sockbuf is overcommitted.
789 	 *
790 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
791 	 * type sockets since that's an error.
792 	 */
793 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
794 		error = EINVAL;
795 		goto out;
796 	}
797 
798 	dontroute =
799 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
800 	    (so->so_proto->pr_flags & PR_ATOMIC);
801 	if (td->td_lwp != NULL)
802 		td->td_lwp->lwp_ru.ru_msgsnd++;
803 	if (control)
804 		clen = control->m_len;
805 #define	gotoerr(errcode)	{ error = errcode; goto release; }
806 
807 restart:
808 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
809 	if (error)
810 		goto out;
811 
812 	do {
813 		if (so->so_state & SS_CANTSENDMORE)
814 			gotoerr(EPIPE);
815 		if (so->so_error) {
816 			error = so->so_error;
817 			so->so_error = 0;
818 			goto release;
819 		}
820 		if ((so->so_state & SS_ISCONNECTED) == 0) {
821 			/*
822 			 * `sendto' and `sendmsg' is allowed on a connection-
823 			 * based socket if it supports implied connect.
824 			 * Return ENOTCONN if not connected and no address is
825 			 * supplied.
826 			 */
827 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
828 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
829 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
830 				    !(resid == 0 && clen != 0))
831 					gotoerr(ENOTCONN);
832 			} else if (addr == NULL)
833 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
834 				   ENOTCONN : EDESTADDRREQ);
835 		}
836 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
837 		    clen > so->so_snd.ssb_hiwat) {
838 			gotoerr(EMSGSIZE);
839 		}
840 		space = ssb_space(&so->so_snd);
841 		if (flags & MSG_OOB)
842 			space += 1024;
843 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
844 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
845 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
846 				gotoerr(EWOULDBLOCK);
847 			ssb_unlock(&so->so_snd);
848 			error = ssb_wait(&so->so_snd);
849 			if (error)
850 				goto out;
851 			goto restart;
852 		}
853 		mp = &top;
854 		space -= clen;
855 		do {
856 		    if (uio == NULL) {
857 			/*
858 			 * Data is prepackaged in "top".
859 			 */
860 			resid = 0;
861 			if (flags & MSG_EOR)
862 				top->m_flags |= M_EOR;
863 		    } else do {
864 			if (resid > INT_MAX)
865 				resid = INT_MAX;
866 			m = m_getl((int)resid, M_WAITOK, MT_DATA,
867 				   top == NULL ? M_PKTHDR : 0, &mlen);
868 			if (top == NULL) {
869 				m->m_pkthdr.len = 0;
870 				m->m_pkthdr.rcvif = NULL;
871 			}
872 			len = imin((int)szmin(mlen, resid), space);
873 			if (resid < MINCLSIZE) {
874 				/*
875 				 * For datagram protocols, leave room
876 				 * for protocol headers in first mbuf.
877 				 */
878 				if (atomic && top == NULL && len < mlen)
879 					MH_ALIGN(m, len);
880 			}
881 			space -= len;
882 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
883 			resid = uio->uio_resid;
884 			m->m_len = len;
885 			*mp = m;
886 			top->m_pkthdr.len += len;
887 			if (error)
888 				goto release;
889 			mp = &m->m_next;
890 			if (resid == 0) {
891 				if (flags & MSG_EOR)
892 					top->m_flags |= M_EOR;
893 				break;
894 			}
895 		    } while (space > 0 && atomic);
896 		    if (dontroute)
897 			    so->so_options |= SO_DONTROUTE;
898 		    if (flags & MSG_OOB) {
899 		    	    pru_flags = PRUS_OOB;
900 		    } else if ((flags & MSG_EOF) &&
901 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
902 			       (resid == 0)) {
903 			    /*
904 			     * If the user set MSG_EOF, the protocol
905 			     * understands this flag and nothing left to
906 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
907 			     */
908 		    	    pru_flags = PRUS_EOF;
909 		    } else if (resid > 0 && space > 0) {
910 			    /* If there is more to send, set PRUS_MORETOCOME */
911 		    	    pru_flags = PRUS_MORETOCOME;
912 		    } else {
913 		    	    pru_flags = 0;
914 		    }
915 		    /*
916 		     * XXX all the SS_CANTSENDMORE checks previously
917 		     * done could be out of date.  We could have recieved
918 		     * a reset packet in an interrupt or maybe we slept
919 		     * while doing page faults in uiomove() etc. We could
920 		     * probably recheck again inside the splnet() protection
921 		     * here, but there are probably other places that this
922 		     * also happens.  We must rethink this.
923 		     */
924 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
925 		    if (dontroute)
926 			    so->so_options &= ~SO_DONTROUTE;
927 		    clen = 0;
928 		    control = NULL;
929 		    top = NULL;
930 		    mp = &top;
931 		    if (error)
932 			    goto release;
933 		} while (resid && space > 0);
934 	} while (resid);
935 
936 release:
937 	ssb_unlock(&so->so_snd);
938 out:
939 	if (top)
940 		m_freem(top);
941 	if (control)
942 		m_freem(control);
943 	return (error);
944 }
945 
946 #ifdef INET
947 /*
948  * A specialization of sosend() for UDP based on protocol-specific knowledge:
949  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
950  *	sosendallatonce() returns true,
951  *	the "atomic" variable is true,
952  *	and sosendudp() blocks until space is available for the entire send.
953  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
954  *	PR_IMPLOPCL flags set.
955  *   UDP has no out-of-band data.
956  *   UDP has no control data.
957  *   UDP does not support MSG_EOR.
958  */
959 int
960 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
961 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
962 {
963 	size_t resid;
964 	int error, pru_flags = 0;
965 	int space;
966 
967 	if (td->td_lwp != NULL)
968 		td->td_lwp->lwp_ru.ru_msgsnd++;
969 	if (control)
970 		m_freem(control);
971 
972 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
973 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
974 
975 restart:
976 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
977 	if (error)
978 		goto out;
979 
980 	if (so->so_state & SS_CANTSENDMORE)
981 		gotoerr(EPIPE);
982 	if (so->so_error) {
983 		error = so->so_error;
984 		so->so_error = 0;
985 		goto release;
986 	}
987 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
988 		gotoerr(EDESTADDRREQ);
989 	if (resid > so->so_snd.ssb_hiwat)
990 		gotoerr(EMSGSIZE);
991 	space = ssb_space(&so->so_snd);
992 	if (uio && (space < 0 || (size_t)space < resid)) {
993 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
994 			gotoerr(EWOULDBLOCK);
995 		ssb_unlock(&so->so_snd);
996 		error = ssb_wait(&so->so_snd);
997 		if (error)
998 			goto out;
999 		goto restart;
1000 	}
1001 
1002 	if (uio) {
1003 		int hdrlen = max_hdr;
1004 
1005 		/*
1006 		 * We try to optimize out the additional mbuf
1007 		 * allocations in M_PREPEND() on output path, e.g.
1008 		 * - udp_output(), when it tries to prepend protocol
1009 		 *   headers.
1010 		 * - Link layer output function, when it tries to
1011 		 *   prepend link layer header.
1012 		 *
1013 		 * This probably will not benefit any data that will
1014 		 * be fragmented, so this optimization is only performed
1015 		 * when the size of data and max size of protocol+link
1016 		 * headers fit into one mbuf cluster.
1017 		 */
1018 		if (uio->uio_resid > MCLBYTES - hdrlen ||
1019 		    !udp_sosend_prepend) {
1020 			top = m_uiomove(uio);
1021 			if (top == NULL)
1022 				goto release;
1023 		} else {
1024 			int nsize;
1025 
1026 			top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1027 			    MT_DATA, M_PKTHDR, &nsize);
1028 			KASSERT(nsize >= uio->uio_resid + hdrlen,
1029 			    ("sosendudp invalid nsize %d, "
1030 			     "resid %zu, hdrlen %d",
1031 			     nsize, uio->uio_resid, hdrlen));
1032 
1033 			top->m_len = uio->uio_resid;
1034 			top->m_pkthdr.len = uio->uio_resid;
1035 			top->m_data += hdrlen;
1036 
1037 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1038 			if (error)
1039 				goto out;
1040 		}
1041 	}
1042 
1043 	if (flags & MSG_DONTROUTE)
1044 		pru_flags |= PRUS_DONTROUTE;
1045 
1046 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1047 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1048 		error = 0;
1049 	} else {
1050 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1051 	}
1052 	top = NULL;		/* sent or freed in lower layer */
1053 
1054 release:
1055 	ssb_unlock(&so->so_snd);
1056 out:
1057 	if (top)
1058 		m_freem(top);
1059 	return (error);
1060 }
1061 
1062 int
1063 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1064 	struct mbuf *top, struct mbuf *control, int flags,
1065 	struct thread *td)
1066 {
1067 	struct mbuf **mp;
1068 	struct mbuf *m;
1069 	size_t resid;
1070 	int space, len;
1071 	int error, mlen;
1072 	int allatonce;
1073 	int pru_flags;
1074 
1075 	if (uio) {
1076 		KKASSERT(top == NULL);
1077 		allatonce = 0;
1078 		resid = uio->uio_resid;
1079 	} else {
1080 		allatonce = 1;
1081 		resid = (size_t)top->m_pkthdr.len;
1082 #ifdef INVARIANTS
1083 		len = 0;
1084 		for (m = top; m; m = m->m_next)
1085 			len += m->m_len;
1086 		KKASSERT(top->m_pkthdr.len == len);
1087 #endif
1088 	}
1089 
1090 	/*
1091 	 * WARNING!  resid is unsigned, space and len are signed.  space
1092 	 * 	     can wind up negative if the sockbuf is overcommitted.
1093 	 *
1094 	 * Also check to make sure that MSG_EOR isn't used on TCP
1095 	 */
1096 	if (flags & MSG_EOR) {
1097 		error = EINVAL;
1098 		goto out;
1099 	}
1100 
1101 	if (control) {
1102 		/* TCP doesn't do control messages (rights, creds, etc) */
1103 		if (control->m_len) {
1104 			error = EINVAL;
1105 			goto out;
1106 		}
1107 		m_freem(control);	/* empty control, just free it */
1108 		control = NULL;
1109 	}
1110 
1111 	if (td->td_lwp != NULL)
1112 		td->td_lwp->lwp_ru.ru_msgsnd++;
1113 
1114 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1115 
1116 restart:
1117 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1118 	if (error)
1119 		goto out;
1120 
1121 	do {
1122 		if (so->so_state & SS_CANTSENDMORE)
1123 			gotoerr(EPIPE);
1124 		if (so->so_error) {
1125 			error = so->so_error;
1126 			so->so_error = 0;
1127 			goto release;
1128 		}
1129 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1130 		    (so->so_state & SS_ISCONFIRMING) == 0)
1131 			gotoerr(ENOTCONN);
1132 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1133 			gotoerr(EMSGSIZE);
1134 
1135 		space = ssb_space_prealloc(&so->so_snd);
1136 		if (flags & MSG_OOB)
1137 			space += 1024;
1138 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1139 		    space < so->so_snd.ssb_lowat) {
1140 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1141 				gotoerr(EWOULDBLOCK);
1142 			ssb_unlock(&so->so_snd);
1143 			error = ssb_wait(&so->so_snd);
1144 			if (error)
1145 				goto out;
1146 			goto restart;
1147 		}
1148 		mp = &top;
1149 		do {
1150 		    int cnt = 0, async = 0;
1151 
1152 		    if (uio == NULL) {
1153 			/*
1154 			 * Data is prepackaged in "top".
1155 			 */
1156 			resid = 0;
1157 		    } else do {
1158 			if (resid > INT_MAX)
1159 				resid = INT_MAX;
1160 			if (tcp_sosend_jcluster) {
1161 				m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1162 					   top == NULL ? M_PKTHDR : 0, &mlen);
1163 			} else {
1164 				m = m_getl((int)resid, M_WAITOK, MT_DATA,
1165 					   top == NULL ? M_PKTHDR : 0, &mlen);
1166 			}
1167 			if (top == NULL) {
1168 				m->m_pkthdr.len = 0;
1169 				m->m_pkthdr.rcvif = NULL;
1170 			}
1171 			len = imin((int)szmin(mlen, resid), space);
1172 			space -= len;
1173 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1174 			resid = uio->uio_resid;
1175 			m->m_len = len;
1176 			*mp = m;
1177 			top->m_pkthdr.len += len;
1178 			if (error)
1179 				goto release;
1180 			mp = &m->m_next;
1181 			if (resid == 0)
1182 				break;
1183 			++cnt;
1184 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1185 
1186 		    if (tcp_sosend_async)
1187 			    async = 1;
1188 
1189 		    if (flags & MSG_OOB) {
1190 		    	    pru_flags = PRUS_OOB;
1191 			    async = 0;
1192 		    } else if ((flags & MSG_EOF) && resid == 0) {
1193 			    pru_flags = PRUS_EOF;
1194 		    } else if (resid > 0 && space > 0) {
1195 			    /* If there is more to send, set PRUS_MORETOCOME */
1196 		    	    pru_flags = PRUS_MORETOCOME;
1197 			    async = 1;
1198 		    } else {
1199 		    	    pru_flags = 0;
1200 		    }
1201 
1202 		    if (flags & MSG_SYNC)
1203 			    async = 0;
1204 
1205 		    /*
1206 		     * XXX all the SS_CANTSENDMORE checks previously
1207 		     * done could be out of date.  We could have recieved
1208 		     * a reset packet in an interrupt or maybe we slept
1209 		     * while doing page faults in uiomove() etc. We could
1210 		     * probably recheck again inside the splnet() protection
1211 		     * here, but there are probably other places that this
1212 		     * also happens.  We must rethink this.
1213 		     */
1214 		    for (m = top; m; m = m->m_next)
1215 			    ssb_preallocstream(&so->so_snd, m);
1216 		    if (!async) {
1217 			    error = so_pru_send(so, pru_flags, top,
1218 			        NULL, NULL, td);
1219 		    } else {
1220 			    so_pru_send_async(so, pru_flags, top,
1221 			        NULL, NULL, td);
1222 			    error = 0;
1223 		    }
1224 
1225 		    top = NULL;
1226 		    mp = &top;
1227 		    if (error)
1228 			    goto release;
1229 		} while (resid && space > 0);
1230 	} while (resid);
1231 
1232 release:
1233 	ssb_unlock(&so->so_snd);
1234 out:
1235 	if (top)
1236 		m_freem(top);
1237 	if (control)
1238 		m_freem(control);
1239 	return (error);
1240 }
1241 #endif
1242 
1243 /*
1244  * Implement receive operations on a socket.
1245  *
1246  * We depend on the way that records are added to the signalsockbuf
1247  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1248  * must begin with an address if the protocol so specifies,
1249  * followed by an optional mbuf or mbufs containing ancillary data,
1250  * and then zero or more mbufs of data.
1251  *
1252  * Although the signalsockbuf is locked, new data may still be appended.
1253  * A token inside the ssb_lock deals with MP issues and still allows
1254  * the network to access the socket if we block in a uio.
1255  *
1256  * The caller may receive the data as a single mbuf chain by supplying
1257  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1258  * only for the count in uio_resid.
1259  */
1260 int
1261 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1262 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1263 {
1264 	struct mbuf *m, *n;
1265 	struct mbuf *free_chain = NULL;
1266 	int flags, len, error, offset;
1267 	struct protosw *pr = so->so_proto;
1268 	int moff, type = 0;
1269 	size_t resid, orig_resid;
1270 	boolean_t free_rights = FALSE;
1271 
1272 	if (uio)
1273 		resid = uio->uio_resid;
1274 	else
1275 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1276 	orig_resid = resid;
1277 
1278 	if (psa)
1279 		*psa = NULL;
1280 	if (controlp)
1281 		*controlp = NULL;
1282 	if (flagsp)
1283 		flags = *flagsp &~ MSG_EOR;
1284 	else
1285 		flags = 0;
1286 	if (flags & MSG_OOB) {
1287 		m = m_get(M_WAITOK, MT_DATA);
1288 		if (m == NULL)
1289 			return (ENOBUFS);
1290 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1291 		if (error)
1292 			goto bad;
1293 		if (sio) {
1294 			do {
1295 				sbappend(sio, m);
1296 				KKASSERT(resid >= (size_t)m->m_len);
1297 				resid -= (size_t)m->m_len;
1298 			} while (resid > 0 && m);
1299 		} else {
1300 			do {
1301 				uio->uio_resid = resid;
1302 				error = uiomove(mtod(m, caddr_t),
1303 						(int)szmin(resid, m->m_len),
1304 						uio);
1305 				resid = uio->uio_resid;
1306 				m = m_free(m);
1307 			} while (uio->uio_resid && error == 0 && m);
1308 		}
1309 bad:
1310 		if (m)
1311 			m_freem(m);
1312 		return (error);
1313 	}
1314 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1315 		so_pru_rcvd(so, 0);
1316 
1317 	/*
1318 	 * The token interlocks against the protocol thread while
1319 	 * ssb_lock is a blocking lock against other userland entities.
1320 	 */
1321 	lwkt_gettoken(&so->so_rcv.ssb_token);
1322 restart:
1323 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1324 	if (error)
1325 		goto done;
1326 
1327 	m = so->so_rcv.ssb_mb;
1328 	/*
1329 	 * If we have less data than requested, block awaiting more
1330 	 * (subject to any timeout) if:
1331 	 *   1. the current count is less than the low water mark, or
1332 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1333 	 *	receive operation at once if we block (resid <= hiwat).
1334 	 *   3. MSG_DONTWAIT is not set
1335 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1336 	 * we have to do the receive in sections, and thus risk returning
1337 	 * a short count if a timeout or signal occurs after we start.
1338 	 */
1339 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1340 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1341 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1342 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1343 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1344 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1345 		if (so->so_error) {
1346 			if (m)
1347 				goto dontblock;
1348 			error = so->so_error;
1349 			if ((flags & MSG_PEEK) == 0)
1350 				so->so_error = 0;
1351 			goto release;
1352 		}
1353 		if (so->so_state & SS_CANTRCVMORE) {
1354 			if (m)
1355 				goto dontblock;
1356 			else
1357 				goto release;
1358 		}
1359 		for (; m; m = m->m_next) {
1360 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1361 				m = so->so_rcv.ssb_mb;
1362 				goto dontblock;
1363 			}
1364 		}
1365 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1366 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1367 			error = ENOTCONN;
1368 			goto release;
1369 		}
1370 		if (resid == 0)
1371 			goto release;
1372 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1373 			error = EWOULDBLOCK;
1374 			goto release;
1375 		}
1376 		ssb_unlock(&so->so_rcv);
1377 		error = ssb_wait(&so->so_rcv);
1378 		if (error)
1379 			goto done;
1380 		goto restart;
1381 	}
1382 dontblock:
1383 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1384 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1385 
1386 	/*
1387 	 * note: m should be == sb_mb here.  Cache the next record while
1388 	 * cleaning up.  Note that calling m_free*() will break out critical
1389 	 * section.
1390 	 */
1391 	KKASSERT(m == so->so_rcv.ssb_mb);
1392 
1393 	/*
1394 	 * Skip any address mbufs prepending the record.
1395 	 */
1396 	if (pr->pr_flags & PR_ADDR) {
1397 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1398 		orig_resid = 0;
1399 		if (psa)
1400 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1401 		if (flags & MSG_PEEK)
1402 			m = m->m_next;
1403 		else
1404 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1405 	}
1406 
1407 	/*
1408 	 * Skip any control mbufs prepending the record.
1409 	 */
1410 	while (m && m->m_type == MT_CONTROL && error == 0) {
1411 		if (flags & MSG_PEEK) {
1412 			if (controlp)
1413 				*controlp = m_copy(m, 0, m->m_len);
1414 			m = m->m_next;	/* XXX race */
1415 		} else {
1416 			const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1417 
1418 			if (controlp) {
1419 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1420 				if (pr->pr_domain->dom_externalize &&
1421 				    cm->cmsg_level == SOL_SOCKET &&
1422 				    cm->cmsg_type == SCM_RIGHTS) {
1423 					error = pr->pr_domain->dom_externalize
1424 					    (m, flags);
1425 				}
1426 				*controlp = m;
1427 				m = n;
1428 			} else {
1429 				if (cm->cmsg_level == SOL_SOCKET &&
1430 				    cm->cmsg_type == SCM_RIGHTS)
1431 					free_rights = TRUE;
1432 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1433 			}
1434 		}
1435 		if (controlp && *controlp) {
1436 			orig_resid = 0;
1437 			controlp = &(*controlp)->m_next;
1438 		}
1439 	}
1440 
1441 	/*
1442 	 * flag OOB data.
1443 	 */
1444 	if (m) {
1445 		type = m->m_type;
1446 		if (type == MT_OOBDATA)
1447 			flags |= MSG_OOB;
1448 	}
1449 
1450 	/*
1451 	 * Copy to the UIO or mbuf return chain (*mp).
1452 	 */
1453 	moff = 0;
1454 	offset = 0;
1455 	while (m && resid > 0 && error == 0) {
1456 		if (m->m_type == MT_OOBDATA) {
1457 			if (type != MT_OOBDATA)
1458 				break;
1459 		} else if (type == MT_OOBDATA)
1460 			break;
1461 		else
1462 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1463 			("receive 3"));
1464 		soclrstate(so, SS_RCVATMARK);
1465 		len = (resid > INT_MAX) ? INT_MAX : resid;
1466 		if (so->so_oobmark && len > so->so_oobmark - offset)
1467 			len = so->so_oobmark - offset;
1468 		if (len > m->m_len - moff)
1469 			len = m->m_len - moff;
1470 
1471 		/*
1472 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1473 		 * The SIO is dealt with when we eat the mbuf, but deal
1474 		 * with the resid here either way.
1475 		 */
1476 		if (uio) {
1477 			uio->uio_resid = resid;
1478 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1479 			resid = uio->uio_resid;
1480 			if (error)
1481 				goto release;
1482 		} else {
1483 			resid -= (size_t)len;
1484 		}
1485 
1486 		/*
1487 		 * Eat the entire mbuf or just a piece of it
1488 		 */
1489 		if (len == m->m_len - moff) {
1490 			if (m->m_flags & M_EOR)
1491 				flags |= MSG_EOR;
1492 			if (flags & MSG_PEEK) {
1493 				m = m->m_next;
1494 				moff = 0;
1495 			} else {
1496 				if (sio) {
1497 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1498 					sbappend(sio, m);
1499 					m = n;
1500 				} else {
1501 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1502 				}
1503 			}
1504 		} else {
1505 			if (flags & MSG_PEEK) {
1506 				moff += len;
1507 			} else {
1508 				if (sio) {
1509 					n = m_copym(m, 0, len, M_WAITOK);
1510 					if (n)
1511 						sbappend(sio, n);
1512 				}
1513 				m->m_data += len;
1514 				m->m_len -= len;
1515 				so->so_rcv.ssb_cc -= len;
1516 			}
1517 		}
1518 		if (so->so_oobmark) {
1519 			if ((flags & MSG_PEEK) == 0) {
1520 				so->so_oobmark -= len;
1521 				if (so->so_oobmark == 0) {
1522 					sosetstate(so, SS_RCVATMARK);
1523 					break;
1524 				}
1525 			} else {
1526 				offset += len;
1527 				if (offset == so->so_oobmark)
1528 					break;
1529 			}
1530 		}
1531 		if (flags & MSG_EOR)
1532 			break;
1533 		/*
1534 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1535 		 * we must not quit until resid == 0 or an error
1536 		 * termination.  If a signal/timeout occurs, return
1537 		 * with a short count but without error.
1538 		 * Keep signalsockbuf locked against other readers.
1539 		 */
1540 		while ((flags & MSG_WAITALL) && m == NULL &&
1541 		       resid > 0 && !sosendallatonce(so) &&
1542 		       so->so_rcv.ssb_mb == NULL) {
1543 			if (so->so_error || so->so_state & SS_CANTRCVMORE)
1544 				break;
1545 			/*
1546 			 * The window might have closed to zero, make
1547 			 * sure we send an ack now that we've drained
1548 			 * the buffer or we might end up blocking until
1549 			 * the idle takes over (5 seconds).
1550 			 */
1551 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1552 				so_pru_rcvd(so, flags);
1553 			error = ssb_wait(&so->so_rcv);
1554 			if (error) {
1555 				ssb_unlock(&so->so_rcv);
1556 				error = 0;
1557 				goto done;
1558 			}
1559 			m = so->so_rcv.ssb_mb;
1560 		}
1561 	}
1562 
1563 	/*
1564 	 * If an atomic read was requested but unread data still remains
1565 	 * in the record, set MSG_TRUNC.
1566 	 */
1567 	if (m && pr->pr_flags & PR_ATOMIC)
1568 		flags |= MSG_TRUNC;
1569 
1570 	/*
1571 	 * Cleanup.  If an atomic read was requested drop any unread data.
1572 	 */
1573 	if ((flags & MSG_PEEK) == 0) {
1574 		if (m && (pr->pr_flags & PR_ATOMIC))
1575 			sbdroprecord(&so->so_rcv.sb);
1576 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1577 			so_pru_rcvd(so, flags);
1578 	}
1579 
1580 	if (orig_resid == resid && orig_resid &&
1581 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1582 		ssb_unlock(&so->so_rcv);
1583 		goto restart;
1584 	}
1585 
1586 	if (flagsp)
1587 		*flagsp |= flags;
1588 release:
1589 	ssb_unlock(&so->so_rcv);
1590 done:
1591 	lwkt_reltoken(&so->so_rcv.ssb_token);
1592 	if (free_chain) {
1593 		if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1594 		    pr->pr_domain->dom_dispose)
1595 			pr->pr_domain->dom_dispose(free_chain);
1596 		m_freem(free_chain);
1597 	}
1598 	return (error);
1599 }
1600 
1601 int
1602 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1603 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1604 {
1605 	struct mbuf *m, *n;
1606 	struct mbuf *free_chain = NULL;
1607 	int flags, len, error, offset;
1608 	struct protosw *pr = so->so_proto;
1609 	int moff;
1610 	int didoob;
1611 	size_t resid, orig_resid, restmp;
1612 
1613 	if (uio)
1614 		resid = uio->uio_resid;
1615 	else
1616 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1617 	orig_resid = resid;
1618 
1619 	if (psa)
1620 		*psa = NULL;
1621 	if (controlp)
1622 		*controlp = NULL;
1623 	if (flagsp)
1624 		flags = *flagsp &~ MSG_EOR;
1625 	else
1626 		flags = 0;
1627 	if (flags & MSG_OOB) {
1628 		m = m_get(M_WAITOK, MT_DATA);
1629 		if (m == NULL)
1630 			return (ENOBUFS);
1631 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1632 		if (error)
1633 			goto bad;
1634 		if (sio) {
1635 			do {
1636 				sbappend(sio, m);
1637 				KKASSERT(resid >= (size_t)m->m_len);
1638 				resid -= (size_t)m->m_len;
1639 			} while (resid > 0 && m);
1640 		} else {
1641 			do {
1642 				uio->uio_resid = resid;
1643 				error = uiomove(mtod(m, caddr_t),
1644 						(int)szmin(resid, m->m_len),
1645 						uio);
1646 				resid = uio->uio_resid;
1647 				m = m_free(m);
1648 			} while (uio->uio_resid && error == 0 && m);
1649 		}
1650 bad:
1651 		if (m)
1652 			m_freem(m);
1653 		return (error);
1654 	}
1655 
1656 	/*
1657 	 * The token interlocks against the protocol thread while
1658 	 * ssb_lock is a blocking lock against other userland entities.
1659 	 *
1660 	 * Lock a limited number of mbufs (not all, so sbcompress() still
1661 	 * works well).  The token is used as an interlock for sbwait() so
1662 	 * release it afterwords.
1663 	 */
1664 restart:
1665 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1666 	if (error)
1667 		goto done;
1668 
1669 	lwkt_gettoken(&so->so_rcv.ssb_token);
1670 	m = so->so_rcv.ssb_mb;
1671 
1672 	/*
1673 	 * If we have less data than requested, block awaiting more
1674 	 * (subject to any timeout) if:
1675 	 *   1. the current count is less than the low water mark, or
1676 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1677 	 *	receive operation at once if we block (resid <= hiwat).
1678 	 *   3. MSG_DONTWAIT is not set
1679 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1680 	 * we have to do the receive in sections, and thus risk returning
1681 	 * a short count if a timeout or signal occurs after we start.
1682 	 */
1683 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1684 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1685 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1686 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1687 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1688 		if (so->so_error) {
1689 			if (m)
1690 				goto dontblock;
1691 			lwkt_reltoken(&so->so_rcv.ssb_token);
1692 			error = so->so_error;
1693 			if ((flags & MSG_PEEK) == 0)
1694 				so->so_error = 0;
1695 			goto release;
1696 		}
1697 		if (so->so_state & SS_CANTRCVMORE) {
1698 			if (m)
1699 				goto dontblock;
1700 			lwkt_reltoken(&so->so_rcv.ssb_token);
1701 			goto release;
1702 		}
1703 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1704 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1705 			lwkt_reltoken(&so->so_rcv.ssb_token);
1706 			error = ENOTCONN;
1707 			goto release;
1708 		}
1709 		if (resid == 0) {
1710 			lwkt_reltoken(&so->so_rcv.ssb_token);
1711 			goto release;
1712 		}
1713 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1714 			lwkt_reltoken(&so->so_rcv.ssb_token);
1715 			error = EWOULDBLOCK;
1716 			goto release;
1717 		}
1718 		ssb_unlock(&so->so_rcv);
1719 		error = ssb_wait(&so->so_rcv);
1720 		lwkt_reltoken(&so->so_rcv.ssb_token);
1721 		if (error)
1722 			goto done;
1723 		goto restart;
1724 	}
1725 
1726 	/*
1727 	 * Token still held
1728 	 */
1729 dontblock:
1730 	n = m;
1731 	restmp = 0;
1732 	while (n && restmp < resid) {
1733 		n->m_flags |= M_SOLOCKED;
1734 		restmp += n->m_len;
1735 		if (n->m_next == NULL)
1736 			n = n->m_nextpkt;
1737 		else
1738 			n = n->m_next;
1739 	}
1740 
1741 	/*
1742 	 * Release token for loop
1743 	 */
1744 	lwkt_reltoken(&so->so_rcv.ssb_token);
1745 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1746 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1747 
1748 	/*
1749 	 * note: m should be == sb_mb here.  Cache the next record while
1750 	 * cleaning up.  Note that calling m_free*() will break out critical
1751 	 * section.
1752 	 */
1753 	KKASSERT(m == so->so_rcv.ssb_mb);
1754 
1755 	/*
1756 	 * Copy to the UIO or mbuf return chain (*mp).
1757 	 *
1758 	 * NOTE: Token is not held for loop
1759 	 */
1760 	moff = 0;
1761 	offset = 0;
1762 	didoob = 0;
1763 
1764 	while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1765 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1766 		    ("receive 3"));
1767 
1768 		soclrstate(so, SS_RCVATMARK);
1769 		len = (resid > INT_MAX) ? INT_MAX : resid;
1770 		if (so->so_oobmark && len > so->so_oobmark - offset)
1771 			len = so->so_oobmark - offset;
1772 		if (len > m->m_len - moff)
1773 			len = m->m_len - moff;
1774 
1775 		/*
1776 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1777 		 * The SIO is dealt with when we eat the mbuf, but deal
1778 		 * with the resid here either way.
1779 		 */
1780 		if (uio) {
1781 			uio->uio_resid = resid;
1782 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1783 			resid = uio->uio_resid;
1784 			if (error)
1785 				goto release;
1786 		} else {
1787 			resid -= (size_t)len;
1788 		}
1789 
1790 		/*
1791 		 * Eat the entire mbuf or just a piece of it
1792 		 */
1793 		offset += len;
1794 		if (len == m->m_len - moff) {
1795 			m = m->m_next;
1796 			moff = 0;
1797 		} else {
1798 			moff += len;
1799 		}
1800 
1801 		/*
1802 		 * Check oobmark
1803 		 */
1804 		if (so->so_oobmark && offset == so->so_oobmark) {
1805 			didoob = 1;
1806 			break;
1807 		}
1808 	}
1809 
1810 	/*
1811 	 * Synchronize sockbuf with data we read.
1812 	 *
1813 	 * NOTE: (m) is junk on entry (it could be left over from the
1814 	 *	 previous loop).
1815 	 */
1816 	if ((flags & MSG_PEEK) == 0) {
1817 		lwkt_gettoken(&so->so_rcv.ssb_token);
1818 		m = so->so_rcv.ssb_mb;
1819 		while (m && offset >= m->m_len) {
1820 			if (so->so_oobmark) {
1821 				so->so_oobmark -= m->m_len;
1822 				if (so->so_oobmark == 0) {
1823 					sosetstate(so, SS_RCVATMARK);
1824 					didoob = 1;
1825 				}
1826 			}
1827 			offset -= m->m_len;
1828 			if (sio) {
1829 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1830 				sbappend(sio, m);
1831 				m = n;
1832 			} else {
1833 				m = sbunlinkmbuf(&so->so_rcv.sb,
1834 						 m, &free_chain);
1835 			}
1836 		}
1837 		if (offset) {
1838 			KKASSERT(m);
1839 			if (sio) {
1840 				n = m_copym(m, 0, offset, M_WAITOK);
1841 				if (n)
1842 					sbappend(sio, n);
1843 			}
1844 			m->m_data += offset;
1845 			m->m_len -= offset;
1846 			so->so_rcv.ssb_cc -= offset;
1847 			if (so->so_oobmark) {
1848 				so->so_oobmark -= offset;
1849 				if (so->so_oobmark == 0) {
1850 					sosetstate(so, SS_RCVATMARK);
1851 					didoob = 1;
1852 				}
1853 			}
1854 			offset = 0;
1855 		}
1856 		lwkt_reltoken(&so->so_rcv.ssb_token);
1857 	}
1858 
1859 	/*
1860 	 * If the MSG_WAITALL flag is set (for non-atomic socket),
1861 	 * we must not quit until resid == 0 or an error termination.
1862 	 *
1863 	 * If a signal/timeout occurs, return with a short count but without
1864 	 * error.
1865 	 *
1866 	 * Keep signalsockbuf locked against other readers.
1867 	 *
1868 	 * XXX if MSG_PEEK we currently do quit.
1869 	 */
1870 	if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1871 	    didoob == 0 && resid > 0 &&
1872 	    !sosendallatonce(so)) {
1873 		lwkt_gettoken(&so->so_rcv.ssb_token);
1874 		error = 0;
1875 		while ((m = so->so_rcv.ssb_mb) == NULL) {
1876 			if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1877 				error = so->so_error;
1878 				break;
1879 			}
1880 			/*
1881 			 * The window might have closed to zero, make
1882 			 * sure we send an ack now that we've drained
1883 			 * the buffer or we might end up blocking until
1884 			 * the idle takes over (5 seconds).
1885 			 */
1886 			if (so->so_pcb)
1887 				so_pru_rcvd_async(so);
1888 			if (so->so_rcv.ssb_mb == NULL)
1889 				error = ssb_wait(&so->so_rcv);
1890 			if (error) {
1891 				lwkt_reltoken(&so->so_rcv.ssb_token);
1892 				ssb_unlock(&so->so_rcv);
1893 				error = 0;
1894 				goto done;
1895 			}
1896 		}
1897 		if (m && error == 0)
1898 			goto dontblock;
1899 		lwkt_reltoken(&so->so_rcv.ssb_token);
1900 	}
1901 
1902 	/*
1903 	 * Token not held here.
1904 	 *
1905 	 * Cleanup.  If an atomic read was requested drop any unread data XXX
1906 	 */
1907 	if ((flags & MSG_PEEK) == 0) {
1908 		if (so->so_pcb)
1909 			so_pru_rcvd_async(so);
1910 	}
1911 
1912 	if (orig_resid == resid && orig_resid &&
1913 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1914 		ssb_unlock(&so->so_rcv);
1915 		goto restart;
1916 	}
1917 
1918 	if (flagsp)
1919 		*flagsp |= flags;
1920 release:
1921 	ssb_unlock(&so->so_rcv);
1922 done:
1923 	if (free_chain)
1924 		m_freem(free_chain);
1925 	return (error);
1926 }
1927 
1928 /*
1929  * Shut a socket down.  Note that we do not get a frontend lock as we
1930  * want to be able to shut the socket down even if another thread is
1931  * blocked in a read(), thus waking it up.
1932  */
1933 int
1934 soshutdown(struct socket *so, int how)
1935 {
1936 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1937 		return (EINVAL);
1938 
1939 	if (how != SHUT_WR) {
1940 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1941 		sorflush(so);
1942 		/*ssb_unlock(&so->so_rcv);*/
1943 	}
1944 	if (how != SHUT_RD)
1945 		return (so_pru_shutdown(so));
1946 	return (0);
1947 }
1948 
1949 void
1950 sorflush(struct socket *so)
1951 {
1952 	struct signalsockbuf *ssb = &so->so_rcv;
1953 	struct protosw *pr = so->so_proto;
1954 	struct signalsockbuf asb;
1955 
1956 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1957 
1958 	lwkt_gettoken(&ssb->ssb_token);
1959 	socantrcvmore(so);
1960 	asb = *ssb;
1961 
1962 	/*
1963 	 * Can't just blow up the ssb structure here
1964 	 */
1965 	bzero(&ssb->sb, sizeof(ssb->sb));
1966 	ssb->ssb_timeo = 0;
1967 	ssb->ssb_lowat = 0;
1968 	ssb->ssb_hiwat = 0;
1969 	ssb->ssb_mbmax = 0;
1970 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1971 
1972 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1973 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1974 	ssb_release(&asb, so);
1975 
1976 	lwkt_reltoken(&ssb->ssb_token);
1977 }
1978 
1979 #ifdef INET
1980 static int
1981 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1982 {
1983 	struct accept_filter_arg	*afap = NULL;
1984 	struct accept_filter	*afp;
1985 	struct so_accf	*af = so->so_accf;
1986 	int	error = 0;
1987 
1988 	/* do not set/remove accept filters on non listen sockets */
1989 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1990 		error = EINVAL;
1991 		goto out;
1992 	}
1993 
1994 	/* removing the filter */
1995 	if (sopt == NULL) {
1996 		if (af != NULL) {
1997 			if (af->so_accept_filter != NULL &&
1998 				af->so_accept_filter->accf_destroy != NULL) {
1999 				af->so_accept_filter->accf_destroy(so);
2000 			}
2001 			if (af->so_accept_filter_str != NULL) {
2002 				kfree(af->so_accept_filter_str, M_ACCF);
2003 			}
2004 			kfree(af, M_ACCF);
2005 			so->so_accf = NULL;
2006 		}
2007 		so->so_options &= ~SO_ACCEPTFILTER;
2008 		return (0);
2009 	}
2010 	/* adding a filter */
2011 	/* must remove previous filter first */
2012 	if (af != NULL) {
2013 		error = EINVAL;
2014 		goto out;
2015 	}
2016 	/* don't put large objects on the kernel stack */
2017 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2018 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2019 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
2020 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2021 	if (error)
2022 		goto out;
2023 	afp = accept_filt_get(afap->af_name);
2024 	if (afp == NULL) {
2025 		error = ENOENT;
2026 		goto out;
2027 	}
2028 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2029 	if (afp->accf_create != NULL) {
2030 		if (afap->af_name[0] != '\0') {
2031 			int len = strlen(afap->af_name) + 1;
2032 
2033 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
2034 							   M_WAITOK);
2035 			strcpy(af->so_accept_filter_str, afap->af_name);
2036 		}
2037 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2038 		if (af->so_accept_filter_arg == NULL) {
2039 			kfree(af->so_accept_filter_str, M_ACCF);
2040 			kfree(af, M_ACCF);
2041 			so->so_accf = NULL;
2042 			error = EINVAL;
2043 			goto out;
2044 		}
2045 	}
2046 	af->so_accept_filter = afp;
2047 	so->so_accf = af;
2048 	so->so_options |= SO_ACCEPTFILTER;
2049 out:
2050 	if (afap != NULL)
2051 		kfree(afap, M_TEMP);
2052 	return (error);
2053 }
2054 #endif /* INET */
2055 
2056 /*
2057  * Perhaps this routine, and sooptcopyout(), below, ought to come in
2058  * an additional variant to handle the case where the option value needs
2059  * to be some kind of integer, but not a specific size.
2060  * In addition to their use here, these functions are also called by the
2061  * protocol-level pr_ctloutput() routines.
2062  */
2063 int
2064 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2065 {
2066 	return soopt_to_kbuf(sopt, buf, len, minlen);
2067 }
2068 
2069 int
2070 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2071 {
2072 	size_t	valsize;
2073 
2074 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2075 	KKASSERT(kva_p(buf));
2076 
2077 	/*
2078 	 * If the user gives us more than we wanted, we ignore it,
2079 	 * but if we don't get the minimum length the caller
2080 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
2081 	 * is set to however much we actually retrieved.
2082 	 */
2083 	if ((valsize = sopt->sopt_valsize) < minlen)
2084 		return EINVAL;
2085 	if (valsize > len)
2086 		sopt->sopt_valsize = valsize = len;
2087 
2088 	bcopy(sopt->sopt_val, buf, valsize);
2089 	return 0;
2090 }
2091 
2092 
2093 int
2094 sosetopt(struct socket *so, struct sockopt *sopt)
2095 {
2096 	int	error, optval;
2097 	struct	linger l;
2098 	struct	timeval tv;
2099 	u_long  val;
2100 	struct signalsockbuf *sotmp;
2101 
2102 	error = 0;
2103 	sopt->sopt_dir = SOPT_SET;
2104 	if (sopt->sopt_level != SOL_SOCKET) {
2105 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2106 			return (so_pr_ctloutput(so, sopt));
2107 		}
2108 		error = ENOPROTOOPT;
2109 	} else {
2110 		switch (sopt->sopt_name) {
2111 #ifdef INET
2112 		case SO_ACCEPTFILTER:
2113 			error = do_setopt_accept_filter(so, sopt);
2114 			if (error)
2115 				goto bad;
2116 			break;
2117 #endif /* INET */
2118 		case SO_LINGER:
2119 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2120 			if (error)
2121 				goto bad;
2122 
2123 			so->so_linger = l.l_linger;
2124 			if (l.l_onoff)
2125 				so->so_options |= SO_LINGER;
2126 			else
2127 				so->so_options &= ~SO_LINGER;
2128 			break;
2129 
2130 		case SO_DEBUG:
2131 		case SO_KEEPALIVE:
2132 		case SO_DONTROUTE:
2133 		case SO_USELOOPBACK:
2134 		case SO_BROADCAST:
2135 		case SO_REUSEADDR:
2136 		case SO_REUSEPORT:
2137 		case SO_OOBINLINE:
2138 		case SO_TIMESTAMP:
2139 		case SO_NOSIGPIPE:
2140 			error = sooptcopyin(sopt, &optval, sizeof optval,
2141 					    sizeof optval);
2142 			if (error)
2143 				goto bad;
2144 			if (optval)
2145 				so->so_options |= sopt->sopt_name;
2146 			else
2147 				so->so_options &= ~sopt->sopt_name;
2148 			break;
2149 
2150 		case SO_SNDBUF:
2151 		case SO_RCVBUF:
2152 		case SO_SNDLOWAT:
2153 		case SO_RCVLOWAT:
2154 			error = sooptcopyin(sopt, &optval, sizeof optval,
2155 					    sizeof optval);
2156 			if (error)
2157 				goto bad;
2158 
2159 			/*
2160 			 * Values < 1 make no sense for any of these
2161 			 * options, so disallow them.
2162 			 */
2163 			if (optval < 1) {
2164 				error = EINVAL;
2165 				goto bad;
2166 			}
2167 
2168 			switch (sopt->sopt_name) {
2169 			case SO_SNDBUF:
2170 			case SO_RCVBUF:
2171 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2172 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2173 				    so,
2174 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2175 					error = ENOBUFS;
2176 					goto bad;
2177 				}
2178 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2179 						&so->so_snd : &so->so_rcv;
2180 				atomic_clear_int(&sotmp->ssb_flags,
2181 						 SSB_AUTOSIZE);
2182 				break;
2183 
2184 			/*
2185 			 * Make sure the low-water is never greater than
2186 			 * the high-water.
2187 			 */
2188 			case SO_SNDLOWAT:
2189 				so->so_snd.ssb_lowat =
2190 				    (optval > so->so_snd.ssb_hiwat) ?
2191 				    so->so_snd.ssb_hiwat : optval;
2192 				atomic_clear_int(&so->so_snd.ssb_flags,
2193 						 SSB_AUTOLOWAT);
2194 				break;
2195 			case SO_RCVLOWAT:
2196 				so->so_rcv.ssb_lowat =
2197 				    (optval > so->so_rcv.ssb_hiwat) ?
2198 				    so->so_rcv.ssb_hiwat : optval;
2199 				atomic_clear_int(&so->so_rcv.ssb_flags,
2200 						 SSB_AUTOLOWAT);
2201 				break;
2202 			}
2203 			break;
2204 
2205 		case SO_SNDTIMEO:
2206 		case SO_RCVTIMEO:
2207 			error = sooptcopyin(sopt, &tv, sizeof tv,
2208 					    sizeof tv);
2209 			if (error)
2210 				goto bad;
2211 
2212 			/* assert(hz > 0); */
2213 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2214 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2215 				error = EDOM;
2216 				goto bad;
2217 			}
2218 			/* assert(tick > 0); */
2219 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2220 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2221 			if (val > INT_MAX) {
2222 				error = EDOM;
2223 				goto bad;
2224 			}
2225 			if (val == 0 && tv.tv_usec != 0)
2226 				val = 1;
2227 
2228 			switch (sopt->sopt_name) {
2229 			case SO_SNDTIMEO:
2230 				so->so_snd.ssb_timeo = val;
2231 				break;
2232 			case SO_RCVTIMEO:
2233 				so->so_rcv.ssb_timeo = val;
2234 				break;
2235 			}
2236 			break;
2237 		default:
2238 			error = ENOPROTOOPT;
2239 			break;
2240 		}
2241 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2242 			(void) so_pr_ctloutput(so, sopt);
2243 		}
2244 	}
2245 bad:
2246 	return (error);
2247 }
2248 
2249 /* Helper routine for getsockopt */
2250 int
2251 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2252 {
2253 	soopt_from_kbuf(sopt, buf, len);
2254 	return 0;
2255 }
2256 
2257 void
2258 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2259 {
2260 	size_t	valsize;
2261 
2262 	if (len == 0) {
2263 		sopt->sopt_valsize = 0;
2264 		return;
2265 	}
2266 
2267 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2268 	KKASSERT(kva_p(buf));
2269 
2270 	/*
2271 	 * Documented get behavior is that we always return a value,
2272 	 * possibly truncated to fit in the user's buffer.
2273 	 * Traditional behavior is that we always tell the user
2274 	 * precisely how much we copied, rather than something useful
2275 	 * like the total amount we had available for her.
2276 	 * Note that this interface is not idempotent; the entire answer must
2277 	 * generated ahead of time.
2278 	 */
2279 	valsize = szmin(len, sopt->sopt_valsize);
2280 	sopt->sopt_valsize = valsize;
2281 	if (sopt->sopt_val != 0) {
2282 		bcopy(buf, sopt->sopt_val, valsize);
2283 	}
2284 }
2285 
2286 int
2287 sogetopt(struct socket *so, struct sockopt *sopt)
2288 {
2289 	int	error, optval;
2290 	long	optval_l;
2291 	struct	linger l;
2292 	struct	timeval tv;
2293 #ifdef INET
2294 	struct accept_filter_arg *afap;
2295 #endif
2296 
2297 	error = 0;
2298 	sopt->sopt_dir = SOPT_GET;
2299 	if (sopt->sopt_level != SOL_SOCKET) {
2300 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2301 			return (so_pr_ctloutput(so, sopt));
2302 		} else
2303 			return (ENOPROTOOPT);
2304 	} else {
2305 		switch (sopt->sopt_name) {
2306 #ifdef INET
2307 		case SO_ACCEPTFILTER:
2308 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2309 				return (EINVAL);
2310 			afap = kmalloc(sizeof(*afap), M_TEMP,
2311 				       M_WAITOK | M_ZERO);
2312 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2313 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2314 				if (so->so_accf->so_accept_filter_str != NULL)
2315 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2316 			}
2317 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2318 			kfree(afap, M_TEMP);
2319 			break;
2320 #endif /* INET */
2321 
2322 		case SO_LINGER:
2323 			l.l_onoff = so->so_options & SO_LINGER;
2324 			l.l_linger = so->so_linger;
2325 			error = sooptcopyout(sopt, &l, sizeof l);
2326 			break;
2327 
2328 		case SO_USELOOPBACK:
2329 		case SO_DONTROUTE:
2330 		case SO_DEBUG:
2331 		case SO_KEEPALIVE:
2332 		case SO_REUSEADDR:
2333 		case SO_REUSEPORT:
2334 		case SO_BROADCAST:
2335 		case SO_OOBINLINE:
2336 		case SO_TIMESTAMP:
2337 		case SO_NOSIGPIPE:
2338 			optval = so->so_options & sopt->sopt_name;
2339 integer:
2340 			error = sooptcopyout(sopt, &optval, sizeof optval);
2341 			break;
2342 
2343 		case SO_TYPE:
2344 			optval = so->so_type;
2345 			goto integer;
2346 
2347 		case SO_ERROR:
2348 			optval = so->so_error;
2349 			so->so_error = 0;
2350 			goto integer;
2351 
2352 		case SO_SNDBUF:
2353 			optval = so->so_snd.ssb_hiwat;
2354 			goto integer;
2355 
2356 		case SO_RCVBUF:
2357 			optval = so->so_rcv.ssb_hiwat;
2358 			goto integer;
2359 
2360 		case SO_SNDLOWAT:
2361 			optval = so->so_snd.ssb_lowat;
2362 			goto integer;
2363 
2364 		case SO_RCVLOWAT:
2365 			optval = so->so_rcv.ssb_lowat;
2366 			goto integer;
2367 
2368 		case SO_SNDTIMEO:
2369 		case SO_RCVTIMEO:
2370 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2371 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2372 
2373 			tv.tv_sec = optval / hz;
2374 			tv.tv_usec = (optval % hz) * ustick;
2375 			error = sooptcopyout(sopt, &tv, sizeof tv);
2376 			break;
2377 
2378 		case SO_SNDSPACE:
2379 			optval_l = ssb_space(&so->so_snd);
2380 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2381 			break;
2382 
2383 		case SO_CPUHINT:
2384 			optval = -1; /* no hint */
2385 			goto integer;
2386 
2387 		default:
2388 			error = ENOPROTOOPT;
2389 			break;
2390 		}
2391 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2392 			so_pr_ctloutput(so, sopt);
2393 		return (error);
2394 	}
2395 }
2396 
2397 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2398 int
2399 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2400 {
2401 	struct mbuf *m, *m_prev;
2402 	int sopt_size = sopt->sopt_valsize, msize;
2403 
2404 	m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2405 		   0, &msize);
2406 	if (m == NULL)
2407 		return (ENOBUFS);
2408 	m->m_len = min(msize, sopt_size);
2409 	sopt_size -= m->m_len;
2410 	*mp = m;
2411 	m_prev = m;
2412 
2413 	while (sopt_size > 0) {
2414 		m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2415 			   MT_DATA, 0, &msize);
2416 		if (m == NULL) {
2417 			m_freem(*mp);
2418 			return (ENOBUFS);
2419 		}
2420 		m->m_len = min(msize, sopt_size);
2421 		sopt_size -= m->m_len;
2422 		m_prev->m_next = m;
2423 		m_prev = m;
2424 	}
2425 	return (0);
2426 }
2427 
2428 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2429 int
2430 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2431 {
2432 	soopt_to_mbuf(sopt, m);
2433 	return 0;
2434 }
2435 
2436 void
2437 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2438 {
2439 	size_t valsize;
2440 	void *val;
2441 
2442 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2443 	KKASSERT(kva_p(m));
2444 	if (sopt->sopt_val == NULL)
2445 		return;
2446 	val = sopt->sopt_val;
2447 	valsize = sopt->sopt_valsize;
2448 	while (m != NULL && valsize >= m->m_len) {
2449 		bcopy(val, mtod(m, char *), m->m_len);
2450 		valsize -= m->m_len;
2451 		val = (caddr_t)val + m->m_len;
2452 		m = m->m_next;
2453 	}
2454 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2455 		panic("ip6_sooptmcopyin");
2456 }
2457 
2458 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2459 int
2460 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2461 {
2462 	return soopt_from_mbuf(sopt, m);
2463 }
2464 
2465 int
2466 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2467 {
2468 	struct mbuf *m0 = m;
2469 	size_t valsize = 0;
2470 	size_t maxsize;
2471 	void *val;
2472 
2473 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2474 	KKASSERT(kva_p(m));
2475 	if (sopt->sopt_val == NULL)
2476 		return 0;
2477 	val = sopt->sopt_val;
2478 	maxsize = sopt->sopt_valsize;
2479 	while (m != NULL && maxsize >= m->m_len) {
2480 		bcopy(mtod(m, char *), val, m->m_len);
2481 	       maxsize -= m->m_len;
2482 	       val = (caddr_t)val + m->m_len;
2483 	       valsize += m->m_len;
2484 	       m = m->m_next;
2485 	}
2486 	if (m != NULL) {
2487 		/* enough soopt buffer should be given from user-land */
2488 		m_freem(m0);
2489 		return (EINVAL);
2490 	}
2491 	sopt->sopt_valsize = valsize;
2492 	return 0;
2493 }
2494 
2495 void
2496 sohasoutofband(struct socket *so)
2497 {
2498 	if (so->so_sigio != NULL)
2499 		pgsigio(so->so_sigio, SIGURG, 0);
2500 	/*
2501 	 * NOTE:
2502 	 * There is no need to use NOTE_OOB as KNOTE hint here:
2503 	 * soread filter depends on so_oobmark and SS_RCVATMARK
2504 	 * so_state.  NOTE_OOB would cause unnecessary penalty
2505 	 * in KNOTE, if there was knote processing contention.
2506 	 */
2507 	KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2508 }
2509 
2510 int
2511 sokqfilter(struct file *fp, struct knote *kn)
2512 {
2513 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2514 	struct signalsockbuf *ssb;
2515 
2516 	switch (kn->kn_filter) {
2517 	case EVFILT_READ:
2518 		if (so->so_options & SO_ACCEPTCONN)
2519 			kn->kn_fop = &solisten_filtops;
2520 		else
2521 			kn->kn_fop = &soread_filtops;
2522 		ssb = &so->so_rcv;
2523 		break;
2524 	case EVFILT_WRITE:
2525 		kn->kn_fop = &sowrite_filtops;
2526 		ssb = &so->so_snd;
2527 		break;
2528 	case EVFILT_EXCEPT:
2529 		kn->kn_fop = &soexcept_filtops;
2530 		ssb = &so->so_rcv;
2531 		break;
2532 	default:
2533 		return (EOPNOTSUPP);
2534 	}
2535 
2536 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2537 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2538 	return (0);
2539 }
2540 
2541 static void
2542 filt_sordetach(struct knote *kn)
2543 {
2544 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2545 
2546 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2547 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2548 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2549 }
2550 
2551 /*ARGSUSED*/
2552 static int
2553 filt_soread(struct knote *kn, long hint __unused)
2554 {
2555 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2556 
2557 	if (kn->kn_sfflags & NOTE_OOB) {
2558 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2559 			kn->kn_fflags |= NOTE_OOB;
2560 			return (1);
2561 		}
2562 		return (0);
2563 	}
2564 	kn->kn_data = so->so_rcv.ssb_cc;
2565 
2566 	if (so->so_state & SS_CANTRCVMORE) {
2567 		/*
2568 		 * Only set NODATA if all data has been exhausted.
2569 		 */
2570 		if (kn->kn_data == 0)
2571 			kn->kn_flags |= EV_NODATA;
2572 		kn->kn_flags |= EV_EOF;
2573 		kn->kn_fflags = so->so_error;
2574 		return (1);
2575 	}
2576 	if (so->so_error)	/* temporary udp error */
2577 		return (1);
2578 	if (kn->kn_sfflags & NOTE_LOWAT)
2579 		return (kn->kn_data >= kn->kn_sdata);
2580 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2581 		!TAILQ_EMPTY(&so->so_comp));
2582 }
2583 
2584 static void
2585 filt_sowdetach(struct knote *kn)
2586 {
2587 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2588 
2589 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2590 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2591 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2592 }
2593 
2594 /*ARGSUSED*/
2595 static int
2596 filt_sowrite(struct knote *kn, long hint __unused)
2597 {
2598 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2599 
2600 	if (so->so_snd.ssb_flags & SSB_PREALLOC)
2601 		kn->kn_data = ssb_space_prealloc(&so->so_snd);
2602 	else
2603 		kn->kn_data = ssb_space(&so->so_snd);
2604 
2605 	if (so->so_state & SS_CANTSENDMORE) {
2606 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2607 		kn->kn_fflags = so->so_error;
2608 		return (1);
2609 	}
2610 	if (so->so_error)	/* temporary udp error */
2611 		return (1);
2612 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2613 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2614 		return (0);
2615 	if (kn->kn_sfflags & NOTE_LOWAT)
2616 		return (kn->kn_data >= kn->kn_sdata);
2617 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2618 }
2619 
2620 /*ARGSUSED*/
2621 static int
2622 filt_solisten(struct knote *kn, long hint __unused)
2623 {
2624 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2625 	int qlen = so->so_qlen;
2626 
2627 	if (soavailconn > 0 && qlen > soavailconn)
2628 		qlen = soavailconn;
2629 	kn->kn_data = qlen;
2630 
2631 	return (!TAILQ_EMPTY(&so->so_comp));
2632 }
2633