xref: /freebsd/sys/kern/uipc_socket.c (revision 6fd05b64b5b65dd4ba9b86482a0634a5f0b96c29)
1 /*
2  * Copyright (c) 2004 The FreeBSD Foundation
3  * Copyright (c) 2004 Robert Watson
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.  All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 4. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
32  */
33 
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
36 
37 #include "opt_inet.h"
38 #include "opt_mac.h"
39 #include "opt_zero.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/fcntl.h>
44 #include <sys/limits.h>
45 #include <sys/lock.h>
46 #include <sys/mac.h>
47 #include <sys/malloc.h>
48 #include <sys/mbuf.h>
49 #include <sys/mutex.h>
50 #include <sys/domain.h>
51 #include <sys/file.h>			/* for struct knote */
52 #include <sys/kernel.h>
53 #include <sys/event.h>
54 #include <sys/poll.h>
55 #include <sys/proc.h>
56 #include <sys/protosw.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/resourcevar.h>
60 #include <sys/signalvar.h>
61 #include <sys/sysctl.h>
62 #include <sys/uio.h>
63 #include <sys/jail.h>
64 
65 #include <vm/uma.h>
66 
67 
68 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
69 		    int flags);
70 
71 #ifdef INET
72 static int	 do_setopt_accept_filter(struct socket *so, struct sockopt *sopt);
73 #endif
74 
75 static void	filt_sordetach(struct knote *kn);
76 static int	filt_soread(struct knote *kn, long hint);
77 static void	filt_sowdetach(struct knote *kn);
78 static int	filt_sowrite(struct knote *kn, long hint);
79 static int	filt_solisten(struct knote *kn, long hint);
80 
81 static struct filterops solisten_filtops =
82 	{ 1, NULL, filt_sordetach, filt_solisten };
83 static struct filterops soread_filtops =
84 	{ 1, NULL, filt_sordetach, filt_soread };
85 static struct filterops sowrite_filtops =
86 	{ 1, NULL, filt_sowdetach, filt_sowrite };
87 
88 uma_zone_t socket_zone;
89 so_gen_t	so_gencnt;	/* generation count for sockets */
90 
91 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
92 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
93 
94 SYSCTL_DECL(_kern_ipc);
95 
96 static int somaxconn = SOMAXCONN;
97 SYSCTL_INT(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLFLAG_RW,
98     &somaxconn, 0, "Maximum pending socket connection queue size");
99 static int numopensockets;
100 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
101     &numopensockets, 0, "Number of open sockets");
102 #ifdef ZERO_COPY_SOCKETS
103 /* These aren't static because they're used in other files. */
104 int so_zero_copy_send = 1;
105 int so_zero_copy_receive = 1;
106 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
107     "Zero copy controls");
108 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
109     &so_zero_copy_receive, 0, "Enable zero copy receive");
110 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
111     &so_zero_copy_send, 0, "Enable zero copy send");
112 #endif /* ZERO_COPY_SOCKETS */
113 
114 /*
115  * accept_mtx locks down per-socket fields relating to accept queues.  See
116  * socketvar.h for an annotation of the protected fields of struct socket.
117  */
118 struct mtx accept_mtx;
119 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
120 
121 /*
122  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
123  * so_gencnt field.
124  *
125  * XXXRW: These variables might be better manipulated using atomic operations
126  * for improved efficiency.
127  */
128 static struct mtx so_global_mtx;
129 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
130 
131 /*
132  * Socket operation routines.
133  * These routines are called by the routines in
134  * sys_socket.c or from a system process, and
135  * implement the semantics of socket operations by
136  * switching out to the protocol specific routines.
137  */
138 
139 /*
140  * Get a socket structure from our zone, and initialize it.
141  * Note that it would probably be better to allocate socket
142  * and PCB at the same time, but I'm not convinced that all
143  * the protocols can be easily modified to do this.
144  *
145  * soalloc() returns a socket with a ref count of 0.
146  */
147 struct socket *
148 soalloc(int mflags)
149 {
150 	struct socket *so;
151 #ifdef MAC
152 	int error;
153 #endif
154 
155 	so = uma_zalloc(socket_zone, mflags | M_ZERO);
156 	if (so != NULL) {
157 #ifdef MAC
158 		error = mac_init_socket(so, mflags);
159 		if (error != 0) {
160 			uma_zfree(socket_zone, so);
161 			so = NULL;
162 			return so;
163 		}
164 #endif
165 		SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
166 		SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
167 		/* sx_init(&so->so_sxlock, "socket sxlock"); */
168 		TAILQ_INIT(&so->so_aiojobq);
169 		mtx_lock(&so_global_mtx);
170 		so->so_gencnt = ++so_gencnt;
171 		++numopensockets;
172 		mtx_unlock(&so_global_mtx);
173 	}
174 	return so;
175 }
176 
177 /*
178  * socreate returns a socket with a ref count of 1.  The socket should be
179  * closed with soclose().
180  */
181 int
182 socreate(dom, aso, type, proto, cred, td)
183 	int dom;
184 	struct socket **aso;
185 	int type;
186 	int proto;
187 	struct ucred *cred;
188 	struct thread *td;
189 {
190 	struct protosw *prp;
191 	struct socket *so;
192 	int error;
193 
194 	if (proto)
195 		prp = pffindproto(dom, proto, type);
196 	else
197 		prp = pffindtype(dom, type);
198 
199 	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL)
200 		return (EPROTONOSUPPORT);
201 
202 	if (jailed(cred) && jail_socket_unixiproute_only &&
203 	    prp->pr_domain->dom_family != PF_LOCAL &&
204 	    prp->pr_domain->dom_family != PF_INET &&
205 	    prp->pr_domain->dom_family != PF_ROUTE) {
206 		return (EPROTONOSUPPORT);
207 	}
208 
209 	if (prp->pr_type != type)
210 		return (EPROTOTYPE);
211 	so = soalloc(M_WAITOK);
212 	if (so == NULL)
213 		return (ENOBUFS);
214 
215 	TAILQ_INIT(&so->so_incomp);
216 	TAILQ_INIT(&so->so_comp);
217 	so->so_type = type;
218 	so->so_cred = crhold(cred);
219 	so->so_proto = prp;
220 #ifdef MAC
221 	mac_create_socket(cred, so);
222 #endif
223 	SOCK_LOCK(so);
224 	soref(so);
225 	SOCK_UNLOCK(so);
226 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
227 	if (error) {
228 		SOCK_LOCK(so);
229 		so->so_state |= SS_NOFDREF;
230 		sorele(so);
231 		return (error);
232 	}
233 	*aso = so;
234 	return (0);
235 }
236 
237 int
238 sobind(so, nam, td)
239 	struct socket *so;
240 	struct sockaddr *nam;
241 	struct thread *td;
242 {
243 
244 	return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
245 }
246 
247 void
248 sodealloc(struct socket *so)
249 {
250 
251 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
252 	mtx_lock(&so_global_mtx);
253 	so->so_gencnt = ++so_gencnt;
254 	mtx_unlock(&so_global_mtx);
255 	if (so->so_rcv.sb_hiwat)
256 		(void)chgsbsize(so->so_cred->cr_uidinfo,
257 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
258 	if (so->so_snd.sb_hiwat)
259 		(void)chgsbsize(so->so_cred->cr_uidinfo,
260 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
261 #ifdef INET
262 	/* remove acccept filter if one is present. */
263 	if (so->so_accf != NULL)
264 		do_setopt_accept_filter(so, NULL);
265 #endif
266 #ifdef MAC
267 	mac_destroy_socket(so);
268 #endif
269 	crfree(so->so_cred);
270 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
271 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
272 	/* sx_destroy(&so->so_sxlock); */
273 	uma_zfree(socket_zone, so);
274 	/*
275 	 * XXXRW: Seems like a shame to grab the mutex again down here, but
276 	 * we don't want to decrement the socket count until after we free
277 	 * the socket, and we can't increment the gencnt on the socket after
278 	 * we free, it so...
279 	 */
280 	mtx_lock(&so_global_mtx);
281 	--numopensockets;
282 	mtx_unlock(&so_global_mtx);
283 }
284 
285 int
286 solisten(so, backlog, td)
287 	struct socket *so;
288 	int backlog;
289 	struct thread *td;
290 {
291 	int error;
292 
293 	/*
294 	 * XXXRW: Ordering issue here -- perhaps we need to set
295 	 * SO_ACCEPTCONN before the call to pru_listen()?
296 	 * XXXRW: General atomic test-and-set concerns here also.
297 	 */
298 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
299 			    SS_ISDISCONNECTING))
300 		return (EINVAL);
301 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, td);
302 	if (error)
303 		return (error);
304 	ACCEPT_LOCK();
305 	if (TAILQ_EMPTY(&so->so_comp)) {
306 		SOCK_LOCK(so);
307 		so->so_options |= SO_ACCEPTCONN;
308 		SOCK_UNLOCK(so);
309 	}
310 	if (backlog < 0 || backlog > somaxconn)
311 		backlog = somaxconn;
312 	so->so_qlimit = backlog;
313 	ACCEPT_UNLOCK();
314 	return (0);
315 }
316 
317 void
318 sofree(so)
319 	struct socket *so;
320 {
321 	struct socket *head;
322 
323 	KASSERT(so->so_count == 0, ("socket %p so_count not 0", so));
324 	SOCK_LOCK_ASSERT(so);
325 
326 	if (so->so_pcb != NULL || (so->so_state & SS_NOFDREF) == 0) {
327 		SOCK_UNLOCK(so);
328 		return;
329 	}
330 
331 	SOCK_UNLOCK(so);
332 	ACCEPT_LOCK();
333 	head = so->so_head;
334 	if (head != NULL) {
335 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
336 		    (so->so_qstate & SQ_INCOMP) != 0,
337 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
338 		    "SQ_INCOMP"));
339 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
340 		    (so->so_qstate & SQ_INCOMP) == 0,
341 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
342 		/*
343 		 * accept(2) is responsible draining the completed
344 		 * connection queue and freeing those sockets, so
345 		 * we just return here if this socket is currently
346 		 * on the completed connection queue.  Otherwise,
347 		 * accept(2) may hang after select(2) has indicating
348 		 * that a listening socket was ready.  If it's an
349 		 * incomplete connection, we remove it from the queue
350 		 * and free it; otherwise, it won't be released until
351 		 * the listening socket is closed.
352 		 */
353 		if ((so->so_qstate & SQ_COMP) != 0) {
354 			ACCEPT_UNLOCK();
355 			return;
356 		}
357 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
358 		head->so_incqlen--;
359 		so->so_qstate &= ~SQ_INCOMP;
360 		so->so_head = NULL;
361 	}
362 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
363 	    (so->so_qstate & SQ_INCOMP) == 0,
364 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
365 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
366 	ACCEPT_UNLOCK();
367 	SOCKBUF_LOCK(&so->so_snd);
368 	so->so_snd.sb_flags |= SB_NOINTR;
369 	(void)sblock(&so->so_snd, M_WAITOK);
370 	/*
371 	 * socantsendmore_locked() drops the socket buffer mutex so that it
372 	 * can safely perform wakeups.  Re-acquire the mutex before
373 	 * continuing.
374 	 */
375 	socantsendmore_locked(so);
376 	SOCKBUF_LOCK(&so->so_snd);
377 	sbunlock(&so->so_snd);
378 	sbrelease_locked(&so->so_snd, so);
379 	SOCKBUF_UNLOCK(&so->so_snd);
380 	sorflush(so);
381 	sodealloc(so);
382 }
383 
384 /*
385  * Close a socket on last file table reference removal.
386  * Initiate disconnect if connected.
387  * Free socket when disconnect complete.
388  *
389  * This function will sorele() the socket.  Note that soclose() may be
390  * called prior to the ref count reaching zero.  The actual socket
391  * structure will not be freed until the ref count reaches zero.
392  */
393 int
394 soclose(so)
395 	struct socket *so;
396 {
397 	int error = 0;
398 
399 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
400 
401 	funsetown(&so->so_sigio);
402 	if (so->so_options & SO_ACCEPTCONN) {
403 		struct socket *sp;
404 		ACCEPT_LOCK();
405 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
406 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
407 			so->so_incqlen--;
408 			sp->so_qstate &= ~SQ_INCOMP;
409 			sp->so_head = NULL;
410 			ACCEPT_UNLOCK();
411 			(void) soabort(sp);
412 			ACCEPT_LOCK();
413 		}
414 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
415 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
416 			so->so_qlen--;
417 			sp->so_qstate &= ~SQ_COMP;
418 			sp->so_head = NULL;
419 			ACCEPT_UNLOCK();
420 			(void) soabort(sp);
421 			ACCEPT_LOCK();
422 		}
423 		ACCEPT_UNLOCK();
424 	}
425 	if (so->so_pcb == NULL)
426 		goto discard;
427 	if (so->so_state & SS_ISCONNECTED) {
428 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
429 			error = sodisconnect(so);
430 			if (error)
431 				goto drop;
432 		}
433 		if (so->so_options & SO_LINGER) {
434 			if ((so->so_state & SS_ISDISCONNECTING) &&
435 			    (so->so_state & SS_NBIO))
436 				goto drop;
437 			while (so->so_state & SS_ISCONNECTED) {
438 				error = tsleep(&so->so_timeo,
439 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
440 				if (error)
441 					break;
442 			}
443 		}
444 	}
445 drop:
446 	if (so->so_pcb != NULL) {
447 		int error2 = (*so->so_proto->pr_usrreqs->pru_detach)(so);
448 		if (error == 0)
449 			error = error2;
450 	}
451 discard:
452 	SOCK_LOCK(so);
453 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
454 	so->so_state |= SS_NOFDREF;
455 	sorele(so);
456 	return (error);
457 }
458 
459 /*
460  * soabort() must not be called with any socket locks held, as it calls
461  * into the protocol, which will call back into the socket code causing
462  * it to acquire additional socket locks that may cause recursion or lock
463  * order reversals.
464  */
465 int
466 soabort(so)
467 	struct socket *so;
468 {
469 	int error;
470 
471 	error = (*so->so_proto->pr_usrreqs->pru_abort)(so);
472 	if (error) {
473 		SOCK_LOCK(so);
474 		sotryfree(so);	/* note: does not decrement the ref count */
475 		return error;
476 	}
477 	return (0);
478 }
479 
480 int
481 soaccept(so, nam)
482 	struct socket *so;
483 	struct sockaddr **nam;
484 {
485 	int error;
486 
487 	SOCK_LOCK(so);
488 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
489 	so->so_state &= ~SS_NOFDREF;
490 	SOCK_UNLOCK(so);
491 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
492 	return (error);
493 }
494 
495 int
496 soconnect(so, nam, td)
497 	struct socket *so;
498 	struct sockaddr *nam;
499 	struct thread *td;
500 {
501 	int error;
502 
503 	if (so->so_options & SO_ACCEPTCONN)
504 		return (EOPNOTSUPP);
505 	/*
506 	 * If protocol is connection-based, can only connect once.
507 	 * Otherwise, if connected, try to disconnect first.
508 	 * This allows user to disconnect by connecting to, e.g.,
509 	 * a null address.
510 	 */
511 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
512 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
513 	    (error = sodisconnect(so))))
514 		error = EISCONN;
515 	else
516 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
517 	return (error);
518 }
519 
520 int
521 soconnect2(so1, so2)
522 	struct socket *so1;
523 	struct socket *so2;
524 {
525 
526 	return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
527 }
528 
529 int
530 sodisconnect(so)
531 	struct socket *so;
532 {
533 	int error;
534 
535 	if ((so->so_state & SS_ISCONNECTED) == 0)
536 		return (ENOTCONN);
537 	if (so->so_state & SS_ISDISCONNECTING)
538 		return (EALREADY);
539 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
540 	return (error);
541 }
542 
543 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? M_NOWAIT : M_WAITOK)
544 /*
545  * Send on a socket.
546  * If send must go all at once and message is larger than
547  * send buffering, then hard error.
548  * Lock against other senders.
549  * If must go all at once and not enough room now, then
550  * inform user that this would block and do nothing.
551  * Otherwise, if nonblocking, send as much as possible.
552  * The data to be sent is described by "uio" if nonzero,
553  * otherwise by the mbuf chain "top" (which must be null
554  * if uio is not).  Data provided in mbuf chain must be small
555  * enough to send all at once.
556  *
557  * Returns nonzero on error, timeout or signal; callers
558  * must check for short counts if EINTR/ERESTART are returned.
559  * Data and control buffers are freed on return.
560  */
561 
562 #ifdef ZERO_COPY_SOCKETS
563 struct so_zerocopy_stats{
564 	int size_ok;
565 	int align_ok;
566 	int found_ifp;
567 };
568 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
569 #include <netinet/in.h>
570 #include <net/route.h>
571 #include <netinet/in_pcb.h>
572 #include <vm/vm.h>
573 #include <vm/vm_page.h>
574 #include <vm/vm_object.h>
575 #endif /*ZERO_COPY_SOCKETS*/
576 
577 int
578 sosend(so, addr, uio, top, control, flags, td)
579 	struct socket *so;
580 	struct sockaddr *addr;
581 	struct uio *uio;
582 	struct mbuf *top;
583 	struct mbuf *control;
584 	int flags;
585 	struct thread *td;
586 {
587 	struct mbuf **mp;
588 	struct mbuf *m;
589 	long space, len = 0, resid;
590 	int clen = 0, error, dontroute;
591 	int atomic = sosendallatonce(so) || top;
592 #ifdef ZERO_COPY_SOCKETS
593 	int cow_send;
594 #endif /* ZERO_COPY_SOCKETS */
595 
596 	if (uio != NULL)
597 		resid = uio->uio_resid;
598 	else
599 		resid = top->m_pkthdr.len;
600 	/*
601 	 * In theory resid should be unsigned.
602 	 * However, space must be signed, as it might be less than 0
603 	 * if we over-committed, and we must use a signed comparison
604 	 * of space and resid.  On the other hand, a negative resid
605 	 * causes us to loop sending 0-length segments to the protocol.
606 	 *
607 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
608 	 * type sockets since that's an error.
609 	 */
610 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
611 		error = EINVAL;
612 		goto out;
613 	}
614 
615 	dontroute =
616 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
617 	    (so->so_proto->pr_flags & PR_ATOMIC);
618 	if (td != NULL)
619 		td->td_proc->p_stats->p_ru.ru_msgsnd++;
620 	if (control != NULL)
621 		clen = control->m_len;
622 #define	snderr(errno)	{ error = (errno); goto release; }
623 
624 	SOCKBUF_LOCK(&so->so_snd);
625 restart:
626 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
627 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
628 	if (error)
629 		goto out_locked;
630 	do {
631 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
632 		if (so->so_snd.sb_state & SBS_CANTSENDMORE)
633 			snderr(EPIPE);
634 		if (so->so_error) {
635 			error = so->so_error;
636 			so->so_error = 0;
637 			goto release;
638 		}
639 		if ((so->so_state & SS_ISCONNECTED) == 0) {
640 			/*
641 			 * `sendto' and `sendmsg' is allowed on a connection-
642 			 * based socket if it supports implied connect.
643 			 * Return ENOTCONN if not connected and no address is
644 			 * supplied.
645 			 */
646 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
647 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
648 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
649 				    !(resid == 0 && clen != 0))
650 					snderr(ENOTCONN);
651 			} else if (addr == NULL)
652 			    snderr(so->so_proto->pr_flags & PR_CONNREQUIRED ?
653 				   ENOTCONN : EDESTADDRREQ);
654 		}
655 		space = sbspace(&so->so_snd);
656 		if (flags & MSG_OOB)
657 			space += 1024;
658 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
659 		    clen > so->so_snd.sb_hiwat)
660 			snderr(EMSGSIZE);
661 		if (space < resid + clen &&
662 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
663 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO))
664 				snderr(EWOULDBLOCK);
665 			sbunlock(&so->so_snd);
666 			error = sbwait(&so->so_snd);
667 			if (error)
668 				goto out_locked;
669 			goto restart;
670 		}
671 		SOCKBUF_UNLOCK(&so->so_snd);
672 		mp = &top;
673 		space -= clen;
674 		do {
675 		    if (uio == NULL) {
676 			/*
677 			 * Data is prepackaged in "top".
678 			 */
679 			resid = 0;
680 			if (flags & MSG_EOR)
681 				top->m_flags |= M_EOR;
682 		    } else do {
683 #ifdef ZERO_COPY_SOCKETS
684 			cow_send = 0;
685 #endif /* ZERO_COPY_SOCKETS */
686 			if (resid >= MINCLSIZE) {
687 #ifdef ZERO_COPY_SOCKETS
688 				if (top == NULL) {
689 					MGETHDR(m, M_TRYWAIT, MT_DATA);
690 					if (m == NULL) {
691 						error = ENOBUFS;
692 						SOCKBUF_LOCK(&so->so_snd);
693 						goto release;
694 					}
695 					m->m_pkthdr.len = 0;
696 					m->m_pkthdr.rcvif = (struct ifnet *)0;
697 				} else {
698 					MGET(m, M_TRYWAIT, MT_DATA);
699 					if (m == NULL) {
700 						error = ENOBUFS;
701 						SOCKBUF_LOCK(&so->so_snd);
702 						goto release;
703 					}
704 				}
705 				if (so_zero_copy_send &&
706 				    resid>=PAGE_SIZE &&
707 				    space>=PAGE_SIZE &&
708 				    uio->uio_iov->iov_len>=PAGE_SIZE) {
709 					so_zerocp_stats.size_ok++;
710 					if (!((vm_offset_t)
711 					  uio->uio_iov->iov_base & PAGE_MASK)){
712 						so_zerocp_stats.align_ok++;
713 						cow_send = socow_setup(m, uio);
714 					}
715 				}
716 				if (!cow_send) {
717 					MCLGET(m, M_TRYWAIT);
718 					if ((m->m_flags & M_EXT) == 0) {
719 						m_free(m);
720 						m = NULL;
721 					} else {
722 						len = min(min(MCLBYTES, resid), space);
723 					}
724 				} else
725 					len = PAGE_SIZE;
726 #else /* ZERO_COPY_SOCKETS */
727 				if (top == NULL) {
728 					m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
729 					m->m_pkthdr.len = 0;
730 					m->m_pkthdr.rcvif = (struct ifnet *)0;
731 				} else
732 					m = m_getcl(M_TRYWAIT, MT_DATA, 0);
733 				len = min(min(MCLBYTES, resid), space);
734 #endif /* ZERO_COPY_SOCKETS */
735 			} else {
736 				if (top == NULL) {
737 					m = m_gethdr(M_TRYWAIT, MT_DATA);
738 					m->m_pkthdr.len = 0;
739 					m->m_pkthdr.rcvif = (struct ifnet *)0;
740 
741 					len = min(min(MHLEN, resid), space);
742 					/*
743 					 * For datagram protocols, leave room
744 					 * for protocol headers in first mbuf.
745 					 */
746 					if (atomic && m && len < MHLEN)
747 						MH_ALIGN(m, len);
748 				} else {
749 					m = m_get(M_TRYWAIT, MT_DATA);
750 					len = min(min(MLEN, resid), space);
751 				}
752 			}
753 			if (m == NULL) {
754 				error = ENOBUFS;
755 				SOCKBUF_LOCK(&so->so_snd);
756 				goto release;
757 			}
758 
759 			space -= len;
760 #ifdef ZERO_COPY_SOCKETS
761 			if (cow_send)
762 				error = 0;
763 			else
764 #endif /* ZERO_COPY_SOCKETS */
765 			error = uiomove(mtod(m, void *), (int)len, uio);
766 			resid = uio->uio_resid;
767 			m->m_len = len;
768 			*mp = m;
769 			top->m_pkthdr.len += len;
770 			if (error) {
771 				SOCKBUF_LOCK(&so->so_snd);
772 				goto release;
773 			}
774 			mp = &m->m_next;
775 			if (resid <= 0) {
776 				if (flags & MSG_EOR)
777 					top->m_flags |= M_EOR;
778 				break;
779 			}
780 		    } while (space > 0 && atomic);
781 		    if (dontroute) {
782 			    SOCK_LOCK(so);
783 			    so->so_options |= SO_DONTROUTE;
784 			    SOCK_UNLOCK(so);
785 		    }
786 		    /*
787 		     * XXX all the SBS_CANTSENDMORE checks previously
788 		     * done could be out of date.  We could have recieved
789 		     * a reset packet in an interrupt or maybe we slept
790 		     * while doing page faults in uiomove() etc. We could
791 		     * probably recheck again inside the splnet() protection
792 		     * here, but there are probably other places that this
793 		     * also happens.  We must rethink this.
794 		     */
795 		    error = (*so->so_proto->pr_usrreqs->pru_send)(so,
796 			(flags & MSG_OOB) ? PRUS_OOB :
797 			/*
798 			 * If the user set MSG_EOF, the protocol
799 			 * understands this flag and nothing left to
800 			 * send then use PRU_SEND_EOF instead of PRU_SEND.
801 			 */
802 			((flags & MSG_EOF) &&
803 			 (so->so_proto->pr_flags & PR_IMPLOPCL) &&
804 			 (resid <= 0)) ?
805 				PRUS_EOF :
806 			/* If there is more to send set PRUS_MORETOCOME */
807 			(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
808 			top, addr, control, td);
809 		    if (dontroute) {
810 			    SOCK_LOCK(so);
811 			    so->so_options &= ~SO_DONTROUTE;
812 			    SOCK_UNLOCK(so);
813 		    }
814 		    clen = 0;
815 		    control = NULL;
816 		    top = NULL;
817 		    mp = &top;
818 		    if (error) {
819 			SOCKBUF_LOCK(&so->so_snd);
820 			goto release;
821 		    }
822 		} while (resid && space > 0);
823 		SOCKBUF_LOCK(&so->so_snd);
824 	} while (resid);
825 
826 release:
827 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
828 	sbunlock(&so->so_snd);
829 out_locked:
830 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
831 	SOCKBUF_UNLOCK(&so->so_snd);
832 out:
833 	if (top != NULL)
834 		m_freem(top);
835 	if (control != NULL)
836 		m_freem(control);
837 	return (error);
838 }
839 
840 /*
841  * The part of soreceive() that implements reading non-inline out-of-band
842  * data from a socket.  For more complete comments, see soreceive(), from
843  * which this code originated.
844  *
845  * XXXRW: Note that soreceive_rcvoob(), unlike the remainder of soreiceve(),
846  * is unable to return an mbuf chain to the caller.
847  */
848 static int
849 soreceive_rcvoob(so, uio, flags)
850 	struct socket *so;
851 	struct uio *uio;
852 	int flags;
853 {
854 	struct protosw *pr = so->so_proto;
855 	struct mbuf *m;
856 	int error;
857 
858 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
859 
860 	m = m_get(M_TRYWAIT, MT_DATA);
861 	if (m == NULL)
862 		return (ENOBUFS);
863 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
864 	if (error)
865 		goto bad;
866 	do {
867 #ifdef ZERO_COPY_SOCKETS
868 		if (so_zero_copy_receive) {
869 			vm_page_t pg;
870 			int disposable;
871 
872 			if ((m->m_flags & M_EXT)
873 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
874 				disposable = 1;
875 			else
876 				disposable = 0;
877 
878 			pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t)));
879 			if (uio->uio_offset == -1)
880 				uio->uio_offset =IDX_TO_OFF(pg->pindex);
881 
882 			error = uiomoveco(mtod(m, void *),
883 					  min(uio->uio_resid, m->m_len),
884 					  uio, pg->object,
885 					  disposable);
886 		} else
887 #endif /* ZERO_COPY_SOCKETS */
888 		error = uiomove(mtod(m, void *),
889 		    (int) min(uio->uio_resid, m->m_len), uio);
890 		m = m_free(m);
891 	} while (uio->uio_resid && error == 0 && m);
892 bad:
893 	if (m != NULL)
894 		m_freem(m);
895 	return (error);
896 }
897 
898 /*
899  * Following replacement or removal of the first mbuf on the first mbuf chain
900  * of a socket buffer, push necessary state changes back into the socket
901  * buffer so that other consumers see the values consistently.  'nextrecord'
902  * is the callers locally stored value of the original value of
903  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
904  * NOTE: 'nextrecord' may be NULL.
905  */
906 static __inline void
907 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
908 {
909 
910 	SOCKBUF_LOCK_ASSERT(sb);
911 	/*
912 	 * First, update for the new value of nextrecord.  If necessary, make
913 	 * it the first record.
914 	 */
915 	if (sb->sb_mb != NULL)
916 		sb->sb_mb->m_nextpkt = nextrecord;
917 	else
918 		sb->sb_mb = nextrecord;
919 
920         /*
921          * Now update any dependent socket buffer fields to reflect the new
922          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
923 	 * addition of a second clause that takes care of the case where
924 	 * sb_mb has been updated, but remains the last record.
925          */
926         if (sb->sb_mb == NULL) {
927                 sb->sb_mbtail = NULL;
928                 sb->sb_lastrecord = NULL;
929         } else if (sb->sb_mb->m_nextpkt == NULL)
930                 sb->sb_lastrecord = sb->sb_mb;
931 }
932 
933 
934 /*
935  * Implement receive operations on a socket.
936  * We depend on the way that records are added to the sockbuf
937  * by sbappend*.  In particular, each record (mbufs linked through m_next)
938  * must begin with an address if the protocol so specifies,
939  * followed by an optional mbuf or mbufs containing ancillary data,
940  * and then zero or more mbufs of data.
941  * In order to avoid blocking network interrupts for the entire time here,
942  * we splx() while doing the actual copy to user space.
943  * Although the sockbuf is locked, new data may still be appended,
944  * and thus we must maintain consistency of the sockbuf during that time.
945  *
946  * The caller may receive the data as a single mbuf chain by supplying
947  * an mbuf **mp0 for use in returning the chain.  The uio is then used
948  * only for the count in uio_resid.
949  */
950 int
951 soreceive(so, psa, uio, mp0, controlp, flagsp)
952 	struct socket *so;
953 	struct sockaddr **psa;
954 	struct uio *uio;
955 	struct mbuf **mp0;
956 	struct mbuf **controlp;
957 	int *flagsp;
958 {
959 	struct mbuf *m, **mp;
960 	int flags, len, error, offset;
961 	struct protosw *pr = so->so_proto;
962 	struct mbuf *nextrecord;
963 	int moff, type = 0;
964 	int orig_resid = uio->uio_resid;
965 
966 	mp = mp0;
967 	if (psa != NULL)
968 		*psa = NULL;
969 	if (controlp != NULL)
970 		*controlp = NULL;
971 	if (flagsp != NULL)
972 		flags = *flagsp &~ MSG_EOR;
973 	else
974 		flags = 0;
975 	if (flags & MSG_OOB)
976 		return (soreceive_rcvoob(so, uio, flags));
977 	if (mp != NULL)
978 		*mp = NULL;
979 	if (so->so_state & SS_ISCONFIRMING && uio->uio_resid)
980 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
981 
982 	SOCKBUF_LOCK(&so->so_rcv);
983 restart:
984 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
985 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
986 	if (error)
987 		goto out;
988 
989 	m = so->so_rcv.sb_mb;
990 	/*
991 	 * If we have less data than requested, block awaiting more
992 	 * (subject to any timeout) if:
993 	 *   1. the current count is less than the low water mark, or
994 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
995 	 *	receive operation at once if we block (resid <= hiwat).
996 	 *   3. MSG_DONTWAIT is not set
997 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
998 	 * we have to do the receive in sections, and thus risk returning
999 	 * a short count if a timeout or signal occurs after we start.
1000 	 */
1001 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1002 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1003 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1004 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1005 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1006 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1007 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1008 		    m, so->so_rcv.sb_cc));
1009 		if (so->so_error) {
1010 			if (m != NULL)
1011 				goto dontblock;
1012 			error = so->so_error;
1013 			if ((flags & MSG_PEEK) == 0)
1014 				so->so_error = 0;
1015 			goto release;
1016 		}
1017 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1018 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1019 			if (m)
1020 				goto dontblock;
1021 			else
1022 				goto release;
1023 		}
1024 		for (; m != NULL; m = m->m_next)
1025 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1026 				m = so->so_rcv.sb_mb;
1027 				goto dontblock;
1028 			}
1029 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1030 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1031 			error = ENOTCONN;
1032 			goto release;
1033 		}
1034 		if (uio->uio_resid == 0)
1035 			goto release;
1036 		if ((so->so_state & SS_NBIO) ||
1037 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1038 			error = EWOULDBLOCK;
1039 			goto release;
1040 		}
1041 		SBLASTRECORDCHK(&so->so_rcv);
1042 		SBLASTMBUFCHK(&so->so_rcv);
1043 		sbunlock(&so->so_rcv);
1044 		error = sbwait(&so->so_rcv);
1045 		if (error)
1046 			goto out;
1047 		goto restart;
1048 	}
1049 dontblock:
1050 	/*
1051 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1052 	 * pointer to the next record in the socket buffer.  We must keep the
1053 	 * various socket buffer pointers and local stack versions of the
1054 	 * pointers in sync, pushing out modifications before dropping the
1055 	 * socket buffer mutex, and re-reading them when picking it up.
1056 	 *
1057 	 * Otherwise, we will race with the network stack appending new data
1058 	 * or records onto the socket buffer by using inconsistent/stale
1059 	 * versions of the field, possibly resulting in socket buffer
1060 	 * corruption.
1061 	 *
1062 	 * By holding the high-level sblock(), we prevent simultaneous
1063 	 * readers from pulling off the front of the socket buffer.
1064 	 */
1065 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1066 	if (uio->uio_td)
1067 		uio->uio_td->td_proc->p_stats->p_ru.ru_msgrcv++;
1068 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1069 	SBLASTRECORDCHK(&so->so_rcv);
1070 	SBLASTMBUFCHK(&so->so_rcv);
1071 	nextrecord = m->m_nextpkt;
1072 	if (pr->pr_flags & PR_ADDR) {
1073 		KASSERT(m->m_type == MT_SONAME,
1074 		    ("m->m_type == %d", m->m_type));
1075 		orig_resid = 0;
1076 		if (psa != NULL)
1077 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1078 			    M_NOWAIT);
1079 		if (flags & MSG_PEEK) {
1080 			m = m->m_next;
1081 		} else {
1082 			sbfree(&so->so_rcv, m);
1083 			so->so_rcv.sb_mb = m_free(m);
1084 			m = so->so_rcv.sb_mb;
1085 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1086 		}
1087 	}
1088 
1089 	/*
1090 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1091 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1092 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1093 	 * perform externalization.
1094 	 */
1095 	if (m != NULL && m->m_type == MT_CONTROL) {
1096 		struct mbuf *cm = NULL;
1097 		struct mbuf **cme = &cm;
1098 
1099 		do {
1100 			if (flags & MSG_PEEK) {
1101 				if (controlp != NULL) {
1102 					*controlp = m_copy(m, 0, m->m_len);
1103 					controlp = &(*controlp)->m_next;
1104 				}
1105 				m = m->m_next;
1106 			} else {
1107 				sbfree(&so->so_rcv, m);
1108 				so->so_rcv.sb_mb = m->m_next;
1109 				m->m_next = NULL;
1110 				if (controlp) {
1111 					/*
1112 					 * Collect mbufs for processing below.
1113 					 */
1114 					*cme = m;
1115 					cme = &(*cme)->m_next;
1116 				} else
1117 					m_free(m);
1118 				m = so->so_rcv.sb_mb;
1119 			}
1120 		} while (m != NULL && m->m_type == MT_CONTROL);
1121 		if ((flags & MSG_PEEK) == 0)
1122 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1123 		if (cm != NULL) {
1124 			if (pr->pr_domain->dom_externalize != NULL) {
1125 				SOCKBUF_UNLOCK(&so->so_rcv);
1126 				error = (*pr->pr_domain->dom_externalize)
1127 				    (cm, controlp);
1128 				SOCKBUF_LOCK(&so->so_rcv);
1129 			} else
1130 				m_freem(cm);
1131 		}
1132 		nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1133 		orig_resid = 0;
1134 	}
1135 	if (m != NULL) {
1136 		if ((flags & MSG_PEEK) == 0) {
1137 			KASSERT(m->m_nextpkt == nextrecord,
1138 			    ("soreceive: post-control, nextrecord !sync"));
1139 			if (nextrecord == NULL) {
1140 				KASSERT(so->so_rcv.sb_mb == m,
1141 				    ("soreceive: post-control, sb_mb!=m"));
1142 				KASSERT(so->so_rcv.sb_lastrecord == m,
1143 				    ("soreceive: post-control, lastrecord!=m"));
1144 			}
1145 		}
1146 		type = m->m_type;
1147 		if (type == MT_OOBDATA)
1148 			flags |= MSG_OOB;
1149 	} else {
1150 		if ((flags & MSG_PEEK) == 0) {
1151 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1152 			    ("soreceive: sb_mb != nextrecord"));
1153 			if (so->so_rcv.sb_mb == NULL) {
1154 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1155 				    ("soreceive: sb_lastercord != NULL"));
1156 			}
1157 		}
1158 	}
1159 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1160 	SBLASTRECORDCHK(&so->so_rcv);
1161 	SBLASTMBUFCHK(&so->so_rcv);
1162 
1163 	/*
1164 	 * Now continue to read any data mbufs off of the head of the socket
1165 	 * buffer until the read request is satisfied.  Note that 'type' is
1166 	 * used to store the type of any mbuf reads that have happened so far
1167 	 * such that soreceive() can stop reading if the type changes, which
1168 	 * causes soreceive() to return only one of regular data and inline
1169 	 * out-of-band data in a single socket receive operation.
1170 	 */
1171 	moff = 0;
1172 	offset = 0;
1173 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1174 		/*
1175 		 * If the type of mbuf has changed since the last mbuf
1176 		 * examined ('type'), end the receive operation.
1177 	 	 */
1178 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1179 		if (m->m_type == MT_OOBDATA) {
1180 			if (type != MT_OOBDATA)
1181 				break;
1182 		} else if (type == MT_OOBDATA)
1183 			break;
1184 		else
1185 		    KASSERT(m->m_type == MT_DATA || m->m_type == MT_HEADER,
1186 			("m->m_type == %d", m->m_type));
1187 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1188 		len = uio->uio_resid;
1189 		if (so->so_oobmark && len > so->so_oobmark - offset)
1190 			len = so->so_oobmark - offset;
1191 		if (len > m->m_len - moff)
1192 			len = m->m_len - moff;
1193 		/*
1194 		 * If mp is set, just pass back the mbufs.
1195 		 * Otherwise copy them out via the uio, then free.
1196 		 * Sockbuf must be consistent here (points to current mbuf,
1197 		 * it points to next record) when we drop priority;
1198 		 * we must note any additions to the sockbuf when we
1199 		 * block interrupts again.
1200 		 */
1201 		if (mp == NULL) {
1202 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1203 			SBLASTRECORDCHK(&so->so_rcv);
1204 			SBLASTMBUFCHK(&so->so_rcv);
1205 			SOCKBUF_UNLOCK(&so->so_rcv);
1206 #ifdef ZERO_COPY_SOCKETS
1207 			if (so_zero_copy_receive) {
1208 				vm_page_t pg;
1209 				int disposable;
1210 
1211 				if ((m->m_flags & M_EXT)
1212 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1213 					disposable = 1;
1214 				else
1215 					disposable = 0;
1216 
1217 				pg = PHYS_TO_VM_PAGE(vtophys(mtod(m, caddr_t) +
1218 					moff));
1219 
1220 				if (uio->uio_offset == -1)
1221 					uio->uio_offset =IDX_TO_OFF(pg->pindex);
1222 
1223 				error = uiomoveco(mtod(m, char *) + moff,
1224 						  (int)len, uio,pg->object,
1225 						  disposable);
1226 			} else
1227 #endif /* ZERO_COPY_SOCKETS */
1228 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1229 			SOCKBUF_LOCK(&so->so_rcv);
1230 			if (error)
1231 				goto release;
1232 		} else
1233 			uio->uio_resid -= len;
1234 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1235 		if (len == m->m_len - moff) {
1236 			if (m->m_flags & M_EOR)
1237 				flags |= MSG_EOR;
1238 			if (flags & MSG_PEEK) {
1239 				m = m->m_next;
1240 				moff = 0;
1241 			} else {
1242 				nextrecord = m->m_nextpkt;
1243 				sbfree(&so->so_rcv, m);
1244 				if (mp != NULL) {
1245 					*mp = m;
1246 					mp = &m->m_next;
1247 					so->so_rcv.sb_mb = m = m->m_next;
1248 					*mp = NULL;
1249 				} else {
1250 					so->so_rcv.sb_mb = m_free(m);
1251 					m = so->so_rcv.sb_mb;
1252 				}
1253 				if (m != NULL) {
1254 					m->m_nextpkt = nextrecord;
1255 					if (nextrecord == NULL)
1256 						so->so_rcv.sb_lastrecord = m;
1257 				} else {
1258 					so->so_rcv.sb_mb = nextrecord;
1259 					SB_EMPTY_FIXUP(&so->so_rcv);
1260 				}
1261 				SBLASTRECORDCHK(&so->so_rcv);
1262 				SBLASTMBUFCHK(&so->so_rcv);
1263 			}
1264 		} else {
1265 			if (flags & MSG_PEEK)
1266 				moff += len;
1267 			else {
1268 				if (mp != NULL) {
1269 					SOCKBUF_UNLOCK(&so->so_rcv);
1270 					*mp = m_copym(m, 0, len, M_TRYWAIT);
1271 					SOCKBUF_LOCK(&so->so_rcv);
1272 				}
1273 				m->m_data += len;
1274 				m->m_len -= len;
1275 				so->so_rcv.sb_cc -= len;
1276 			}
1277 		}
1278 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1279 		if (so->so_oobmark) {
1280 			if ((flags & MSG_PEEK) == 0) {
1281 				so->so_oobmark -= len;
1282 				if (so->so_oobmark == 0) {
1283 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1284 					break;
1285 				}
1286 			} else {
1287 				offset += len;
1288 				if (offset == so->so_oobmark)
1289 					break;
1290 			}
1291 		}
1292 		if (flags & MSG_EOR)
1293 			break;
1294 		/*
1295 		 * If the MSG_WAITALL flag is set (for non-atomic socket),
1296 		 * we must not quit until "uio->uio_resid == 0" or an error
1297 		 * termination.  If a signal/timeout occurs, return
1298 		 * with a short count but without error.
1299 		 * Keep sockbuf locked against other readers.
1300 		 */
1301 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1302 		    !sosendallatonce(so) && nextrecord == NULL) {
1303 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1304 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1305 				break;
1306 			/*
1307 			 * Notify the protocol that some data has been
1308 			 * drained before blocking.
1309 			 */
1310 			if (pr->pr_flags & PR_WANTRCVD && so->so_pcb != NULL) {
1311 				SOCKBUF_UNLOCK(&so->so_rcv);
1312 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1313 				SOCKBUF_LOCK(&so->so_rcv);
1314 			}
1315 			SBLASTRECORDCHK(&so->so_rcv);
1316 			SBLASTMBUFCHK(&so->so_rcv);
1317 			error = sbwait(&so->so_rcv);
1318 			if (error)
1319 				goto release;
1320 			m = so->so_rcv.sb_mb;
1321 			if (m != NULL)
1322 				nextrecord = m->m_nextpkt;
1323 		}
1324 	}
1325 
1326 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1327 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1328 		flags |= MSG_TRUNC;
1329 		if ((flags & MSG_PEEK) == 0)
1330 			(void) sbdroprecord_locked(&so->so_rcv);
1331 	}
1332 	if ((flags & MSG_PEEK) == 0) {
1333 		if (m == NULL) {
1334 			/*
1335 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1336 			 * part makes sure sb_lastrecord is up-to-date if
1337 			 * there is still data in the socket buffer.
1338 			 */
1339 			so->so_rcv.sb_mb = nextrecord;
1340 			if (so->so_rcv.sb_mb == NULL) {
1341 				so->so_rcv.sb_mbtail = NULL;
1342 				so->so_rcv.sb_lastrecord = NULL;
1343 			} else if (nextrecord->m_nextpkt == NULL)
1344 				so->so_rcv.sb_lastrecord = nextrecord;
1345 		}
1346 		SBLASTRECORDCHK(&so->so_rcv);
1347 		SBLASTMBUFCHK(&so->so_rcv);
1348 		if (pr->pr_flags & PR_WANTRCVD && so->so_pcb) {
1349 			SOCKBUF_UNLOCK(&so->so_rcv);
1350 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1351 			SOCKBUF_LOCK(&so->so_rcv);
1352 		}
1353 	}
1354 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1355 	if (orig_resid == uio->uio_resid && orig_resid &&
1356 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1357 		sbunlock(&so->so_rcv);
1358 		goto restart;
1359 	}
1360 
1361 	if (flagsp != NULL)
1362 		*flagsp |= flags;
1363 release:
1364 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1365 	sbunlock(&so->so_rcv);
1366 out:
1367 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1368 	SOCKBUF_UNLOCK(&so->so_rcv);
1369 	return (error);
1370 }
1371 
1372 int
1373 soshutdown(so, how)
1374 	struct socket *so;
1375 	int how;
1376 {
1377 	struct protosw *pr = so->so_proto;
1378 
1379 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1380 		return (EINVAL);
1381 
1382 	if (how != SHUT_WR)
1383 		sorflush(so);
1384 	if (how != SHUT_RD)
1385 		return ((*pr->pr_usrreqs->pru_shutdown)(so));
1386 	return (0);
1387 }
1388 
1389 void
1390 sorflush(so)
1391 	struct socket *so;
1392 {
1393 	struct sockbuf *sb = &so->so_rcv;
1394 	struct protosw *pr = so->so_proto;
1395 	struct sockbuf asb;
1396 
1397 	/*
1398 	 * XXXRW: This is quite ugly.  The existing code made a copy of the
1399 	 * socket buffer, then zero'd the original to clear the buffer
1400 	 * fields.  However, with mutexes in the socket buffer, this causes
1401 	 * problems.  We only clear the zeroable bits of the original;
1402 	 * however, we have to initialize and destroy the mutex in the copy
1403 	 * so that dom_dispose() and sbrelease() can lock t as needed.
1404 	 */
1405 	SOCKBUF_LOCK(sb);
1406 	sb->sb_flags |= SB_NOINTR;
1407 	(void) sblock(sb, M_WAITOK);
1408 	/*
1409 	 * socantrcvmore_locked() drops the socket buffer mutex so that it
1410 	 * can safely perform wakeups.  Re-acquire the mutex before
1411 	 * continuing.
1412 	 */
1413 	socantrcvmore_locked(so);
1414 	SOCKBUF_LOCK(sb);
1415 	sbunlock(sb);
1416 	/*
1417 	 * Invalidate/clear most of the sockbuf structure, but leave
1418 	 * selinfo and mutex data unchanged.
1419 	 */
1420 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1421 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
1422 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1423 	bzero(&sb->sb_startzero,
1424 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1425 	SOCKBUF_UNLOCK(sb);
1426 
1427 	SOCKBUF_LOCK_INIT(&asb, "so_rcv");
1428 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1429 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1430 	sbrelease(&asb, so);
1431 	SOCKBUF_LOCK_DESTROY(&asb);
1432 }
1433 
1434 #ifdef INET
1435 static int
1436 do_setopt_accept_filter(so, sopt)
1437 	struct	socket *so;
1438 	struct	sockopt *sopt;
1439 {
1440 	struct accept_filter_arg	*afap = NULL;
1441 	struct accept_filter	*afp;
1442 	struct so_accf	*af = so->so_accf;
1443 	int	error = 0;
1444 
1445 	/* do not set/remove accept filters on non listen sockets */
1446 	if ((so->so_options & SO_ACCEPTCONN) == 0) {
1447 		error = EINVAL;
1448 		goto out;
1449 	}
1450 
1451 	/* removing the filter */
1452 	if (sopt == NULL) {
1453 		if (af != NULL) {
1454 			if (af->so_accept_filter != NULL &&
1455 				af->so_accept_filter->accf_destroy != NULL) {
1456 				af->so_accept_filter->accf_destroy(so);
1457 			}
1458 			if (af->so_accept_filter_str != NULL) {
1459 				FREE(af->so_accept_filter_str, M_ACCF);
1460 			}
1461 			FREE(af, M_ACCF);
1462 			so->so_accf = NULL;
1463 		}
1464 		so->so_options &= ~SO_ACCEPTFILTER;
1465 		return (0);
1466 	}
1467 	/* adding a filter */
1468 	/* must remove previous filter first */
1469 	if (af != NULL) {
1470 		error = EINVAL;
1471 		goto out;
1472 	}
1473 	/* don't put large objects on the kernel stack */
1474 	MALLOC(afap, struct accept_filter_arg *, sizeof(*afap), M_TEMP, M_WAITOK);
1475 	error = sooptcopyin(sopt, afap, sizeof *afap, sizeof *afap);
1476 	afap->af_name[sizeof(afap->af_name)-1] = '\0';
1477 	afap->af_arg[sizeof(afap->af_arg)-1] = '\0';
1478 	if (error)
1479 		goto out;
1480 	afp = accept_filt_get(afap->af_name);
1481 	if (afp == NULL) {
1482 		error = ENOENT;
1483 		goto out;
1484 	}
1485 	MALLOC(af, struct so_accf *, sizeof(*af), M_ACCF, M_WAITOK | M_ZERO);
1486 	if (afp->accf_create != NULL) {
1487 		if (afap->af_name[0] != '\0') {
1488 			int len = strlen(afap->af_name) + 1;
1489 
1490 			MALLOC(af->so_accept_filter_str, char *, len, M_ACCF, M_WAITOK);
1491 			strcpy(af->so_accept_filter_str, afap->af_name);
1492 		}
1493 		af->so_accept_filter_arg = afp->accf_create(so, afap->af_arg);
1494 		if (af->so_accept_filter_arg == NULL) {
1495 			FREE(af->so_accept_filter_str, M_ACCF);
1496 			FREE(af, M_ACCF);
1497 			so->so_accf = NULL;
1498 			error = EINVAL;
1499 			goto out;
1500 		}
1501 	}
1502 	af->so_accept_filter = afp;
1503 	so->so_accf = af;
1504 	so->so_options |= SO_ACCEPTFILTER;
1505 out:
1506 	if (afap != NULL)
1507 		FREE(afap, M_TEMP);
1508 	return (error);
1509 }
1510 #endif /* INET */
1511 
1512 /*
1513  * Perhaps this routine, and sooptcopyout(), below, ought to come in
1514  * an additional variant to handle the case where the option value needs
1515  * to be some kind of integer, but not a specific size.
1516  * In addition to their use here, these functions are also called by the
1517  * protocol-level pr_ctloutput() routines.
1518  */
1519 int
1520 sooptcopyin(sopt, buf, len, minlen)
1521 	struct	sockopt *sopt;
1522 	void	*buf;
1523 	size_t	len;
1524 	size_t	minlen;
1525 {
1526 	size_t	valsize;
1527 
1528 	/*
1529 	 * If the user gives us more than we wanted, we ignore it,
1530 	 * but if we don't get the minimum length the caller
1531 	 * wants, we return EINVAL.  On success, sopt->sopt_valsize
1532 	 * is set to however much we actually retrieved.
1533 	 */
1534 	if ((valsize = sopt->sopt_valsize) < minlen)
1535 		return EINVAL;
1536 	if (valsize > len)
1537 		sopt->sopt_valsize = valsize = len;
1538 
1539 	if (sopt->sopt_td != NULL)
1540 		return (copyin(sopt->sopt_val, buf, valsize));
1541 
1542 	bcopy(sopt->sopt_val, buf, valsize);
1543 	return 0;
1544 }
1545 
1546 /*
1547  * Kernel version of setsockopt(2)/
1548  * XXX: optlen is size_t, not socklen_t
1549  */
1550 int
1551 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1552     size_t optlen)
1553 {
1554 	struct sockopt sopt;
1555 
1556 	sopt.sopt_level = level;
1557 	sopt.sopt_name = optname;
1558 	sopt.sopt_dir = SOPT_SET;
1559 	sopt.sopt_val = optval;
1560 	sopt.sopt_valsize = optlen;
1561 	sopt.sopt_td = NULL;
1562 	return (sosetopt(so, &sopt));
1563 }
1564 
1565 int
1566 sosetopt(so, sopt)
1567 	struct socket *so;
1568 	struct sockopt *sopt;
1569 {
1570 	int	error, optval;
1571 	struct	linger l;
1572 	struct	timeval tv;
1573 	u_long  val;
1574 #ifdef MAC
1575 	struct mac extmac;
1576 #endif
1577 
1578 	error = 0;
1579 	if (sopt->sopt_level != SOL_SOCKET) {
1580 		if (so->so_proto && so->so_proto->pr_ctloutput)
1581 			return ((*so->so_proto->pr_ctloutput)
1582 				  (so, sopt));
1583 		error = ENOPROTOOPT;
1584 	} else {
1585 		switch (sopt->sopt_name) {
1586 #ifdef INET
1587 		case SO_ACCEPTFILTER:
1588 			error = do_setopt_accept_filter(so, sopt);
1589 			if (error)
1590 				goto bad;
1591 			break;
1592 #endif
1593 		case SO_LINGER:
1594 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1595 			if (error)
1596 				goto bad;
1597 
1598 			SOCK_LOCK(so);
1599 			so->so_linger = l.l_linger;
1600 			if (l.l_onoff)
1601 				so->so_options |= SO_LINGER;
1602 			else
1603 				so->so_options &= ~SO_LINGER;
1604 			SOCK_UNLOCK(so);
1605 			break;
1606 
1607 		case SO_DEBUG:
1608 		case SO_KEEPALIVE:
1609 		case SO_DONTROUTE:
1610 		case SO_USELOOPBACK:
1611 		case SO_BROADCAST:
1612 		case SO_REUSEADDR:
1613 		case SO_REUSEPORT:
1614 		case SO_OOBINLINE:
1615 		case SO_TIMESTAMP:
1616 		case SO_BINTIME:
1617 		case SO_NOSIGPIPE:
1618 			error = sooptcopyin(sopt, &optval, sizeof optval,
1619 					    sizeof optval);
1620 			if (error)
1621 				goto bad;
1622 			SOCK_LOCK(so);
1623 			if (optval)
1624 				so->so_options |= sopt->sopt_name;
1625 			else
1626 				so->so_options &= ~sopt->sopt_name;
1627 			SOCK_UNLOCK(so);
1628 			break;
1629 
1630 		case SO_SNDBUF:
1631 		case SO_RCVBUF:
1632 		case SO_SNDLOWAT:
1633 		case SO_RCVLOWAT:
1634 			error = sooptcopyin(sopt, &optval, sizeof optval,
1635 					    sizeof optval);
1636 			if (error)
1637 				goto bad;
1638 
1639 			/*
1640 			 * Values < 1 make no sense for any of these
1641 			 * options, so disallow them.
1642 			 */
1643 			if (optval < 1) {
1644 				error = EINVAL;
1645 				goto bad;
1646 			}
1647 
1648 			switch (sopt->sopt_name) {
1649 			case SO_SNDBUF:
1650 			case SO_RCVBUF:
1651 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
1652 				    &so->so_snd : &so->so_rcv, (u_long)optval,
1653 				    so, curthread) == 0) {
1654 					error = ENOBUFS;
1655 					goto bad;
1656 				}
1657 				break;
1658 
1659 			/*
1660 			 * Make sure the low-water is never greater than
1661 			 * the high-water.
1662 			 */
1663 			case SO_SNDLOWAT:
1664 				SOCKBUF_LOCK(&so->so_snd);
1665 				so->so_snd.sb_lowat =
1666 				    (optval > so->so_snd.sb_hiwat) ?
1667 				    so->so_snd.sb_hiwat : optval;
1668 				SOCKBUF_UNLOCK(&so->so_snd);
1669 				break;
1670 			case SO_RCVLOWAT:
1671 				SOCKBUF_LOCK(&so->so_rcv);
1672 				so->so_rcv.sb_lowat =
1673 				    (optval > so->so_rcv.sb_hiwat) ?
1674 				    so->so_rcv.sb_hiwat : optval;
1675 				SOCKBUF_UNLOCK(&so->so_rcv);
1676 				break;
1677 			}
1678 			break;
1679 
1680 		case SO_SNDTIMEO:
1681 		case SO_RCVTIMEO:
1682 			error = sooptcopyin(sopt, &tv, sizeof tv,
1683 					    sizeof tv);
1684 			if (error)
1685 				goto bad;
1686 
1687 			/* assert(hz > 0); */
1688 			if (tv.tv_sec < 0 || tv.tv_sec > SHRT_MAX / hz ||
1689 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
1690 				error = EDOM;
1691 				goto bad;
1692 			}
1693 			/* assert(tick > 0); */
1694 			/* assert(ULONG_MAX - SHRT_MAX >= 1000000); */
1695 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
1696 			if (val > SHRT_MAX) {
1697 				error = EDOM;
1698 				goto bad;
1699 			}
1700 			if (val == 0 && tv.tv_usec != 0)
1701 				val = 1;
1702 
1703 			switch (sopt->sopt_name) {
1704 			case SO_SNDTIMEO:
1705 				so->so_snd.sb_timeo = val;
1706 				break;
1707 			case SO_RCVTIMEO:
1708 				so->so_rcv.sb_timeo = val;
1709 				break;
1710 			}
1711 			break;
1712 		case SO_LABEL:
1713 #ifdef MAC
1714 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
1715 			    sizeof extmac);
1716 			if (error)
1717 				goto bad;
1718 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
1719 			    so, &extmac);
1720 #else
1721 			error = EOPNOTSUPP;
1722 #endif
1723 			break;
1724 		default:
1725 			error = ENOPROTOOPT;
1726 			break;
1727 		}
1728 		if (error == 0 && so->so_proto != NULL &&
1729 		    so->so_proto->pr_ctloutput != NULL) {
1730 			(void) ((*so->so_proto->pr_ctloutput)
1731 				  (so, sopt));
1732 		}
1733 	}
1734 bad:
1735 	return (error);
1736 }
1737 
1738 /* Helper routine for getsockopt */
1739 int
1740 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
1741 {
1742 	int	error;
1743 	size_t	valsize;
1744 
1745 	error = 0;
1746 
1747 	/*
1748 	 * Documented get behavior is that we always return a value,
1749 	 * possibly truncated to fit in the user's buffer.
1750 	 * Traditional behavior is that we always tell the user
1751 	 * precisely how much we copied, rather than something useful
1752 	 * like the total amount we had available for her.
1753 	 * Note that this interface is not idempotent; the entire answer must
1754 	 * generated ahead of time.
1755 	 */
1756 	valsize = min(len, sopt->sopt_valsize);
1757 	sopt->sopt_valsize = valsize;
1758 	if (sopt->sopt_val != NULL) {
1759 		if (sopt->sopt_td != NULL)
1760 			error = copyout(buf, sopt->sopt_val, valsize);
1761 		else
1762 			bcopy(buf, sopt->sopt_val, valsize);
1763 	}
1764 	return error;
1765 }
1766 
1767 int
1768 sogetopt(so, sopt)
1769 	struct socket *so;
1770 	struct sockopt *sopt;
1771 {
1772 	int	error, optval;
1773 	struct	linger l;
1774 	struct	timeval tv;
1775 #ifdef INET
1776 	struct accept_filter_arg *afap;
1777 #endif
1778 #ifdef MAC
1779 	struct mac extmac;
1780 #endif
1781 
1782 	error = 0;
1783 	if (sopt->sopt_level != SOL_SOCKET) {
1784 		if (so->so_proto && so->so_proto->pr_ctloutput) {
1785 			return ((*so->so_proto->pr_ctloutput)
1786 				  (so, sopt));
1787 		} else
1788 			return (ENOPROTOOPT);
1789 	} else {
1790 		switch (sopt->sopt_name) {
1791 #ifdef INET
1792 		case SO_ACCEPTFILTER:
1793 			if ((so->so_options & SO_ACCEPTCONN) == 0)
1794 				return (EINVAL);
1795 			MALLOC(afap, struct accept_filter_arg *, sizeof(*afap),
1796 				M_TEMP, M_WAITOK | M_ZERO);
1797 			if ((so->so_options & SO_ACCEPTFILTER) != 0) {
1798 				strcpy(afap->af_name, so->so_accf->so_accept_filter->accf_name);
1799 				if (so->so_accf->so_accept_filter_str != NULL)
1800 					strcpy(afap->af_arg, so->so_accf->so_accept_filter_str);
1801 			}
1802 			error = sooptcopyout(sopt, afap, sizeof(*afap));
1803 			FREE(afap, M_TEMP);
1804 			break;
1805 #endif
1806 
1807 		case SO_LINGER:
1808 			/*
1809 			 * XXXRW: We grab the lock here to get a consistent
1810 			 * snapshot of both fields.  This may not really
1811 			 * be necessary.
1812 			 */
1813 			SOCK_LOCK(so);
1814 			l.l_onoff = so->so_options & SO_LINGER;
1815 			l.l_linger = so->so_linger;
1816 			SOCK_UNLOCK(so);
1817 			error = sooptcopyout(sopt, &l, sizeof l);
1818 			break;
1819 
1820 		case SO_USELOOPBACK:
1821 		case SO_DONTROUTE:
1822 		case SO_DEBUG:
1823 		case SO_KEEPALIVE:
1824 		case SO_REUSEADDR:
1825 		case SO_REUSEPORT:
1826 		case SO_BROADCAST:
1827 		case SO_OOBINLINE:
1828 		case SO_TIMESTAMP:
1829 		case SO_BINTIME:
1830 		case SO_NOSIGPIPE:
1831 			optval = so->so_options & sopt->sopt_name;
1832 integer:
1833 			error = sooptcopyout(sopt, &optval, sizeof optval);
1834 			break;
1835 
1836 		case SO_TYPE:
1837 			optval = so->so_type;
1838 			goto integer;
1839 
1840 		case SO_ERROR:
1841 			optval = so->so_error;
1842 			so->so_error = 0;
1843 			goto integer;
1844 
1845 		case SO_SNDBUF:
1846 			optval = so->so_snd.sb_hiwat;
1847 			goto integer;
1848 
1849 		case SO_RCVBUF:
1850 			optval = so->so_rcv.sb_hiwat;
1851 			goto integer;
1852 
1853 		case SO_SNDLOWAT:
1854 			optval = so->so_snd.sb_lowat;
1855 			goto integer;
1856 
1857 		case SO_RCVLOWAT:
1858 			optval = so->so_rcv.sb_lowat;
1859 			goto integer;
1860 
1861 		case SO_SNDTIMEO:
1862 		case SO_RCVTIMEO:
1863 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
1864 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
1865 
1866 			tv.tv_sec = optval / hz;
1867 			tv.tv_usec = (optval % hz) * tick;
1868 			error = sooptcopyout(sopt, &tv, sizeof tv);
1869 			break;
1870 		case SO_LABEL:
1871 #ifdef MAC
1872 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1873 			    sizeof(extmac));
1874 			if (error)
1875 				return (error);
1876 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
1877 			    so, &extmac);
1878 			if (error)
1879 				return (error);
1880 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
1881 #else
1882 			error = EOPNOTSUPP;
1883 #endif
1884 			break;
1885 		case SO_PEERLABEL:
1886 #ifdef MAC
1887 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
1888 			    sizeof(extmac));
1889 			if (error)
1890 				return (error);
1891 			error = mac_getsockopt_peerlabel(
1892 			    sopt->sopt_td->td_ucred, so, &extmac);
1893 			if (error)
1894 				return (error);
1895 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
1896 #else
1897 			error = EOPNOTSUPP;
1898 #endif
1899 			break;
1900 		default:
1901 			error = ENOPROTOOPT;
1902 			break;
1903 		}
1904 		return (error);
1905 	}
1906 }
1907 
1908 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
1909 int
1910 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
1911 {
1912 	struct mbuf *m, *m_prev;
1913 	int sopt_size = sopt->sopt_valsize;
1914 
1915 	MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1916 	if (m == NULL)
1917 		return ENOBUFS;
1918 	if (sopt_size > MLEN) {
1919 		MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
1920 		if ((m->m_flags & M_EXT) == 0) {
1921 			m_free(m);
1922 			return ENOBUFS;
1923 		}
1924 		m->m_len = min(MCLBYTES, sopt_size);
1925 	} else {
1926 		m->m_len = min(MLEN, sopt_size);
1927 	}
1928 	sopt_size -= m->m_len;
1929 	*mp = m;
1930 	m_prev = m;
1931 
1932 	while (sopt_size) {
1933 		MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
1934 		if (m == NULL) {
1935 			m_freem(*mp);
1936 			return ENOBUFS;
1937 		}
1938 		if (sopt_size > MLEN) {
1939 			MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
1940 			    M_DONTWAIT);
1941 			if ((m->m_flags & M_EXT) == 0) {
1942 				m_freem(m);
1943 				m_freem(*mp);
1944 				return ENOBUFS;
1945 			}
1946 			m->m_len = min(MCLBYTES, sopt_size);
1947 		} else {
1948 			m->m_len = min(MLEN, sopt_size);
1949 		}
1950 		sopt_size -= m->m_len;
1951 		m_prev->m_next = m;
1952 		m_prev = m;
1953 	}
1954 	return 0;
1955 }
1956 
1957 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
1958 int
1959 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
1960 {
1961 	struct mbuf *m0 = m;
1962 
1963 	if (sopt->sopt_val == NULL)
1964 		return 0;
1965 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1966 		if (sopt->sopt_td != NULL) {
1967 			int error;
1968 
1969 			error = copyin(sopt->sopt_val, mtod(m, char *),
1970 				       m->m_len);
1971 			if (error != 0) {
1972 				m_freem(m0);
1973 				return(error);
1974 			}
1975 		} else
1976 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
1977 		sopt->sopt_valsize -= m->m_len;
1978 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
1979 		m = m->m_next;
1980 	}
1981 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
1982 		panic("ip6_sooptmcopyin");
1983 	return 0;
1984 }
1985 
1986 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
1987 int
1988 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
1989 {
1990 	struct mbuf *m0 = m;
1991 	size_t valsize = 0;
1992 
1993 	if (sopt->sopt_val == NULL)
1994 		return 0;
1995 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
1996 		if (sopt->sopt_td != NULL) {
1997 			int error;
1998 
1999 			error = copyout(mtod(m, char *), sopt->sopt_val,
2000 				       m->m_len);
2001 			if (error != 0) {
2002 				m_freem(m0);
2003 				return(error);
2004 			}
2005 		} else
2006 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2007 	       sopt->sopt_valsize -= m->m_len;
2008 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2009 	       valsize += m->m_len;
2010 	       m = m->m_next;
2011 	}
2012 	if (m != NULL) {
2013 		/* enough soopt buffer should be given from user-land */
2014 		m_freem(m0);
2015 		return(EINVAL);
2016 	}
2017 	sopt->sopt_valsize = valsize;
2018 	return 0;
2019 }
2020 
2021 void
2022 sohasoutofband(so)
2023 	struct socket *so;
2024 {
2025 	if (so->so_sigio != NULL)
2026 		pgsigio(&so->so_sigio, SIGURG, 0);
2027 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2028 }
2029 
2030 int
2031 sopoll(struct socket *so, int events, struct ucred *active_cred,
2032     struct thread *td)
2033 {
2034 	int revents = 0;
2035 
2036 	if (events & (POLLIN | POLLRDNORM))
2037 		if (soreadable(so))
2038 			revents |= events & (POLLIN | POLLRDNORM);
2039 
2040 	if (events & POLLINIGNEOF)
2041 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2042 		    !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2043 			revents |= POLLINIGNEOF;
2044 
2045 	if (events & (POLLOUT | POLLWRNORM))
2046 		if (sowriteable(so))
2047 			revents |= events & (POLLOUT | POLLWRNORM);
2048 
2049 	if (events & (POLLPRI | POLLRDBAND))
2050 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2051 			revents |= events & (POLLPRI | POLLRDBAND);
2052 
2053 	if (revents == 0) {
2054 		if (events &
2055 		    (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2056 		     POLLRDBAND)) {
2057 			SOCKBUF_LOCK(&so->so_rcv);
2058 			selrecord(td, &so->so_rcv.sb_sel);
2059 			so->so_rcv.sb_flags |= SB_SEL;
2060 			SOCKBUF_UNLOCK(&so->so_rcv);
2061 		}
2062 
2063 		if (events & (POLLOUT | POLLWRNORM)) {
2064 			SOCKBUF_LOCK(&so->so_snd);
2065 			selrecord(td, &so->so_snd.sb_sel);
2066 			so->so_snd.sb_flags |= SB_SEL;
2067 			SOCKBUF_UNLOCK(&so->so_snd);
2068 		}
2069 	}
2070 
2071 	return (revents);
2072 }
2073 
2074 int
2075 soo_kqfilter(struct file *fp, struct knote *kn)
2076 {
2077 	struct socket *so = kn->kn_fp->f_data;
2078 	struct sockbuf *sb;
2079 
2080 	switch (kn->kn_filter) {
2081 	case EVFILT_READ:
2082 		if (so->so_options & SO_ACCEPTCONN)
2083 			kn->kn_fop = &solisten_filtops;
2084 		else
2085 			kn->kn_fop = &soread_filtops;
2086 		sb = &so->so_rcv;
2087 		break;
2088 	case EVFILT_WRITE:
2089 		kn->kn_fop = &sowrite_filtops;
2090 		sb = &so->so_snd;
2091 		break;
2092 	default:
2093 		return (1);
2094 	}
2095 
2096 	SOCKBUF_LOCK(sb);
2097 	SLIST_INSERT_HEAD(&sb->sb_sel.si_note, kn, kn_selnext);
2098 	sb->sb_flags |= SB_KNOTE;
2099 	SOCKBUF_UNLOCK(sb);
2100 	return (0);
2101 }
2102 
2103 static void
2104 filt_sordetach(struct knote *kn)
2105 {
2106 	struct socket *so = kn->kn_fp->f_data;
2107 
2108 	SOCKBUF_LOCK(&so->so_rcv);
2109 	SLIST_REMOVE(&so->so_rcv.sb_sel.si_note, kn, knote, kn_selnext);
2110 	if (SLIST_EMPTY(&so->so_rcv.sb_sel.si_note))
2111 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2112 	SOCKBUF_UNLOCK(&so->so_rcv);
2113 }
2114 
2115 /*ARGSUSED*/
2116 static int
2117 filt_soread(struct knote *kn, long hint)
2118 {
2119 	struct socket *so = kn->kn_fp->f_data;
2120 	int need_lock, result;
2121 
2122 	/*
2123 	 * XXXRW: Conditional locking because filt_soread() can be called
2124 	 * either from KNOTE() in the socket context where the socket buffer
2125 	 * lock is already held, or from kqueue() itself.
2126 	 */
2127 	need_lock = !SOCKBUF_OWNED(&so->so_rcv);
2128 	if (need_lock)
2129 		SOCKBUF_LOCK(&so->so_rcv);
2130 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2131 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2132 		kn->kn_flags |= EV_EOF;
2133 		kn->kn_fflags = so->so_error;
2134 		result = 1;
2135 	} else if (so->so_error)	/* temporary udp error */
2136 		result = 1;
2137 	else if (kn->kn_sfflags & NOTE_LOWAT)
2138 		result = (kn->kn_data >= kn->kn_sdata);
2139 	else
2140 		result = (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2141 	if (need_lock)
2142 		SOCKBUF_UNLOCK(&so->so_rcv);
2143 	return (result);
2144 }
2145 
2146 static void
2147 filt_sowdetach(struct knote *kn)
2148 {
2149 	struct socket *so = kn->kn_fp->f_data;
2150 
2151 	SOCKBUF_LOCK(&so->so_snd);
2152 	SLIST_REMOVE(&so->so_snd.sb_sel.si_note, kn, knote, kn_selnext);
2153 	if (SLIST_EMPTY(&so->so_snd.sb_sel.si_note))
2154 		so->so_snd.sb_flags &= ~SB_KNOTE;
2155 	SOCKBUF_UNLOCK(&so->so_snd);
2156 }
2157 
2158 /*ARGSUSED*/
2159 static int
2160 filt_sowrite(struct knote *kn, long hint)
2161 {
2162 	struct socket *so = kn->kn_fp->f_data;
2163 	int need_lock, result;
2164 
2165 	/*
2166 	 * XXXRW: Conditional locking because filt_soread() can be called
2167 	 * either from KNOTE() in the socket context where the socket buffer
2168 	 * lock is already held, or from kqueue() itself.
2169 	 */
2170 	need_lock = !SOCKBUF_OWNED(&so->so_snd);
2171 	if (need_lock)
2172 		SOCKBUF_LOCK(&so->so_snd);
2173 	kn->kn_data = sbspace(&so->so_snd);
2174 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2175 		kn->kn_flags |= EV_EOF;
2176 		kn->kn_fflags = so->so_error;
2177 		result = 1;
2178 	} else if (so->so_error)	/* temporary udp error */
2179 		result = 1;
2180 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2181 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2182 		result = 0;
2183 	else if (kn->kn_sfflags & NOTE_LOWAT)
2184 		result = (kn->kn_data >= kn->kn_sdata);
2185 	else
2186 		result = (kn->kn_data >= so->so_snd.sb_lowat);
2187 	if (need_lock)
2188 		SOCKBUF_UNLOCK(&so->so_snd);
2189 	return (result);
2190 }
2191 
2192 /*ARGSUSED*/
2193 static int
2194 filt_solisten(struct knote *kn, long hint)
2195 {
2196 	struct socket *so = kn->kn_fp->f_data;
2197 
2198 	kn->kn_data = so->so_qlen;
2199 	return (! TAILQ_EMPTY(&so->so_comp));
2200 }
2201 
2202 int
2203 socheckuid(struct socket *so, uid_t uid)
2204 {
2205 
2206 	if (so == NULL)
2207 		return (EPERM);
2208 	if (so->so_cred->cr_uid == uid)
2209 		return (0);
2210 	return (EPERM);
2211 }
2212