xref: /dflybsd-src/sys/kern/uipc_socket.c (revision d78d3a2272f5ecf9e0b570e362128240417a1b85)
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/socketvar2.h>
93 #include <sys/spinlock2.h>
94 
95 #include <machine/limits.h>
96 
97 #ifdef INET
98 extern int tcp_sosend_agglim;
99 extern int tcp_sosend_async;
100 extern int tcp_sosend_jcluster;
101 extern int udp_sosend_async;
102 extern int udp_sosend_prepend;
103 
104 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
105 #endif /* INET */
106 
107 static void 	filt_sordetach(struct knote *kn);
108 static int 	filt_soread(struct knote *kn, long hint);
109 static void 	filt_sowdetach(struct knote *kn);
110 static int	filt_sowrite(struct knote *kn, long hint);
111 static int	filt_solisten(struct knote *kn, long hint);
112 
113 static int	soclose_sync(struct socket *so, int fflag);
114 static void	soclose_fast(struct socket *so);
115 
116 static struct filterops solisten_filtops =
117 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_solisten };
118 static struct filterops soread_filtops =
119 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
120 static struct filterops sowrite_filtops =
121 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sowdetach, filt_sowrite };
122 static struct filterops soexcept_filtops =
123 	{ FILTEROP_ISFD|FILTEROP_MPSAFE, NULL, filt_sordetach, filt_soread };
124 
125 MALLOC_DEFINE(M_SOCKET, "socket", "socket struct");
126 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
127 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
128 
129 
130 static int somaxconn = SOMAXCONN;
131 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
132     &somaxconn, 0, "Maximum pending socket connection queue size");
133 
134 static int use_soclose_fast = 1;
135 SYSCTL_INT(_kern_ipc, OID_AUTO, soclose_fast, CTLFLAG_RW,
136     &use_soclose_fast, 0, "Fast socket close");
137 
138 int use_soaccept_pred_fast = 1;
139 SYSCTL_INT(_kern_ipc, OID_AUTO, soaccept_pred_fast, CTLFLAG_RW,
140     &use_soaccept_pred_fast, 0, "Fast socket accept predication");
141 
142 int use_sendfile_async = 1;
143 SYSCTL_INT(_kern_ipc, OID_AUTO, sendfile_async, CTLFLAG_RW,
144     &use_sendfile_async, 0, "sendfile uses asynchronized pru_send");
145 
146 int use_soconnect_async = 1;
147 SYSCTL_INT(_kern_ipc, OID_AUTO, soconnect_async, CTLFLAG_RW,
148     &use_soconnect_async, 0, "soconnect uses asynchronized pru_connect");
149 
150 static int use_socreate_fast = 1;
151 SYSCTL_INT(_kern_ipc, OID_AUTO, socreate_fast, CTLFLAG_RW,
152     &use_socreate_fast, 0, "Fast socket creation");
153 
154 static int soavailconn = 32;
155 SYSCTL_INT(_kern_ipc, OID_AUTO, soavailconn, CTLFLAG_RW,
156     &soavailconn, 0, "Maximum available socket connection queue size");
157 
158 /*
159  * Socket operation routines.
160  * These routines are called by the routines in
161  * sys_socket.c or from a system process, and
162  * implement the semantics of socket operations by
163  * switching out to the protocol specific routines.
164  */
165 
166 /*
167  * Get a socket structure, and initialize it.
168  * Note that it would probably be better to allocate socket
169  * and PCB at the same time, but I'm not convinced that all
170  * the protocols can be easily modified to do this.
171  */
172 struct socket *
173 soalloc(int waitok, struct protosw *pr)
174 {
175 	globaldata_t gd = mycpu;
176 	struct socket *so;
177 	unsigned waitmask;
178 
179 	waitmask = waitok ? M_WAITOK : M_NOWAIT;
180 	so = kmalloc(sizeof(struct socket), M_SOCKET, M_ZERO|waitmask);
181 	if (so) {
182 		/* XXX race condition for reentrant kernel */
183 		so->so_proto = pr;
184 		TAILQ_INIT(&so->so_aiojobq);
185 		TAILQ_INIT(&so->so_rcv.ssb_mlist);
186 		TAILQ_INIT(&so->so_snd.ssb_mlist);
187 		lwkt_token_init(&so->so_rcv.ssb_token, "rcvtok");
188 		lwkt_token_init(&so->so_snd.ssb_token, "sndtok");
189 		spin_init(&so->so_rcvd_spin, "soalloc");
190 		netmsg_init(&so->so_rcvd_msg.base, so, &netisr_adone_rport,
191 		    MSGF_DROPABLE | MSGF_PRIORITY,
192 		    so->so_proto->pr_usrreqs->pru_rcvd);
193 		so->so_rcvd_msg.nm_pru_flags |= PRUR_ASYNC;
194 		so->so_state = SS_NOFDREF;
195 		so->so_refs = 1;
196 		so->so_inum = gd->gd_anoninum++ * ncpus + gd->gd_cpuid + 2;
197 	}
198 	return so;
199 }
200 
201 int
202 socreate(int dom, struct socket **aso, int type,
203 	int proto, struct thread *td)
204 {
205 	struct proc *p = td->td_proc;
206 	struct protosw *prp;
207 	struct socket *so;
208 	struct pru_attach_info ai;
209 	int error;
210 
211 	if (proto)
212 		prp = pffindproto(dom, proto, type);
213 	else
214 		prp = pffindtype(dom, type);
215 
216 	if (prp == NULL || prp->pr_usrreqs->pru_attach == 0)
217 		return (EPROTONOSUPPORT);
218 
219 	if (p->p_ucred->cr_prison && jail_socket_unixiproute_only &&
220 	    prp->pr_domain->dom_family != PF_LOCAL &&
221 	    prp->pr_domain->dom_family != PF_INET &&
222 	    prp->pr_domain->dom_family != PF_INET6 &&
223 	    prp->pr_domain->dom_family != PF_ROUTE) {
224 		return (EPROTONOSUPPORT);
225 	}
226 
227 	if (prp->pr_type != type)
228 		return (EPROTOTYPE);
229 	so = soalloc(p != NULL, prp);
230 	if (so == NULL)
231 		return (ENOBUFS);
232 
233 	/*
234 	 * Callers of socreate() presumably will connect up a descriptor
235 	 * and call soclose() if they cannot.  This represents our so_refs
236 	 * (which should be 1) from soalloc().
237 	 */
238 	soclrstate(so, SS_NOFDREF);
239 
240 	/*
241 	 * Set a default port for protocol processing.  No action will occur
242 	 * on the socket on this port until an inpcb is attached to it and
243 	 * is able to match incoming packets, or until the socket becomes
244 	 * available to userland.
245 	 *
246 	 * We normally default the socket to the protocol thread on cpu 0,
247 	 * if protocol does not provide its own method to initialize the
248 	 * default port.
249 	 *
250 	 * If PR_SYNC_PORT is set (unix domain sockets) there is no protocol
251 	 * thread and all pr_*()/pru_*() calls are executed synchronously.
252 	 */
253 	if (prp->pr_flags & PR_SYNC_PORT)
254 		so->so_port = &netisr_sync_port;
255 	else if (prp->pr_initport != NULL)
256 		so->so_port = prp->pr_initport();
257 	else
258 		so->so_port = netisr_cpuport(0);
259 
260 	TAILQ_INIT(&so->so_incomp);
261 	TAILQ_INIT(&so->so_comp);
262 	so->so_type = type;
263 	so->so_cred = crhold(p->p_ucred);
264 	ai.sb_rlimit = &p->p_rlimit[RLIMIT_SBSIZE];
265 	ai.p_ucred = p->p_ucred;
266 	ai.fd_rdir = p->p_fd->fd_rdir;
267 
268 	/*
269 	 * Auto-sizing of socket buffers is managed by the protocols and
270 	 * the appropriate flags must be set in the pru_attach function.
271 	 */
272 	if (use_socreate_fast && prp->pr_usrreqs->pru_preattach)
273 		error = so_pru_attach_fast(so, proto, &ai);
274 	else
275 		error = so_pru_attach(so, proto, &ai);
276 	if (error) {
277 		sosetstate(so, SS_NOFDREF);
278 		sofree(so);	/* from soalloc */
279 		return error;
280 	}
281 
282 	/*
283 	 * NOTE: Returns referenced socket.
284 	 */
285 	*aso = so;
286 	return (0);
287 }
288 
289 int
290 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
291 {
292 	int error;
293 
294 	error = so_pru_bind(so, nam, td);
295 	return (error);
296 }
297 
298 static void
299 sodealloc(struct socket *so)
300 {
301 	KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) == 0);
302 
303 #ifdef INVARIANTS
304 	if (so->so_options & SO_ACCEPTCONN) {
305 		KASSERT(TAILQ_EMPTY(&so->so_comp), ("so_comp is not empty"));
306 		KASSERT(TAILQ_EMPTY(&so->so_incomp),
307 		    ("so_incomp is not empty"));
308 	}
309 #endif
310 
311 	if (so->so_rcv.ssb_hiwat)
312 		(void)chgsbsize(so->so_cred->cr_uidinfo,
313 		    &so->so_rcv.ssb_hiwat, 0, RLIM_INFINITY);
314 	if (so->so_snd.ssb_hiwat)
315 		(void)chgsbsize(so->so_cred->cr_uidinfo,
316 		    &so->so_snd.ssb_hiwat, 0, RLIM_INFINITY);
317 #ifdef INET
318 	/* remove accept filter if present */
319 	if (so->so_accf != NULL)
320 		do_setopt_accept_filter(so, NULL);
321 #endif /* INET */
322 	crfree(so->so_cred);
323 	if (so->so_faddr != NULL)
324 		kfree(so->so_faddr, M_SONAME);
325 	kfree(so, M_SOCKET);
326 }
327 
328 int
329 solisten(struct socket *so, int backlog, struct thread *td)
330 {
331 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING))
332 		return (EINVAL);
333 
334 	lwkt_gettoken(&so->so_rcv.ssb_token);
335 	if (TAILQ_EMPTY(&so->so_comp))
336 		so->so_options |= SO_ACCEPTCONN;
337 	lwkt_reltoken(&so->so_rcv.ssb_token);
338 	if (backlog < 0 || backlog > somaxconn)
339 		backlog = somaxconn;
340 	so->so_qlimit = backlog;
341 	return so_pru_listen(so, td);
342 }
343 
344 static void
345 soqflush(struct socket *so)
346 {
347 	lwkt_getpooltoken(so);
348 	if (so->so_options & SO_ACCEPTCONN) {
349 		struct socket *sp;
350 
351 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
352 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
353 			    SS_INCOMP);
354 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
355 			so->so_incqlen--;
356 			soclrstate(sp, SS_INCOMP);
357 			soabort_async(sp, TRUE);
358 		}
359 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
360 			KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) ==
361 			    SS_COMP);
362 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
363 			so->so_qlen--;
364 			soclrstate(sp, SS_COMP);
365 			soabort_async(sp, TRUE);
366 		}
367 	}
368 	lwkt_relpooltoken(so);
369 }
370 
371 /*
372  * Destroy a disconnected socket.  This routine is a NOP if entities
373  * still have a reference on the socket:
374  *
375  *	so_pcb -	The protocol stack still has a reference
376  *	SS_NOFDREF -	There is no longer a file pointer reference
377  */
378 void
379 sofree(struct socket *so)
380 {
381 	struct socket *head;
382 
383 	/*
384 	 * This is a bit hackish at the moment.  We need to interlock
385 	 * any accept queue we are on before we potentially lose the
386 	 * last reference to avoid races against a re-reference from
387 	 * someone operating on the queue.
388 	 */
389 	while ((head = so->so_head) != NULL) {
390 		lwkt_getpooltoken(head);
391 		if (so->so_head == head)
392 			break;
393 		lwkt_relpooltoken(head);
394 	}
395 
396 	/*
397 	 * Arbitrage the last free.
398 	 */
399 	KKASSERT(so->so_refs > 0);
400 	if (atomic_fetchadd_int(&so->so_refs, -1) != 1) {
401 		if (head)
402 			lwkt_relpooltoken(head);
403 		return;
404 	}
405 
406 	KKASSERT(so->so_pcb == NULL && (so->so_state & SS_NOFDREF));
407 	KKASSERT((so->so_state & SS_ASSERTINPROG) == 0);
408 
409 	if (head != NULL) {
410 		/*
411 		 * We're done, remove ourselves from the accept queue we are
412 		 * on, if we are on one.
413 		 */
414 		if (so->so_state & SS_INCOMP) {
415 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
416 			    SS_INCOMP);
417 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
418 			head->so_incqlen--;
419 		} else if (so->so_state & SS_COMP) {
420 			/*
421 			 * We must not decommission a socket that's
422 			 * on the accept(2) queue.  If we do, then
423 			 * accept(2) may hang after select(2) indicated
424 			 * that the listening socket was ready.
425 			 */
426 			KKASSERT((so->so_state & (SS_INCOMP | SS_COMP)) ==
427 			    SS_COMP);
428 			lwkt_relpooltoken(head);
429 			return;
430 		} else {
431 			panic("sofree: not queued");
432 		}
433 		soclrstate(so, SS_INCOMP);
434 		so->so_head = NULL;
435 		lwkt_relpooltoken(head);
436 	} else {
437 		/* Flush accept queues, if we are accepting. */
438 		soqflush(so);
439 	}
440 	ssb_release(&so->so_snd, so);
441 	sorflush(so);
442 	sodealloc(so);
443 }
444 
445 /*
446  * Close a socket on last file table reference removal.
447  * Initiate disconnect if connected.
448  * Free socket when disconnect complete.
449  */
450 int
451 soclose(struct socket *so, int fflag)
452 {
453 	int error;
454 
455 	funsetown(&so->so_sigio);
456 	sosetstate(so, SS_ISCLOSING);
457 	if (!use_soclose_fast ||
458 	    (so->so_proto->pr_flags & PR_SYNC_PORT) ||
459 	    ((so->so_state & SS_ISCONNECTED) &&
460 	     (so->so_options & SO_LINGER) &&
461 	     so->so_linger != 0)) {
462 		error = soclose_sync(so, fflag);
463 	} else {
464 		soclose_fast(so);
465 		error = 0;
466 	}
467 	return error;
468 }
469 
470 void
471 sodiscard(struct socket *so)
472 {
473 	if (so->so_state & SS_NOFDREF)
474 		panic("soclose: NOFDREF");
475 	sosetstate(so, SS_NOFDREF);	/* take ref */
476 }
477 
478 /*
479  * Append the completed queue of head to head_inh (inherting listen socket).
480  */
481 void
482 soinherit(struct socket *head, struct socket *head_inh)
483 {
484 	boolean_t do_wakeup = FALSE;
485 
486 	KASSERT(head->so_options & SO_ACCEPTCONN,
487 	    ("head does not accept connection"));
488 	KASSERT(head_inh->so_options & SO_ACCEPTCONN,
489 	    ("head_inh does not accept connection"));
490 
491 	lwkt_getpooltoken(head);
492 	lwkt_getpooltoken(head_inh);
493 
494 	if (head->so_qlen > 0)
495 		do_wakeup = TRUE;
496 
497 	while (!TAILQ_EMPTY(&head->so_comp)) {
498 		struct ucred *old_cr;
499 		struct socket *sp;
500 
501 		sp = TAILQ_FIRST(&head->so_comp);
502 		KKASSERT((sp->so_state & (SS_INCOMP | SS_COMP)) == SS_COMP);
503 
504 		/*
505 		 * Remove this socket from the current listen socket
506 		 * completed queue.
507 		 */
508 		TAILQ_REMOVE(&head->so_comp, sp, so_list);
509 		head->so_qlen--;
510 
511 		/* Save the old ucred for later free. */
512 		old_cr = sp->so_cred;
513 
514 		/*
515 		 * Install this socket to the inheriting listen socket
516 		 * completed queue.
517 		 */
518 		sp->so_cred = crhold(head_inh->so_cred); /* non-blocking */
519 		sp->so_head = head_inh;
520 
521 		TAILQ_INSERT_TAIL(&head_inh->so_comp, sp, so_list);
522 		head_inh->so_qlen++;
523 
524 		/*
525 		 * NOTE:
526 		 * crfree() may block and release the tokens temporarily.
527 		 * However, we are fine here, since the transition is done.
528 		 */
529 		crfree(old_cr);
530 	}
531 
532 	lwkt_relpooltoken(head_inh);
533 	lwkt_relpooltoken(head);
534 
535 	if (do_wakeup) {
536 		/*
537 		 * "New" connections have arrived
538 		 */
539 		sorwakeup(head_inh);
540 		wakeup(&head_inh->so_timeo);
541 	}
542 }
543 
544 static int
545 soclose_sync(struct socket *so, int fflag)
546 {
547 	int error = 0;
548 
549 	if ((so->so_proto->pr_flags & PR_SYNC_PORT) == 0)
550 		so_pru_sync(so); /* unpend async prus */
551 
552 	if (so->so_pcb == NULL)
553 		goto discard;
554 
555 	if (so->so_state & SS_ISCONNECTED) {
556 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
557 			error = sodisconnect(so);
558 			if (error)
559 				goto drop;
560 		}
561 		if (so->so_options & SO_LINGER) {
562 			if ((so->so_state & SS_ISDISCONNECTING) &&
563 			    (fflag & FNONBLOCK))
564 				goto drop;
565 			while (so->so_state & SS_ISCONNECTED) {
566 				error = tsleep(&so->so_timeo, PCATCH,
567 					       "soclos", so->so_linger * hz);
568 				if (error)
569 					break;
570 			}
571 		}
572 	}
573 drop:
574 	if (so->so_pcb) {
575 		int error2;
576 
577 		error2 = so_pru_detach(so);
578 		if (error2 == EJUSTRETURN) {
579 			/*
580 			 * Protocol will call sodiscard()
581 			 * and sofree() for us.
582 			 */
583 			return error;
584 		}
585 		if (error == 0)
586 			error = error2;
587 	}
588 discard:
589 	sodiscard(so);
590 	sofree(so);		/* dispose of ref */
591 
592 	return (error);
593 }
594 
595 static void
596 soclose_fast_handler(netmsg_t msg)
597 {
598 	struct socket *so = msg->base.nm_so;
599 
600 	if (so->so_pcb == NULL)
601 		goto discard;
602 
603 	if ((so->so_state & SS_ISCONNECTED) &&
604 	    (so->so_state & SS_ISDISCONNECTING) == 0)
605 		so_pru_disconnect_direct(so);
606 
607 	if (so->so_pcb) {
608 		int error;
609 
610 		error = so_pru_detach_direct(so);
611 		if (error == EJUSTRETURN) {
612 			/*
613 			 * Protocol will call sodiscard()
614 			 * and sofree() for us.
615 			 */
616 			return;
617 		}
618 	}
619 discard:
620 	sodiscard(so);
621 	sofree(so);
622 }
623 
624 static void
625 soclose_fast(struct socket *so)
626 {
627 	struct netmsg_base *base = &so->so_clomsg;
628 
629 	netmsg_init(base, so, &netisr_apanic_rport, 0,
630 	    soclose_fast_handler);
631 	if (so->so_port == netisr_curport())
632 		lwkt_sendmsg_oncpu(so->so_port, &base->lmsg);
633 	else
634 		lwkt_sendmsg(so->so_port, &base->lmsg);
635 }
636 
637 /*
638  * Abort and destroy a socket.  Only one abort can be in progress
639  * at any given moment.
640  */
641 void
642 soabort_async(struct socket *so, boolean_t clr_head)
643 {
644 	/*
645 	 * Keep a reference before clearing the so_head
646 	 * to avoid racing socket close in netisr.
647 	 */
648 	soreference(so);
649 	if (clr_head)
650 		so->so_head = NULL;
651 	so_pru_abort_async(so);
652 }
653 
654 void
655 soabort_direct(struct socket *so)
656 {
657 	soreference(so);
658 	so_pru_abort_direct(so);
659 }
660 
661 /*
662  * so is passed in ref'd, which becomes owned by
663  * the cleared SS_NOFDREF flag.
664  */
665 void
666 soaccept_generic(struct socket *so)
667 {
668 	if ((so->so_state & SS_NOFDREF) == 0)
669 		panic("soaccept: !NOFDREF");
670 	soclrstate(so, SS_NOFDREF);	/* owned by lack of SS_NOFDREF */
671 }
672 
673 int
674 soaccept(struct socket *so, struct sockaddr **nam)
675 {
676 	int error;
677 
678 	soaccept_generic(so);
679 	error = so_pru_accept(so, nam);
680 	return (error);
681 }
682 
683 int
684 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td,
685     boolean_t sync)
686 {
687 	int error;
688 
689 	if (so->so_options & SO_ACCEPTCONN)
690 		return (EOPNOTSUPP);
691 	/*
692 	 * If protocol is connection-based, can only connect once.
693 	 * Otherwise, if connected, try to disconnect first.
694 	 * This allows user to disconnect by connecting to, e.g.,
695 	 * a null address.
696 	 */
697 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
698 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
699 	    (error = sodisconnect(so)))) {
700 		error = EISCONN;
701 	} else {
702 		/*
703 		 * Prevent accumulated error from previous connection
704 		 * from biting us.
705 		 */
706 		so->so_error = 0;
707 		if (!sync && so->so_proto->pr_usrreqs->pru_preconnect)
708 			error = so_pru_connect_async(so, nam, td);
709 		else
710 			error = so_pru_connect(so, nam, td);
711 	}
712 	return (error);
713 }
714 
715 int
716 soconnect2(struct socket *so1, struct socket *so2)
717 {
718 	int error;
719 
720 	error = so_pru_connect2(so1, so2);
721 	return (error);
722 }
723 
724 int
725 sodisconnect(struct socket *so)
726 {
727 	int error;
728 
729 	if ((so->so_state & SS_ISCONNECTED) == 0) {
730 		error = ENOTCONN;
731 		goto bad;
732 	}
733 	if (so->so_state & SS_ISDISCONNECTING) {
734 		error = EALREADY;
735 		goto bad;
736 	}
737 	error = so_pru_disconnect(so);
738 bad:
739 	return (error);
740 }
741 
742 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
743 /*
744  * Send on a socket.
745  * If send must go all at once and message is larger than
746  * send buffering, then hard error.
747  * Lock against other senders.
748  * If must go all at once and not enough room now, then
749  * inform user that this would block and do nothing.
750  * Otherwise, if nonblocking, send as much as possible.
751  * The data to be sent is described by "uio" if nonzero,
752  * otherwise by the mbuf chain "top" (which must be null
753  * if uio is not).  Data provided in mbuf chain must be small
754  * enough to send all at once.
755  *
756  * Returns nonzero on error, timeout or signal; callers
757  * must check for short counts if EINTR/ERESTART are returned.
758  * Data and control buffers are freed on return.
759  */
760 int
761 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
762 	struct mbuf *top, struct mbuf *control, int flags,
763 	struct thread *td)
764 {
765 	struct mbuf **mp;
766 	struct mbuf *m;
767 	size_t resid;
768 	int space, len;
769 	int clen = 0, error, dontroute, mlen;
770 	int atomic = sosendallatonce(so) || top;
771 	int pru_flags;
772 
773 	if (uio) {
774 		resid = uio->uio_resid;
775 	} else {
776 		resid = (size_t)top->m_pkthdr.len;
777 #ifdef INVARIANTS
778 		len = 0;
779 		for (m = top; m; m = m->m_next)
780 			len += m->m_len;
781 		KKASSERT(top->m_pkthdr.len == len);
782 #endif
783 	}
784 
785 	/*
786 	 * WARNING!  resid is unsigned, space and len are signed.  space
787 	 * 	     can wind up negative if the sockbuf is overcommitted.
788 	 *
789 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
790 	 * type sockets since that's an error.
791 	 */
792 	if (so->so_type == SOCK_STREAM && (flags & MSG_EOR)) {
793 		error = EINVAL;
794 		goto out;
795 	}
796 
797 	dontroute =
798 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
799 	    (so->so_proto->pr_flags & PR_ATOMIC);
800 	if (td->td_lwp != NULL)
801 		td->td_lwp->lwp_ru.ru_msgsnd++;
802 	if (control)
803 		clen = control->m_len;
804 #define	gotoerr(errcode)	{ error = errcode; goto release; }
805 
806 restart:
807 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
808 	if (error)
809 		goto out;
810 
811 	do {
812 		if (so->so_state & SS_CANTSENDMORE)
813 			gotoerr(EPIPE);
814 		if (so->so_error) {
815 			error = so->so_error;
816 			so->so_error = 0;
817 			goto release;
818 		}
819 		if ((so->so_state & SS_ISCONNECTED) == 0) {
820 			/*
821 			 * `sendto' and `sendmsg' is allowed on a connection-
822 			 * based socket if it supports implied connect.
823 			 * Return ENOTCONN if not connected and no address is
824 			 * supplied.
825 			 */
826 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
827 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
828 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
829 				    !(resid == 0 && clen != 0))
830 					gotoerr(ENOTCONN);
831 			} else if (addr == NULL)
832 			    gotoerr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
833 				   ENOTCONN : EDESTADDRREQ);
834 		}
835 		if ((atomic && resid > so->so_snd.ssb_hiwat) ||
836 		    clen > so->so_snd.ssb_hiwat) {
837 			gotoerr(EMSGSIZE);
838 		}
839 		space = ssb_space(&so->so_snd);
840 		if (flags & MSG_OOB)
841 			space += 1024;
842 		if ((space < 0 || (size_t)space < resid + clen) && uio &&
843 		    (atomic || space < so->so_snd.ssb_lowat || space < clen)) {
844 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
845 				gotoerr(EWOULDBLOCK);
846 			ssb_unlock(&so->so_snd);
847 			error = ssb_wait(&so->so_snd);
848 			if (error)
849 				goto out;
850 			goto restart;
851 		}
852 		mp = &top;
853 		space -= clen;
854 		do {
855 		    if (uio == NULL) {
856 			/*
857 			 * Data is prepackaged in "top".
858 			 */
859 			resid = 0;
860 			if (flags & MSG_EOR)
861 				top->m_flags |= M_EOR;
862 		    } else do {
863 			if (resid > INT_MAX)
864 				resid = INT_MAX;
865 			m = m_getl((int)resid, M_WAITOK, MT_DATA,
866 				   top == NULL ? M_PKTHDR : 0, &mlen);
867 			if (top == NULL) {
868 				m->m_pkthdr.len = 0;
869 				m->m_pkthdr.rcvif = NULL;
870 			}
871 			len = imin((int)szmin(mlen, resid), space);
872 			if (resid < MINCLSIZE) {
873 				/*
874 				 * For datagram protocols, leave room
875 				 * for protocol headers in first mbuf.
876 				 */
877 				if (atomic && top == NULL && len < mlen)
878 					MH_ALIGN(m, len);
879 			}
880 			space -= len;
881 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
882 			resid = uio->uio_resid;
883 			m->m_len = len;
884 			*mp = m;
885 			top->m_pkthdr.len += len;
886 			if (error)
887 				goto release;
888 			mp = &m->m_next;
889 			if (resid == 0) {
890 				if (flags & MSG_EOR)
891 					top->m_flags |= M_EOR;
892 				break;
893 			}
894 		    } while (space > 0 && atomic);
895 		    if (dontroute)
896 			    so->so_options |= SO_DONTROUTE;
897 		    if (flags & MSG_OOB) {
898 		    	    pru_flags = PRUS_OOB;
899 		    } else if ((flags & MSG_EOF) &&
900 		    	       (so->so_proto->pr_flags & PR_IMPLOPCL) &&
901 			       (resid == 0)) {
902 			    /*
903 			     * If the user set MSG_EOF, the protocol
904 			     * understands this flag and nothing left to
905 			     * send then use PRU_SEND_EOF instead of PRU_SEND.
906 			     */
907 		    	    pru_flags = PRUS_EOF;
908 		    } else if (resid > 0 && space > 0) {
909 			    /* If there is more to send, set PRUS_MORETOCOME */
910 		    	    pru_flags = PRUS_MORETOCOME;
911 		    } else {
912 		    	    pru_flags = 0;
913 		    }
914 		    /*
915 		     * XXX all the SS_CANTSENDMORE checks previously
916 		     * done could be out of date.  We could have recieved
917 		     * a reset packet in an interrupt or maybe we slept
918 		     * while doing page faults in uiomove() etc. We could
919 		     * probably recheck again inside the splnet() protection
920 		     * here, but there are probably other places that this
921 		     * also happens.  We must rethink this.
922 		     */
923 		    error = so_pru_send(so, pru_flags, top, addr, control, td);
924 		    if (dontroute)
925 			    so->so_options &= ~SO_DONTROUTE;
926 		    clen = 0;
927 		    control = NULL;
928 		    top = NULL;
929 		    mp = &top;
930 		    if (error)
931 			    goto release;
932 		} while (resid && space > 0);
933 	} while (resid);
934 
935 release:
936 	ssb_unlock(&so->so_snd);
937 out:
938 	if (top)
939 		m_freem(top);
940 	if (control)
941 		m_freem(control);
942 	return (error);
943 }
944 
945 #ifdef INET
946 /*
947  * A specialization of sosend() for UDP based on protocol-specific knowledge:
948  *   so->so_proto->pr_flags has the PR_ATOMIC field set.  This means that
949  *	sosendallatonce() returns true,
950  *	the "atomic" variable is true,
951  *	and sosendudp() blocks until space is available for the entire send.
952  *   so->so_proto->pr_flags does not have the PR_CONNREQUIRED or
953  *	PR_IMPLOPCL flags set.
954  *   UDP has no out-of-band data.
955  *   UDP has no control data.
956  *   UDP does not support MSG_EOR.
957  */
958 int
959 sosendudp(struct socket *so, struct sockaddr *addr, struct uio *uio,
960 	  struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
961 {
962 	size_t resid;
963 	int error, pru_flags = 0;
964 	int space;
965 
966 	if (td->td_lwp != NULL)
967 		td->td_lwp->lwp_ru.ru_msgsnd++;
968 	if (control)
969 		m_freem(control);
970 
971 	KASSERT((uio && !top) || (top && !uio), ("bad arguments to sosendudp"));
972 	resid = uio ? uio->uio_resid : (size_t)top->m_pkthdr.len;
973 
974 restart:
975 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
976 	if (error)
977 		goto out;
978 
979 	if (so->so_state & SS_CANTSENDMORE)
980 		gotoerr(EPIPE);
981 	if (so->so_error) {
982 		error = so->so_error;
983 		so->so_error = 0;
984 		goto release;
985 	}
986 	if (!(so->so_state & SS_ISCONNECTED) && addr == NULL)
987 		gotoerr(EDESTADDRREQ);
988 	if (resid > so->so_snd.ssb_hiwat)
989 		gotoerr(EMSGSIZE);
990 	space = ssb_space(&so->so_snd);
991 	if (uio && (space < 0 || (size_t)space < resid)) {
992 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
993 			gotoerr(EWOULDBLOCK);
994 		ssb_unlock(&so->so_snd);
995 		error = ssb_wait(&so->so_snd);
996 		if (error)
997 			goto out;
998 		goto restart;
999 	}
1000 
1001 	if (uio) {
1002 		int hdrlen = max_hdr;
1003 
1004 		/*
1005 		 * We try to optimize out the additional mbuf
1006 		 * allocations in M_PREPEND() on output path, e.g.
1007 		 * - udp_output(), when it tries to prepend protocol
1008 		 *   headers.
1009 		 * - Link layer output function, when it tries to
1010 		 *   prepend link layer header.
1011 		 *
1012 		 * This probably will not benefit any data that will
1013 		 * be fragmented, so this optimization is only performed
1014 		 * when the size of data and max size of protocol+link
1015 		 * headers fit into one mbuf cluster.
1016 		 */
1017 		if (uio->uio_resid > MCLBYTES - hdrlen ||
1018 		    !udp_sosend_prepend) {
1019 			top = m_uiomove(uio);
1020 			if (top == NULL)
1021 				goto release;
1022 		} else {
1023 			int nsize;
1024 
1025 			top = m_getl(uio->uio_resid + hdrlen, M_WAITOK,
1026 			    MT_DATA, M_PKTHDR, &nsize);
1027 			KASSERT(nsize >= uio->uio_resid + hdrlen,
1028 			    ("sosendudp invalid nsize %d, "
1029 			     "resid %zu, hdrlen %d",
1030 			     nsize, uio->uio_resid, hdrlen));
1031 
1032 			top->m_len = uio->uio_resid;
1033 			top->m_pkthdr.len = uio->uio_resid;
1034 			top->m_data += hdrlen;
1035 
1036 			error = uiomove(mtod(top, caddr_t), top->m_len, uio);
1037 			if (error)
1038 				goto out;
1039 		}
1040 	}
1041 
1042 	if (flags & MSG_DONTROUTE)
1043 		pru_flags |= PRUS_DONTROUTE;
1044 
1045 	if (udp_sosend_async && (flags & MSG_SYNC) == 0) {
1046 		so_pru_send_async(so, pru_flags, top, addr, NULL, td);
1047 		error = 0;
1048 	} else {
1049 		error = so_pru_send(so, pru_flags, top, addr, NULL, td);
1050 	}
1051 	top = NULL;		/* sent or freed in lower layer */
1052 
1053 release:
1054 	ssb_unlock(&so->so_snd);
1055 out:
1056 	if (top)
1057 		m_freem(top);
1058 	return (error);
1059 }
1060 
1061 int
1062 sosendtcp(struct socket *so, struct sockaddr *addr, struct uio *uio,
1063 	struct mbuf *top, struct mbuf *control, int flags,
1064 	struct thread *td)
1065 {
1066 	struct mbuf **mp;
1067 	struct mbuf *m;
1068 	size_t resid;
1069 	int space, len;
1070 	int error, mlen;
1071 	int allatonce;
1072 	int pru_flags;
1073 
1074 	if (uio) {
1075 		KKASSERT(top == NULL);
1076 		allatonce = 0;
1077 		resid = uio->uio_resid;
1078 	} else {
1079 		allatonce = 1;
1080 		resid = (size_t)top->m_pkthdr.len;
1081 #ifdef INVARIANTS
1082 		len = 0;
1083 		for (m = top; m; m = m->m_next)
1084 			len += m->m_len;
1085 		KKASSERT(top->m_pkthdr.len == len);
1086 #endif
1087 	}
1088 
1089 	/*
1090 	 * WARNING!  resid is unsigned, space and len are signed.  space
1091 	 * 	     can wind up negative if the sockbuf is overcommitted.
1092 	 *
1093 	 * Also check to make sure that MSG_EOR isn't used on TCP
1094 	 */
1095 	if (flags & MSG_EOR) {
1096 		error = EINVAL;
1097 		goto out;
1098 	}
1099 
1100 	if (control) {
1101 		/* TCP doesn't do control messages (rights, creds, etc) */
1102 		if (control->m_len) {
1103 			error = EINVAL;
1104 			goto out;
1105 		}
1106 		m_freem(control);	/* empty control, just free it */
1107 		control = NULL;
1108 	}
1109 
1110 	if (td->td_lwp != NULL)
1111 		td->td_lwp->lwp_ru.ru_msgsnd++;
1112 
1113 #define	gotoerr(errcode)	{ error = errcode; goto release; }
1114 
1115 restart:
1116 	error = ssb_lock(&so->so_snd, SBLOCKWAIT(flags));
1117 	if (error)
1118 		goto out;
1119 
1120 	do {
1121 		if (so->so_state & SS_CANTSENDMORE)
1122 			gotoerr(EPIPE);
1123 		if (so->so_error) {
1124 			error = so->so_error;
1125 			so->so_error = 0;
1126 			goto release;
1127 		}
1128 		if ((so->so_state & SS_ISCONNECTED) == 0 &&
1129 		    (so->so_state & SS_ISCONFIRMING) == 0)
1130 			gotoerr(ENOTCONN);
1131 		if (allatonce && resid > so->so_snd.ssb_hiwat)
1132 			gotoerr(EMSGSIZE);
1133 
1134 		space = ssb_space_prealloc(&so->so_snd);
1135 		if (flags & MSG_OOB)
1136 			space += 1024;
1137 		if ((space < 0 || (size_t)space < resid) && !allatonce &&
1138 		    space < so->so_snd.ssb_lowat) {
1139 			if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT))
1140 				gotoerr(EWOULDBLOCK);
1141 			ssb_unlock(&so->so_snd);
1142 			error = ssb_wait(&so->so_snd);
1143 			if (error)
1144 				goto out;
1145 			goto restart;
1146 		}
1147 		mp = &top;
1148 		do {
1149 		    int cnt = 0, async = 0;
1150 
1151 		    if (uio == NULL) {
1152 			/*
1153 			 * Data is prepackaged in "top".
1154 			 */
1155 			resid = 0;
1156 		    } else do {
1157 			if (resid > INT_MAX)
1158 				resid = INT_MAX;
1159 			if (tcp_sosend_jcluster) {
1160 				m = m_getlj((int)resid, M_WAITOK, MT_DATA,
1161 					   top == NULL ? M_PKTHDR : 0, &mlen);
1162 			} else {
1163 				m = m_getl((int)resid, M_WAITOK, MT_DATA,
1164 					   top == NULL ? M_PKTHDR : 0, &mlen);
1165 			}
1166 			if (top == NULL) {
1167 				m->m_pkthdr.len = 0;
1168 				m->m_pkthdr.rcvif = NULL;
1169 			}
1170 			len = imin((int)szmin(mlen, resid), space);
1171 			space -= len;
1172 			error = uiomove(mtod(m, caddr_t), (size_t)len, uio);
1173 			resid = uio->uio_resid;
1174 			m->m_len = len;
1175 			*mp = m;
1176 			top->m_pkthdr.len += len;
1177 			if (error)
1178 				goto release;
1179 			mp = &m->m_next;
1180 			if (resid == 0)
1181 				break;
1182 			++cnt;
1183 		    } while (space > 0 && cnt < tcp_sosend_agglim);
1184 
1185 		    if (tcp_sosend_async)
1186 			    async = 1;
1187 
1188 		    if (flags & MSG_OOB) {
1189 		    	    pru_flags = PRUS_OOB;
1190 			    async = 0;
1191 		    } else if ((flags & MSG_EOF) && resid == 0) {
1192 			    pru_flags = PRUS_EOF;
1193 		    } else if (resid > 0 && space > 0) {
1194 			    /* If there is more to send, set PRUS_MORETOCOME */
1195 		    	    pru_flags = PRUS_MORETOCOME;
1196 			    async = 1;
1197 		    } else {
1198 		    	    pru_flags = 0;
1199 		    }
1200 
1201 		    if (flags & MSG_SYNC)
1202 			    async = 0;
1203 
1204 		    /*
1205 		     * XXX all the SS_CANTSENDMORE checks previously
1206 		     * done could be out of date.  We could have recieved
1207 		     * a reset packet in an interrupt or maybe we slept
1208 		     * while doing page faults in uiomove() etc. We could
1209 		     * probably recheck again inside the splnet() protection
1210 		     * here, but there are probably other places that this
1211 		     * also happens.  We must rethink this.
1212 		     */
1213 		    for (m = top; m; m = m->m_next)
1214 			    ssb_preallocstream(&so->so_snd, m);
1215 		    if (!async) {
1216 			    error = so_pru_send(so, pru_flags, top,
1217 			        NULL, NULL, td);
1218 		    } else {
1219 			    so_pru_send_async(so, pru_flags, top,
1220 			        NULL, NULL, td);
1221 			    error = 0;
1222 		    }
1223 
1224 		    top = NULL;
1225 		    mp = &top;
1226 		    if (error)
1227 			    goto release;
1228 		} while (resid && space > 0);
1229 	} while (resid);
1230 
1231 release:
1232 	ssb_unlock(&so->so_snd);
1233 out:
1234 	if (top)
1235 		m_freem(top);
1236 	if (control)
1237 		m_freem(control);
1238 	return (error);
1239 }
1240 #endif
1241 
1242 /*
1243  * Implement receive operations on a socket.
1244  *
1245  * We depend on the way that records are added to the signalsockbuf
1246  * by sbappend*.  In particular, each record (mbufs linked through m_next)
1247  * must begin with an address if the protocol so specifies,
1248  * followed by an optional mbuf or mbufs containing ancillary data,
1249  * and then zero or more mbufs of data.
1250  *
1251  * Although the signalsockbuf is locked, new data may still be appended.
1252  * A token inside the ssb_lock deals with MP issues and still allows
1253  * the network to access the socket if we block in a uio.
1254  *
1255  * The caller may receive the data as a single mbuf chain by supplying
1256  * an mbuf **mp0 for use in returning the chain.  The uio is then used
1257  * only for the count in uio_resid.
1258  */
1259 int
1260 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1261 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1262 {
1263 	struct mbuf *m, *n;
1264 	struct mbuf *free_chain = NULL;
1265 	int flags, len, error, offset;
1266 	struct protosw *pr = so->so_proto;
1267 	int moff, type = 0;
1268 	size_t resid, orig_resid;
1269 	boolean_t free_rights = FALSE;
1270 
1271 	if (uio)
1272 		resid = uio->uio_resid;
1273 	else
1274 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1275 	orig_resid = resid;
1276 
1277 	if (psa)
1278 		*psa = NULL;
1279 	if (controlp)
1280 		*controlp = NULL;
1281 	if (flagsp)
1282 		flags = *flagsp &~ MSG_EOR;
1283 	else
1284 		flags = 0;
1285 	if (flags & MSG_OOB) {
1286 		m = m_get(M_WAITOK, MT_DATA);
1287 		if (m == NULL)
1288 			return (ENOBUFS);
1289 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1290 		if (error)
1291 			goto bad;
1292 		if (sio) {
1293 			do {
1294 				sbappend(sio, m);
1295 				KKASSERT(resid >= (size_t)m->m_len);
1296 				resid -= (size_t)m->m_len;
1297 			} while (resid > 0 && m);
1298 		} else {
1299 			do {
1300 				uio->uio_resid = resid;
1301 				error = uiomove(mtod(m, caddr_t),
1302 						(int)szmin(resid, m->m_len),
1303 						uio);
1304 				resid = uio->uio_resid;
1305 				m = m_free(m);
1306 			} while (uio->uio_resid && error == 0 && m);
1307 		}
1308 bad:
1309 		if (m)
1310 			m_freem(m);
1311 		return (error);
1312 	}
1313 	if ((so->so_state & SS_ISCONFIRMING) && resid)
1314 		so_pru_rcvd(so, 0);
1315 
1316 	/*
1317 	 * The token interlocks against the protocol thread while
1318 	 * ssb_lock is a blocking lock against other userland entities.
1319 	 */
1320 	lwkt_gettoken(&so->so_rcv.ssb_token);
1321 restart:
1322 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1323 	if (error)
1324 		goto done;
1325 
1326 	m = so->so_rcv.ssb_mb;
1327 	/*
1328 	 * If we have less data than requested, block awaiting more
1329 	 * (subject to any timeout) if:
1330 	 *   1. the current count is less than the low water mark, or
1331 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1332 	 *	receive operation at once if we block (resid <= hiwat).
1333 	 *   3. MSG_DONTWAIT is not set
1334 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1335 	 * we have to do the receive in sections, and thus risk returning
1336 	 * a short count if a timeout or signal occurs after we start.
1337 	 */
1338 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1339 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1340 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1341 	    ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)) &&
1342 	    m->m_nextpkt == 0 && (pr->pr_flags & PR_ATOMIC) == 0)) {
1343 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1344 		if (so->so_error || so->so_rerror) {
1345 			if (m)
1346 				goto dontblock;
1347 			if (so->so_error)
1348 				error = so->so_error;
1349 			else
1350 				error = so->so_rerror;
1351 			if ((flags & MSG_PEEK) == 0) {
1352 				if (so->so_error)
1353 					so->so_error = 0;
1354 				else
1355 					so->so_rerror = 0;
1356 			}
1357 			goto release;
1358 		}
1359 		if (so->so_state & SS_CANTRCVMORE) {
1360 			if (m)
1361 				goto dontblock;
1362 			else
1363 				goto release;
1364 		}
1365 		for (; m; m = m->m_next) {
1366 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1367 				m = so->so_rcv.ssb_mb;
1368 				goto dontblock;
1369 			}
1370 		}
1371 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1372 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1373 			error = ENOTCONN;
1374 			goto release;
1375 		}
1376 		if (resid == 0)
1377 			goto release;
1378 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1379 			error = EWOULDBLOCK;
1380 			goto release;
1381 		}
1382 		ssb_unlock(&so->so_rcv);
1383 		error = ssb_wait(&so->so_rcv);
1384 		if (error)
1385 			goto done;
1386 		goto restart;
1387 	}
1388 dontblock:
1389 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1390 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1391 
1392 	/*
1393 	 * note: m should be == sb_mb here.  Cache the next record while
1394 	 * cleaning up.  Note that calling m_free*() will break out critical
1395 	 * section.
1396 	 */
1397 	KKASSERT(m == so->so_rcv.ssb_mb);
1398 
1399 	/*
1400 	 * Skip any address mbufs prepending the record.
1401 	 */
1402 	if (pr->pr_flags & PR_ADDR) {
1403 		KASSERT(m->m_type == MT_SONAME, ("receive 1a"));
1404 		orig_resid = 0;
1405 		if (psa)
1406 			*psa = dup_sockaddr(mtod(m, struct sockaddr *));
1407 		if (flags & MSG_PEEK)
1408 			m = m->m_next;
1409 		else
1410 			m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1411 	}
1412 
1413 	/*
1414 	 * Skip any control mbufs prepending the record.
1415 	 */
1416 	while (m && m->m_type == MT_CONTROL && error == 0) {
1417 		if (flags & MSG_PEEK) {
1418 			if (controlp)
1419 				*controlp = m_copy(m, 0, m->m_len);
1420 			m = m->m_next;	/* XXX race */
1421 		} else {
1422 			const struct cmsghdr *cm = mtod(m, struct cmsghdr *);
1423 
1424 			if (controlp) {
1425 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1426 				if (pr->pr_domain->dom_externalize &&
1427 				    cm->cmsg_level == SOL_SOCKET &&
1428 				    cm->cmsg_type == SCM_RIGHTS) {
1429 					error = pr->pr_domain->dom_externalize
1430 					    (m, flags);
1431 				}
1432 				*controlp = m;
1433 				m = n;
1434 			} else {
1435 				if (cm->cmsg_level == SOL_SOCKET &&
1436 				    cm->cmsg_type == SCM_RIGHTS)
1437 					free_rights = TRUE;
1438 				m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1439 			}
1440 		}
1441 		if (controlp && *controlp) {
1442 			orig_resid = 0;
1443 			controlp = &(*controlp)->m_next;
1444 		}
1445 	}
1446 
1447 	/*
1448 	 * flag OOB data.
1449 	 */
1450 	if (m) {
1451 		type = m->m_type;
1452 		if (type == MT_OOBDATA)
1453 			flags |= MSG_OOB;
1454 	}
1455 
1456 	/*
1457 	 * Copy to the UIO or mbuf return chain (*mp).
1458 	 */
1459 	moff = 0;
1460 	offset = 0;
1461 	while (m && resid > 0 && error == 0) {
1462 		if (m->m_type == MT_OOBDATA) {
1463 			if (type != MT_OOBDATA)
1464 				break;
1465 		} else if (type == MT_OOBDATA)
1466 			break;
1467 		else
1468 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1469 			("receive 3"));
1470 		soclrstate(so, SS_RCVATMARK);
1471 		len = (resid > INT_MAX) ? INT_MAX : resid;
1472 		if (so->so_oobmark && len > so->so_oobmark - offset)
1473 			len = so->so_oobmark - offset;
1474 		if (len > m->m_len - moff)
1475 			len = m->m_len - moff;
1476 
1477 		/*
1478 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1479 		 * The SIO is dealt with when we eat the mbuf, but deal
1480 		 * with the resid here either way.
1481 		 */
1482 		if (uio) {
1483 			uio->uio_resid = resid;
1484 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1485 			resid = uio->uio_resid;
1486 			if (error)
1487 				goto release;
1488 		} else {
1489 			resid -= (size_t)len;
1490 		}
1491 
1492 		/*
1493 		 * Eat the entire mbuf or just a piece of it
1494 		 */
1495 		if (len == m->m_len - moff) {
1496 			if (m->m_flags & M_EOR)
1497 				flags |= MSG_EOR;
1498 			if (flags & MSG_PEEK) {
1499 				m = m->m_next;
1500 				moff = 0;
1501 			} else {
1502 				if (sio) {
1503 					n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1504 					sbappend(sio, m);
1505 					m = n;
1506 				} else {
1507 					m = sbunlinkmbuf(&so->so_rcv.sb, m, &free_chain);
1508 				}
1509 			}
1510 		} else {
1511 			if (flags & MSG_PEEK) {
1512 				moff += len;
1513 			} else {
1514 				if (sio) {
1515 					n = m_copym(m, 0, len, M_WAITOK);
1516 					if (n)
1517 						sbappend(sio, n);
1518 				}
1519 				m->m_data += len;
1520 				m->m_len -= len;
1521 				so->so_rcv.ssb_cc -= len;
1522 			}
1523 		}
1524 		if (so->so_oobmark) {
1525 			if ((flags & MSG_PEEK) == 0) {
1526 				so->so_oobmark -= len;
1527 				if (so->so_oobmark == 0) {
1528 					sosetstate(so, SS_RCVATMARK);
1529 					break;
1530 				}
1531 			} else {
1532 				offset += len;
1533 				if (offset == so->so_oobmark)
1534 					break;
1535 			}
1536 		}
1537 		if (flags & MSG_EOR)
1538 			break;
1539 		/*
1540 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1541 		 * we must not quit until resid == 0 or an error
1542 		 * termination.  If a signal/timeout occurs, return
1543 		 * with a short count but without error.
1544 		 * Keep signalsockbuf locked against other readers.
1545 		 */
1546 		while ((flags & MSG_WAITALL) && m == NULL &&
1547 		       resid > 0 && !sosendallatonce(so) &&
1548 		       so->so_rcv.ssb_mb == NULL) {
1549 			if (so->so_error || so->so_rerror ||
1550 			    so->so_state & SS_CANTRCVMORE)
1551 				break;
1552 			/*
1553 			 * The window might have closed to zero, make
1554 			 * sure we send an ack now that we've drained
1555 			 * the buffer or we might end up blocking until
1556 			 * the idle takes over (5 seconds).
1557 			 */
1558 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb)
1559 				so_pru_rcvd(so, flags);
1560 			error = ssb_wait(&so->so_rcv);
1561 			if (error) {
1562 				ssb_unlock(&so->so_rcv);
1563 				error = 0;
1564 				goto done;
1565 			}
1566 			m = so->so_rcv.ssb_mb;
1567 		}
1568 	}
1569 
1570 	/*
1571 	 * If an atomic read was requested but unread data still remains
1572 	 * in the record, set MSG_TRUNC.
1573 	 */
1574 	if (m && pr->pr_flags & PR_ATOMIC)
1575 		flags |= MSG_TRUNC;
1576 
1577 	/*
1578 	 * Cleanup.  If an atomic read was requested drop any unread data.
1579 	 */
1580 	if ((flags & MSG_PEEK) == 0) {
1581 		if (m && (pr->pr_flags & PR_ATOMIC))
1582 			sbdroprecord(&so->so_rcv.sb);
1583 		if ((pr->pr_flags & PR_WANTRCVD) && so->so_pcb)
1584 			so_pru_rcvd(so, flags);
1585 	}
1586 
1587 	if (orig_resid == resid && orig_resid &&
1588 	    (flags & MSG_EOR) == 0 && (so->so_state & SS_CANTRCVMORE) == 0) {
1589 		ssb_unlock(&so->so_rcv);
1590 		goto restart;
1591 	}
1592 
1593 	if (flagsp)
1594 		*flagsp |= flags;
1595 release:
1596 	ssb_unlock(&so->so_rcv);
1597 done:
1598 	lwkt_reltoken(&so->so_rcv.ssb_token);
1599 	if (free_chain) {
1600 		if (free_rights && (pr->pr_flags & PR_RIGHTS) &&
1601 		    pr->pr_domain->dom_dispose)
1602 			pr->pr_domain->dom_dispose(free_chain);
1603 		m_freem(free_chain);
1604 	}
1605 	return (error);
1606 }
1607 
1608 int
1609 sorecvtcp(struct socket *so, struct sockaddr **psa, struct uio *uio,
1610 	  struct sockbuf *sio, struct mbuf **controlp, int *flagsp)
1611 {
1612 	struct mbuf *m, *n;
1613 	struct mbuf *free_chain = NULL;
1614 	int flags, len, error, offset;
1615 	struct protosw *pr = so->so_proto;
1616 	int moff;
1617 	int didoob;
1618 	size_t resid, orig_resid, restmp;
1619 
1620 	if (uio)
1621 		resid = uio->uio_resid;
1622 	else
1623 		resid = (size_t)(sio->sb_climit - sio->sb_cc);
1624 	orig_resid = resid;
1625 
1626 	if (psa)
1627 		*psa = NULL;
1628 	if (controlp)
1629 		*controlp = NULL;
1630 	if (flagsp)
1631 		flags = *flagsp &~ MSG_EOR;
1632 	else
1633 		flags = 0;
1634 	if (flags & MSG_OOB) {
1635 		m = m_get(M_WAITOK, MT_DATA);
1636 		if (m == NULL)
1637 			return (ENOBUFS);
1638 		error = so_pru_rcvoob(so, m, flags & MSG_PEEK);
1639 		if (error)
1640 			goto bad;
1641 		if (sio) {
1642 			do {
1643 				sbappend(sio, m);
1644 				KKASSERT(resid >= (size_t)m->m_len);
1645 				resid -= (size_t)m->m_len;
1646 			} while (resid > 0 && m);
1647 		} else {
1648 			do {
1649 				uio->uio_resid = resid;
1650 				error = uiomove(mtod(m, caddr_t),
1651 						(int)szmin(resid, m->m_len),
1652 						uio);
1653 				resid = uio->uio_resid;
1654 				m = m_free(m);
1655 			} while (uio->uio_resid && error == 0 && m);
1656 		}
1657 bad:
1658 		if (m)
1659 			m_freem(m);
1660 		return (error);
1661 	}
1662 
1663 	/*
1664 	 * The token interlocks against the protocol thread while
1665 	 * ssb_lock is a blocking lock against other userland entities.
1666 	 *
1667 	 * Lock a limited number of mbufs (not all, so sbcompress() still
1668 	 * works well).  The token is used as an interlock for sbwait() so
1669 	 * release it afterwords.
1670 	 */
1671 restart:
1672 	error = ssb_lock(&so->so_rcv, SBLOCKWAIT(flags));
1673 	if (error)
1674 		goto done;
1675 
1676 	lwkt_gettoken(&so->so_rcv.ssb_token);
1677 	m = so->so_rcv.ssb_mb;
1678 
1679 	/*
1680 	 * If we have less data than requested, block awaiting more
1681 	 * (subject to any timeout) if:
1682 	 *   1. the current count is less than the low water mark, or
1683 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1684 	 *	receive operation at once if we block (resid <= hiwat).
1685 	 *   3. MSG_DONTWAIT is not set
1686 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1687 	 * we have to do the receive in sections, and thus risk returning
1688 	 * a short count if a timeout or signal occurs after we start.
1689 	 */
1690 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1691 	    (size_t)so->so_rcv.ssb_cc < resid) &&
1692 	    (so->so_rcv.ssb_cc < so->so_rcv.ssb_lowat ||
1693 	   ((flags & MSG_WAITALL) && resid <= (size_t)so->so_rcv.ssb_hiwat)))) {
1694 		KASSERT(m != NULL || !so->so_rcv.ssb_cc, ("receive 1"));
1695 		if (so->so_error) {
1696 			if (m)
1697 				goto dontblock;
1698 			lwkt_reltoken(&so->so_rcv.ssb_token);
1699 			error = so->so_error;
1700 			if ((flags & MSG_PEEK) == 0)
1701 				so->so_error = 0;
1702 			goto release;
1703 		}
1704 		if (so->so_state & SS_CANTRCVMORE) {
1705 			if (m)
1706 				goto dontblock;
1707 			lwkt_reltoken(&so->so_rcv.ssb_token);
1708 			goto release;
1709 		}
1710 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1711 		    (pr->pr_flags & PR_CONNREQUIRED)) {
1712 			lwkt_reltoken(&so->so_rcv.ssb_token);
1713 			error = ENOTCONN;
1714 			goto release;
1715 		}
1716 		if (resid == 0) {
1717 			lwkt_reltoken(&so->so_rcv.ssb_token);
1718 			goto release;
1719 		}
1720 		if (flags & (MSG_FNONBLOCKING|MSG_DONTWAIT)) {
1721 			lwkt_reltoken(&so->so_rcv.ssb_token);
1722 			error = EWOULDBLOCK;
1723 			goto release;
1724 		}
1725 		ssb_unlock(&so->so_rcv);
1726 		error = ssb_wait(&so->so_rcv);
1727 		lwkt_reltoken(&so->so_rcv.ssb_token);
1728 		if (error)
1729 			goto done;
1730 		goto restart;
1731 	}
1732 
1733 	/*
1734 	 * Token still held
1735 	 */
1736 dontblock:
1737 	n = m;
1738 	restmp = 0;
1739 	while (n && restmp < resid) {
1740 		n->m_flags |= M_SOLOCKED;
1741 		restmp += n->m_len;
1742 		if (n->m_next == NULL)
1743 			n = n->m_nextpkt;
1744 		else
1745 			n = n->m_next;
1746 	}
1747 
1748 	/*
1749 	 * Release token for loop
1750 	 */
1751 	lwkt_reltoken(&so->so_rcv.ssb_token);
1752 	if (uio && uio->uio_td && uio->uio_td->td_proc)
1753 		uio->uio_td->td_lwp->lwp_ru.ru_msgrcv++;
1754 
1755 	/*
1756 	 * note: m should be == sb_mb here.  Cache the next record while
1757 	 * cleaning up.  Note that calling m_free*() will break out critical
1758 	 * section.
1759 	 */
1760 	KKASSERT(m == so->so_rcv.ssb_mb);
1761 
1762 	/*
1763 	 * Copy to the UIO or mbuf return chain (*mp).
1764 	 *
1765 	 * NOTE: Token is not held for loop
1766 	 */
1767 	moff = 0;
1768 	offset = 0;
1769 	didoob = 0;
1770 
1771 	while (m && (m->m_flags & M_SOLOCKED) && resid > 0 && error == 0) {
1772 		KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1773 		    ("receive 3"));
1774 
1775 		soclrstate(so, SS_RCVATMARK);
1776 		len = (resid > INT_MAX) ? INT_MAX : resid;
1777 		if (so->so_oobmark && len > so->so_oobmark - offset)
1778 			len = so->so_oobmark - offset;
1779 		if (len > m->m_len - moff)
1780 			len = m->m_len - moff;
1781 
1782 		/*
1783 		 * Copy out to the UIO or pass the mbufs back to the SIO.
1784 		 * The SIO is dealt with when we eat the mbuf, but deal
1785 		 * with the resid here either way.
1786 		 */
1787 		if (uio) {
1788 			uio->uio_resid = resid;
1789 			error = uiomove(mtod(m, caddr_t) + moff, len, uio);
1790 			resid = uio->uio_resid;
1791 			if (error)
1792 				goto release;
1793 		} else {
1794 			resid -= (size_t)len;
1795 		}
1796 
1797 		/*
1798 		 * Eat the entire mbuf or just a piece of it
1799 		 */
1800 		offset += len;
1801 		if (len == m->m_len - moff) {
1802 			m = m->m_next;
1803 			moff = 0;
1804 		} else {
1805 			moff += len;
1806 		}
1807 
1808 		/*
1809 		 * Check oobmark
1810 		 */
1811 		if (so->so_oobmark && offset == so->so_oobmark) {
1812 			didoob = 1;
1813 			break;
1814 		}
1815 	}
1816 
1817 	/*
1818 	 * Synchronize sockbuf with data we read.
1819 	 *
1820 	 * NOTE: (m) is junk on entry (it could be left over from the
1821 	 *	 previous loop).
1822 	 */
1823 	if ((flags & MSG_PEEK) == 0) {
1824 		lwkt_gettoken(&so->so_rcv.ssb_token);
1825 		m = so->so_rcv.ssb_mb;
1826 		while (m && offset >= m->m_len) {
1827 			if (so->so_oobmark) {
1828 				so->so_oobmark -= m->m_len;
1829 				if (so->so_oobmark == 0) {
1830 					sosetstate(so, SS_RCVATMARK);
1831 					didoob = 1;
1832 				}
1833 			}
1834 			offset -= m->m_len;
1835 			if (sio) {
1836 				n = sbunlinkmbuf(&so->so_rcv.sb, m, NULL);
1837 				sbappend(sio, m);
1838 				m = n;
1839 			} else {
1840 				m = sbunlinkmbuf(&so->so_rcv.sb,
1841 						 m, &free_chain);
1842 			}
1843 		}
1844 		if (offset) {
1845 			KKASSERT(m);
1846 			if (sio) {
1847 				n = m_copym(m, 0, offset, M_WAITOK);
1848 				if (n)
1849 					sbappend(sio, n);
1850 			}
1851 			m->m_data += offset;
1852 			m->m_len -= offset;
1853 			so->so_rcv.ssb_cc -= offset;
1854 			if (so->so_oobmark) {
1855 				so->so_oobmark -= offset;
1856 				if (so->so_oobmark == 0) {
1857 					sosetstate(so, SS_RCVATMARK);
1858 					didoob = 1;
1859 				}
1860 			}
1861 			offset = 0;
1862 		}
1863 		lwkt_reltoken(&so->so_rcv.ssb_token);
1864 	}
1865 
1866 	/*
1867 	 * If the MSG_WAITALL flag is set (for non-atomic socket),
1868 	 * we must not quit until resid == 0 or an error termination.
1869 	 *
1870 	 * If a signal/timeout occurs, return with a short count but without
1871 	 * error.
1872 	 *
1873 	 * Keep signalsockbuf locked against other readers.
1874 	 *
1875 	 * XXX if MSG_PEEK we currently do quit.
1876 	 */
1877 	if ((flags & MSG_WAITALL) && !(flags & MSG_PEEK) &&
1878 	    didoob == 0 && resid > 0 &&
1879 	    !sosendallatonce(so)) {
1880 		lwkt_gettoken(&so->so_rcv.ssb_token);
1881 		error = 0;
1882 		while ((m = so->so_rcv.ssb_mb) == NULL) {
1883 			if (so->so_error || (so->so_state & SS_CANTRCVMORE)) {
1884 				error = so->so_error;
1885 				break;
1886 			}
1887 			/*
1888 			 * The window might have closed to zero, make
1889 			 * sure we send an ack now that we've drained
1890 			 * the buffer or we might end up blocking until
1891 			 * the idle takes over (5 seconds).
1892 			 */
1893 			if (so->so_pcb)
1894 				so_pru_rcvd_async(so);
1895 			if (so->so_rcv.ssb_mb == NULL)
1896 				error = ssb_wait(&so->so_rcv);
1897 			if (error) {
1898 				lwkt_reltoken(&so->so_rcv.ssb_token);
1899 				ssb_unlock(&so->so_rcv);
1900 				error = 0;
1901 				goto done;
1902 			}
1903 		}
1904 		if (m && error == 0)
1905 			goto dontblock;
1906 		lwkt_reltoken(&so->so_rcv.ssb_token);
1907 	}
1908 
1909 	/*
1910 	 * Token not held here.
1911 	 *
1912 	 * Cleanup.  If an atomic read was requested drop any unread data XXX
1913 	 */
1914 	if ((flags & MSG_PEEK) == 0) {
1915 		if (so->so_pcb)
1916 			so_pru_rcvd_async(so);
1917 	}
1918 
1919 	if (orig_resid == resid && orig_resid &&
1920 	    (so->so_state & SS_CANTRCVMORE) == 0) {
1921 		ssb_unlock(&so->so_rcv);
1922 		goto restart;
1923 	}
1924 
1925 	if (flagsp)
1926 		*flagsp |= flags;
1927 release:
1928 	ssb_unlock(&so->so_rcv);
1929 done:
1930 	if (free_chain)
1931 		m_freem(free_chain);
1932 	return (error);
1933 }
1934 
1935 /*
1936  * Shut a socket down.  Note that we do not get a frontend lock as we
1937  * want to be able to shut the socket down even if another thread is
1938  * blocked in a read(), thus waking it up.
1939  */
1940 int
1941 soshutdown(struct socket *so, int how)
1942 {
1943 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1944 		return (EINVAL);
1945 
1946 	if (how != SHUT_WR) {
1947 		/*ssb_lock(&so->so_rcv, M_WAITOK);*/
1948 		sorflush(so);
1949 		/*ssb_unlock(&so->so_rcv);*/
1950 	}
1951 	if (how != SHUT_RD)
1952 		return (so_pru_shutdown(so));
1953 	return (0);
1954 }
1955 
1956 void
1957 sorflush(struct socket *so)
1958 {
1959 	struct signalsockbuf *ssb = &so->so_rcv;
1960 	struct protosw *pr = so->so_proto;
1961 	struct signalsockbuf asb;
1962 
1963 	atomic_set_int(&ssb->ssb_flags, SSB_NOINTR);
1964 
1965 	lwkt_gettoken(&ssb->ssb_token);
1966 	socantrcvmore(so);
1967 	asb = *ssb;
1968 
1969 	/*
1970 	 * Can't just blow up the ssb structure here
1971 	 */
1972 	bzero(&ssb->sb, sizeof(ssb->sb));
1973 	ssb->ssb_timeo = 0;
1974 	ssb->ssb_lowat = 0;
1975 	ssb->ssb_hiwat = 0;
1976 	ssb->ssb_mbmax = 0;
1977 	atomic_clear_int(&ssb->ssb_flags, SSB_CLEAR_MASK);
1978 
1979 	if ((pr->pr_flags & PR_RIGHTS) && pr->pr_domain->dom_dispose)
1980 		(*pr->pr_domain->dom_dispose)(asb.ssb_mb);
1981 	ssb_release(&asb, so);
1982 
1983 	lwkt_reltoken(&ssb->ssb_token);
1984 }
1985 
1986 #ifdef INET
1987 static int
1988 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt)
1989 {
1990 	struct accept_filter_arg	*afap = NULL;
1991 	struct accept_filter	*afp;
1992 	struct so_accf	*af = so->so_accf;
1993 	int	error = 0;
1994 
1995 	/* do not set/remove accept filters on non listen sockets */
1996 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1997 		error = EINVAL;
1998 		goto out;
1999 	}
2000 
2001 	/* removing the filter */
2002 	if (sopt == NULL) {
2003 		if (af != NULL) {
2004 			if (af->so_accept_filter != NULL &&
2005 				af->so_accept_filter->accf_destroy != NULL) {
2006 				af->so_accept_filter->accf_destroy(so);
2007 			}
2008 			if (af->so_accept_filter_str != NULL) {
2009 				kfree(af->so_accept_filter_str, M_ACCF);
2010 			}
2011 			kfree(af, M_ACCF);
2012 			so->so_accf = NULL;
2013 		}
2014 		so->so_options &= ~SO_ACCEPTFILTER;
2015 		return (0);
2016 	}
2017 	/* adding a filter */
2018 	/* must remove previous filter first */
2019 	if (af != NULL) {
2020 		error = EINVAL;
2021 		goto out;
2022 	}
2023 	/* don't put large objects on the kernel stack */
2024 	afap = kmalloc(sizeof(*afap), M_TEMP, M_WAITOK);
2025 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
2026 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
2027 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
2028 	if (error)
2029 		goto out;
2030 	afp = accept_filt_get(afap->af_name);
2031 	if (afp == NULL) {
2032 		error = ENOENT;
2033 		goto out;
2034 	}
2035 	af = kmalloc(sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
2036 	if (afp->accf_create != NULL) {
2037 		if (afap->af_name[0] != '\0') {
2038 			int len = strlen(afap->af_name) + 1;
2039 
2040 			af->so_accept_filter_str = kmalloc(len, M_ACCF,
2041 							   M_WAITOK);
2042 			strcpy(af->so_accept_filter_str, afap->af_name);
2043 		}
2044 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
2045 		if (af->so_accept_filter_arg == NULL) {
2046 			kfree(af->so_accept_filter_str, M_ACCF);
2047 			kfree(af, M_ACCF);
2048 			so->so_accf = NULL;
2049 			error = EINVAL;
2050 			goto out;
2051 		}
2052 	}
2053 	af->so_accept_filter = afp;
2054 	so->so_accf = af;
2055 	so->so_options |= SO_ACCEPTFILTER;
2056 out:
2057 	if (afap != NULL)
2058 		kfree(afap, M_TEMP);
2059 	return (error);
2060 }
2061 #endif /* INET */
2062 
2063 /*
2064  * Perhaps this routine, and sooptcopyout(), below, ought to come in
2065  * an additional variant to handle the case where the option value needs
2066  * to be some kind of integer, but not a specific size.
2067  * In addition to their use here, these functions are also called by the
2068  * protocol-level pr_ctloutput() routines.
2069  */
2070 int
2071 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2072 {
2073 	return soopt_to_kbuf(sopt, buf, len, minlen);
2074 }
2075 
2076 int
2077 soopt_to_kbuf(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2078 {
2079 	size_t	valsize;
2080 
2081 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2082 	KKASSERT(kva_p(buf));
2083 
2084 	/*
2085 	 * If the user gives us more than we wanted, we ignore it,
2086 	 * but if we don't get the minimum length the caller
2087 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
2088 	 * is set to however much we actually retrieved.
2089 	 */
2090 	if ((valsize = sopt->sopt_valsize) < minlen)
2091 		return EINVAL;
2092 	if (valsize > len)
2093 		sopt->sopt_valsize = valsize = len;
2094 
2095 	bcopy(sopt->sopt_val, buf, valsize);
2096 	return 0;
2097 }
2098 
2099 
2100 int
2101 sosetopt(struct socket *so, struct sockopt *sopt)
2102 {
2103 	int	error, optval;
2104 	struct	linger l;
2105 	struct	timeval tv;
2106 	u_long  val;
2107 	struct signalsockbuf *sotmp;
2108 
2109 	error = 0;
2110 	sopt->sopt_dir = SOPT_SET;
2111 	if (sopt->sopt_level != SOL_SOCKET) {
2112 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2113 			return (so_pr_ctloutput(so, sopt));
2114 		}
2115 		error = ENOPROTOOPT;
2116 	} else {
2117 		switch (sopt->sopt_name) {
2118 #ifdef INET
2119 		case SO_ACCEPTFILTER:
2120 			error = do_setopt_accept_filter(so, sopt);
2121 			if (error)
2122 				goto bad;
2123 			break;
2124 #endif /* INET */
2125 		case SO_LINGER:
2126 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2127 			if (error)
2128 				goto bad;
2129 
2130 			so->so_linger = l.l_linger;
2131 			if (l.l_onoff)
2132 				so->so_options |= SO_LINGER;
2133 			else
2134 				so->so_options &= ~SO_LINGER;
2135 			break;
2136 
2137 		case SO_DEBUG:
2138 		case SO_KEEPALIVE:
2139 		case SO_DONTROUTE:
2140 		case SO_USELOOPBACK:
2141 		case SO_BROADCAST:
2142 		case SO_REUSEADDR:
2143 		case SO_REUSEPORT:
2144 		case SO_OOBINLINE:
2145 		case SO_TIMESTAMP:
2146 		case SO_NOSIGPIPE:
2147 		case SO_RERROR:
2148 			error = sooptcopyin(sopt, &optval, sizeof optval,
2149 					    sizeof optval);
2150 			if (error)
2151 				goto bad;
2152 			if (optval)
2153 				so->so_options |= sopt->sopt_name;
2154 			else
2155 				so->so_options &= ~sopt->sopt_name;
2156 			break;
2157 
2158 		case SO_SNDBUF:
2159 		case SO_RCVBUF:
2160 		case SO_SNDLOWAT:
2161 		case SO_RCVLOWAT:
2162 			error = sooptcopyin(sopt, &optval, sizeof optval,
2163 					    sizeof optval);
2164 			if (error)
2165 				goto bad;
2166 
2167 			/*
2168 			 * Values < 1 make no sense for any of these
2169 			 * options, so disallow them.
2170 			 */
2171 			if (optval < 1) {
2172 				error = EINVAL;
2173 				goto bad;
2174 			}
2175 
2176 			switch (sopt->sopt_name) {
2177 			case SO_SNDBUF:
2178 			case SO_RCVBUF:
2179 				if (ssb_reserve(sopt->sopt_name == SO_SNDBUF ?
2180 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2181 				    so,
2182 				    &curproc->p_rlimit[RLIMIT_SBSIZE]) == 0) {
2183 					error = ENOBUFS;
2184 					goto bad;
2185 				}
2186 				sotmp = (sopt->sopt_name == SO_SNDBUF) ?
2187 						&so->so_snd : &so->so_rcv;
2188 				atomic_clear_int(&sotmp->ssb_flags,
2189 						 SSB_AUTOSIZE);
2190 				break;
2191 
2192 			/*
2193 			 * Make sure the low-water is never greater than
2194 			 * the high-water.
2195 			 */
2196 			case SO_SNDLOWAT:
2197 				so->so_snd.ssb_lowat =
2198 				    (optval > so->so_snd.ssb_hiwat) ?
2199 				    so->so_snd.ssb_hiwat : optval;
2200 				atomic_clear_int(&so->so_snd.ssb_flags,
2201 						 SSB_AUTOLOWAT);
2202 				break;
2203 			case SO_RCVLOWAT:
2204 				so->so_rcv.ssb_lowat =
2205 				    (optval > so->so_rcv.ssb_hiwat) ?
2206 				    so->so_rcv.ssb_hiwat : optval;
2207 				atomic_clear_int(&so->so_rcv.ssb_flags,
2208 						 SSB_AUTOLOWAT);
2209 				break;
2210 			}
2211 			break;
2212 
2213 		case SO_SNDTIMEO:
2214 		case SO_RCVTIMEO:
2215 			error = sooptcopyin(sopt, &tv, sizeof tv,
2216 					    sizeof tv);
2217 			if (error)
2218 				goto bad;
2219 
2220 			/* assert(hz > 0); */
2221 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2222 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2223 				error = EDOM;
2224 				goto bad;
2225 			}
2226 			/* assert(tick > 0); */
2227 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2228 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / ustick;
2229 			if (val > INT_MAX) {
2230 				error = EDOM;
2231 				goto bad;
2232 			}
2233 			if (val == 0 && tv.tv_usec != 0)
2234 				val = 1;
2235 
2236 			switch (sopt->sopt_name) {
2237 			case SO_SNDTIMEO:
2238 				so->so_snd.ssb_timeo = val;
2239 				break;
2240 			case SO_RCVTIMEO:
2241 				so->so_rcv.ssb_timeo = val;
2242 				break;
2243 			}
2244 			break;
2245 		default:
2246 			error = ENOPROTOOPT;
2247 			break;
2248 		}
2249 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput) {
2250 			(void) so_pr_ctloutput(so, sopt);
2251 		}
2252 	}
2253 bad:
2254 	return (error);
2255 }
2256 
2257 /* Helper routine for getsockopt */
2258 int
2259 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2260 {
2261 	soopt_from_kbuf(sopt, buf, len);
2262 	return 0;
2263 }
2264 
2265 void
2266 soopt_from_kbuf(struct sockopt *sopt, const void *buf, size_t len)
2267 {
2268 	size_t	valsize;
2269 
2270 	if (len == 0) {
2271 		sopt->sopt_valsize = 0;
2272 		return;
2273 	}
2274 
2275 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2276 	KKASSERT(kva_p(buf));
2277 
2278 	/*
2279 	 * Documented get behavior is that we always return a value,
2280 	 * possibly truncated to fit in the user's buffer.
2281 	 * Traditional behavior is that we always tell the user
2282 	 * precisely how much we copied, rather than something useful
2283 	 * like the total amount we had available for her.
2284 	 * Note that this interface is not idempotent; the entire answer must
2285 	 * generated ahead of time.
2286 	 */
2287 	valsize = szmin(len, sopt->sopt_valsize);
2288 	sopt->sopt_valsize = valsize;
2289 	if (sopt->sopt_val != 0) {
2290 		bcopy(buf, sopt->sopt_val, valsize);
2291 	}
2292 }
2293 
2294 int
2295 sogetopt(struct socket *so, struct sockopt *sopt)
2296 {
2297 	int	error, optval;
2298 	long	optval_l;
2299 	struct	linger l;
2300 	struct	timeval tv;
2301 #ifdef INET
2302 	struct accept_filter_arg *afap;
2303 #endif
2304 
2305 	error = 0;
2306 	sopt->sopt_dir = SOPT_GET;
2307 	if (sopt->sopt_level != SOL_SOCKET) {
2308 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2309 			return (so_pr_ctloutput(so, sopt));
2310 		} else
2311 			return (ENOPROTOOPT);
2312 	} else {
2313 		switch (sopt->sopt_name) {
2314 #ifdef INET
2315 		case SO_ACCEPTFILTER:
2316 			if ((so->so_options & SO_ACCEPTCONN) == 0)
2317 				return (EINVAL);
2318 			afap = kmalloc(sizeof(*afap), M_TEMP,
2319 				       M_WAITOK | M_ZERO);
2320 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
2321 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
2322 				if (so->so_accf->so_accept_filter_str != NULL)
2323 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
2324 			}
2325 			error = sooptcopyout(sopt, afap, sizeof(*afap));
2326 			kfree(afap, M_TEMP);
2327 			break;
2328 #endif /* INET */
2329 
2330 		case SO_LINGER:
2331 			l.l_onoff = so->so_options & SO_LINGER;
2332 			l.l_linger = so->so_linger;
2333 			error = sooptcopyout(sopt, &l, sizeof l);
2334 			break;
2335 
2336 		case SO_USELOOPBACK:
2337 		case SO_DONTROUTE:
2338 		case SO_DEBUG:
2339 		case SO_KEEPALIVE:
2340 		case SO_REUSEADDR:
2341 		case SO_REUSEPORT:
2342 		case SO_BROADCAST:
2343 		case SO_OOBINLINE:
2344 		case SO_TIMESTAMP:
2345 		case SO_NOSIGPIPE:
2346 		case SO_RERROR:
2347 			optval = so->so_options & sopt->sopt_name;
2348 integer:
2349 			error = sooptcopyout(sopt, &optval, sizeof optval);
2350 			break;
2351 
2352 		case SO_TYPE:
2353 			optval = so->so_type;
2354 			goto integer;
2355 
2356 		case SO_ERROR:
2357 			if (so->so_error) {
2358 				optval = so->so_error;
2359 				so->so_error = 0;
2360 			} else {
2361 				optval = so->so_rerror;
2362 				so->so_rerror = 0;
2363 			}
2364 			goto integer;
2365 
2366 		case SO_SNDBUF:
2367 			optval = so->so_snd.ssb_hiwat;
2368 			goto integer;
2369 
2370 		case SO_RCVBUF:
2371 			optval = so->so_rcv.ssb_hiwat;
2372 			goto integer;
2373 
2374 		case SO_SNDLOWAT:
2375 			optval = so->so_snd.ssb_lowat;
2376 			goto integer;
2377 
2378 		case SO_RCVLOWAT:
2379 			optval = so->so_rcv.ssb_lowat;
2380 			goto integer;
2381 
2382 		case SO_SNDTIMEO:
2383 		case SO_RCVTIMEO:
2384 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2385 				  so->so_snd.ssb_timeo : so->so_rcv.ssb_timeo);
2386 
2387 			tv.tv_sec = optval / hz;
2388 			tv.tv_usec = (optval % hz) * ustick;
2389 			error = sooptcopyout(sopt, &tv, sizeof tv);
2390 			break;
2391 
2392 		case SO_SNDSPACE:
2393 			optval_l = ssb_space(&so->so_snd);
2394 			error = sooptcopyout(sopt, &optval_l, sizeof(optval_l));
2395 			break;
2396 
2397 		case SO_CPUHINT:
2398 			optval = -1; /* no hint */
2399 			goto integer;
2400 
2401 		default:
2402 			error = ENOPROTOOPT;
2403 			break;
2404 		}
2405 		if (error == 0 && so->so_proto && so->so_proto->pr_ctloutput)
2406 			so_pr_ctloutput(so, sopt);
2407 		return (error);
2408 	}
2409 }
2410 
2411 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2412 int
2413 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2414 {
2415 	struct mbuf *m, *m_prev;
2416 	int sopt_size = sopt->sopt_valsize, msize;
2417 
2418 	m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA,
2419 		   0, &msize);
2420 	if (m == NULL)
2421 		return (ENOBUFS);
2422 	m->m_len = min(msize, sopt_size);
2423 	sopt_size -= m->m_len;
2424 	*mp = m;
2425 	m_prev = m;
2426 
2427 	while (sopt_size > 0) {
2428 		m = m_getl(sopt_size, sopt->sopt_td ? M_WAITOK : M_NOWAIT,
2429 			   MT_DATA, 0, &msize);
2430 		if (m == NULL) {
2431 			m_freem(*mp);
2432 			return (ENOBUFS);
2433 		}
2434 		m->m_len = min(msize, sopt_size);
2435 		sopt_size -= m->m_len;
2436 		m_prev->m_next = m;
2437 		m_prev = m;
2438 	}
2439 	return (0);
2440 }
2441 
2442 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2443 int
2444 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2445 {
2446 	soopt_to_mbuf(sopt, m);
2447 	return 0;
2448 }
2449 
2450 void
2451 soopt_to_mbuf(struct sockopt *sopt, struct mbuf *m)
2452 {
2453 	size_t valsize;
2454 	void *val;
2455 
2456 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2457 	KKASSERT(kva_p(m));
2458 	if (sopt->sopt_val == NULL)
2459 		return;
2460 	val = sopt->sopt_val;
2461 	valsize = sopt->sopt_valsize;
2462 	while (m != NULL && valsize >= m->m_len) {
2463 		bcopy(val, mtod(m, char *), m->m_len);
2464 		valsize -= m->m_len;
2465 		val = (caddr_t)val + m->m_len;
2466 		m = m->m_next;
2467 	}
2468 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2469 		panic("ip6_sooptmcopyin");
2470 }
2471 
2472 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2473 int
2474 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2475 {
2476 	return soopt_from_mbuf(sopt, m);
2477 }
2478 
2479 int
2480 soopt_from_mbuf(struct sockopt *sopt, struct mbuf *m)
2481 {
2482 	struct mbuf *m0 = m;
2483 	size_t valsize = 0;
2484 	size_t maxsize;
2485 	void *val;
2486 
2487 	KKASSERT(!sopt->sopt_val || kva_p(sopt->sopt_val));
2488 	KKASSERT(kva_p(m));
2489 	if (sopt->sopt_val == NULL)
2490 		return 0;
2491 	val = sopt->sopt_val;
2492 	maxsize = sopt->sopt_valsize;
2493 	while (m != NULL && maxsize >= m->m_len) {
2494 		bcopy(mtod(m, char *), val, m->m_len);
2495 	       maxsize -= m->m_len;
2496 	       val = (caddr_t)val + m->m_len;
2497 	       valsize += m->m_len;
2498 	       m = m->m_next;
2499 	}
2500 	if (m != NULL) {
2501 		/* enough soopt buffer should be given from user-land */
2502 		m_freem(m0);
2503 		return (EINVAL);
2504 	}
2505 	sopt->sopt_valsize = valsize;
2506 	return 0;
2507 }
2508 
2509 void
2510 sohasoutofband(struct socket *so)
2511 {
2512 	if (so->so_sigio != NULL)
2513 		pgsigio(so->so_sigio, SIGURG, 0);
2514 	/*
2515 	 * NOTE:
2516 	 * There is no need to use NOTE_OOB as KNOTE hint here:
2517 	 * soread filter depends on so_oobmark and SS_RCVATMARK
2518 	 * so_state.  NOTE_OOB would cause unnecessary penalty
2519 	 * in KNOTE, if there was knote processing contention.
2520 	 */
2521 	KNOTE(&so->so_rcv.ssb_kq.ki_note, 0);
2522 }
2523 
2524 int
2525 sokqfilter(struct file *fp, struct knote *kn)
2526 {
2527 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2528 	struct signalsockbuf *ssb;
2529 
2530 	switch (kn->kn_filter) {
2531 	case EVFILT_READ:
2532 		if (so->so_options & SO_ACCEPTCONN)
2533 			kn->kn_fop = &solisten_filtops;
2534 		else
2535 			kn->kn_fop = &soread_filtops;
2536 		ssb = &so->so_rcv;
2537 		break;
2538 	case EVFILT_WRITE:
2539 		kn->kn_fop = &sowrite_filtops;
2540 		ssb = &so->so_snd;
2541 		break;
2542 	case EVFILT_EXCEPT:
2543 		kn->kn_fop = &soexcept_filtops;
2544 		ssb = &so->so_rcv;
2545 		break;
2546 	default:
2547 		return (EOPNOTSUPP);
2548 	}
2549 
2550 	knote_insert(&ssb->ssb_kq.ki_note, kn);
2551 	atomic_set_int(&ssb->ssb_flags, SSB_KNOTE);
2552 	return (0);
2553 }
2554 
2555 static void
2556 filt_sordetach(struct knote *kn)
2557 {
2558 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2559 
2560 	knote_remove(&so->so_rcv.ssb_kq.ki_note, kn);
2561 	if (SLIST_EMPTY(&so->so_rcv.ssb_kq.ki_note))
2562 		atomic_clear_int(&so->so_rcv.ssb_flags, SSB_KNOTE);
2563 }
2564 
2565 /*ARGSUSED*/
2566 static int
2567 filt_soread(struct knote *kn, long hint __unused)
2568 {
2569 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2570 
2571 	if (kn->kn_sfflags & NOTE_OOB) {
2572 		if ((so->so_oobmark || (so->so_state & SS_RCVATMARK))) {
2573 			kn->kn_fflags |= NOTE_OOB;
2574 			return (1);
2575 		}
2576 		return (0);
2577 	}
2578 	kn->kn_data = so->so_rcv.ssb_cc;
2579 
2580 	if (so->so_state & SS_CANTRCVMORE) {
2581 		/*
2582 		 * Only set NODATA if all data has been exhausted.
2583 		 */
2584 		if (kn->kn_data == 0)
2585 			kn->kn_flags |= EV_NODATA;
2586 		kn->kn_flags |= EV_EOF;
2587 		kn->kn_fflags = so->so_error;
2588 		return (1);
2589 	}
2590 	if (so->so_error || so->so_rerror)
2591 		return (1);
2592 	if (kn->kn_sfflags & NOTE_LOWAT)
2593 		return (kn->kn_data >= kn->kn_sdata);
2594 	return ((kn->kn_data >= so->so_rcv.ssb_lowat) ||
2595 		!TAILQ_EMPTY(&so->so_comp));
2596 }
2597 
2598 static void
2599 filt_sowdetach(struct knote *kn)
2600 {
2601 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2602 
2603 	knote_remove(&so->so_snd.ssb_kq.ki_note, kn);
2604 	if (SLIST_EMPTY(&so->so_snd.ssb_kq.ki_note))
2605 		atomic_clear_int(&so->so_snd.ssb_flags, SSB_KNOTE);
2606 }
2607 
2608 /*ARGSUSED*/
2609 static int
2610 filt_sowrite(struct knote *kn, long hint __unused)
2611 {
2612 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2613 
2614 	if (so->so_snd.ssb_flags & SSB_PREALLOC)
2615 		kn->kn_data = ssb_space_prealloc(&so->so_snd);
2616 	else
2617 		kn->kn_data = ssb_space(&so->so_snd);
2618 
2619 	if (so->so_state & SS_CANTSENDMORE) {
2620 		kn->kn_flags |= (EV_EOF | EV_NODATA);
2621 		kn->kn_fflags = so->so_error;
2622 		return (1);
2623 	}
2624 	if (so->so_error)	/* temporary udp error */
2625 		return (1);
2626 	if (((so->so_state & SS_ISCONNECTED) == 0) &&
2627 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2628 		return (0);
2629 	if (kn->kn_sfflags & NOTE_LOWAT)
2630 		return (kn->kn_data >= kn->kn_sdata);
2631 	return (kn->kn_data >= so->so_snd.ssb_lowat);
2632 }
2633 
2634 /*ARGSUSED*/
2635 static int
2636 filt_solisten(struct knote *kn, long hint __unused)
2637 {
2638 	struct socket *so = (struct socket *)kn->kn_fp->f_data;
2639 	int qlen = so->so_qlen;
2640 
2641 	if (soavailconn > 0 && qlen > soavailconn)
2642 		qlen = soavailconn;
2643 	kn->kn_data = qlen;
2644 
2645 	return (!TAILQ_EMPTY(&so->so_comp));
2646 }
2647