xref: /freebsd/sys/kern/uipc_socket.c (revision 9fd69f37d28cfd7438cac3eeb45fe9dd46b4d7dd)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004 The FreeBSD Foundation
5  * Copyright (c) 2004-2008 Robert N. M. Watson
6  * All rights reserved.
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  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
33  */
34 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  */
96 
97 #include <sys/cdefs.h>
98 __FBSDID("$FreeBSD$");
99 
100 #include "opt_inet.h"
101 #include "opt_inet6.h"
102 #include "opt_zero.h"
103 #include "opt_compat.h"
104 
105 #include <sys/param.h>
106 #include <sys/systm.h>
107 #include <sys/fcntl.h>
108 #include <sys/limits.h>
109 #include <sys/lock.h>
110 #include <sys/mac.h>
111 #include <sys/malloc.h>
112 #include <sys/mbuf.h>
113 #include <sys/mutex.h>
114 #include <sys/domain.h>
115 #include <sys/file.h>			/* for struct knote */
116 #include <sys/kernel.h>
117 #include <sys/event.h>
118 #include <sys/eventhandler.h>
119 #include <sys/poll.h>
120 #include <sys/proc.h>
121 #include <sys/protosw.h>
122 #include <sys/socket.h>
123 #include <sys/socketvar.h>
124 #include <sys/resourcevar.h>
125 #include <net/route.h>
126 #include <sys/signalvar.h>
127 #include <sys/stat.h>
128 #include <sys/sx.h>
129 #include <sys/sysctl.h>
130 #include <sys/uio.h>
131 #include <sys/jail.h>
132 
133 #include <net/vnet.h>
134 
135 #include <security/mac/mac_framework.h>
136 
137 #include <vm/uma.h>
138 
139 #ifdef COMPAT_IA32
140 #include <sys/mount.h>
141 #include <sys/sysent.h>
142 #include <compat/freebsd32/freebsd32.h>
143 #endif
144 
145 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
146 		    int flags);
147 
148 static void	filt_sordetach(struct knote *kn);
149 static int	filt_soread(struct knote *kn, long hint);
150 static void	filt_sowdetach(struct knote *kn);
151 static int	filt_sowrite(struct knote *kn, long hint);
152 static int	filt_solisten(struct knote *kn, long hint);
153 
154 static struct filterops solisten_filtops = {
155 	.f_isfd = 1,
156 	.f_detach = filt_sordetach,
157 	.f_event = filt_solisten,
158 };
159 static struct filterops soread_filtops = {
160 	.f_isfd = 1,
161 	.f_detach = filt_sordetach,
162 	.f_event = filt_soread,
163 };
164 static struct filterops sowrite_filtops = {
165 	.f_isfd = 1,
166 	.f_detach = filt_sowdetach,
167 	.f_event = filt_sowrite,
168 };
169 
170 uma_zone_t socket_zone;
171 so_gen_t	so_gencnt;	/* generation count for sockets */
172 
173 int	maxsockets;
174 
175 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
176 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
177 
178 static int somaxconn = SOMAXCONN;
179 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
180 /* XXX: we dont have SYSCTL_USHORT */
181 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
182     0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
183     "queue size");
184 static int numopensockets;
185 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
186     &numopensockets, 0, "Number of open sockets");
187 #ifdef ZERO_COPY_SOCKETS
188 /* These aren't static because they're used in other files. */
189 int so_zero_copy_send = 1;
190 int so_zero_copy_receive = 1;
191 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
192     "Zero copy controls");
193 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
194     &so_zero_copy_receive, 0, "Enable zero copy receive");
195 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
196     &so_zero_copy_send, 0, "Enable zero copy send");
197 #endif /* ZERO_COPY_SOCKETS */
198 
199 /*
200  * accept_mtx locks down per-socket fields relating to accept queues.  See
201  * socketvar.h for an annotation of the protected fields of struct socket.
202  */
203 struct mtx accept_mtx;
204 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
205 
206 /*
207  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
208  * so_gencnt field.
209  */
210 static struct mtx so_global_mtx;
211 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
212 
213 /*
214  * General IPC sysctl name space, used by sockets and a variety of other IPC
215  * types.
216  */
217 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
218 
219 /*
220  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
221  * of the change so that they can update their dependent limits as required.
222  */
223 static int
224 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
225 {
226 	int error, newmaxsockets;
227 
228 	newmaxsockets = maxsockets;
229 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
230 	if (error == 0 && req->newptr) {
231 		if (newmaxsockets > maxsockets) {
232 			maxsockets = newmaxsockets;
233 			if (maxsockets > ((maxfiles / 4) * 3)) {
234 				maxfiles = (maxsockets * 5) / 4;
235 				maxfilesperproc = (maxfiles * 9) / 10;
236 			}
237 			EVENTHANDLER_INVOKE(maxsockets_change);
238 		} else
239 			error = EINVAL;
240 	}
241 	return (error);
242 }
243 
244 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
245     &maxsockets, 0, sysctl_maxsockets, "IU",
246     "Maximum number of sockets avaliable");
247 
248 /*
249  * Initialise maxsockets.  This SYSINIT must be run after
250  * tunable_mbinit().
251  */
252 static void
253 init_maxsockets(void *ignored)
254 {
255 
256 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
257 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
258 }
259 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
260 
261 /*
262  * Socket operation routines.  These routines are called by the routines in
263  * sys_socket.c or from a system process, and implement the semantics of
264  * socket operations by switching out to the protocol specific routines.
265  */
266 
267 /*
268  * Get a socket structure from our zone, and initialize it.  Note that it
269  * would probably be better to allocate socket and PCB at the same time, but
270  * I'm not convinced that all the protocols can be easily modified to do
271  * this.
272  *
273  * soalloc() returns a socket with a ref count of 0.
274  */
275 static struct socket *
276 soalloc(struct vnet *vnet)
277 {
278 	struct socket *so;
279 
280 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
281 	if (so == NULL)
282 		return (NULL);
283 #ifdef MAC
284 	if (mac_socket_init(so, M_NOWAIT) != 0) {
285 		uma_zfree(socket_zone, so);
286 		return (NULL);
287 	}
288 #endif
289 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
290 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
291 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
292 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
293 	TAILQ_INIT(&so->so_aiojobq);
294 	mtx_lock(&so_global_mtx);
295 	so->so_gencnt = ++so_gencnt;
296 	++numopensockets;
297 #ifdef VIMAGE
298 	vnet->vnet_sockcnt++;
299 	so->so_vnet = vnet;
300 #endif
301 	mtx_unlock(&so_global_mtx);
302 	return (so);
303 }
304 
305 /*
306  * Free the storage associated with a socket at the socket layer, tear down
307  * locks, labels, etc.  All protocol state is assumed already to have been
308  * torn down (and possibly never set up) by the caller.
309  */
310 static void
311 sodealloc(struct socket *so)
312 {
313 
314 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
315 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
316 
317 	mtx_lock(&so_global_mtx);
318 	so->so_gencnt = ++so_gencnt;
319 	--numopensockets;	/* Could be below, but faster here. */
320 #ifdef VIMAGE
321 	so->so_vnet->vnet_sockcnt--;
322 #endif
323 	mtx_unlock(&so_global_mtx);
324 	if (so->so_rcv.sb_hiwat)
325 		(void)chgsbsize(so->so_cred->cr_uidinfo,
326 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
327 	if (so->so_snd.sb_hiwat)
328 		(void)chgsbsize(so->so_cred->cr_uidinfo,
329 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
330 #ifdef INET
331 	/* remove acccept filter if one is present. */
332 	if (so->so_accf != NULL)
333 		do_setopt_accept_filter(so, NULL);
334 #endif
335 #ifdef MAC
336 	mac_socket_destroy(so);
337 #endif
338 	crfree(so->so_cred);
339 	sx_destroy(&so->so_snd.sb_sx);
340 	sx_destroy(&so->so_rcv.sb_sx);
341 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
342 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
343 	uma_zfree(socket_zone, so);
344 }
345 
346 /*
347  * socreate returns a socket with a ref count of 1.  The socket should be
348  * closed with soclose().
349  */
350 int
351 socreate(int dom, struct socket **aso, int type, int proto,
352     struct ucred *cred, struct thread *td)
353 {
354 	struct protosw *prp;
355 	struct socket *so;
356 	int error;
357 
358 	if (proto)
359 		prp = pffindproto(dom, proto, type);
360 	else
361 		prp = pffindtype(dom, type);
362 
363 	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
364 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
365 		return (EPROTONOSUPPORT);
366 
367 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
368 		return (EPROTONOSUPPORT);
369 
370 	if (prp->pr_type != type)
371 		return (EPROTOTYPE);
372 	so = soalloc(CRED_TO_VNET(cred));
373 	if (so == NULL)
374 		return (ENOBUFS);
375 
376 	TAILQ_INIT(&so->so_incomp);
377 	TAILQ_INIT(&so->so_comp);
378 	so->so_type = type;
379 	so->so_cred = crhold(cred);
380 	if ((prp->pr_domain->dom_family == PF_INET) ||
381 	    (prp->pr_domain->dom_family == PF_ROUTE))
382 		so->so_fibnum = td->td_proc->p_fibnum;
383 	else
384 		so->so_fibnum = 0;
385 	so->so_proto = prp;
386 #ifdef MAC
387 	mac_socket_create(cred, so);
388 #endif
389 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
390 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
391 	so->so_count = 1;
392 	/*
393 	 * Auto-sizing of socket buffers is managed by the protocols and
394 	 * the appropriate flags must be set in the pru_attach function.
395 	 */
396 	CURVNET_SET(so->so_vnet);
397 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
398 	CURVNET_RESTORE();
399 	if (error) {
400 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
401 		    so->so_count));
402 		so->so_count = 0;
403 		sodealloc(so);
404 		return (error);
405 	}
406 	*aso = so;
407 	return (0);
408 }
409 
410 #ifdef REGRESSION
411 static int regression_sonewconn_earlytest = 1;
412 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
413     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
414 #endif
415 
416 /*
417  * When an attempt at a new connection is noted on a socket which accepts
418  * connections, sonewconn is called.  If the connection is possible (subject
419  * to space constraints, etc.) then we allocate a new structure, propoerly
420  * linked into the data structure of the original socket, and return this.
421  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
422  *
423  * Note: the ref count on the socket is 0 on return.
424  */
425 struct socket *
426 sonewconn(struct socket *head, int connstatus)
427 {
428 	struct socket *so;
429 	int over;
430 
431 	ACCEPT_LOCK();
432 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
433 	ACCEPT_UNLOCK();
434 #ifdef REGRESSION
435 	if (regression_sonewconn_earlytest && over)
436 #else
437 	if (over)
438 #endif
439 		return (NULL);
440 	VNET_ASSERT(head->so_vnet);
441 	so = soalloc(head->so_vnet);
442 	if (so == NULL)
443 		return (NULL);
444 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
445 		connstatus = 0;
446 	so->so_head = head;
447 	so->so_type = head->so_type;
448 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
449 	so->so_linger = head->so_linger;
450 	so->so_state = head->so_state | SS_NOFDREF;
451 	so->so_fibnum = head->so_fibnum;
452 	so->so_proto = head->so_proto;
453 	so->so_cred = crhold(head->so_cred);
454 #ifdef MAC
455 	mac_socket_newconn(head, so);
456 #endif
457 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
458 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
459 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
460 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
461 		sodealloc(so);
462 		return (NULL);
463 	}
464 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
465 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
466 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
467 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
468 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
469 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
470 	so->so_state |= connstatus;
471 	ACCEPT_LOCK();
472 	if (connstatus) {
473 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
474 		so->so_qstate |= SQ_COMP;
475 		head->so_qlen++;
476 	} else {
477 		/*
478 		 * Keep removing sockets from the head until there's room for
479 		 * us to insert on the tail.  In pre-locking revisions, this
480 		 * was a simple if(), but as we could be racing with other
481 		 * threads and soabort() requires dropping locks, we must
482 		 * loop waiting for the condition to be true.
483 		 */
484 		while (head->so_incqlen > head->so_qlimit) {
485 			struct socket *sp;
486 			sp = TAILQ_FIRST(&head->so_incomp);
487 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
488 			head->so_incqlen--;
489 			sp->so_qstate &= ~SQ_INCOMP;
490 			sp->so_head = NULL;
491 			ACCEPT_UNLOCK();
492 			soabort(sp);
493 			ACCEPT_LOCK();
494 		}
495 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
496 		so->so_qstate |= SQ_INCOMP;
497 		head->so_incqlen++;
498 	}
499 	ACCEPT_UNLOCK();
500 	if (connstatus) {
501 		sorwakeup(head);
502 		wakeup_one(&head->so_timeo);
503 	}
504 	return (so);
505 }
506 
507 int
508 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
509 {
510 	int error;
511 
512 	CURVNET_SET(so->so_vnet);
513 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
514 	CURVNET_RESTORE();
515 	return error;
516 }
517 
518 /*
519  * solisten() transitions a socket from a non-listening state to a listening
520  * state, but can also be used to update the listen queue depth on an
521  * existing listen socket.  The protocol will call back into the sockets
522  * layer using solisten_proto_check() and solisten_proto() to check and set
523  * socket-layer listen state.  Call backs are used so that the protocol can
524  * acquire both protocol and socket layer locks in whatever order is required
525  * by the protocol.
526  *
527  * Protocol implementors are advised to hold the socket lock across the
528  * socket-layer test and set to avoid races at the socket layer.
529  */
530 int
531 solisten(struct socket *so, int backlog, struct thread *td)
532 {
533 
534 	return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
535 }
536 
537 int
538 solisten_proto_check(struct socket *so)
539 {
540 
541 	SOCK_LOCK_ASSERT(so);
542 
543 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
544 	    SS_ISDISCONNECTING))
545 		return (EINVAL);
546 	return (0);
547 }
548 
549 void
550 solisten_proto(struct socket *so, int backlog)
551 {
552 
553 	SOCK_LOCK_ASSERT(so);
554 
555 	if (backlog < 0 || backlog > somaxconn)
556 		backlog = somaxconn;
557 	so->so_qlimit = backlog;
558 	so->so_options |= SO_ACCEPTCONN;
559 }
560 
561 /*
562  * Attempt to free a socket.  This should really be sotryfree().
563  *
564  * sofree() will succeed if:
565  *
566  * - There are no outstanding file descriptor references or related consumers
567  *   (so_count == 0).
568  *
569  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
570  *
571  * - The protocol does not have an outstanding strong reference on the socket
572  *   (SS_PROTOREF).
573  *
574  * - The socket is not in a completed connection queue, so a process has been
575  *   notified that it is present.  If it is removed, the user process may
576  *   block in accept() despite select() saying the socket was ready.
577  *
578  * Otherwise, it will quietly abort so that a future call to sofree(), when
579  * conditions are right, can succeed.
580  */
581 void
582 sofree(struct socket *so)
583 {
584 	struct protosw *pr = so->so_proto;
585 	struct socket *head;
586 
587 	ACCEPT_LOCK_ASSERT();
588 	SOCK_LOCK_ASSERT(so);
589 
590 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
591 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
592 		SOCK_UNLOCK(so);
593 		ACCEPT_UNLOCK();
594 		return;
595 	}
596 
597 	head = so->so_head;
598 	if (head != NULL) {
599 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
600 		    (so->so_qstate & SQ_INCOMP) != 0,
601 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
602 		    "SQ_INCOMP"));
603 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
604 		    (so->so_qstate & SQ_INCOMP) == 0,
605 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
606 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
607 		head->so_incqlen--;
608 		so->so_qstate &= ~SQ_INCOMP;
609 		so->so_head = NULL;
610 	}
611 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
612 	    (so->so_qstate & SQ_INCOMP) == 0,
613 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
614 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
615 	if (so->so_options & SO_ACCEPTCONN) {
616 		KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
617 		KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
618 	}
619 	SOCK_UNLOCK(so);
620 	ACCEPT_UNLOCK();
621 
622 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
623 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
624 	if (pr->pr_usrreqs->pru_detach != NULL)
625 		(*pr->pr_usrreqs->pru_detach)(so);
626 
627 	/*
628 	 * From this point on, we assume that no other references to this
629 	 * socket exist anywhere else in the stack.  Therefore, no locks need
630 	 * to be acquired or held.
631 	 *
632 	 * We used to do a lot of socket buffer and socket locking here, as
633 	 * well as invoke sorflush() and perform wakeups.  The direct call to
634 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
635 	 * necessary from sorflush().
636 	 *
637 	 * Notice that the socket buffer and kqueue state are torn down
638 	 * before calling pru_detach.  This means that protocols shold not
639 	 * assume they can perform socket wakeups, etc, in their detach code.
640 	 */
641 	sbdestroy(&so->so_snd, so);
642 	sbdestroy(&so->so_rcv, so);
643 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
644 	knlist_destroy(&so->so_snd.sb_sel.si_note);
645 	sodealloc(so);
646 }
647 
648 /*
649  * Close a socket on last file table reference removal.  Initiate disconnect
650  * if connected.  Free socket when disconnect complete.
651  *
652  * This function will sorele() the socket.  Note that soclose() may be called
653  * prior to the ref count reaching zero.  The actual socket structure will
654  * not be freed until the ref count reaches zero.
655  */
656 int
657 soclose(struct socket *so)
658 {
659 	int error = 0;
660 
661 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
662 
663 	CURVNET_SET(so->so_vnet);
664 	funsetown(&so->so_sigio);
665 	if (so->so_state & SS_ISCONNECTED) {
666 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
667 			error = sodisconnect(so);
668 			if (error)
669 				goto drop;
670 		}
671 		if (so->so_options & SO_LINGER) {
672 			if ((so->so_state & SS_ISDISCONNECTING) &&
673 			    (so->so_state & SS_NBIO))
674 				goto drop;
675 			while (so->so_state & SS_ISCONNECTED) {
676 				error = tsleep(&so->so_timeo,
677 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
678 				if (error)
679 					break;
680 			}
681 		}
682 	}
683 
684 drop:
685 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
686 		(*so->so_proto->pr_usrreqs->pru_close)(so);
687 	if (so->so_options & SO_ACCEPTCONN) {
688 		struct socket *sp;
689 		ACCEPT_LOCK();
690 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
691 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
692 			so->so_incqlen--;
693 			sp->so_qstate &= ~SQ_INCOMP;
694 			sp->so_head = NULL;
695 			ACCEPT_UNLOCK();
696 			soabort(sp);
697 			ACCEPT_LOCK();
698 		}
699 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
700 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
701 			so->so_qlen--;
702 			sp->so_qstate &= ~SQ_COMP;
703 			sp->so_head = NULL;
704 			ACCEPT_UNLOCK();
705 			soabort(sp);
706 			ACCEPT_LOCK();
707 		}
708 		ACCEPT_UNLOCK();
709 	}
710 	ACCEPT_LOCK();
711 	SOCK_LOCK(so);
712 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
713 	so->so_state |= SS_NOFDREF;
714 	sorele(so);
715 	CURVNET_RESTORE();
716 	return (error);
717 }
718 
719 /*
720  * soabort() is used to abruptly tear down a connection, such as when a
721  * resource limit is reached (listen queue depth exceeded), or if a listen
722  * socket is closed while there are sockets waiting to be accepted.
723  *
724  * This interface is tricky, because it is called on an unreferenced socket,
725  * and must be called only by a thread that has actually removed the socket
726  * from the listen queue it was on, or races with other threads are risked.
727  *
728  * This interface will call into the protocol code, so must not be called
729  * with any socket locks held.  Protocols do call it while holding their own
730  * recursible protocol mutexes, but this is something that should be subject
731  * to review in the future.
732  */
733 void
734 soabort(struct socket *so)
735 {
736 
737 	/*
738 	 * In as much as is possible, assert that no references to this
739 	 * socket are held.  This is not quite the same as asserting that the
740 	 * current thread is responsible for arranging for no references, but
741 	 * is as close as we can get for now.
742 	 */
743 	KASSERT(so->so_count == 0, ("soabort: so_count"));
744 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
745 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
746 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
747 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
748 
749 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
750 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
751 	ACCEPT_LOCK();
752 	SOCK_LOCK(so);
753 	sofree(so);
754 }
755 
756 int
757 soaccept(struct socket *so, struct sockaddr **nam)
758 {
759 	int error;
760 
761 	SOCK_LOCK(so);
762 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
763 	so->so_state &= ~SS_NOFDREF;
764 	SOCK_UNLOCK(so);
765 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
766 	return (error);
767 }
768 
769 int
770 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
771 {
772 	int error;
773 
774 	if (so->so_options & SO_ACCEPTCONN)
775 		return (EOPNOTSUPP);
776 
777 	CURVNET_SET(so->so_vnet);
778 	/*
779 	 * If protocol is connection-based, can only connect once.
780 	 * Otherwise, if connected, try to disconnect first.  This allows
781 	 * user to disconnect by connecting to, e.g., a null address.
782 	 */
783 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
784 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
785 	    (error = sodisconnect(so)))) {
786 		error = EISCONN;
787 	} else {
788 		/*
789 		 * Prevent accumulated error from previous connection from
790 		 * biting us.
791 		 */
792 		so->so_error = 0;
793 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
794 	}
795 	CURVNET_RESTORE();
796 
797 	return (error);
798 }
799 
800 int
801 soconnect2(struct socket *so1, struct socket *so2)
802 {
803 
804 	return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
805 }
806 
807 int
808 sodisconnect(struct socket *so)
809 {
810 	int error;
811 
812 	if ((so->so_state & SS_ISCONNECTED) == 0)
813 		return (ENOTCONN);
814 	if (so->so_state & SS_ISDISCONNECTING)
815 		return (EALREADY);
816 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
817 	return (error);
818 }
819 
820 #ifdef ZERO_COPY_SOCKETS
821 struct so_zerocopy_stats{
822 	int size_ok;
823 	int align_ok;
824 	int found_ifp;
825 };
826 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
827 #include <netinet/in.h>
828 #include <net/route.h>
829 #include <netinet/in_pcb.h>
830 #include <vm/vm.h>
831 #include <vm/vm_page.h>
832 #include <vm/vm_object.h>
833 
834 /*
835  * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
836  * sosend_dgram() and sosend_generic() use m_uiotombuf().
837  *
838  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
839  * all of the data referenced by the uio.  If desired, it uses zero-copy.
840  * *space will be updated to reflect data copied in.
841  *
842  * NB: If atomic I/O is requested, the caller must already have checked that
843  * space can hold resid bytes.
844  *
845  * NB: In the event of an error, the caller may need to free the partial
846  * chain pointed to by *mpp.  The contents of both *uio and *space may be
847  * modified even in the case of an error.
848  */
849 static int
850 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
851     int flags)
852 {
853 	struct mbuf *m, **mp, *top;
854 	long len, resid;
855 	int error;
856 #ifdef ZERO_COPY_SOCKETS
857 	int cow_send;
858 #endif
859 
860 	*retmp = top = NULL;
861 	mp = &top;
862 	len = 0;
863 	resid = uio->uio_resid;
864 	error = 0;
865 	do {
866 #ifdef ZERO_COPY_SOCKETS
867 		cow_send = 0;
868 #endif /* ZERO_COPY_SOCKETS */
869 		if (resid >= MINCLSIZE) {
870 #ifdef ZERO_COPY_SOCKETS
871 			if (top == NULL) {
872 				m = m_gethdr(M_WAITOK, MT_DATA);
873 				m->m_pkthdr.len = 0;
874 				m->m_pkthdr.rcvif = NULL;
875 			} else
876 				m = m_get(M_WAITOK, MT_DATA);
877 			if (so_zero_copy_send &&
878 			    resid>=PAGE_SIZE &&
879 			    *space>=PAGE_SIZE &&
880 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
881 				so_zerocp_stats.size_ok++;
882 				so_zerocp_stats.align_ok++;
883 				cow_send = socow_setup(m, uio);
884 				len = cow_send;
885 			}
886 			if (!cow_send) {
887 				m_clget(m, M_WAITOK);
888 				len = min(min(MCLBYTES, resid), *space);
889 			}
890 #else /* ZERO_COPY_SOCKETS */
891 			if (top == NULL) {
892 				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
893 				m->m_pkthdr.len = 0;
894 				m->m_pkthdr.rcvif = NULL;
895 			} else
896 				m = m_getcl(M_WAIT, MT_DATA, 0);
897 			len = min(min(MCLBYTES, resid), *space);
898 #endif /* ZERO_COPY_SOCKETS */
899 		} else {
900 			if (top == NULL) {
901 				m = m_gethdr(M_WAIT, MT_DATA);
902 				m->m_pkthdr.len = 0;
903 				m->m_pkthdr.rcvif = NULL;
904 
905 				len = min(min(MHLEN, resid), *space);
906 				/*
907 				 * For datagram protocols, leave room
908 				 * for protocol headers in first mbuf.
909 				 */
910 				if (atomic && m && len < MHLEN)
911 					MH_ALIGN(m, len);
912 			} else {
913 				m = m_get(M_WAIT, MT_DATA);
914 				len = min(min(MLEN, resid), *space);
915 			}
916 		}
917 		if (m == NULL) {
918 			error = ENOBUFS;
919 			goto out;
920 		}
921 
922 		*space -= len;
923 #ifdef ZERO_COPY_SOCKETS
924 		if (cow_send)
925 			error = 0;
926 		else
927 #endif /* ZERO_COPY_SOCKETS */
928 		error = uiomove(mtod(m, void *), (int)len, uio);
929 		resid = uio->uio_resid;
930 		m->m_len = len;
931 		*mp = m;
932 		top->m_pkthdr.len += len;
933 		if (error)
934 			goto out;
935 		mp = &m->m_next;
936 		if (resid <= 0) {
937 			if (flags & MSG_EOR)
938 				top->m_flags |= M_EOR;
939 			break;
940 		}
941 	} while (*space > 0 && atomic);
942 out:
943 	*retmp = top;
944 	return (error);
945 }
946 #endif /*ZERO_COPY_SOCKETS*/
947 
948 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
949 
950 int
951 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
952     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
953 {
954 	long space, resid;
955 	int clen = 0, error, dontroute;
956 #ifdef ZERO_COPY_SOCKETS
957 	int atomic = sosendallatonce(so) || top;
958 #endif
959 
960 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
961 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
962 	    ("sodgram_send: !PR_ATOMIC"));
963 
964 	if (uio != NULL)
965 		resid = uio->uio_resid;
966 	else
967 		resid = top->m_pkthdr.len;
968 	/*
969 	 * In theory resid should be unsigned.  However, space must be
970 	 * signed, as it might be less than 0 if we over-committed, and we
971 	 * must use a signed comparison of space and resid.  On the other
972 	 * hand, a negative resid causes us to loop sending 0-length
973 	 * segments to the protocol.
974 	 */
975 	if (resid < 0) {
976 		error = EINVAL;
977 		goto out;
978 	}
979 
980 	dontroute =
981 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
982 	if (td != NULL)
983 		td->td_ru.ru_msgsnd++;
984 	if (control != NULL)
985 		clen = control->m_len;
986 
987 	SOCKBUF_LOCK(&so->so_snd);
988 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
989 		SOCKBUF_UNLOCK(&so->so_snd);
990 		error = EPIPE;
991 		goto out;
992 	}
993 	if (so->so_error) {
994 		error = so->so_error;
995 		so->so_error = 0;
996 		SOCKBUF_UNLOCK(&so->so_snd);
997 		goto out;
998 	}
999 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1000 		/*
1001 		 * `sendto' and `sendmsg' is allowed on a connection-based
1002 		 * socket if it supports implied connect.  Return ENOTCONN if
1003 		 * not connected and no address is supplied.
1004 		 */
1005 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1006 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1007 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1008 			    !(resid == 0 && clen != 0)) {
1009 				SOCKBUF_UNLOCK(&so->so_snd);
1010 				error = ENOTCONN;
1011 				goto out;
1012 			}
1013 		} else if (addr == NULL) {
1014 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1015 				error = ENOTCONN;
1016 			else
1017 				error = EDESTADDRREQ;
1018 			SOCKBUF_UNLOCK(&so->so_snd);
1019 			goto out;
1020 		}
1021 	}
1022 
1023 	/*
1024 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1025 	 * problem and need fixing.
1026 	 */
1027 	space = sbspace(&so->so_snd);
1028 	if (flags & MSG_OOB)
1029 		space += 1024;
1030 	space -= clen;
1031 	SOCKBUF_UNLOCK(&so->so_snd);
1032 	if (resid > space) {
1033 		error = EMSGSIZE;
1034 		goto out;
1035 	}
1036 	if (uio == NULL) {
1037 		resid = 0;
1038 		if (flags & MSG_EOR)
1039 			top->m_flags |= M_EOR;
1040 	} else {
1041 #ifdef ZERO_COPY_SOCKETS
1042 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1043 		if (error)
1044 			goto out;
1045 #else
1046 		/*
1047 		 * Copy the data from userland into a mbuf chain.
1048 		 * If no data is to be copied in, a single empty mbuf
1049 		 * is returned.
1050 		 */
1051 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1052 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1053 		if (top == NULL) {
1054 			error = EFAULT;	/* only possible error */
1055 			goto out;
1056 		}
1057 		space -= resid - uio->uio_resid;
1058 #endif
1059 		resid = uio->uio_resid;
1060 	}
1061 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1062 	/*
1063 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1064 	 * than with.
1065 	 */
1066 	if (dontroute) {
1067 		SOCK_LOCK(so);
1068 		so->so_options |= SO_DONTROUTE;
1069 		SOCK_UNLOCK(so);
1070 	}
1071 	/*
1072 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1073 	 * of date.  We could have recieved a reset packet in an interrupt or
1074 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1075 	 * probably recheck again inside the locking protection here, but
1076 	 * there are probably other places that this also happens.  We must
1077 	 * rethink this.
1078 	 */
1079 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1080 	    (flags & MSG_OOB) ? PRUS_OOB :
1081 	/*
1082 	 * If the user set MSG_EOF, the protocol understands this flag and
1083 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1084 	 */
1085 	    ((flags & MSG_EOF) &&
1086 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1087 	     (resid <= 0)) ?
1088 		PRUS_EOF :
1089 		/* If there is more to send set PRUS_MORETOCOME */
1090 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1091 		top, addr, control, td);
1092 	if (dontroute) {
1093 		SOCK_LOCK(so);
1094 		so->so_options &= ~SO_DONTROUTE;
1095 		SOCK_UNLOCK(so);
1096 	}
1097 	clen = 0;
1098 	control = NULL;
1099 	top = NULL;
1100 out:
1101 	if (top != NULL)
1102 		m_freem(top);
1103 	if (control != NULL)
1104 		m_freem(control);
1105 	return (error);
1106 }
1107 
1108 /*
1109  * Send on a socket.  If send must go all at once and message is larger than
1110  * send buffering, then hard error.  Lock against other senders.  If must go
1111  * all at once and not enough room now, then inform user that this would
1112  * block and do nothing.  Otherwise, if nonblocking, send as much as
1113  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1114  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1115  * in mbuf chain must be small enough to send all at once.
1116  *
1117  * Returns nonzero on error, timeout or signal; callers must check for short
1118  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1119  * on return.
1120  */
1121 int
1122 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1123     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1124 {
1125 	long space, resid;
1126 	int clen = 0, error, dontroute;
1127 	int atomic = sosendallatonce(so) || top;
1128 
1129 	if (uio != NULL)
1130 		resid = uio->uio_resid;
1131 	else
1132 		resid = top->m_pkthdr.len;
1133 	/*
1134 	 * In theory resid should be unsigned.  However, space must be
1135 	 * signed, as it might be less than 0 if we over-committed, and we
1136 	 * must use a signed comparison of space and resid.  On the other
1137 	 * hand, a negative resid causes us to loop sending 0-length
1138 	 * segments to the protocol.
1139 	 *
1140 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1141 	 * type sockets since that's an error.
1142 	 */
1143 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1144 		error = EINVAL;
1145 		goto out;
1146 	}
1147 
1148 	dontroute =
1149 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1150 	    (so->so_proto->pr_flags & PR_ATOMIC);
1151 	if (td != NULL)
1152 		td->td_ru.ru_msgsnd++;
1153 	if (control != NULL)
1154 		clen = control->m_len;
1155 
1156 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1157 	if (error)
1158 		goto out;
1159 
1160 restart:
1161 	do {
1162 		SOCKBUF_LOCK(&so->so_snd);
1163 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1164 			SOCKBUF_UNLOCK(&so->so_snd);
1165 			error = EPIPE;
1166 			goto release;
1167 		}
1168 		if (so->so_error) {
1169 			error = so->so_error;
1170 			so->so_error = 0;
1171 			SOCKBUF_UNLOCK(&so->so_snd);
1172 			goto release;
1173 		}
1174 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1175 			/*
1176 			 * `sendto' and `sendmsg' is allowed on a connection-
1177 			 * based socket if it supports implied connect.
1178 			 * Return ENOTCONN if not connected and no address is
1179 			 * supplied.
1180 			 */
1181 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1182 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1183 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1184 				    !(resid == 0 && clen != 0)) {
1185 					SOCKBUF_UNLOCK(&so->so_snd);
1186 					error = ENOTCONN;
1187 					goto release;
1188 				}
1189 			} else if (addr == NULL) {
1190 				SOCKBUF_UNLOCK(&so->so_snd);
1191 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1192 					error = ENOTCONN;
1193 				else
1194 					error = EDESTADDRREQ;
1195 				goto release;
1196 			}
1197 		}
1198 		space = sbspace(&so->so_snd);
1199 		if (flags & MSG_OOB)
1200 			space += 1024;
1201 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1202 		    clen > so->so_snd.sb_hiwat) {
1203 			SOCKBUF_UNLOCK(&so->so_snd);
1204 			error = EMSGSIZE;
1205 			goto release;
1206 		}
1207 		if (space < resid + clen &&
1208 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1209 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1210 				SOCKBUF_UNLOCK(&so->so_snd);
1211 				error = EWOULDBLOCK;
1212 				goto release;
1213 			}
1214 			error = sbwait(&so->so_snd);
1215 			SOCKBUF_UNLOCK(&so->so_snd);
1216 			if (error)
1217 				goto release;
1218 			goto restart;
1219 		}
1220 		SOCKBUF_UNLOCK(&so->so_snd);
1221 		space -= clen;
1222 		do {
1223 			if (uio == NULL) {
1224 				resid = 0;
1225 				if (flags & MSG_EOR)
1226 					top->m_flags |= M_EOR;
1227 			} else {
1228 #ifdef ZERO_COPY_SOCKETS
1229 				error = sosend_copyin(uio, &top, atomic,
1230 				    &space, flags);
1231 				if (error != 0)
1232 					goto release;
1233 #else
1234 				/*
1235 				 * Copy the data from userland into a mbuf
1236 				 * chain.  If no data is to be copied in,
1237 				 * a single empty mbuf is returned.
1238 				 */
1239 				top = m_uiotombuf(uio, M_WAITOK, space,
1240 				    (atomic ? max_hdr : 0),
1241 				    (atomic ? M_PKTHDR : 0) |
1242 				    ((flags & MSG_EOR) ? M_EOR : 0));
1243 				if (top == NULL) {
1244 					error = EFAULT; /* only possible error */
1245 					goto release;
1246 				}
1247 				space -= resid - uio->uio_resid;
1248 #endif
1249 				resid = uio->uio_resid;
1250 			}
1251 			if (dontroute) {
1252 				SOCK_LOCK(so);
1253 				so->so_options |= SO_DONTROUTE;
1254 				SOCK_UNLOCK(so);
1255 			}
1256 			/*
1257 			 * XXX all the SBS_CANTSENDMORE checks previously
1258 			 * done could be out of date.  We could have recieved
1259 			 * a reset packet in an interrupt or maybe we slept
1260 			 * while doing page faults in uiomove() etc.  We
1261 			 * could probably recheck again inside the locking
1262 			 * protection here, but there are probably other
1263 			 * places that this also happens.  We must rethink
1264 			 * this.
1265 			 */
1266 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1267 			    (flags & MSG_OOB) ? PRUS_OOB :
1268 			/*
1269 			 * If the user set MSG_EOF, the protocol understands
1270 			 * this flag and nothing left to send then use
1271 			 * PRU_SEND_EOF instead of PRU_SEND.
1272 			 */
1273 			    ((flags & MSG_EOF) &&
1274 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1275 			     (resid <= 0)) ?
1276 				PRUS_EOF :
1277 			/* If there is more to send set PRUS_MORETOCOME. */
1278 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1279 			    top, addr, control, td);
1280 			if (dontroute) {
1281 				SOCK_LOCK(so);
1282 				so->so_options &= ~SO_DONTROUTE;
1283 				SOCK_UNLOCK(so);
1284 			}
1285 			clen = 0;
1286 			control = NULL;
1287 			top = NULL;
1288 			if (error)
1289 				goto release;
1290 		} while (resid && space > 0);
1291 	} while (resid);
1292 
1293 release:
1294 	sbunlock(&so->so_snd);
1295 out:
1296 	if (top != NULL)
1297 		m_freem(top);
1298 	if (control != NULL)
1299 		m_freem(control);
1300 	return (error);
1301 }
1302 
1303 int
1304 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1305     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1306 {
1307 	int error;
1308 
1309 	CURVNET_SET(so->so_vnet);
1310 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1311 	    control, flags, td);
1312 	CURVNET_RESTORE();
1313 	return (error);
1314 }
1315 
1316 /*
1317  * The part of soreceive() that implements reading non-inline out-of-band
1318  * data from a socket.  For more complete comments, see soreceive(), from
1319  * which this code originated.
1320  *
1321  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1322  * unable to return an mbuf chain to the caller.
1323  */
1324 static int
1325 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1326 {
1327 	struct protosw *pr = so->so_proto;
1328 	struct mbuf *m;
1329 	int error;
1330 
1331 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1332 
1333 	m = m_get(M_WAIT, MT_DATA);
1334 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1335 	if (error)
1336 		goto bad;
1337 	do {
1338 #ifdef ZERO_COPY_SOCKETS
1339 		if (so_zero_copy_receive) {
1340 			int disposable;
1341 
1342 			if ((m->m_flags & M_EXT)
1343 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1344 				disposable = 1;
1345 			else
1346 				disposable = 0;
1347 
1348 			error = uiomoveco(mtod(m, void *),
1349 					  min(uio->uio_resid, m->m_len),
1350 					  uio, disposable);
1351 		} else
1352 #endif /* ZERO_COPY_SOCKETS */
1353 		error = uiomove(mtod(m, void *),
1354 		    (int) min(uio->uio_resid, m->m_len), uio);
1355 		m = m_free(m);
1356 	} while (uio->uio_resid && error == 0 && m);
1357 bad:
1358 	if (m != NULL)
1359 		m_freem(m);
1360 	return (error);
1361 }
1362 
1363 /*
1364  * Following replacement or removal of the first mbuf on the first mbuf chain
1365  * of a socket buffer, push necessary state changes back into the socket
1366  * buffer so that other consumers see the values consistently.  'nextrecord'
1367  * is the callers locally stored value of the original value of
1368  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1369  * NOTE: 'nextrecord' may be NULL.
1370  */
1371 static __inline void
1372 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1373 {
1374 
1375 	SOCKBUF_LOCK_ASSERT(sb);
1376 	/*
1377 	 * First, update for the new value of nextrecord.  If necessary, make
1378 	 * it the first record.
1379 	 */
1380 	if (sb->sb_mb != NULL)
1381 		sb->sb_mb->m_nextpkt = nextrecord;
1382 	else
1383 		sb->sb_mb = nextrecord;
1384 
1385         /*
1386          * Now update any dependent socket buffer fields to reflect the new
1387          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1388 	 * addition of a second clause that takes care of the case where
1389 	 * sb_mb has been updated, but remains the last record.
1390          */
1391         if (sb->sb_mb == NULL) {
1392                 sb->sb_mbtail = NULL;
1393                 sb->sb_lastrecord = NULL;
1394         } else if (sb->sb_mb->m_nextpkt == NULL)
1395                 sb->sb_lastrecord = sb->sb_mb;
1396 }
1397 
1398 
1399 /*
1400  * Implement receive operations on a socket.  We depend on the way that
1401  * records are added to the sockbuf by sbappend.  In particular, each record
1402  * (mbufs linked through m_next) must begin with an address if the protocol
1403  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1404  * data, and then zero or more mbufs of data.  In order to allow parallelism
1405  * between network receive and copying to user space, as well as avoid
1406  * sleeping with a mutex held, we release the socket buffer mutex during the
1407  * user space copy.  Although the sockbuf is locked, new data may still be
1408  * appended, and thus we must maintain consistency of the sockbuf during that
1409  * time.
1410  *
1411  * The caller may receive the data as a single mbuf chain by supplying an
1412  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1413  * the count in uio_resid.
1414  */
1415 int
1416 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1417     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1418 {
1419 	struct mbuf *m, **mp;
1420 	int flags, len, error, offset;
1421 	struct protosw *pr = so->so_proto;
1422 	struct mbuf *nextrecord;
1423 	int moff, type = 0;
1424 	int orig_resid = uio->uio_resid;
1425 
1426 	mp = mp0;
1427 	if (psa != NULL)
1428 		*psa = NULL;
1429 	if (controlp != NULL)
1430 		*controlp = NULL;
1431 	if (flagsp != NULL)
1432 		flags = *flagsp &~ MSG_EOR;
1433 	else
1434 		flags = 0;
1435 	if (flags & MSG_OOB)
1436 		return (soreceive_rcvoob(so, uio, flags));
1437 	if (mp != NULL)
1438 		*mp = NULL;
1439 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1440 	    && uio->uio_resid)
1441 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1442 
1443 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1444 	if (error)
1445 		return (error);
1446 
1447 restart:
1448 	SOCKBUF_LOCK(&so->so_rcv);
1449 	m = so->so_rcv.sb_mb;
1450 	/*
1451 	 * If we have less data than requested, block awaiting more (subject
1452 	 * to any timeout) if:
1453 	 *   1. the current count is less than the low water mark, or
1454 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1455 	 *	receive operation at once if we block (resid <= hiwat).
1456 	 *   3. MSG_DONTWAIT is not set
1457 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1458 	 * we have to do the receive in sections, and thus risk returning a
1459 	 * short count if a timeout or signal occurs after we start.
1460 	 */
1461 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1462 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1463 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1464 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1465 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1466 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1467 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1468 		    m, so->so_rcv.sb_cc));
1469 		if (so->so_error) {
1470 			if (m != NULL)
1471 				goto dontblock;
1472 			error = so->so_error;
1473 			if ((flags & MSG_PEEK) == 0)
1474 				so->so_error = 0;
1475 			SOCKBUF_UNLOCK(&so->so_rcv);
1476 			goto release;
1477 		}
1478 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1479 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1480 			if (m == NULL) {
1481 				SOCKBUF_UNLOCK(&so->so_rcv);
1482 				goto release;
1483 			} else
1484 				goto dontblock;
1485 		}
1486 		for (; m != NULL; m = m->m_next)
1487 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1488 				m = so->so_rcv.sb_mb;
1489 				goto dontblock;
1490 			}
1491 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1492 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1493 			SOCKBUF_UNLOCK(&so->so_rcv);
1494 			error = ENOTCONN;
1495 			goto release;
1496 		}
1497 		if (uio->uio_resid == 0) {
1498 			SOCKBUF_UNLOCK(&so->so_rcv);
1499 			goto release;
1500 		}
1501 		if ((so->so_state & SS_NBIO) ||
1502 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1503 			SOCKBUF_UNLOCK(&so->so_rcv);
1504 			error = EWOULDBLOCK;
1505 			goto release;
1506 		}
1507 		SBLASTRECORDCHK(&so->so_rcv);
1508 		SBLASTMBUFCHK(&so->so_rcv);
1509 		error = sbwait(&so->so_rcv);
1510 		SOCKBUF_UNLOCK(&so->so_rcv);
1511 		if (error)
1512 			goto release;
1513 		goto restart;
1514 	}
1515 dontblock:
1516 	/*
1517 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1518 	 * pointer to the next record in the socket buffer.  We must keep the
1519 	 * various socket buffer pointers and local stack versions of the
1520 	 * pointers in sync, pushing out modifications before dropping the
1521 	 * socket buffer mutex, and re-reading them when picking it up.
1522 	 *
1523 	 * Otherwise, we will race with the network stack appending new data
1524 	 * or records onto the socket buffer by using inconsistent/stale
1525 	 * versions of the field, possibly resulting in socket buffer
1526 	 * corruption.
1527 	 *
1528 	 * By holding the high-level sblock(), we prevent simultaneous
1529 	 * readers from pulling off the front of the socket buffer.
1530 	 */
1531 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1532 	if (uio->uio_td)
1533 		uio->uio_td->td_ru.ru_msgrcv++;
1534 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1535 	SBLASTRECORDCHK(&so->so_rcv);
1536 	SBLASTMBUFCHK(&so->so_rcv);
1537 	nextrecord = m->m_nextpkt;
1538 	if (pr->pr_flags & PR_ADDR) {
1539 		KASSERT(m->m_type == MT_SONAME,
1540 		    ("m->m_type == %d", m->m_type));
1541 		orig_resid = 0;
1542 		if (psa != NULL)
1543 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1544 			    M_NOWAIT);
1545 		if (flags & MSG_PEEK) {
1546 			m = m->m_next;
1547 		} else {
1548 			sbfree(&so->so_rcv, m);
1549 			so->so_rcv.sb_mb = m_free(m);
1550 			m = so->so_rcv.sb_mb;
1551 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1552 		}
1553 	}
1554 
1555 	/*
1556 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1557 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1558 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1559 	 * perform externalization (or freeing if controlp == NULL).
1560 	 */
1561 	if (m != NULL && m->m_type == MT_CONTROL) {
1562 		struct mbuf *cm = NULL, *cmn;
1563 		struct mbuf **cme = &cm;
1564 
1565 		do {
1566 			if (flags & MSG_PEEK) {
1567 				if (controlp != NULL) {
1568 					*controlp = m_copy(m, 0, m->m_len);
1569 					controlp = &(*controlp)->m_next;
1570 				}
1571 				m = m->m_next;
1572 			} else {
1573 				sbfree(&so->so_rcv, m);
1574 				so->so_rcv.sb_mb = m->m_next;
1575 				m->m_next = NULL;
1576 				*cme = m;
1577 				cme = &(*cme)->m_next;
1578 				m = so->so_rcv.sb_mb;
1579 			}
1580 		} while (m != NULL && m->m_type == MT_CONTROL);
1581 		if ((flags & MSG_PEEK) == 0)
1582 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1583 		while (cm != NULL) {
1584 			cmn = cm->m_next;
1585 			cm->m_next = NULL;
1586 			if (pr->pr_domain->dom_externalize != NULL) {
1587 				SOCKBUF_UNLOCK(&so->so_rcv);
1588 				error = (*pr->pr_domain->dom_externalize)
1589 				    (cm, controlp);
1590 				SOCKBUF_LOCK(&so->so_rcv);
1591 			} else if (controlp != NULL)
1592 				*controlp = cm;
1593 			else
1594 				m_freem(cm);
1595 			if (controlp != NULL) {
1596 				orig_resid = 0;
1597 				while (*controlp != NULL)
1598 					controlp = &(*controlp)->m_next;
1599 			}
1600 			cm = cmn;
1601 		}
1602 		if (m != NULL)
1603 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1604 		else
1605 			nextrecord = so->so_rcv.sb_mb;
1606 		orig_resid = 0;
1607 	}
1608 	if (m != NULL) {
1609 		if ((flags & MSG_PEEK) == 0) {
1610 			KASSERT(m->m_nextpkt == nextrecord,
1611 			    ("soreceive: post-control, nextrecord !sync"));
1612 			if (nextrecord == NULL) {
1613 				KASSERT(so->so_rcv.sb_mb == m,
1614 				    ("soreceive: post-control, sb_mb!=m"));
1615 				KASSERT(so->so_rcv.sb_lastrecord == m,
1616 				    ("soreceive: post-control, lastrecord!=m"));
1617 			}
1618 		}
1619 		type = m->m_type;
1620 		if (type == MT_OOBDATA)
1621 			flags |= MSG_OOB;
1622 	} else {
1623 		if ((flags & MSG_PEEK) == 0) {
1624 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1625 			    ("soreceive: sb_mb != nextrecord"));
1626 			if (so->so_rcv.sb_mb == NULL) {
1627 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1628 				    ("soreceive: sb_lastercord != NULL"));
1629 			}
1630 		}
1631 	}
1632 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1633 	SBLASTRECORDCHK(&so->so_rcv);
1634 	SBLASTMBUFCHK(&so->so_rcv);
1635 
1636 	/*
1637 	 * Now continue to read any data mbufs off of the head of the socket
1638 	 * buffer until the read request is satisfied.  Note that 'type' is
1639 	 * used to store the type of any mbuf reads that have happened so far
1640 	 * such that soreceive() can stop reading if the type changes, which
1641 	 * causes soreceive() to return only one of regular data and inline
1642 	 * out-of-band data in a single socket receive operation.
1643 	 */
1644 	moff = 0;
1645 	offset = 0;
1646 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1647 		/*
1648 		 * If the type of mbuf has changed since the last mbuf
1649 		 * examined ('type'), end the receive operation.
1650 	 	 */
1651 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1652 		if (m->m_type == MT_OOBDATA) {
1653 			if (type != MT_OOBDATA)
1654 				break;
1655 		} else if (type == MT_OOBDATA)
1656 			break;
1657 		else
1658 		    KASSERT(m->m_type == MT_DATA,
1659 			("m->m_type == %d", m->m_type));
1660 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1661 		len = uio->uio_resid;
1662 		if (so->so_oobmark && len > so->so_oobmark - offset)
1663 			len = so->so_oobmark - offset;
1664 		if (len > m->m_len - moff)
1665 			len = m->m_len - moff;
1666 		/*
1667 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1668 		 * them out via the uio, then free.  Sockbuf must be
1669 		 * consistent here (points to current mbuf, it points to next
1670 		 * record) when we drop priority; we must note any additions
1671 		 * to the sockbuf when we block interrupts again.
1672 		 */
1673 		if (mp == NULL) {
1674 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1675 			SBLASTRECORDCHK(&so->so_rcv);
1676 			SBLASTMBUFCHK(&so->so_rcv);
1677 			SOCKBUF_UNLOCK(&so->so_rcv);
1678 #ifdef ZERO_COPY_SOCKETS
1679 			if (so_zero_copy_receive) {
1680 				int disposable;
1681 
1682 				if ((m->m_flags & M_EXT)
1683 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1684 					disposable = 1;
1685 				else
1686 					disposable = 0;
1687 
1688 				error = uiomoveco(mtod(m, char *) + moff,
1689 						  (int)len, uio,
1690 						  disposable);
1691 			} else
1692 #endif /* ZERO_COPY_SOCKETS */
1693 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1694 			SOCKBUF_LOCK(&so->so_rcv);
1695 			if (error) {
1696 				/*
1697 				 * The MT_SONAME mbuf has already been removed
1698 				 * from the record, so it is necessary to
1699 				 * remove the data mbufs, if any, to preserve
1700 				 * the invariant in the case of PR_ADDR that
1701 				 * requires MT_SONAME mbufs at the head of
1702 				 * each record.
1703 				 */
1704 				if (m && pr->pr_flags & PR_ATOMIC &&
1705 				    ((flags & MSG_PEEK) == 0))
1706 					(void)sbdroprecord_locked(&so->so_rcv);
1707 				SOCKBUF_UNLOCK(&so->so_rcv);
1708 				goto release;
1709 			}
1710 		} else
1711 			uio->uio_resid -= len;
1712 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1713 		if (len == m->m_len - moff) {
1714 			if (m->m_flags & M_EOR)
1715 				flags |= MSG_EOR;
1716 			if (flags & MSG_PEEK) {
1717 				m = m->m_next;
1718 				moff = 0;
1719 			} else {
1720 				nextrecord = m->m_nextpkt;
1721 				sbfree(&so->so_rcv, m);
1722 				if (mp != NULL) {
1723 					*mp = m;
1724 					mp = &m->m_next;
1725 					so->so_rcv.sb_mb = m = m->m_next;
1726 					*mp = NULL;
1727 				} else {
1728 					so->so_rcv.sb_mb = m_free(m);
1729 					m = so->so_rcv.sb_mb;
1730 				}
1731 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1732 				SBLASTRECORDCHK(&so->so_rcv);
1733 				SBLASTMBUFCHK(&so->so_rcv);
1734 			}
1735 		} else {
1736 			if (flags & MSG_PEEK)
1737 				moff += len;
1738 			else {
1739 				if (mp != NULL) {
1740 					int copy_flag;
1741 
1742 					if (flags & MSG_DONTWAIT)
1743 						copy_flag = M_DONTWAIT;
1744 					else
1745 						copy_flag = M_WAIT;
1746 					if (copy_flag == M_WAIT)
1747 						SOCKBUF_UNLOCK(&so->so_rcv);
1748 					*mp = m_copym(m, 0, len, copy_flag);
1749 					if (copy_flag == M_WAIT)
1750 						SOCKBUF_LOCK(&so->so_rcv);
1751  					if (*mp == NULL) {
1752  						/*
1753  						 * m_copym() couldn't
1754 						 * allocate an mbuf.  Adjust
1755 						 * uio_resid back (it was
1756 						 * adjusted down by len
1757 						 * bytes, which we didn't end
1758 						 * up "copying" over).
1759  						 */
1760  						uio->uio_resid += len;
1761  						break;
1762  					}
1763 				}
1764 				m->m_data += len;
1765 				m->m_len -= len;
1766 				so->so_rcv.sb_cc -= len;
1767 			}
1768 		}
1769 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1770 		if (so->so_oobmark) {
1771 			if ((flags & MSG_PEEK) == 0) {
1772 				so->so_oobmark -= len;
1773 				if (so->so_oobmark == 0) {
1774 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1775 					break;
1776 				}
1777 			} else {
1778 				offset += len;
1779 				if (offset == so->so_oobmark)
1780 					break;
1781 			}
1782 		}
1783 		if (flags & MSG_EOR)
1784 			break;
1785 		/*
1786 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1787 		 * must not quit until "uio->uio_resid == 0" or an error
1788 		 * termination.  If a signal/timeout occurs, return with a
1789 		 * short count but without error.  Keep sockbuf locked
1790 		 * against other readers.
1791 		 */
1792 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1793 		    !sosendallatonce(so) && nextrecord == NULL) {
1794 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1795 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1796 				break;
1797 			/*
1798 			 * Notify the protocol that some data has been
1799 			 * drained before blocking.
1800 			 */
1801 			if (pr->pr_flags & PR_WANTRCVD) {
1802 				SOCKBUF_UNLOCK(&so->so_rcv);
1803 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1804 				SOCKBUF_LOCK(&so->so_rcv);
1805 			}
1806 			SBLASTRECORDCHK(&so->so_rcv);
1807 			SBLASTMBUFCHK(&so->so_rcv);
1808 			error = sbwait(&so->so_rcv);
1809 			if (error) {
1810 				SOCKBUF_UNLOCK(&so->so_rcv);
1811 				goto release;
1812 			}
1813 			m = so->so_rcv.sb_mb;
1814 			if (m != NULL)
1815 				nextrecord = m->m_nextpkt;
1816 		}
1817 	}
1818 
1819 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1820 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1821 		flags |= MSG_TRUNC;
1822 		if ((flags & MSG_PEEK) == 0)
1823 			(void) sbdroprecord_locked(&so->so_rcv);
1824 	}
1825 	if ((flags & MSG_PEEK) == 0) {
1826 		if (m == NULL) {
1827 			/*
1828 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1829 			 * part makes sure sb_lastrecord is up-to-date if
1830 			 * there is still data in the socket buffer.
1831 			 */
1832 			so->so_rcv.sb_mb = nextrecord;
1833 			if (so->so_rcv.sb_mb == NULL) {
1834 				so->so_rcv.sb_mbtail = NULL;
1835 				so->so_rcv.sb_lastrecord = NULL;
1836 			} else if (nextrecord->m_nextpkt == NULL)
1837 				so->so_rcv.sb_lastrecord = nextrecord;
1838 		}
1839 		SBLASTRECORDCHK(&so->so_rcv);
1840 		SBLASTMBUFCHK(&so->so_rcv);
1841 		/*
1842 		 * If soreceive() is being done from the socket callback,
1843 		 * then don't need to generate ACK to peer to update window,
1844 		 * since ACK will be generated on return to TCP.
1845 		 */
1846 		if (!(flags & MSG_SOCALLBCK) &&
1847 		    (pr->pr_flags & PR_WANTRCVD)) {
1848 			SOCKBUF_UNLOCK(&so->so_rcv);
1849 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1850 			SOCKBUF_LOCK(&so->so_rcv);
1851 		}
1852 	}
1853 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1854 	if (orig_resid == uio->uio_resid && orig_resid &&
1855 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1856 		SOCKBUF_UNLOCK(&so->so_rcv);
1857 		goto restart;
1858 	}
1859 	SOCKBUF_UNLOCK(&so->so_rcv);
1860 
1861 	if (flagsp != NULL)
1862 		*flagsp |= flags;
1863 release:
1864 	sbunlock(&so->so_rcv);
1865 	return (error);
1866 }
1867 
1868 /*
1869  * Optimized version of soreceive() for stream (TCP) sockets.
1870  */
1871 #ifdef TCP_SORECEIVE_STREAM
1872 int
1873 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1874     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1875 {
1876 	int len = 0, error = 0, flags, oresid;
1877 	struct sockbuf *sb;
1878 	struct mbuf *m, *n = NULL;
1879 
1880 	/* We only do stream sockets. */
1881 	if (so->so_type != SOCK_STREAM)
1882 		return (EINVAL);
1883 	if (psa != NULL)
1884 		*psa = NULL;
1885 	if (controlp != NULL)
1886 		return (EINVAL);
1887 	if (flagsp != NULL)
1888 		flags = *flagsp &~ MSG_EOR;
1889 	else
1890 		flags = 0;
1891 	if (flags & MSG_OOB)
1892 		return (soreceive_rcvoob(so, uio, flags));
1893 	if (mp0 != NULL)
1894 		*mp0 = NULL;
1895 
1896 	sb = &so->so_rcv;
1897 
1898 	/* Prevent other readers from entering the socket. */
1899 	error = sblock(sb, SBLOCKWAIT(flags));
1900 	if (error)
1901 		goto out;
1902 	SOCKBUF_LOCK(sb);
1903 
1904 	/* Easy one, no space to copyout anything. */
1905 	if (uio->uio_resid == 0) {
1906 		error = EINVAL;
1907 		goto out;
1908 	}
1909 	oresid = uio->uio_resid;
1910 
1911 	/* We will never ever get anything unless we are connected. */
1912 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1913 		/* When disconnecting there may be still some data left. */
1914 		if (sb->sb_cc > 0)
1915 			goto deliver;
1916 		if (!(so->so_state & SS_ISDISCONNECTED))
1917 			error = ENOTCONN;
1918 		goto out;
1919 	}
1920 
1921 	/* Socket buffer is empty and we shall not block. */
1922 	if (sb->sb_cc == 0 &&
1923 	    ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1924 		error = EAGAIN;
1925 		goto out;
1926 	}
1927 
1928 restart:
1929 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1930 
1931 	/* Abort if socket has reported problems. */
1932 	if (so->so_error) {
1933 		if (sb->sb_cc > 0)
1934 			goto deliver;
1935 		if (oresid > uio->uio_resid)
1936 			goto out;
1937 		error = so->so_error;
1938 		if (!(flags & MSG_PEEK))
1939 			so->so_error = 0;
1940 		goto out;
1941 	}
1942 
1943 	/* Door is closed.  Deliver what is left, if any. */
1944 	if (sb->sb_state & SBS_CANTRCVMORE) {
1945 		if (sb->sb_cc > 0)
1946 			goto deliver;
1947 		else
1948 			goto out;
1949 	}
1950 
1951 	/* Socket buffer got some data that we shall deliver now. */
1952 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1953 	    ((sb->sb_flags & SS_NBIO) ||
1954 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1955 	     sb->sb_cc >= sb->sb_lowat ||
1956 	     sb->sb_cc >= uio->uio_resid ||
1957 	     sb->sb_cc >= sb->sb_hiwat) ) {
1958 		goto deliver;
1959 	}
1960 
1961 	/* On MSG_WAITALL we must wait until all data or error arrives. */
1962 	if ((flags & MSG_WAITALL) &&
1963 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
1964 		goto deliver;
1965 
1966 	/*
1967 	 * Wait and block until (more) data comes in.
1968 	 * NB: Drops the sockbuf lock during wait.
1969 	 */
1970 	error = sbwait(sb);
1971 	if (error)
1972 		goto out;
1973 	goto restart;
1974 
1975 deliver:
1976 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1977 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1978 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1979 
1980 	/* Statistics. */
1981 	if (uio->uio_td)
1982 		uio->uio_td->td_ru.ru_msgrcv++;
1983 
1984 	/* Fill uio until full or current end of socket buffer is reached. */
1985 	len = min(uio->uio_resid, sb->sb_cc);
1986 	if (mp0 != NULL) {
1987 		/* Dequeue as many mbufs as possible. */
1988 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1989 			for (*mp0 = m = sb->sb_mb;
1990 			     m != NULL && m->m_len <= len;
1991 			     m = m->m_next) {
1992 				len -= m->m_len;
1993 				uio->uio_resid -= m->m_len;
1994 				sbfree(sb, m);
1995 				n = m;
1996 			}
1997 			sb->sb_mb = m;
1998 			if (sb->sb_mb == NULL)
1999 				SB_EMPTY_FIXUP(sb);
2000 			n->m_next = NULL;
2001 		}
2002 		/* Copy the remainder. */
2003 		if (len > 0) {
2004 			KASSERT(sb->sb_mb != NULL,
2005 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2006 
2007 			m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2008 			if (m == NULL)
2009 				len = 0;	/* Don't flush data from sockbuf. */
2010 			else
2011 				uio->uio_resid -= m->m_len;
2012 			if (*mp0 != NULL)
2013 				n->m_next = m;
2014 			else
2015 				*mp0 = m;
2016 			if (*mp0 == NULL) {
2017 				error = ENOBUFS;
2018 				goto out;
2019 			}
2020 		}
2021 	} else {
2022 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2023 		SOCKBUF_UNLOCK(sb);
2024 		error = m_mbuftouio(uio, sb->sb_mb, len);
2025 		SOCKBUF_LOCK(sb);
2026 		if (error)
2027 			goto out;
2028 	}
2029 	SBLASTRECORDCHK(sb);
2030 	SBLASTMBUFCHK(sb);
2031 
2032 	/*
2033 	 * Remove the delivered data from the socket buffer unless we
2034 	 * were only peeking.
2035 	 */
2036 	if (!(flags & MSG_PEEK)) {
2037 		if (len > 0)
2038 			sbdrop_locked(sb, len);
2039 
2040 		/* Notify protocol that we drained some data. */
2041 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2042 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2043 		     !(flags & MSG_SOCALLBCK))) {
2044 			SOCKBUF_UNLOCK(sb);
2045 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2046 			SOCKBUF_LOCK(sb);
2047 		}
2048 	}
2049 
2050 	/*
2051 	 * For MSG_WAITALL we may have to loop again and wait for
2052 	 * more data to come in.
2053 	 */
2054 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2055 		goto restart;
2056 out:
2057 	SOCKBUF_LOCK_ASSERT(sb);
2058 	SBLASTRECORDCHK(sb);
2059 	SBLASTMBUFCHK(sb);
2060 	SOCKBUF_UNLOCK(sb);
2061 	sbunlock(sb);
2062 	return (error);
2063 }
2064 #endif /* TCP_SORECEIVE_STREAM */
2065 
2066 /*
2067  * Optimized version of soreceive() for simple datagram cases from userspace.
2068  * Unlike in the stream case, we're able to drop a datagram if copyout()
2069  * fails, and because we handle datagrams atomically, we don't need to use a
2070  * sleep lock to prevent I/O interlacing.
2071  */
2072 int
2073 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2074     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2075 {
2076 	struct mbuf *m, *m2;
2077 	int flags, len, error;
2078 	struct protosw *pr = so->so_proto;
2079 	struct mbuf *nextrecord;
2080 
2081 	if (psa != NULL)
2082 		*psa = NULL;
2083 	if (controlp != NULL)
2084 		*controlp = NULL;
2085 	if (flagsp != NULL)
2086 		flags = *flagsp &~ MSG_EOR;
2087 	else
2088 		flags = 0;
2089 
2090 	/*
2091 	 * For any complicated cases, fall back to the full
2092 	 * soreceive_generic().
2093 	 */
2094 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2095 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2096 		    flagsp));
2097 
2098 	/*
2099 	 * Enforce restrictions on use.
2100 	 */
2101 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2102 	    ("soreceive_dgram: wantrcvd"));
2103 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2104 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2105 	    ("soreceive_dgram: SBS_RCVATMARK"));
2106 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2107 	    ("soreceive_dgram: P_CONNREQUIRED"));
2108 
2109 	/*
2110 	 * Loop blocking while waiting for a datagram.
2111 	 */
2112 	SOCKBUF_LOCK(&so->so_rcv);
2113 	while ((m = so->so_rcv.sb_mb) == NULL) {
2114 		KASSERT(so->so_rcv.sb_cc == 0,
2115 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2116 		    so->so_rcv.sb_cc));
2117 		if (so->so_error) {
2118 			error = so->so_error;
2119 			so->so_error = 0;
2120 			SOCKBUF_UNLOCK(&so->so_rcv);
2121 			return (error);
2122 		}
2123 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2124 		    uio->uio_resid == 0) {
2125 			SOCKBUF_UNLOCK(&so->so_rcv);
2126 			return (0);
2127 		}
2128 		if ((so->so_state & SS_NBIO) ||
2129 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2130 			SOCKBUF_UNLOCK(&so->so_rcv);
2131 			return (EWOULDBLOCK);
2132 		}
2133 		SBLASTRECORDCHK(&so->so_rcv);
2134 		SBLASTMBUFCHK(&so->so_rcv);
2135 		error = sbwait(&so->so_rcv);
2136 		if (error) {
2137 			SOCKBUF_UNLOCK(&so->so_rcv);
2138 			return (error);
2139 		}
2140 	}
2141 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2142 
2143 	if (uio->uio_td)
2144 		uio->uio_td->td_ru.ru_msgrcv++;
2145 	SBLASTRECORDCHK(&so->so_rcv);
2146 	SBLASTMBUFCHK(&so->so_rcv);
2147 	nextrecord = m->m_nextpkt;
2148 	if (nextrecord == NULL) {
2149 		KASSERT(so->so_rcv.sb_lastrecord == m,
2150 		    ("soreceive_dgram: lastrecord != m"));
2151 	}
2152 
2153 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2154 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2155 
2156 	/*
2157 	 * Pull 'm' and its chain off the front of the packet queue.
2158 	 */
2159 	so->so_rcv.sb_mb = NULL;
2160 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2161 
2162 	/*
2163 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2164 	 */
2165 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2166 		sbfree(&so->so_rcv, m2);
2167 
2168 	/*
2169 	 * Do a few last checks before we let go of the lock.
2170 	 */
2171 	SBLASTRECORDCHK(&so->so_rcv);
2172 	SBLASTMBUFCHK(&so->so_rcv);
2173 	SOCKBUF_UNLOCK(&so->so_rcv);
2174 
2175 	if (pr->pr_flags & PR_ADDR) {
2176 		KASSERT(m->m_type == MT_SONAME,
2177 		    ("m->m_type == %d", m->m_type));
2178 		if (psa != NULL)
2179 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2180 			    M_NOWAIT);
2181 		m = m_free(m);
2182 	}
2183 	if (m == NULL) {
2184 		/* XXXRW: Can this happen? */
2185 		return (0);
2186 	}
2187 
2188 	/*
2189 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2190 	 * queue.
2191 	 *
2192 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2193 	 * in the first mbuf chain on the socket buffer.  We call into the
2194 	 * protocol to perform externalization (or freeing if controlp ==
2195 	 * NULL).
2196 	 */
2197 	if (m->m_type == MT_CONTROL) {
2198 		struct mbuf *cm = NULL, *cmn;
2199 		struct mbuf **cme = &cm;
2200 
2201 		do {
2202 			m2 = m->m_next;
2203 			m->m_next = NULL;
2204 			*cme = m;
2205 			cme = &(*cme)->m_next;
2206 			m = m2;
2207 		} while (m != NULL && m->m_type == MT_CONTROL);
2208 		while (cm != NULL) {
2209 			cmn = cm->m_next;
2210 			cm->m_next = NULL;
2211 			if (pr->pr_domain->dom_externalize != NULL) {
2212 				error = (*pr->pr_domain->dom_externalize)
2213 				    (cm, controlp);
2214 			} else if (controlp != NULL)
2215 				*controlp = cm;
2216 			else
2217 				m_freem(cm);
2218 			if (controlp != NULL) {
2219 				while (*controlp != NULL)
2220 					controlp = &(*controlp)->m_next;
2221 			}
2222 			cm = cmn;
2223 		}
2224 	}
2225 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2226 
2227 	while (m != NULL && uio->uio_resid > 0) {
2228 		len = uio->uio_resid;
2229 		if (len > m->m_len)
2230 			len = m->m_len;
2231 		error = uiomove(mtod(m, char *), (int)len, uio);
2232 		if (error) {
2233 			m_freem(m);
2234 			return (error);
2235 		}
2236 		m = m_free(m);
2237 	}
2238 	if (m != NULL)
2239 		flags |= MSG_TRUNC;
2240 	m_freem(m);
2241 	if (flagsp != NULL)
2242 		*flagsp |= flags;
2243 	return (0);
2244 }
2245 
2246 int
2247 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2248     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2249 {
2250 
2251 	return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2252 	    controlp, flagsp));
2253 }
2254 
2255 int
2256 soshutdown(struct socket *so, int how)
2257 {
2258 	struct protosw *pr = so->so_proto;
2259 	int error;
2260 
2261 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2262 		return (EINVAL);
2263 	if (pr->pr_usrreqs->pru_flush != NULL) {
2264 	        (*pr->pr_usrreqs->pru_flush)(so, how);
2265 	}
2266 	if (how != SHUT_WR)
2267 		sorflush(so);
2268 	if (how != SHUT_RD) {
2269 		CURVNET_SET(so->so_vnet);
2270 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2271 		CURVNET_RESTORE();
2272 		return (error);
2273 	}
2274 	return (0);
2275 }
2276 
2277 void
2278 sorflush(struct socket *so)
2279 {
2280 	struct sockbuf *sb = &so->so_rcv;
2281 	struct protosw *pr = so->so_proto;
2282 	struct sockbuf asb;
2283 
2284 	/*
2285 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2286 	 * held, and in order to generally avoid holding the lock for a long
2287 	 * time, we make a copy of the socket buffer and clear the original
2288 	 * (except locks, state).  The new socket buffer copy won't have
2289 	 * initialized locks so we can only call routines that won't use or
2290 	 * assert those locks.
2291 	 *
2292 	 * Dislodge threads currently blocked in receive and wait to acquire
2293 	 * a lock against other simultaneous readers before clearing the
2294 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2295 	 * despite any existing socket disposition on interruptable waiting.
2296 	 */
2297 	CURVNET_SET(so->so_vnet);
2298 	socantrcvmore(so);
2299 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2300 
2301 	/*
2302 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2303 	 * and mutex data unchanged.
2304 	 */
2305 	SOCKBUF_LOCK(sb);
2306 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2307 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2308 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2309 	bzero(&sb->sb_startzero,
2310 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2311 	SOCKBUF_UNLOCK(sb);
2312 	sbunlock(sb);
2313 
2314 	/*
2315 	 * Dispose of special rights and flush the socket buffer.  Don't call
2316 	 * any unsafe routines (that rely on locks being initialized) on asb.
2317 	 */
2318 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2319 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2320 	sbrelease_internal(&asb, so);
2321 	CURVNET_RESTORE();
2322 }
2323 
2324 /*
2325  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2326  * additional variant to handle the case where the option value needs to be
2327  * some kind of integer, but not a specific size.  In addition to their use
2328  * here, these functions are also called by the protocol-level pr_ctloutput()
2329  * routines.
2330  */
2331 int
2332 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2333 {
2334 	size_t	valsize;
2335 
2336 	/*
2337 	 * If the user gives us more than we wanted, we ignore it, but if we
2338 	 * don't get the minimum length the caller wants, we return EINVAL.
2339 	 * On success, sopt->sopt_valsize is set to however much we actually
2340 	 * retrieved.
2341 	 */
2342 	if ((valsize = sopt->sopt_valsize) < minlen)
2343 		return EINVAL;
2344 	if (valsize > len)
2345 		sopt->sopt_valsize = valsize = len;
2346 
2347 	if (sopt->sopt_td != NULL)
2348 		return (copyin(sopt->sopt_val, buf, valsize));
2349 
2350 	bcopy(sopt->sopt_val, buf, valsize);
2351 	return (0);
2352 }
2353 
2354 /*
2355  * Kernel version of setsockopt(2).
2356  *
2357  * XXX: optlen is size_t, not socklen_t
2358  */
2359 int
2360 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2361     size_t optlen)
2362 {
2363 	struct sockopt sopt;
2364 
2365 	sopt.sopt_level = level;
2366 	sopt.sopt_name = optname;
2367 	sopt.sopt_dir = SOPT_SET;
2368 	sopt.sopt_val = optval;
2369 	sopt.sopt_valsize = optlen;
2370 	sopt.sopt_td = NULL;
2371 	return (sosetopt(so, &sopt));
2372 }
2373 
2374 int
2375 sosetopt(struct socket *so, struct sockopt *sopt)
2376 {
2377 	int	error, optval;
2378 	struct	linger l;
2379 	struct	timeval tv;
2380 	u_long  val;
2381 #ifdef MAC
2382 	struct mac extmac;
2383 #endif
2384 
2385 	error = 0;
2386 	if (sopt->sopt_level != SOL_SOCKET) {
2387 		if (so->so_proto && so->so_proto->pr_ctloutput)
2388 			return ((*so->so_proto->pr_ctloutput)
2389 				  (so, sopt));
2390 		error = ENOPROTOOPT;
2391 	} else {
2392 		switch (sopt->sopt_name) {
2393 #ifdef INET
2394 		case SO_ACCEPTFILTER:
2395 			error = do_setopt_accept_filter(so, sopt);
2396 			if (error)
2397 				goto bad;
2398 			break;
2399 #endif
2400 		case SO_LINGER:
2401 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2402 			if (error)
2403 				goto bad;
2404 
2405 			SOCK_LOCK(so);
2406 			so->so_linger = l.l_linger;
2407 			if (l.l_onoff)
2408 				so->so_options |= SO_LINGER;
2409 			else
2410 				so->so_options &= ~SO_LINGER;
2411 			SOCK_UNLOCK(so);
2412 			break;
2413 
2414 		case SO_DEBUG:
2415 		case SO_KEEPALIVE:
2416 		case SO_DONTROUTE:
2417 		case SO_USELOOPBACK:
2418 		case SO_BROADCAST:
2419 		case SO_REUSEADDR:
2420 		case SO_REUSEPORT:
2421 		case SO_OOBINLINE:
2422 		case SO_TIMESTAMP:
2423 		case SO_BINTIME:
2424 		case SO_NOSIGPIPE:
2425 		case SO_NO_DDP:
2426 		case SO_NO_OFFLOAD:
2427 			error = sooptcopyin(sopt, &optval, sizeof optval,
2428 					    sizeof optval);
2429 			if (error)
2430 				goto bad;
2431 			SOCK_LOCK(so);
2432 			if (optval)
2433 				so->so_options |= sopt->sopt_name;
2434 			else
2435 				so->so_options &= ~sopt->sopt_name;
2436 			SOCK_UNLOCK(so);
2437 			break;
2438 
2439 		case SO_SETFIB:
2440 			error = sooptcopyin(sopt, &optval, sizeof optval,
2441 					    sizeof optval);
2442 			if (optval < 1 || optval > rt_numfibs) {
2443 				error = EINVAL;
2444 				goto bad;
2445 			}
2446 			if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2447 			    (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2448 				so->so_fibnum = optval;
2449 				/* Note: ignore error */
2450 				if (so->so_proto && so->so_proto->pr_ctloutput)
2451 					(*so->so_proto->pr_ctloutput)(so, sopt);
2452 			} else {
2453 				so->so_fibnum = 0;
2454 			}
2455 			break;
2456 		case SO_SNDBUF:
2457 		case SO_RCVBUF:
2458 		case SO_SNDLOWAT:
2459 		case SO_RCVLOWAT:
2460 			error = sooptcopyin(sopt, &optval, sizeof optval,
2461 					    sizeof optval);
2462 			if (error)
2463 				goto bad;
2464 
2465 			/*
2466 			 * Values < 1 make no sense for any of these options,
2467 			 * so disallow them.
2468 			 */
2469 			if (optval < 1) {
2470 				error = EINVAL;
2471 				goto bad;
2472 			}
2473 
2474 			switch (sopt->sopt_name) {
2475 			case SO_SNDBUF:
2476 			case SO_RCVBUF:
2477 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2478 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2479 				    so, curthread) == 0) {
2480 					error = ENOBUFS;
2481 					goto bad;
2482 				}
2483 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2484 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2485 				break;
2486 
2487 			/*
2488 			 * Make sure the low-water is never greater than the
2489 			 * high-water.
2490 			 */
2491 			case SO_SNDLOWAT:
2492 				SOCKBUF_LOCK(&so->so_snd);
2493 				so->so_snd.sb_lowat =
2494 				    (optval > so->so_snd.sb_hiwat) ?
2495 				    so->so_snd.sb_hiwat : optval;
2496 				SOCKBUF_UNLOCK(&so->so_snd);
2497 				break;
2498 			case SO_RCVLOWAT:
2499 				SOCKBUF_LOCK(&so->so_rcv);
2500 				so->so_rcv.sb_lowat =
2501 				    (optval > so->so_rcv.sb_hiwat) ?
2502 				    so->so_rcv.sb_hiwat : optval;
2503 				SOCKBUF_UNLOCK(&so->so_rcv);
2504 				break;
2505 			}
2506 			break;
2507 
2508 		case SO_SNDTIMEO:
2509 		case SO_RCVTIMEO:
2510 #ifdef COMPAT_IA32
2511 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2512 				struct timeval32 tv32;
2513 
2514 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2515 				    sizeof tv32);
2516 				CP(tv32, tv, tv_sec);
2517 				CP(tv32, tv, tv_usec);
2518 			} else
2519 #endif
2520 				error = sooptcopyin(sopt, &tv, sizeof tv,
2521 				    sizeof tv);
2522 			if (error)
2523 				goto bad;
2524 
2525 			/* assert(hz > 0); */
2526 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2527 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2528 				error = EDOM;
2529 				goto bad;
2530 			}
2531 			/* assert(tick > 0); */
2532 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2533 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2534 			if (val > INT_MAX) {
2535 				error = EDOM;
2536 				goto bad;
2537 			}
2538 			if (val == 0 && tv.tv_usec != 0)
2539 				val = 1;
2540 
2541 			switch (sopt->sopt_name) {
2542 			case SO_SNDTIMEO:
2543 				so->so_snd.sb_timeo = val;
2544 				break;
2545 			case SO_RCVTIMEO:
2546 				so->so_rcv.sb_timeo = val;
2547 				break;
2548 			}
2549 			break;
2550 
2551 		case SO_LABEL:
2552 #ifdef MAC
2553 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2554 			    sizeof extmac);
2555 			if (error)
2556 				goto bad;
2557 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2558 			    so, &extmac);
2559 #else
2560 			error = EOPNOTSUPP;
2561 #endif
2562 			break;
2563 
2564 		default:
2565 			error = ENOPROTOOPT;
2566 			break;
2567 		}
2568 		if (error == 0 && so->so_proto != NULL &&
2569 		    so->so_proto->pr_ctloutput != NULL) {
2570 			(void) ((*so->so_proto->pr_ctloutput)
2571 				  (so, sopt));
2572 		}
2573 	}
2574 bad:
2575 	return (error);
2576 }
2577 
2578 /*
2579  * Helper routine for getsockopt.
2580  */
2581 int
2582 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2583 {
2584 	int	error;
2585 	size_t	valsize;
2586 
2587 	error = 0;
2588 
2589 	/*
2590 	 * Documented get behavior is that we always return a value, possibly
2591 	 * truncated to fit in the user's buffer.  Traditional behavior is
2592 	 * that we always tell the user precisely how much we copied, rather
2593 	 * than something useful like the total amount we had available for
2594 	 * her.  Note that this interface is not idempotent; the entire
2595 	 * answer must generated ahead of time.
2596 	 */
2597 	valsize = min(len, sopt->sopt_valsize);
2598 	sopt->sopt_valsize = valsize;
2599 	if (sopt->sopt_val != NULL) {
2600 		if (sopt->sopt_td != NULL)
2601 			error = copyout(buf, sopt->sopt_val, valsize);
2602 		else
2603 			bcopy(buf, sopt->sopt_val, valsize);
2604 	}
2605 	return (error);
2606 }
2607 
2608 int
2609 sogetopt(struct socket *so, struct sockopt *sopt)
2610 {
2611 	int	error, optval;
2612 	struct	linger l;
2613 	struct	timeval tv;
2614 #ifdef MAC
2615 	struct mac extmac;
2616 #endif
2617 
2618 	error = 0;
2619 	if (sopt->sopt_level != SOL_SOCKET) {
2620 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2621 			return ((*so->so_proto->pr_ctloutput)
2622 				  (so, sopt));
2623 		} else
2624 			return (ENOPROTOOPT);
2625 	} else {
2626 		switch (sopt->sopt_name) {
2627 #ifdef INET
2628 		case SO_ACCEPTFILTER:
2629 			error = do_getopt_accept_filter(so, sopt);
2630 			break;
2631 #endif
2632 		case SO_LINGER:
2633 			SOCK_LOCK(so);
2634 			l.l_onoff = so->so_options & SO_LINGER;
2635 			l.l_linger = so->so_linger;
2636 			SOCK_UNLOCK(so);
2637 			error = sooptcopyout(sopt, &l, sizeof l);
2638 			break;
2639 
2640 		case SO_USELOOPBACK:
2641 		case SO_DONTROUTE:
2642 		case SO_DEBUG:
2643 		case SO_KEEPALIVE:
2644 		case SO_REUSEADDR:
2645 		case SO_REUSEPORT:
2646 		case SO_BROADCAST:
2647 		case SO_OOBINLINE:
2648 		case SO_ACCEPTCONN:
2649 		case SO_TIMESTAMP:
2650 		case SO_BINTIME:
2651 		case SO_NOSIGPIPE:
2652 			optval = so->so_options & sopt->sopt_name;
2653 integer:
2654 			error = sooptcopyout(sopt, &optval, sizeof optval);
2655 			break;
2656 
2657 		case SO_TYPE:
2658 			optval = so->so_type;
2659 			goto integer;
2660 
2661 		case SO_ERROR:
2662 			SOCK_LOCK(so);
2663 			optval = so->so_error;
2664 			so->so_error = 0;
2665 			SOCK_UNLOCK(so);
2666 			goto integer;
2667 
2668 		case SO_SNDBUF:
2669 			optval = so->so_snd.sb_hiwat;
2670 			goto integer;
2671 
2672 		case SO_RCVBUF:
2673 			optval = so->so_rcv.sb_hiwat;
2674 			goto integer;
2675 
2676 		case SO_SNDLOWAT:
2677 			optval = so->so_snd.sb_lowat;
2678 			goto integer;
2679 
2680 		case SO_RCVLOWAT:
2681 			optval = so->so_rcv.sb_lowat;
2682 			goto integer;
2683 
2684 		case SO_SNDTIMEO:
2685 		case SO_RCVTIMEO:
2686 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2687 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2688 
2689 			tv.tv_sec = optval / hz;
2690 			tv.tv_usec = (optval % hz) * tick;
2691 #ifdef COMPAT_IA32
2692 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2693 				struct timeval32 tv32;
2694 
2695 				CP(tv, tv32, tv_sec);
2696 				CP(tv, tv32, tv_usec);
2697 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2698 			} else
2699 #endif
2700 				error = sooptcopyout(sopt, &tv, sizeof tv);
2701 			break;
2702 
2703 		case SO_LABEL:
2704 #ifdef MAC
2705 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2706 			    sizeof(extmac));
2707 			if (error)
2708 				return (error);
2709 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2710 			    so, &extmac);
2711 			if (error)
2712 				return (error);
2713 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2714 #else
2715 			error = EOPNOTSUPP;
2716 #endif
2717 			break;
2718 
2719 		case SO_PEERLABEL:
2720 #ifdef MAC
2721 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2722 			    sizeof(extmac));
2723 			if (error)
2724 				return (error);
2725 			error = mac_getsockopt_peerlabel(
2726 			    sopt->sopt_td->td_ucred, so, &extmac);
2727 			if (error)
2728 				return (error);
2729 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2730 #else
2731 			error = EOPNOTSUPP;
2732 #endif
2733 			break;
2734 
2735 		case SO_LISTENQLIMIT:
2736 			optval = so->so_qlimit;
2737 			goto integer;
2738 
2739 		case SO_LISTENQLEN:
2740 			optval = so->so_qlen;
2741 			goto integer;
2742 
2743 		case SO_LISTENINCQLEN:
2744 			optval = so->so_incqlen;
2745 			goto integer;
2746 
2747 		default:
2748 			error = ENOPROTOOPT;
2749 			break;
2750 		}
2751 		return (error);
2752 	}
2753 }
2754 
2755 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2756 int
2757 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2758 {
2759 	struct mbuf *m, *m_prev;
2760 	int sopt_size = sopt->sopt_valsize;
2761 
2762 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2763 	if (m == NULL)
2764 		return ENOBUFS;
2765 	if (sopt_size > MLEN) {
2766 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2767 		if ((m->m_flags & M_EXT) == 0) {
2768 			m_free(m);
2769 			return ENOBUFS;
2770 		}
2771 		m->m_len = min(MCLBYTES, sopt_size);
2772 	} else {
2773 		m->m_len = min(MLEN, sopt_size);
2774 	}
2775 	sopt_size -= m->m_len;
2776 	*mp = m;
2777 	m_prev = m;
2778 
2779 	while (sopt_size) {
2780 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2781 		if (m == NULL) {
2782 			m_freem(*mp);
2783 			return ENOBUFS;
2784 		}
2785 		if (sopt_size > MLEN) {
2786 			MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2787 			    M_DONTWAIT);
2788 			if ((m->m_flags & M_EXT) == 0) {
2789 				m_freem(m);
2790 				m_freem(*mp);
2791 				return ENOBUFS;
2792 			}
2793 			m->m_len = min(MCLBYTES, sopt_size);
2794 		} else {
2795 			m->m_len = min(MLEN, sopt_size);
2796 		}
2797 		sopt_size -= m->m_len;
2798 		m_prev->m_next = m;
2799 		m_prev = m;
2800 	}
2801 	return (0);
2802 }
2803 
2804 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2805 int
2806 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2807 {
2808 	struct mbuf *m0 = m;
2809 
2810 	if (sopt->sopt_val == NULL)
2811 		return (0);
2812 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2813 		if (sopt->sopt_td != NULL) {
2814 			int error;
2815 
2816 			error = copyin(sopt->sopt_val, mtod(m, char *),
2817 				       m->m_len);
2818 			if (error != 0) {
2819 				m_freem(m0);
2820 				return(error);
2821 			}
2822 		} else
2823 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2824 		sopt->sopt_valsize -= m->m_len;
2825 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2826 		m = m->m_next;
2827 	}
2828 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2829 		panic("ip6_sooptmcopyin");
2830 	return (0);
2831 }
2832 
2833 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2834 int
2835 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2836 {
2837 	struct mbuf *m0 = m;
2838 	size_t valsize = 0;
2839 
2840 	if (sopt->sopt_val == NULL)
2841 		return (0);
2842 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2843 		if (sopt->sopt_td != NULL) {
2844 			int error;
2845 
2846 			error = copyout(mtod(m, char *), sopt->sopt_val,
2847 				       m->m_len);
2848 			if (error != 0) {
2849 				m_freem(m0);
2850 				return(error);
2851 			}
2852 		} else
2853 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2854 	       sopt->sopt_valsize -= m->m_len;
2855 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2856 	       valsize += m->m_len;
2857 	       m = m->m_next;
2858 	}
2859 	if (m != NULL) {
2860 		/* enough soopt buffer should be given from user-land */
2861 		m_freem(m0);
2862 		return(EINVAL);
2863 	}
2864 	sopt->sopt_valsize = valsize;
2865 	return (0);
2866 }
2867 
2868 /*
2869  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2870  * out-of-band data, which will then notify socket consumers.
2871  */
2872 void
2873 sohasoutofband(struct socket *so)
2874 {
2875 
2876 	if (so->so_sigio != NULL)
2877 		pgsigio(&so->so_sigio, SIGURG, 0);
2878 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2879 }
2880 
2881 int
2882 sopoll(struct socket *so, int events, struct ucred *active_cred,
2883     struct thread *td)
2884 {
2885 
2886 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2887 	    td));
2888 }
2889 
2890 int
2891 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2892     struct thread *td)
2893 {
2894 	int revents = 0;
2895 
2896 	SOCKBUF_LOCK(&so->so_snd);
2897 	SOCKBUF_LOCK(&so->so_rcv);
2898 	if (events & (POLLIN | POLLRDNORM))
2899 		if (soreadabledata(so))
2900 			revents |= events & (POLLIN | POLLRDNORM);
2901 
2902 	if (events & (POLLOUT | POLLWRNORM))
2903 		if (sowriteable(so))
2904 			revents |= events & (POLLOUT | POLLWRNORM);
2905 
2906 	if (events & (POLLPRI | POLLRDBAND))
2907 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2908 			revents |= events & (POLLPRI | POLLRDBAND);
2909 
2910 	if ((events & POLLINIGNEOF) == 0) {
2911 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2912 			revents |= events & (POLLIN | POLLRDNORM);
2913 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2914 				revents |= POLLHUP;
2915 		}
2916 	}
2917 
2918 	if (revents == 0) {
2919 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2920 			selrecord(td, &so->so_rcv.sb_sel);
2921 			so->so_rcv.sb_flags |= SB_SEL;
2922 		}
2923 
2924 		if (events & (POLLOUT | POLLWRNORM)) {
2925 			selrecord(td, &so->so_snd.sb_sel);
2926 			so->so_snd.sb_flags |= SB_SEL;
2927 		}
2928 	}
2929 
2930 	SOCKBUF_UNLOCK(&so->so_rcv);
2931 	SOCKBUF_UNLOCK(&so->so_snd);
2932 	return (revents);
2933 }
2934 
2935 int
2936 soo_kqfilter(struct file *fp, struct knote *kn)
2937 {
2938 	struct socket *so = kn->kn_fp->f_data;
2939 	struct sockbuf *sb;
2940 
2941 	switch (kn->kn_filter) {
2942 	case EVFILT_READ:
2943 		if (so->so_options & SO_ACCEPTCONN)
2944 			kn->kn_fop = &solisten_filtops;
2945 		else
2946 			kn->kn_fop = &soread_filtops;
2947 		sb = &so->so_rcv;
2948 		break;
2949 	case EVFILT_WRITE:
2950 		kn->kn_fop = &sowrite_filtops;
2951 		sb = &so->so_snd;
2952 		break;
2953 	default:
2954 		return (EINVAL);
2955 	}
2956 
2957 	SOCKBUF_LOCK(sb);
2958 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2959 	sb->sb_flags |= SB_KNOTE;
2960 	SOCKBUF_UNLOCK(sb);
2961 	return (0);
2962 }
2963 
2964 /*
2965  * Some routines that return EOPNOTSUPP for entry points that are not
2966  * supported by a protocol.  Fill in as needed.
2967  */
2968 int
2969 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2970 {
2971 
2972 	return EOPNOTSUPP;
2973 }
2974 
2975 int
2976 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2977 {
2978 
2979 	return EOPNOTSUPP;
2980 }
2981 
2982 int
2983 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2984 {
2985 
2986 	return EOPNOTSUPP;
2987 }
2988 
2989 int
2990 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2991 {
2992 
2993 	return EOPNOTSUPP;
2994 }
2995 
2996 int
2997 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2998 {
2999 
3000 	return EOPNOTSUPP;
3001 }
3002 
3003 int
3004 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3005     struct ifnet *ifp, struct thread *td)
3006 {
3007 
3008 	return EOPNOTSUPP;
3009 }
3010 
3011 int
3012 pru_disconnect_notsupp(struct socket *so)
3013 {
3014 
3015 	return EOPNOTSUPP;
3016 }
3017 
3018 int
3019 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3020 {
3021 
3022 	return EOPNOTSUPP;
3023 }
3024 
3025 int
3026 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3027 {
3028 
3029 	return EOPNOTSUPP;
3030 }
3031 
3032 int
3033 pru_rcvd_notsupp(struct socket *so, int flags)
3034 {
3035 
3036 	return EOPNOTSUPP;
3037 }
3038 
3039 int
3040 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3041 {
3042 
3043 	return EOPNOTSUPP;
3044 }
3045 
3046 int
3047 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3048     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3049 {
3050 
3051 	return EOPNOTSUPP;
3052 }
3053 
3054 /*
3055  * This isn't really a ``null'' operation, but it's the default one and
3056  * doesn't do anything destructive.
3057  */
3058 int
3059 pru_sense_null(struct socket *so, struct stat *sb)
3060 {
3061 
3062 	sb->st_blksize = so->so_snd.sb_hiwat;
3063 	return 0;
3064 }
3065 
3066 int
3067 pru_shutdown_notsupp(struct socket *so)
3068 {
3069 
3070 	return EOPNOTSUPP;
3071 }
3072 
3073 int
3074 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3075 {
3076 
3077 	return EOPNOTSUPP;
3078 }
3079 
3080 int
3081 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3082     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3083 {
3084 
3085 	return EOPNOTSUPP;
3086 }
3087 
3088 int
3089 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3090     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3091 {
3092 
3093 	return EOPNOTSUPP;
3094 }
3095 
3096 int
3097 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3098     struct thread *td)
3099 {
3100 
3101 	return EOPNOTSUPP;
3102 }
3103 
3104 static void
3105 filt_sordetach(struct knote *kn)
3106 {
3107 	struct socket *so = kn->kn_fp->f_data;
3108 
3109 	SOCKBUF_LOCK(&so->so_rcv);
3110 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3111 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3112 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3113 	SOCKBUF_UNLOCK(&so->so_rcv);
3114 }
3115 
3116 /*ARGSUSED*/
3117 static int
3118 filt_soread(struct knote *kn, long hint)
3119 {
3120 	struct socket *so;
3121 
3122 	so = kn->kn_fp->f_data;
3123 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3124 
3125 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3126 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3127 		kn->kn_flags |= EV_EOF;
3128 		kn->kn_fflags = so->so_error;
3129 		return (1);
3130 	} else if (so->so_error)	/* temporary udp error */
3131 		return (1);
3132 	else if (kn->kn_sfflags & NOTE_LOWAT)
3133 		return (kn->kn_data >= kn->kn_sdata);
3134 	else
3135 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3136 }
3137 
3138 static void
3139 filt_sowdetach(struct knote *kn)
3140 {
3141 	struct socket *so = kn->kn_fp->f_data;
3142 
3143 	SOCKBUF_LOCK(&so->so_snd);
3144 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3145 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3146 		so->so_snd.sb_flags &= ~SB_KNOTE;
3147 	SOCKBUF_UNLOCK(&so->so_snd);
3148 }
3149 
3150 /*ARGSUSED*/
3151 static int
3152 filt_sowrite(struct knote *kn, long hint)
3153 {
3154 	struct socket *so;
3155 
3156 	so = kn->kn_fp->f_data;
3157 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3158 	kn->kn_data = sbspace(&so->so_snd);
3159 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3160 		kn->kn_flags |= EV_EOF;
3161 		kn->kn_fflags = so->so_error;
3162 		return (1);
3163 	} else if (so->so_error)	/* temporary udp error */
3164 		return (1);
3165 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3166 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3167 		return (0);
3168 	else if (kn->kn_sfflags & NOTE_LOWAT)
3169 		return (kn->kn_data >= kn->kn_sdata);
3170 	else
3171 		return (kn->kn_data >= so->so_snd.sb_lowat);
3172 }
3173 
3174 /*ARGSUSED*/
3175 static int
3176 filt_solisten(struct knote *kn, long hint)
3177 {
3178 	struct socket *so = kn->kn_fp->f_data;
3179 
3180 	kn->kn_data = so->so_qlen;
3181 	return (! TAILQ_EMPTY(&so->so_comp));
3182 }
3183 
3184 int
3185 socheckuid(struct socket *so, uid_t uid)
3186 {
3187 
3188 	if (so == NULL)
3189 		return (EPERM);
3190 	if (so->so_cred->cr_uid != uid)
3191 		return (EPERM);
3192 	return (0);
3193 }
3194 
3195 static int
3196 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3197 {
3198 	int error;
3199 	int val;
3200 
3201 	val = somaxconn;
3202 	error = sysctl_handle_int(oidp, &val, 0, req);
3203 	if (error || !req->newptr )
3204 		return (error);
3205 
3206 	if (val < 1 || val > USHRT_MAX)
3207 		return (EINVAL);
3208 
3209 	somaxconn = val;
3210 	return (0);
3211 }
3212 
3213 /*
3214  * These functions are used by protocols to notify the socket layer (and its
3215  * consumers) of state changes in the sockets driven by protocol-side events.
3216  */
3217 
3218 /*
3219  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3220  *
3221  * Normal sequence from the active (originating) side is that
3222  * soisconnecting() is called during processing of connect() call, resulting
3223  * in an eventual call to soisconnected() if/when the connection is
3224  * established.  When the connection is torn down soisdisconnecting() is
3225  * called during processing of disconnect() call, and soisdisconnected() is
3226  * called when the connection to the peer is totally severed.  The semantics
3227  * of these routines are such that connectionless protocols can call
3228  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3229  * calls when setting up a ``connection'' takes no time.
3230  *
3231  * From the passive side, a socket is created with two queues of sockets:
3232  * so_incomp for connections in progress and so_comp for connections already
3233  * made and awaiting user acceptance.  As a protocol is preparing incoming
3234  * connections, it creates a socket structure queued on so_incomp by calling
3235  * sonewconn().  When the connection is established, soisconnected() is
3236  * called, and transfers the socket structure to so_comp, making it available
3237  * to accept().
3238  *
3239  * If a socket is closed with sockets on either so_incomp or so_comp, these
3240  * sockets are dropped.
3241  *
3242  * If higher-level protocols are implemented in the kernel, the wakeups done
3243  * here will sometimes cause software-interrupt process scheduling.
3244  */
3245 void
3246 soisconnecting(struct socket *so)
3247 {
3248 
3249 	SOCK_LOCK(so);
3250 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3251 	so->so_state |= SS_ISCONNECTING;
3252 	SOCK_UNLOCK(so);
3253 }
3254 
3255 void
3256 soisconnected(struct socket *so)
3257 {
3258 	struct socket *head;
3259 	int ret;
3260 
3261 restart:
3262 	ACCEPT_LOCK();
3263 	SOCK_LOCK(so);
3264 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3265 	so->so_state |= SS_ISCONNECTED;
3266 	head = so->so_head;
3267 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3268 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3269 			SOCK_UNLOCK(so);
3270 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3271 			head->so_incqlen--;
3272 			so->so_qstate &= ~SQ_INCOMP;
3273 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3274 			head->so_qlen++;
3275 			so->so_qstate |= SQ_COMP;
3276 			ACCEPT_UNLOCK();
3277 			sorwakeup(head);
3278 			wakeup_one(&head->so_timeo);
3279 		} else {
3280 			ACCEPT_UNLOCK();
3281 			soupcall_set(so, SO_RCV,
3282 			    head->so_accf->so_accept_filter->accf_callback,
3283 			    head->so_accf->so_accept_filter_arg);
3284 			so->so_options &= ~SO_ACCEPTFILTER;
3285 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3286 			    head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3287 			if (ret == SU_ISCONNECTED)
3288 				soupcall_clear(so, SO_RCV);
3289 			SOCK_UNLOCK(so);
3290 			if (ret == SU_ISCONNECTED)
3291 				goto restart;
3292 		}
3293 		return;
3294 	}
3295 	SOCK_UNLOCK(so);
3296 	ACCEPT_UNLOCK();
3297 	wakeup(&so->so_timeo);
3298 	sorwakeup(so);
3299 	sowwakeup(so);
3300 }
3301 
3302 void
3303 soisdisconnecting(struct socket *so)
3304 {
3305 
3306 	/*
3307 	 * Note: This code assumes that SOCK_LOCK(so) and
3308 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3309 	 */
3310 	SOCKBUF_LOCK(&so->so_rcv);
3311 	so->so_state &= ~SS_ISCONNECTING;
3312 	so->so_state |= SS_ISDISCONNECTING;
3313 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3314 	sorwakeup_locked(so);
3315 	SOCKBUF_LOCK(&so->so_snd);
3316 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3317 	sowwakeup_locked(so);
3318 	wakeup(&so->so_timeo);
3319 }
3320 
3321 void
3322 soisdisconnected(struct socket *so)
3323 {
3324 
3325 	/*
3326 	 * Note: This code assumes that SOCK_LOCK(so) and
3327 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3328 	 */
3329 	SOCKBUF_LOCK(&so->so_rcv);
3330 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3331 	so->so_state |= SS_ISDISCONNECTED;
3332 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3333 	sorwakeup_locked(so);
3334 	SOCKBUF_LOCK(&so->so_snd);
3335 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3336 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3337 	sowwakeup_locked(so);
3338 	wakeup(&so->so_timeo);
3339 }
3340 
3341 /*
3342  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3343  */
3344 struct sockaddr *
3345 sodupsockaddr(const struct sockaddr *sa, int mflags)
3346 {
3347 	struct sockaddr *sa2;
3348 
3349 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3350 	if (sa2)
3351 		bcopy(sa, sa2, sa->sa_len);
3352 	return sa2;
3353 }
3354 
3355 /*
3356  * Register per-socket buffer upcalls.
3357  */
3358 void
3359 soupcall_set(struct socket *so, int which,
3360     int (*func)(struct socket *, void *, int), void *arg)
3361 {
3362 	struct sockbuf *sb;
3363 
3364 	switch (which) {
3365 	case SO_RCV:
3366 		sb = &so->so_rcv;
3367 		break;
3368 	case SO_SND:
3369 		sb = &so->so_snd;
3370 		break;
3371 	default:
3372 		panic("soupcall_set: bad which");
3373 	}
3374 	SOCKBUF_LOCK_ASSERT(sb);
3375 #if 0
3376 	/* XXX: accf_http actually wants to do this on purpose. */
3377 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3378 #endif
3379 	sb->sb_upcall = func;
3380 	sb->sb_upcallarg = arg;
3381 	sb->sb_flags |= SB_UPCALL;
3382 }
3383 
3384 void
3385 soupcall_clear(struct socket *so, int which)
3386 {
3387 	struct sockbuf *sb;
3388 
3389 	switch (which) {
3390 	case SO_RCV:
3391 		sb = &so->so_rcv;
3392 		break;
3393 	case SO_SND:
3394 		sb = &so->so_snd;
3395 		break;
3396 	default:
3397 		panic("soupcall_clear: bad which");
3398 	}
3399 	SOCKBUF_LOCK_ASSERT(sb);
3400 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3401 	sb->sb_upcall = NULL;
3402 	sb->sb_upcallarg = NULL;
3403 	sb->sb_flags &= ~SB_UPCALL;
3404 }
3405 
3406 /*
3407  * Create an external-format (``xsocket'') structure using the information in
3408  * the kernel-format socket structure pointed to by so.  This is done to
3409  * reduce the spew of irrelevant information over this interface, to isolate
3410  * user code from changes in the kernel structure, and potentially to provide
3411  * information-hiding if we decide that some of this information should be
3412  * hidden from users.
3413  */
3414 void
3415 sotoxsocket(struct socket *so, struct xsocket *xso)
3416 {
3417 
3418 	xso->xso_len = sizeof *xso;
3419 	xso->xso_so = so;
3420 	xso->so_type = so->so_type;
3421 	xso->so_options = so->so_options;
3422 	xso->so_linger = so->so_linger;
3423 	xso->so_state = so->so_state;
3424 	xso->so_pcb = so->so_pcb;
3425 	xso->xso_protocol = so->so_proto->pr_protocol;
3426 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3427 	xso->so_qlen = so->so_qlen;
3428 	xso->so_incqlen = so->so_incqlen;
3429 	xso->so_qlimit = so->so_qlimit;
3430 	xso->so_timeo = so->so_timeo;
3431 	xso->so_error = so->so_error;
3432 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3433 	xso->so_oobmark = so->so_oobmark;
3434 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3435 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3436 	xso->so_uid = so->so_cred->cr_uid;
3437 }
3438 
3439 
3440 /*
3441  * Socket accessor functions to provide external consumers with
3442  * a safe interface to socket state
3443  *
3444  */
3445 
3446 void
3447 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3448 {
3449 
3450 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3451 		func(so, arg);
3452 }
3453 
3454 struct sockbuf *
3455 so_sockbuf_rcv(struct socket *so)
3456 {
3457 
3458 	return (&so->so_rcv);
3459 }
3460 
3461 struct sockbuf *
3462 so_sockbuf_snd(struct socket *so)
3463 {
3464 
3465 	return (&so->so_snd);
3466 }
3467 
3468 int
3469 so_state_get(const struct socket *so)
3470 {
3471 
3472 	return (so->so_state);
3473 }
3474 
3475 void
3476 so_state_set(struct socket *so, int val)
3477 {
3478 
3479 	so->so_state = val;
3480 }
3481 
3482 int
3483 so_options_get(const struct socket *so)
3484 {
3485 
3486 	return (so->so_options);
3487 }
3488 
3489 void
3490 so_options_set(struct socket *so, int val)
3491 {
3492 
3493 	so->so_options = val;
3494 }
3495 
3496 int
3497 so_error_get(const struct socket *so)
3498 {
3499 
3500 	return (so->so_error);
3501 }
3502 
3503 void
3504 so_error_set(struct socket *so, int val)
3505 {
3506 
3507 	so->so_error = val;
3508 }
3509 
3510 int
3511 so_linger_get(const struct socket *so)
3512 {
3513 
3514 	return (so->so_linger);
3515 }
3516 
3517 void
3518 so_linger_set(struct socket *so, int val)
3519 {
3520 
3521 	so->so_linger = val;
3522 }
3523 
3524 struct protosw *
3525 so_protosw_get(const struct socket *so)
3526 {
3527 
3528 	return (so->so_proto);
3529 }
3530 
3531 void
3532 so_protosw_set(struct socket *so, struct protosw *val)
3533 {
3534 
3535 	so->so_proto = val;
3536 }
3537 
3538 void
3539 so_sorwakeup(struct socket *so)
3540 {
3541 
3542 	sorwakeup(so);
3543 }
3544 
3545 void
3546 so_sowwakeup(struct socket *so)
3547 {
3548 
3549 	sowwakeup(so);
3550 }
3551 
3552 void
3553 so_sorwakeup_locked(struct socket *so)
3554 {
3555 
3556 	sorwakeup_locked(so);
3557 }
3558 
3559 void
3560 so_sowwakeup_locked(struct socket *so)
3561 {
3562 
3563 	sowwakeup_locked(so);
3564 }
3565 
3566 void
3567 so_lock(struct socket *so)
3568 {
3569 	SOCK_LOCK(so);
3570 }
3571 
3572 void
3573 so_unlock(struct socket *so)
3574 {
3575 	SOCK_UNLOCK(so);
3576 }
3577