xref: /freebsd/sys/kern/uipc_socket.c (revision aa64588d28258aef88cc33b8043112e8856948d0)
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_FREEBSD32
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 				if (error == ENOTCONN)
670 					error = 0;
671 				goto drop;
672 			}
673 		}
674 		if (so->so_options & SO_LINGER) {
675 			if ((so->so_state & SS_ISDISCONNECTING) &&
676 			    (so->so_state & SS_NBIO))
677 				goto drop;
678 			while (so->so_state & SS_ISCONNECTED) {
679 				error = tsleep(&so->so_timeo,
680 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
681 				if (error)
682 					break;
683 			}
684 		}
685 	}
686 
687 drop:
688 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
689 		(*so->so_proto->pr_usrreqs->pru_close)(so);
690 	if (so->so_options & SO_ACCEPTCONN) {
691 		struct socket *sp;
692 		ACCEPT_LOCK();
693 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
694 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
695 			so->so_incqlen--;
696 			sp->so_qstate &= ~SQ_INCOMP;
697 			sp->so_head = NULL;
698 			ACCEPT_UNLOCK();
699 			soabort(sp);
700 			ACCEPT_LOCK();
701 		}
702 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
703 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
704 			so->so_qlen--;
705 			sp->so_qstate &= ~SQ_COMP;
706 			sp->so_head = NULL;
707 			ACCEPT_UNLOCK();
708 			soabort(sp);
709 			ACCEPT_LOCK();
710 		}
711 		ACCEPT_UNLOCK();
712 	}
713 	ACCEPT_LOCK();
714 	SOCK_LOCK(so);
715 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
716 	so->so_state |= SS_NOFDREF;
717 	sorele(so);
718 	CURVNET_RESTORE();
719 	return (error);
720 }
721 
722 /*
723  * soabort() is used to abruptly tear down a connection, such as when a
724  * resource limit is reached (listen queue depth exceeded), or if a listen
725  * socket is closed while there are sockets waiting to be accepted.
726  *
727  * This interface is tricky, because it is called on an unreferenced socket,
728  * and must be called only by a thread that has actually removed the socket
729  * from the listen queue it was on, or races with other threads are risked.
730  *
731  * This interface will call into the protocol code, so must not be called
732  * with any socket locks held.  Protocols do call it while holding their own
733  * recursible protocol mutexes, but this is something that should be subject
734  * to review in the future.
735  */
736 void
737 soabort(struct socket *so)
738 {
739 
740 	/*
741 	 * In as much as is possible, assert that no references to this
742 	 * socket are held.  This is not quite the same as asserting that the
743 	 * current thread is responsible for arranging for no references, but
744 	 * is as close as we can get for now.
745 	 */
746 	KASSERT(so->so_count == 0, ("soabort: so_count"));
747 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
748 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
749 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
750 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
751 
752 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
753 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
754 	ACCEPT_LOCK();
755 	SOCK_LOCK(so);
756 	sofree(so);
757 }
758 
759 int
760 soaccept(struct socket *so, struct sockaddr **nam)
761 {
762 	int error;
763 
764 	SOCK_LOCK(so);
765 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
766 	so->so_state &= ~SS_NOFDREF;
767 	SOCK_UNLOCK(so);
768 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
769 	return (error);
770 }
771 
772 int
773 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
774 {
775 	int error;
776 
777 	if (so->so_options & SO_ACCEPTCONN)
778 		return (EOPNOTSUPP);
779 
780 	CURVNET_SET(so->so_vnet);
781 	/*
782 	 * If protocol is connection-based, can only connect once.
783 	 * Otherwise, if connected, try to disconnect first.  This allows
784 	 * user to disconnect by connecting to, e.g., a null address.
785 	 */
786 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
787 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
788 	    (error = sodisconnect(so)))) {
789 		error = EISCONN;
790 	} else {
791 		/*
792 		 * Prevent accumulated error from previous connection from
793 		 * biting us.
794 		 */
795 		so->so_error = 0;
796 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
797 	}
798 	CURVNET_RESTORE();
799 
800 	return (error);
801 }
802 
803 int
804 soconnect2(struct socket *so1, struct socket *so2)
805 {
806 
807 	return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
808 }
809 
810 int
811 sodisconnect(struct socket *so)
812 {
813 	int error;
814 
815 	if ((so->so_state & SS_ISCONNECTED) == 0)
816 		return (ENOTCONN);
817 	if (so->so_state & SS_ISDISCONNECTING)
818 		return (EALREADY);
819 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
820 	return (error);
821 }
822 
823 #ifdef ZERO_COPY_SOCKETS
824 struct so_zerocopy_stats{
825 	int size_ok;
826 	int align_ok;
827 	int found_ifp;
828 };
829 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
830 #include <netinet/in.h>
831 #include <net/route.h>
832 #include <netinet/in_pcb.h>
833 #include <vm/vm.h>
834 #include <vm/vm_page.h>
835 #include <vm/vm_object.h>
836 
837 /*
838  * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
839  * sosend_dgram() and sosend_generic() use m_uiotombuf().
840  *
841  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
842  * all of the data referenced by the uio.  If desired, it uses zero-copy.
843  * *space will be updated to reflect data copied in.
844  *
845  * NB: If atomic I/O is requested, the caller must already have checked that
846  * space can hold resid bytes.
847  *
848  * NB: In the event of an error, the caller may need to free the partial
849  * chain pointed to by *mpp.  The contents of both *uio and *space may be
850  * modified even in the case of an error.
851  */
852 static int
853 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
854     int flags)
855 {
856 	struct mbuf *m, **mp, *top;
857 	long len, resid;
858 	int error;
859 #ifdef ZERO_COPY_SOCKETS
860 	int cow_send;
861 #endif
862 
863 	*retmp = top = NULL;
864 	mp = &top;
865 	len = 0;
866 	resid = uio->uio_resid;
867 	error = 0;
868 	do {
869 #ifdef ZERO_COPY_SOCKETS
870 		cow_send = 0;
871 #endif /* ZERO_COPY_SOCKETS */
872 		if (resid >= MINCLSIZE) {
873 #ifdef ZERO_COPY_SOCKETS
874 			if (top == NULL) {
875 				m = m_gethdr(M_WAITOK, MT_DATA);
876 				m->m_pkthdr.len = 0;
877 				m->m_pkthdr.rcvif = NULL;
878 			} else
879 				m = m_get(M_WAITOK, MT_DATA);
880 			if (so_zero_copy_send &&
881 			    resid>=PAGE_SIZE &&
882 			    *space>=PAGE_SIZE &&
883 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
884 				so_zerocp_stats.size_ok++;
885 				so_zerocp_stats.align_ok++;
886 				cow_send = socow_setup(m, uio);
887 				len = cow_send;
888 			}
889 			if (!cow_send) {
890 				m_clget(m, M_WAITOK);
891 				len = min(min(MCLBYTES, resid), *space);
892 			}
893 #else /* ZERO_COPY_SOCKETS */
894 			if (top == NULL) {
895 				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
896 				m->m_pkthdr.len = 0;
897 				m->m_pkthdr.rcvif = NULL;
898 			} else
899 				m = m_getcl(M_WAIT, MT_DATA, 0);
900 			len = min(min(MCLBYTES, resid), *space);
901 #endif /* ZERO_COPY_SOCKETS */
902 		} else {
903 			if (top == NULL) {
904 				m = m_gethdr(M_WAIT, MT_DATA);
905 				m->m_pkthdr.len = 0;
906 				m->m_pkthdr.rcvif = NULL;
907 
908 				len = min(min(MHLEN, resid), *space);
909 				/*
910 				 * For datagram protocols, leave room
911 				 * for protocol headers in first mbuf.
912 				 */
913 				if (atomic && m && len < MHLEN)
914 					MH_ALIGN(m, len);
915 			} else {
916 				m = m_get(M_WAIT, MT_DATA);
917 				len = min(min(MLEN, resid), *space);
918 			}
919 		}
920 		if (m == NULL) {
921 			error = ENOBUFS;
922 			goto out;
923 		}
924 
925 		*space -= len;
926 #ifdef ZERO_COPY_SOCKETS
927 		if (cow_send)
928 			error = 0;
929 		else
930 #endif /* ZERO_COPY_SOCKETS */
931 		error = uiomove(mtod(m, void *), (int)len, uio);
932 		resid = uio->uio_resid;
933 		m->m_len = len;
934 		*mp = m;
935 		top->m_pkthdr.len += len;
936 		if (error)
937 			goto out;
938 		mp = &m->m_next;
939 		if (resid <= 0) {
940 			if (flags & MSG_EOR)
941 				top->m_flags |= M_EOR;
942 			break;
943 		}
944 	} while (*space > 0 && atomic);
945 out:
946 	*retmp = top;
947 	return (error);
948 }
949 #endif /*ZERO_COPY_SOCKETS*/
950 
951 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
952 
953 int
954 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
955     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
956 {
957 	long space, resid;
958 	int clen = 0, error, dontroute;
959 #ifdef ZERO_COPY_SOCKETS
960 	int atomic = sosendallatonce(so) || top;
961 #endif
962 
963 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
964 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
965 	    ("sodgram_send: !PR_ATOMIC"));
966 
967 	if (uio != NULL)
968 		resid = uio->uio_resid;
969 	else
970 		resid = top->m_pkthdr.len;
971 	/*
972 	 * In theory resid should be unsigned.  However, space must be
973 	 * signed, as it might be less than 0 if we over-committed, and we
974 	 * must use a signed comparison of space and resid.  On the other
975 	 * hand, a negative resid causes us to loop sending 0-length
976 	 * segments to the protocol.
977 	 */
978 	if (resid < 0) {
979 		error = EINVAL;
980 		goto out;
981 	}
982 
983 	dontroute =
984 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
985 	if (td != NULL)
986 		td->td_ru.ru_msgsnd++;
987 	if (control != NULL)
988 		clen = control->m_len;
989 
990 	SOCKBUF_LOCK(&so->so_snd);
991 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
992 		SOCKBUF_UNLOCK(&so->so_snd);
993 		error = EPIPE;
994 		goto out;
995 	}
996 	if (so->so_error) {
997 		error = so->so_error;
998 		so->so_error = 0;
999 		SOCKBUF_UNLOCK(&so->so_snd);
1000 		goto out;
1001 	}
1002 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1003 		/*
1004 		 * `sendto' and `sendmsg' is allowed on a connection-based
1005 		 * socket if it supports implied connect.  Return ENOTCONN if
1006 		 * not connected and no address is supplied.
1007 		 */
1008 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1009 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1010 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1011 			    !(resid == 0 && clen != 0)) {
1012 				SOCKBUF_UNLOCK(&so->so_snd);
1013 				error = ENOTCONN;
1014 				goto out;
1015 			}
1016 		} else if (addr == NULL) {
1017 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1018 				error = ENOTCONN;
1019 			else
1020 				error = EDESTADDRREQ;
1021 			SOCKBUF_UNLOCK(&so->so_snd);
1022 			goto out;
1023 		}
1024 	}
1025 
1026 	/*
1027 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1028 	 * problem and need fixing.
1029 	 */
1030 	space = sbspace(&so->so_snd);
1031 	if (flags & MSG_OOB)
1032 		space += 1024;
1033 	space -= clen;
1034 	SOCKBUF_UNLOCK(&so->so_snd);
1035 	if (resid > space) {
1036 		error = EMSGSIZE;
1037 		goto out;
1038 	}
1039 	if (uio == NULL) {
1040 		resid = 0;
1041 		if (flags & MSG_EOR)
1042 			top->m_flags |= M_EOR;
1043 	} else {
1044 #ifdef ZERO_COPY_SOCKETS
1045 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1046 		if (error)
1047 			goto out;
1048 #else
1049 		/*
1050 		 * Copy the data from userland into a mbuf chain.
1051 		 * If no data is to be copied in, a single empty mbuf
1052 		 * is returned.
1053 		 */
1054 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1055 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1056 		if (top == NULL) {
1057 			error = EFAULT;	/* only possible error */
1058 			goto out;
1059 		}
1060 		space -= resid - uio->uio_resid;
1061 #endif
1062 		resid = uio->uio_resid;
1063 	}
1064 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1065 	/*
1066 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1067 	 * than with.
1068 	 */
1069 	if (dontroute) {
1070 		SOCK_LOCK(so);
1071 		so->so_options |= SO_DONTROUTE;
1072 		SOCK_UNLOCK(so);
1073 	}
1074 	/*
1075 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1076 	 * of date.  We could have recieved a reset packet in an interrupt or
1077 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1078 	 * probably recheck again inside the locking protection here, but
1079 	 * there are probably other places that this also happens.  We must
1080 	 * rethink this.
1081 	 */
1082 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1083 	    (flags & MSG_OOB) ? PRUS_OOB :
1084 	/*
1085 	 * If the user set MSG_EOF, the protocol understands this flag and
1086 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1087 	 */
1088 	    ((flags & MSG_EOF) &&
1089 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1090 	     (resid <= 0)) ?
1091 		PRUS_EOF :
1092 		/* If there is more to send set PRUS_MORETOCOME */
1093 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1094 		top, addr, control, td);
1095 	if (dontroute) {
1096 		SOCK_LOCK(so);
1097 		so->so_options &= ~SO_DONTROUTE;
1098 		SOCK_UNLOCK(so);
1099 	}
1100 	clen = 0;
1101 	control = NULL;
1102 	top = NULL;
1103 out:
1104 	if (top != NULL)
1105 		m_freem(top);
1106 	if (control != NULL)
1107 		m_freem(control);
1108 	return (error);
1109 }
1110 
1111 /*
1112  * Send on a socket.  If send must go all at once and message is larger than
1113  * send buffering, then hard error.  Lock against other senders.  If must go
1114  * all at once and not enough room now, then inform user that this would
1115  * block and do nothing.  Otherwise, if nonblocking, send as much as
1116  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1117  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1118  * in mbuf chain must be small enough to send all at once.
1119  *
1120  * Returns nonzero on error, timeout or signal; callers must check for short
1121  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1122  * on return.
1123  */
1124 int
1125 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1126     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1127 {
1128 	long space, resid;
1129 	int clen = 0, error, dontroute;
1130 	int atomic = sosendallatonce(so) || top;
1131 
1132 	if (uio != NULL)
1133 		resid = uio->uio_resid;
1134 	else
1135 		resid = top->m_pkthdr.len;
1136 	/*
1137 	 * In theory resid should be unsigned.  However, space must be
1138 	 * signed, as it might be less than 0 if we over-committed, and we
1139 	 * must use a signed comparison of space and resid.  On the other
1140 	 * hand, a negative resid causes us to loop sending 0-length
1141 	 * segments to the protocol.
1142 	 *
1143 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1144 	 * type sockets since that's an error.
1145 	 */
1146 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1147 		error = EINVAL;
1148 		goto out;
1149 	}
1150 
1151 	dontroute =
1152 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1153 	    (so->so_proto->pr_flags & PR_ATOMIC);
1154 	if (td != NULL)
1155 		td->td_ru.ru_msgsnd++;
1156 	if (control != NULL)
1157 		clen = control->m_len;
1158 
1159 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1160 	if (error)
1161 		goto out;
1162 
1163 restart:
1164 	do {
1165 		SOCKBUF_LOCK(&so->so_snd);
1166 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1167 			SOCKBUF_UNLOCK(&so->so_snd);
1168 			error = EPIPE;
1169 			goto release;
1170 		}
1171 		if (so->so_error) {
1172 			error = so->so_error;
1173 			so->so_error = 0;
1174 			SOCKBUF_UNLOCK(&so->so_snd);
1175 			goto release;
1176 		}
1177 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1178 			/*
1179 			 * `sendto' and `sendmsg' is allowed on a connection-
1180 			 * based socket if it supports implied connect.
1181 			 * Return ENOTCONN if not connected and no address is
1182 			 * supplied.
1183 			 */
1184 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1185 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1186 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1187 				    !(resid == 0 && clen != 0)) {
1188 					SOCKBUF_UNLOCK(&so->so_snd);
1189 					error = ENOTCONN;
1190 					goto release;
1191 				}
1192 			} else if (addr == NULL) {
1193 				SOCKBUF_UNLOCK(&so->so_snd);
1194 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1195 					error = ENOTCONN;
1196 				else
1197 					error = EDESTADDRREQ;
1198 				goto release;
1199 			}
1200 		}
1201 		space = sbspace(&so->so_snd);
1202 		if (flags & MSG_OOB)
1203 			space += 1024;
1204 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1205 		    clen > so->so_snd.sb_hiwat) {
1206 			SOCKBUF_UNLOCK(&so->so_snd);
1207 			error = EMSGSIZE;
1208 			goto release;
1209 		}
1210 		if (space < resid + clen &&
1211 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1212 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1213 				SOCKBUF_UNLOCK(&so->so_snd);
1214 				error = EWOULDBLOCK;
1215 				goto release;
1216 			}
1217 			error = sbwait(&so->so_snd);
1218 			SOCKBUF_UNLOCK(&so->so_snd);
1219 			if (error)
1220 				goto release;
1221 			goto restart;
1222 		}
1223 		SOCKBUF_UNLOCK(&so->so_snd);
1224 		space -= clen;
1225 		do {
1226 			if (uio == NULL) {
1227 				resid = 0;
1228 				if (flags & MSG_EOR)
1229 					top->m_flags |= M_EOR;
1230 			} else {
1231 #ifdef ZERO_COPY_SOCKETS
1232 				error = sosend_copyin(uio, &top, atomic,
1233 				    &space, flags);
1234 				if (error != 0)
1235 					goto release;
1236 #else
1237 				/*
1238 				 * Copy the data from userland into a mbuf
1239 				 * chain.  If no data is to be copied in,
1240 				 * a single empty mbuf is returned.
1241 				 */
1242 				top = m_uiotombuf(uio, M_WAITOK, space,
1243 				    (atomic ? max_hdr : 0),
1244 				    (atomic ? M_PKTHDR : 0) |
1245 				    ((flags & MSG_EOR) ? M_EOR : 0));
1246 				if (top == NULL) {
1247 					error = EFAULT; /* only possible error */
1248 					goto release;
1249 				}
1250 				space -= resid - uio->uio_resid;
1251 #endif
1252 				resid = uio->uio_resid;
1253 			}
1254 			if (dontroute) {
1255 				SOCK_LOCK(so);
1256 				so->so_options |= SO_DONTROUTE;
1257 				SOCK_UNLOCK(so);
1258 			}
1259 			/*
1260 			 * XXX all the SBS_CANTSENDMORE checks previously
1261 			 * done could be out of date.  We could have recieved
1262 			 * a reset packet in an interrupt or maybe we slept
1263 			 * while doing page faults in uiomove() etc.  We
1264 			 * could probably recheck again inside the locking
1265 			 * protection here, but there are probably other
1266 			 * places that this also happens.  We must rethink
1267 			 * this.
1268 			 */
1269 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1270 			    (flags & MSG_OOB) ? PRUS_OOB :
1271 			/*
1272 			 * If the user set MSG_EOF, the protocol understands
1273 			 * this flag and nothing left to send then use
1274 			 * PRU_SEND_EOF instead of PRU_SEND.
1275 			 */
1276 			    ((flags & MSG_EOF) &&
1277 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1278 			     (resid <= 0)) ?
1279 				PRUS_EOF :
1280 			/* If there is more to send set PRUS_MORETOCOME. */
1281 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1282 			    top, addr, control, td);
1283 			if (dontroute) {
1284 				SOCK_LOCK(so);
1285 				so->so_options &= ~SO_DONTROUTE;
1286 				SOCK_UNLOCK(so);
1287 			}
1288 			clen = 0;
1289 			control = NULL;
1290 			top = NULL;
1291 			if (error)
1292 				goto release;
1293 		} while (resid && space > 0);
1294 	} while (resid);
1295 
1296 release:
1297 	sbunlock(&so->so_snd);
1298 out:
1299 	if (top != NULL)
1300 		m_freem(top);
1301 	if (control != NULL)
1302 		m_freem(control);
1303 	return (error);
1304 }
1305 
1306 int
1307 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1308     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1309 {
1310 	int error;
1311 
1312 	CURVNET_SET(so->so_vnet);
1313 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1314 	    control, flags, td);
1315 	CURVNET_RESTORE();
1316 	return (error);
1317 }
1318 
1319 /*
1320  * The part of soreceive() that implements reading non-inline out-of-band
1321  * data from a socket.  For more complete comments, see soreceive(), from
1322  * which this code originated.
1323  *
1324  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1325  * unable to return an mbuf chain to the caller.
1326  */
1327 static int
1328 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1329 {
1330 	struct protosw *pr = so->so_proto;
1331 	struct mbuf *m;
1332 	int error;
1333 
1334 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1335 
1336 	m = m_get(M_WAIT, MT_DATA);
1337 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1338 	if (error)
1339 		goto bad;
1340 	do {
1341 #ifdef ZERO_COPY_SOCKETS
1342 		if (so_zero_copy_receive) {
1343 			int disposable;
1344 
1345 			if ((m->m_flags & M_EXT)
1346 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1347 				disposable = 1;
1348 			else
1349 				disposable = 0;
1350 
1351 			error = uiomoveco(mtod(m, void *),
1352 					  min(uio->uio_resid, m->m_len),
1353 					  uio, disposable);
1354 		} else
1355 #endif /* ZERO_COPY_SOCKETS */
1356 		error = uiomove(mtod(m, void *),
1357 		    (int) min(uio->uio_resid, m->m_len), uio);
1358 		m = m_free(m);
1359 	} while (uio->uio_resid && error == 0 && m);
1360 bad:
1361 	if (m != NULL)
1362 		m_freem(m);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * Following replacement or removal of the first mbuf on the first mbuf chain
1368  * of a socket buffer, push necessary state changes back into the socket
1369  * buffer so that other consumers see the values consistently.  'nextrecord'
1370  * is the callers locally stored value of the original value of
1371  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1372  * NOTE: 'nextrecord' may be NULL.
1373  */
1374 static __inline void
1375 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1376 {
1377 
1378 	SOCKBUF_LOCK_ASSERT(sb);
1379 	/*
1380 	 * First, update for the new value of nextrecord.  If necessary, make
1381 	 * it the first record.
1382 	 */
1383 	if (sb->sb_mb != NULL)
1384 		sb->sb_mb->m_nextpkt = nextrecord;
1385 	else
1386 		sb->sb_mb = nextrecord;
1387 
1388         /*
1389          * Now update any dependent socket buffer fields to reflect the new
1390          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1391 	 * addition of a second clause that takes care of the case where
1392 	 * sb_mb has been updated, but remains the last record.
1393          */
1394         if (sb->sb_mb == NULL) {
1395                 sb->sb_mbtail = NULL;
1396                 sb->sb_lastrecord = NULL;
1397         } else if (sb->sb_mb->m_nextpkt == NULL)
1398                 sb->sb_lastrecord = sb->sb_mb;
1399 }
1400 
1401 
1402 /*
1403  * Implement receive operations on a socket.  We depend on the way that
1404  * records are added to the sockbuf by sbappend.  In particular, each record
1405  * (mbufs linked through m_next) must begin with an address if the protocol
1406  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1407  * data, and then zero or more mbufs of data.  In order to allow parallelism
1408  * between network receive and copying to user space, as well as avoid
1409  * sleeping with a mutex held, we release the socket buffer mutex during the
1410  * user space copy.  Although the sockbuf is locked, new data may still be
1411  * appended, and thus we must maintain consistency of the sockbuf during that
1412  * time.
1413  *
1414  * The caller may receive the data as a single mbuf chain by supplying an
1415  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1416  * the count in uio_resid.
1417  */
1418 int
1419 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1420     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1421 {
1422 	struct mbuf *m, **mp;
1423 	int flags, len, error, offset;
1424 	struct protosw *pr = so->so_proto;
1425 	struct mbuf *nextrecord;
1426 	int moff, type = 0;
1427 	int orig_resid = uio->uio_resid;
1428 
1429 	mp = mp0;
1430 	if (psa != NULL)
1431 		*psa = NULL;
1432 	if (controlp != NULL)
1433 		*controlp = NULL;
1434 	if (flagsp != NULL)
1435 		flags = *flagsp &~ MSG_EOR;
1436 	else
1437 		flags = 0;
1438 	if (flags & MSG_OOB)
1439 		return (soreceive_rcvoob(so, uio, flags));
1440 	if (mp != NULL)
1441 		*mp = NULL;
1442 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1443 	    && uio->uio_resid)
1444 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1445 
1446 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1447 	if (error)
1448 		return (error);
1449 
1450 restart:
1451 	SOCKBUF_LOCK(&so->so_rcv);
1452 	m = so->so_rcv.sb_mb;
1453 	/*
1454 	 * If we have less data than requested, block awaiting more (subject
1455 	 * to any timeout) if:
1456 	 *   1. the current count is less than the low water mark, or
1457 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1458 	 *	receive operation at once if we block (resid <= hiwat).
1459 	 *   3. MSG_DONTWAIT is not set
1460 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1461 	 * we have to do the receive in sections, and thus risk returning a
1462 	 * short count if a timeout or signal occurs after we start.
1463 	 */
1464 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1465 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1466 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1467 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1468 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1469 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1470 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1471 		    m, so->so_rcv.sb_cc));
1472 		if (so->so_error) {
1473 			if (m != NULL)
1474 				goto dontblock;
1475 			error = so->so_error;
1476 			if ((flags & MSG_PEEK) == 0)
1477 				so->so_error = 0;
1478 			SOCKBUF_UNLOCK(&so->so_rcv);
1479 			goto release;
1480 		}
1481 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1482 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1483 			if (m == NULL) {
1484 				SOCKBUF_UNLOCK(&so->so_rcv);
1485 				goto release;
1486 			} else
1487 				goto dontblock;
1488 		}
1489 		for (; m != NULL; m = m->m_next)
1490 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1491 				m = so->so_rcv.sb_mb;
1492 				goto dontblock;
1493 			}
1494 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1495 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1496 			SOCKBUF_UNLOCK(&so->so_rcv);
1497 			error = ENOTCONN;
1498 			goto release;
1499 		}
1500 		if (uio->uio_resid == 0) {
1501 			SOCKBUF_UNLOCK(&so->so_rcv);
1502 			goto release;
1503 		}
1504 		if ((so->so_state & SS_NBIO) ||
1505 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1506 			SOCKBUF_UNLOCK(&so->so_rcv);
1507 			error = EWOULDBLOCK;
1508 			goto release;
1509 		}
1510 		SBLASTRECORDCHK(&so->so_rcv);
1511 		SBLASTMBUFCHK(&so->so_rcv);
1512 		error = sbwait(&so->so_rcv);
1513 		SOCKBUF_UNLOCK(&so->so_rcv);
1514 		if (error)
1515 			goto release;
1516 		goto restart;
1517 	}
1518 dontblock:
1519 	/*
1520 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1521 	 * pointer to the next record in the socket buffer.  We must keep the
1522 	 * various socket buffer pointers and local stack versions of the
1523 	 * pointers in sync, pushing out modifications before dropping the
1524 	 * socket buffer mutex, and re-reading them when picking it up.
1525 	 *
1526 	 * Otherwise, we will race with the network stack appending new data
1527 	 * or records onto the socket buffer by using inconsistent/stale
1528 	 * versions of the field, possibly resulting in socket buffer
1529 	 * corruption.
1530 	 *
1531 	 * By holding the high-level sblock(), we prevent simultaneous
1532 	 * readers from pulling off the front of the socket buffer.
1533 	 */
1534 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1535 	if (uio->uio_td)
1536 		uio->uio_td->td_ru.ru_msgrcv++;
1537 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1538 	SBLASTRECORDCHK(&so->so_rcv);
1539 	SBLASTMBUFCHK(&so->so_rcv);
1540 	nextrecord = m->m_nextpkt;
1541 	if (pr->pr_flags & PR_ADDR) {
1542 		KASSERT(m->m_type == MT_SONAME,
1543 		    ("m->m_type == %d", m->m_type));
1544 		orig_resid = 0;
1545 		if (psa != NULL)
1546 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1547 			    M_NOWAIT);
1548 		if (flags & MSG_PEEK) {
1549 			m = m->m_next;
1550 		} else {
1551 			sbfree(&so->so_rcv, m);
1552 			so->so_rcv.sb_mb = m_free(m);
1553 			m = so->so_rcv.sb_mb;
1554 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1555 		}
1556 	}
1557 
1558 	/*
1559 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1560 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1561 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1562 	 * perform externalization (or freeing if controlp == NULL).
1563 	 */
1564 	if (m != NULL && m->m_type == MT_CONTROL) {
1565 		struct mbuf *cm = NULL, *cmn;
1566 		struct mbuf **cme = &cm;
1567 
1568 		do {
1569 			if (flags & MSG_PEEK) {
1570 				if (controlp != NULL) {
1571 					*controlp = m_copy(m, 0, m->m_len);
1572 					controlp = &(*controlp)->m_next;
1573 				}
1574 				m = m->m_next;
1575 			} else {
1576 				sbfree(&so->so_rcv, m);
1577 				so->so_rcv.sb_mb = m->m_next;
1578 				m->m_next = NULL;
1579 				*cme = m;
1580 				cme = &(*cme)->m_next;
1581 				m = so->so_rcv.sb_mb;
1582 			}
1583 		} while (m != NULL && m->m_type == MT_CONTROL);
1584 		if ((flags & MSG_PEEK) == 0)
1585 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1586 		while (cm != NULL) {
1587 			cmn = cm->m_next;
1588 			cm->m_next = NULL;
1589 			if (pr->pr_domain->dom_externalize != NULL) {
1590 				SOCKBUF_UNLOCK(&so->so_rcv);
1591 				error = (*pr->pr_domain->dom_externalize)
1592 				    (cm, controlp);
1593 				SOCKBUF_LOCK(&so->so_rcv);
1594 			} else if (controlp != NULL)
1595 				*controlp = cm;
1596 			else
1597 				m_freem(cm);
1598 			if (controlp != NULL) {
1599 				orig_resid = 0;
1600 				while (*controlp != NULL)
1601 					controlp = &(*controlp)->m_next;
1602 			}
1603 			cm = cmn;
1604 		}
1605 		if (m != NULL)
1606 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1607 		else
1608 			nextrecord = so->so_rcv.sb_mb;
1609 		orig_resid = 0;
1610 	}
1611 	if (m != NULL) {
1612 		if ((flags & MSG_PEEK) == 0) {
1613 			KASSERT(m->m_nextpkt == nextrecord,
1614 			    ("soreceive: post-control, nextrecord !sync"));
1615 			if (nextrecord == NULL) {
1616 				KASSERT(so->so_rcv.sb_mb == m,
1617 				    ("soreceive: post-control, sb_mb!=m"));
1618 				KASSERT(so->so_rcv.sb_lastrecord == m,
1619 				    ("soreceive: post-control, lastrecord!=m"));
1620 			}
1621 		}
1622 		type = m->m_type;
1623 		if (type == MT_OOBDATA)
1624 			flags |= MSG_OOB;
1625 	} else {
1626 		if ((flags & MSG_PEEK) == 0) {
1627 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1628 			    ("soreceive: sb_mb != nextrecord"));
1629 			if (so->so_rcv.sb_mb == NULL) {
1630 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1631 				    ("soreceive: sb_lastercord != NULL"));
1632 			}
1633 		}
1634 	}
1635 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1636 	SBLASTRECORDCHK(&so->so_rcv);
1637 	SBLASTMBUFCHK(&so->so_rcv);
1638 
1639 	/*
1640 	 * Now continue to read any data mbufs off of the head of the socket
1641 	 * buffer until the read request is satisfied.  Note that 'type' is
1642 	 * used to store the type of any mbuf reads that have happened so far
1643 	 * such that soreceive() can stop reading if the type changes, which
1644 	 * causes soreceive() to return only one of regular data and inline
1645 	 * out-of-band data in a single socket receive operation.
1646 	 */
1647 	moff = 0;
1648 	offset = 0;
1649 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1650 		/*
1651 		 * If the type of mbuf has changed since the last mbuf
1652 		 * examined ('type'), end the receive operation.
1653 	 	 */
1654 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1655 		if (m->m_type == MT_OOBDATA) {
1656 			if (type != MT_OOBDATA)
1657 				break;
1658 		} else if (type == MT_OOBDATA)
1659 			break;
1660 		else
1661 		    KASSERT(m->m_type == MT_DATA,
1662 			("m->m_type == %d", m->m_type));
1663 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1664 		len = uio->uio_resid;
1665 		if (so->so_oobmark && len > so->so_oobmark - offset)
1666 			len = so->so_oobmark - offset;
1667 		if (len > m->m_len - moff)
1668 			len = m->m_len - moff;
1669 		/*
1670 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1671 		 * them out via the uio, then free.  Sockbuf must be
1672 		 * consistent here (points to current mbuf, it points to next
1673 		 * record) when we drop priority; we must note any additions
1674 		 * to the sockbuf when we block interrupts again.
1675 		 */
1676 		if (mp == NULL) {
1677 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1678 			SBLASTRECORDCHK(&so->so_rcv);
1679 			SBLASTMBUFCHK(&so->so_rcv);
1680 			SOCKBUF_UNLOCK(&so->so_rcv);
1681 #ifdef ZERO_COPY_SOCKETS
1682 			if (so_zero_copy_receive) {
1683 				int disposable;
1684 
1685 				if ((m->m_flags & M_EXT)
1686 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1687 					disposable = 1;
1688 				else
1689 					disposable = 0;
1690 
1691 				error = uiomoveco(mtod(m, char *) + moff,
1692 						  (int)len, uio,
1693 						  disposable);
1694 			} else
1695 #endif /* ZERO_COPY_SOCKETS */
1696 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1697 			SOCKBUF_LOCK(&so->so_rcv);
1698 			if (error) {
1699 				/*
1700 				 * The MT_SONAME mbuf has already been removed
1701 				 * from the record, so it is necessary to
1702 				 * remove the data mbufs, if any, to preserve
1703 				 * the invariant in the case of PR_ADDR that
1704 				 * requires MT_SONAME mbufs at the head of
1705 				 * each record.
1706 				 */
1707 				if (m && pr->pr_flags & PR_ATOMIC &&
1708 				    ((flags & MSG_PEEK) == 0))
1709 					(void)sbdroprecord_locked(&so->so_rcv);
1710 				SOCKBUF_UNLOCK(&so->so_rcv);
1711 				goto release;
1712 			}
1713 		} else
1714 			uio->uio_resid -= len;
1715 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1716 		if (len == m->m_len - moff) {
1717 			if (m->m_flags & M_EOR)
1718 				flags |= MSG_EOR;
1719 			if (flags & MSG_PEEK) {
1720 				m = m->m_next;
1721 				moff = 0;
1722 			} else {
1723 				nextrecord = m->m_nextpkt;
1724 				sbfree(&so->so_rcv, m);
1725 				if (mp != NULL) {
1726 					*mp = m;
1727 					mp = &m->m_next;
1728 					so->so_rcv.sb_mb = m = m->m_next;
1729 					*mp = NULL;
1730 				} else {
1731 					so->so_rcv.sb_mb = m_free(m);
1732 					m = so->so_rcv.sb_mb;
1733 				}
1734 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1735 				SBLASTRECORDCHK(&so->so_rcv);
1736 				SBLASTMBUFCHK(&so->so_rcv);
1737 			}
1738 		} else {
1739 			if (flags & MSG_PEEK)
1740 				moff += len;
1741 			else {
1742 				if (mp != NULL) {
1743 					int copy_flag;
1744 
1745 					if (flags & MSG_DONTWAIT)
1746 						copy_flag = M_DONTWAIT;
1747 					else
1748 						copy_flag = M_WAIT;
1749 					if (copy_flag == M_WAIT)
1750 						SOCKBUF_UNLOCK(&so->so_rcv);
1751 					*mp = m_copym(m, 0, len, copy_flag);
1752 					if (copy_flag == M_WAIT)
1753 						SOCKBUF_LOCK(&so->so_rcv);
1754  					if (*mp == NULL) {
1755  						/*
1756  						 * m_copym() couldn't
1757 						 * allocate an mbuf.  Adjust
1758 						 * uio_resid back (it was
1759 						 * adjusted down by len
1760 						 * bytes, which we didn't end
1761 						 * up "copying" over).
1762  						 */
1763  						uio->uio_resid += len;
1764  						break;
1765  					}
1766 				}
1767 				m->m_data += len;
1768 				m->m_len -= len;
1769 				so->so_rcv.sb_cc -= len;
1770 			}
1771 		}
1772 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1773 		if (so->so_oobmark) {
1774 			if ((flags & MSG_PEEK) == 0) {
1775 				so->so_oobmark -= len;
1776 				if (so->so_oobmark == 0) {
1777 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1778 					break;
1779 				}
1780 			} else {
1781 				offset += len;
1782 				if (offset == so->so_oobmark)
1783 					break;
1784 			}
1785 		}
1786 		if (flags & MSG_EOR)
1787 			break;
1788 		/*
1789 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1790 		 * must not quit until "uio->uio_resid == 0" or an error
1791 		 * termination.  If a signal/timeout occurs, return with a
1792 		 * short count but without error.  Keep sockbuf locked
1793 		 * against other readers.
1794 		 */
1795 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1796 		    !sosendallatonce(so) && nextrecord == NULL) {
1797 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1798 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1799 				break;
1800 			/*
1801 			 * Notify the protocol that some data has been
1802 			 * drained before blocking.
1803 			 */
1804 			if (pr->pr_flags & PR_WANTRCVD) {
1805 				SOCKBUF_UNLOCK(&so->so_rcv);
1806 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1807 				SOCKBUF_LOCK(&so->so_rcv);
1808 			}
1809 			SBLASTRECORDCHK(&so->so_rcv);
1810 			SBLASTMBUFCHK(&so->so_rcv);
1811 			error = sbwait(&so->so_rcv);
1812 			if (error) {
1813 				SOCKBUF_UNLOCK(&so->so_rcv);
1814 				goto release;
1815 			}
1816 			m = so->so_rcv.sb_mb;
1817 			if (m != NULL)
1818 				nextrecord = m->m_nextpkt;
1819 		}
1820 	}
1821 
1822 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1823 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1824 		flags |= MSG_TRUNC;
1825 		if ((flags & MSG_PEEK) == 0)
1826 			(void) sbdroprecord_locked(&so->so_rcv);
1827 	}
1828 	if ((flags & MSG_PEEK) == 0) {
1829 		if (m == NULL) {
1830 			/*
1831 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1832 			 * part makes sure sb_lastrecord is up-to-date if
1833 			 * there is still data in the socket buffer.
1834 			 */
1835 			so->so_rcv.sb_mb = nextrecord;
1836 			if (so->so_rcv.sb_mb == NULL) {
1837 				so->so_rcv.sb_mbtail = NULL;
1838 				so->so_rcv.sb_lastrecord = NULL;
1839 			} else if (nextrecord->m_nextpkt == NULL)
1840 				so->so_rcv.sb_lastrecord = nextrecord;
1841 		}
1842 		SBLASTRECORDCHK(&so->so_rcv);
1843 		SBLASTMBUFCHK(&so->so_rcv);
1844 		/*
1845 		 * If soreceive() is being done from the socket callback,
1846 		 * then don't need to generate ACK to peer to update window,
1847 		 * since ACK will be generated on return to TCP.
1848 		 */
1849 		if (!(flags & MSG_SOCALLBCK) &&
1850 		    (pr->pr_flags & PR_WANTRCVD)) {
1851 			SOCKBUF_UNLOCK(&so->so_rcv);
1852 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1853 			SOCKBUF_LOCK(&so->so_rcv);
1854 		}
1855 	}
1856 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1857 	if (orig_resid == uio->uio_resid && orig_resid &&
1858 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1859 		SOCKBUF_UNLOCK(&so->so_rcv);
1860 		goto restart;
1861 	}
1862 	SOCKBUF_UNLOCK(&so->so_rcv);
1863 
1864 	if (flagsp != NULL)
1865 		*flagsp |= flags;
1866 release:
1867 	sbunlock(&so->so_rcv);
1868 	return (error);
1869 }
1870 
1871 /*
1872  * Optimized version of soreceive() for stream (TCP) sockets.
1873  */
1874 #ifdef TCP_SORECEIVE_STREAM
1875 int
1876 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1877     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1878 {
1879 	int len = 0, error = 0, flags, oresid;
1880 	struct sockbuf *sb;
1881 	struct mbuf *m, *n = NULL;
1882 
1883 	/* We only do stream sockets. */
1884 	if (so->so_type != SOCK_STREAM)
1885 		return (EINVAL);
1886 	if (psa != NULL)
1887 		*psa = NULL;
1888 	if (controlp != NULL)
1889 		return (EINVAL);
1890 	if (flagsp != NULL)
1891 		flags = *flagsp &~ MSG_EOR;
1892 	else
1893 		flags = 0;
1894 	if (flags & MSG_OOB)
1895 		return (soreceive_rcvoob(so, uio, flags));
1896 	if (mp0 != NULL)
1897 		*mp0 = NULL;
1898 
1899 	sb = &so->so_rcv;
1900 
1901 	/* Prevent other readers from entering the socket. */
1902 	error = sblock(sb, SBLOCKWAIT(flags));
1903 	if (error)
1904 		goto out;
1905 	SOCKBUF_LOCK(sb);
1906 
1907 	/* Easy one, no space to copyout anything. */
1908 	if (uio->uio_resid == 0) {
1909 		error = EINVAL;
1910 		goto out;
1911 	}
1912 	oresid = uio->uio_resid;
1913 
1914 	/* We will never ever get anything unless we are connected. */
1915 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1916 		/* When disconnecting there may be still some data left. */
1917 		if (sb->sb_cc > 0)
1918 			goto deliver;
1919 		if (!(so->so_state & SS_ISDISCONNECTED))
1920 			error = ENOTCONN;
1921 		goto out;
1922 	}
1923 
1924 	/* Socket buffer is empty and we shall not block. */
1925 	if (sb->sb_cc == 0 &&
1926 	    ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1927 		error = EAGAIN;
1928 		goto out;
1929 	}
1930 
1931 restart:
1932 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1933 
1934 	/* Abort if socket has reported problems. */
1935 	if (so->so_error) {
1936 		if (sb->sb_cc > 0)
1937 			goto deliver;
1938 		if (oresid > uio->uio_resid)
1939 			goto out;
1940 		error = so->so_error;
1941 		if (!(flags & MSG_PEEK))
1942 			so->so_error = 0;
1943 		goto out;
1944 	}
1945 
1946 	/* Door is closed.  Deliver what is left, if any. */
1947 	if (sb->sb_state & SBS_CANTRCVMORE) {
1948 		if (sb->sb_cc > 0)
1949 			goto deliver;
1950 		else
1951 			goto out;
1952 	}
1953 
1954 	/* Socket buffer got some data that we shall deliver now. */
1955 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1956 	    ((sb->sb_flags & SS_NBIO) ||
1957 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1958 	     sb->sb_cc >= sb->sb_lowat ||
1959 	     sb->sb_cc >= uio->uio_resid ||
1960 	     sb->sb_cc >= sb->sb_hiwat) ) {
1961 		goto deliver;
1962 	}
1963 
1964 	/* On MSG_WAITALL we must wait until all data or error arrives. */
1965 	if ((flags & MSG_WAITALL) &&
1966 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
1967 		goto deliver;
1968 
1969 	/*
1970 	 * Wait and block until (more) data comes in.
1971 	 * NB: Drops the sockbuf lock during wait.
1972 	 */
1973 	error = sbwait(sb);
1974 	if (error)
1975 		goto out;
1976 	goto restart;
1977 
1978 deliver:
1979 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1980 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1981 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1982 
1983 	/* Statistics. */
1984 	if (uio->uio_td)
1985 		uio->uio_td->td_ru.ru_msgrcv++;
1986 
1987 	/* Fill uio until full or current end of socket buffer is reached. */
1988 	len = min(uio->uio_resid, sb->sb_cc);
1989 	if (mp0 != NULL) {
1990 		/* Dequeue as many mbufs as possible. */
1991 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1992 			for (*mp0 = m = sb->sb_mb;
1993 			     m != NULL && m->m_len <= len;
1994 			     m = m->m_next) {
1995 				len -= m->m_len;
1996 				uio->uio_resid -= m->m_len;
1997 				sbfree(sb, m);
1998 				n = m;
1999 			}
2000 			sb->sb_mb = m;
2001 			if (sb->sb_mb == NULL)
2002 				SB_EMPTY_FIXUP(sb);
2003 			n->m_next = NULL;
2004 		}
2005 		/* Copy the remainder. */
2006 		if (len > 0) {
2007 			KASSERT(sb->sb_mb != NULL,
2008 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2009 
2010 			m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2011 			if (m == NULL)
2012 				len = 0;	/* Don't flush data from sockbuf. */
2013 			else
2014 				uio->uio_resid -= m->m_len;
2015 			if (*mp0 != NULL)
2016 				n->m_next = m;
2017 			else
2018 				*mp0 = m;
2019 			if (*mp0 == NULL) {
2020 				error = ENOBUFS;
2021 				goto out;
2022 			}
2023 		}
2024 	} else {
2025 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2026 		SOCKBUF_UNLOCK(sb);
2027 		error = m_mbuftouio(uio, sb->sb_mb, len);
2028 		SOCKBUF_LOCK(sb);
2029 		if (error)
2030 			goto out;
2031 	}
2032 	SBLASTRECORDCHK(sb);
2033 	SBLASTMBUFCHK(sb);
2034 
2035 	/*
2036 	 * Remove the delivered data from the socket buffer unless we
2037 	 * were only peeking.
2038 	 */
2039 	if (!(flags & MSG_PEEK)) {
2040 		if (len > 0)
2041 			sbdrop_locked(sb, len);
2042 
2043 		/* Notify protocol that we drained some data. */
2044 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2045 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2046 		     !(flags & MSG_SOCALLBCK))) {
2047 			SOCKBUF_UNLOCK(sb);
2048 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2049 			SOCKBUF_LOCK(sb);
2050 		}
2051 	}
2052 
2053 	/*
2054 	 * For MSG_WAITALL we may have to loop again and wait for
2055 	 * more data to come in.
2056 	 */
2057 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2058 		goto restart;
2059 out:
2060 	SOCKBUF_LOCK_ASSERT(sb);
2061 	SBLASTRECORDCHK(sb);
2062 	SBLASTMBUFCHK(sb);
2063 	SOCKBUF_UNLOCK(sb);
2064 	sbunlock(sb);
2065 	return (error);
2066 }
2067 #endif /* TCP_SORECEIVE_STREAM */
2068 
2069 /*
2070  * Optimized version of soreceive() for simple datagram cases from userspace.
2071  * Unlike in the stream case, we're able to drop a datagram if copyout()
2072  * fails, and because we handle datagrams atomically, we don't need to use a
2073  * sleep lock to prevent I/O interlacing.
2074  */
2075 int
2076 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2077     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2078 {
2079 	struct mbuf *m, *m2;
2080 	int flags, len, error;
2081 	struct protosw *pr = so->so_proto;
2082 	struct mbuf *nextrecord;
2083 
2084 	if (psa != NULL)
2085 		*psa = NULL;
2086 	if (controlp != NULL)
2087 		*controlp = NULL;
2088 	if (flagsp != NULL)
2089 		flags = *flagsp &~ MSG_EOR;
2090 	else
2091 		flags = 0;
2092 
2093 	/*
2094 	 * For any complicated cases, fall back to the full
2095 	 * soreceive_generic().
2096 	 */
2097 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2098 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2099 		    flagsp));
2100 
2101 	/*
2102 	 * Enforce restrictions on use.
2103 	 */
2104 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2105 	    ("soreceive_dgram: wantrcvd"));
2106 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2107 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2108 	    ("soreceive_dgram: SBS_RCVATMARK"));
2109 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2110 	    ("soreceive_dgram: P_CONNREQUIRED"));
2111 
2112 	/*
2113 	 * Loop blocking while waiting for a datagram.
2114 	 */
2115 	SOCKBUF_LOCK(&so->so_rcv);
2116 	while ((m = so->so_rcv.sb_mb) == NULL) {
2117 		KASSERT(so->so_rcv.sb_cc == 0,
2118 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2119 		    so->so_rcv.sb_cc));
2120 		if (so->so_error) {
2121 			error = so->so_error;
2122 			so->so_error = 0;
2123 			SOCKBUF_UNLOCK(&so->so_rcv);
2124 			return (error);
2125 		}
2126 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2127 		    uio->uio_resid == 0) {
2128 			SOCKBUF_UNLOCK(&so->so_rcv);
2129 			return (0);
2130 		}
2131 		if ((so->so_state & SS_NBIO) ||
2132 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2133 			SOCKBUF_UNLOCK(&so->so_rcv);
2134 			return (EWOULDBLOCK);
2135 		}
2136 		SBLASTRECORDCHK(&so->so_rcv);
2137 		SBLASTMBUFCHK(&so->so_rcv);
2138 		error = sbwait(&so->so_rcv);
2139 		if (error) {
2140 			SOCKBUF_UNLOCK(&so->so_rcv);
2141 			return (error);
2142 		}
2143 	}
2144 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2145 
2146 	if (uio->uio_td)
2147 		uio->uio_td->td_ru.ru_msgrcv++;
2148 	SBLASTRECORDCHK(&so->so_rcv);
2149 	SBLASTMBUFCHK(&so->so_rcv);
2150 	nextrecord = m->m_nextpkt;
2151 	if (nextrecord == NULL) {
2152 		KASSERT(so->so_rcv.sb_lastrecord == m,
2153 		    ("soreceive_dgram: lastrecord != m"));
2154 	}
2155 
2156 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2157 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2158 
2159 	/*
2160 	 * Pull 'm' and its chain off the front of the packet queue.
2161 	 */
2162 	so->so_rcv.sb_mb = NULL;
2163 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2164 
2165 	/*
2166 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2167 	 */
2168 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2169 		sbfree(&so->so_rcv, m2);
2170 
2171 	/*
2172 	 * Do a few last checks before we let go of the lock.
2173 	 */
2174 	SBLASTRECORDCHK(&so->so_rcv);
2175 	SBLASTMBUFCHK(&so->so_rcv);
2176 	SOCKBUF_UNLOCK(&so->so_rcv);
2177 
2178 	if (pr->pr_flags & PR_ADDR) {
2179 		KASSERT(m->m_type == MT_SONAME,
2180 		    ("m->m_type == %d", m->m_type));
2181 		if (psa != NULL)
2182 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2183 			    M_NOWAIT);
2184 		m = m_free(m);
2185 	}
2186 	if (m == NULL) {
2187 		/* XXXRW: Can this happen? */
2188 		return (0);
2189 	}
2190 
2191 	/*
2192 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2193 	 * queue.
2194 	 *
2195 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2196 	 * in the first mbuf chain on the socket buffer.  We call into the
2197 	 * protocol to perform externalization (or freeing if controlp ==
2198 	 * NULL).
2199 	 */
2200 	if (m->m_type == MT_CONTROL) {
2201 		struct mbuf *cm = NULL, *cmn;
2202 		struct mbuf **cme = &cm;
2203 
2204 		do {
2205 			m2 = m->m_next;
2206 			m->m_next = NULL;
2207 			*cme = m;
2208 			cme = &(*cme)->m_next;
2209 			m = m2;
2210 		} while (m != NULL && m->m_type == MT_CONTROL);
2211 		while (cm != NULL) {
2212 			cmn = cm->m_next;
2213 			cm->m_next = NULL;
2214 			if (pr->pr_domain->dom_externalize != NULL) {
2215 				error = (*pr->pr_domain->dom_externalize)
2216 				    (cm, controlp);
2217 			} else if (controlp != NULL)
2218 				*controlp = cm;
2219 			else
2220 				m_freem(cm);
2221 			if (controlp != NULL) {
2222 				while (*controlp != NULL)
2223 					controlp = &(*controlp)->m_next;
2224 			}
2225 			cm = cmn;
2226 		}
2227 	}
2228 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2229 
2230 	while (m != NULL && uio->uio_resid > 0) {
2231 		len = uio->uio_resid;
2232 		if (len > m->m_len)
2233 			len = m->m_len;
2234 		error = uiomove(mtod(m, char *), (int)len, uio);
2235 		if (error) {
2236 			m_freem(m);
2237 			return (error);
2238 		}
2239 		m = m_free(m);
2240 	}
2241 	if (m != NULL)
2242 		flags |= MSG_TRUNC;
2243 	m_freem(m);
2244 	if (flagsp != NULL)
2245 		*flagsp |= flags;
2246 	return (0);
2247 }
2248 
2249 int
2250 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2251     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2252 {
2253 
2254 	return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2255 	    controlp, flagsp));
2256 }
2257 
2258 int
2259 soshutdown(struct socket *so, int how)
2260 {
2261 	struct protosw *pr = so->so_proto;
2262 	int error;
2263 
2264 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2265 		return (EINVAL);
2266 	if (pr->pr_usrreqs->pru_flush != NULL) {
2267 	        (*pr->pr_usrreqs->pru_flush)(so, how);
2268 	}
2269 	if (how != SHUT_WR)
2270 		sorflush(so);
2271 	if (how != SHUT_RD) {
2272 		CURVNET_SET(so->so_vnet);
2273 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2274 		CURVNET_RESTORE();
2275 		return (error);
2276 	}
2277 	return (0);
2278 }
2279 
2280 void
2281 sorflush(struct socket *so)
2282 {
2283 	struct sockbuf *sb = &so->so_rcv;
2284 	struct protosw *pr = so->so_proto;
2285 	struct sockbuf asb;
2286 
2287 	/*
2288 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2289 	 * held, and in order to generally avoid holding the lock for a long
2290 	 * time, we make a copy of the socket buffer and clear the original
2291 	 * (except locks, state).  The new socket buffer copy won't have
2292 	 * initialized locks so we can only call routines that won't use or
2293 	 * assert those locks.
2294 	 *
2295 	 * Dislodge threads currently blocked in receive and wait to acquire
2296 	 * a lock against other simultaneous readers before clearing the
2297 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2298 	 * despite any existing socket disposition on interruptable waiting.
2299 	 */
2300 	CURVNET_SET(so->so_vnet);
2301 	socantrcvmore(so);
2302 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2303 
2304 	/*
2305 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2306 	 * and mutex data unchanged.
2307 	 */
2308 	SOCKBUF_LOCK(sb);
2309 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2310 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2311 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2312 	bzero(&sb->sb_startzero,
2313 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2314 	SOCKBUF_UNLOCK(sb);
2315 	sbunlock(sb);
2316 
2317 	/*
2318 	 * Dispose of special rights and flush the socket buffer.  Don't call
2319 	 * any unsafe routines (that rely on locks being initialized) on asb.
2320 	 */
2321 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2322 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2323 	sbrelease_internal(&asb, so);
2324 	CURVNET_RESTORE();
2325 }
2326 
2327 /*
2328  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2329  * additional variant to handle the case where the option value needs to be
2330  * some kind of integer, but not a specific size.  In addition to their use
2331  * here, these functions are also called by the protocol-level pr_ctloutput()
2332  * routines.
2333  */
2334 int
2335 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2336 {
2337 	size_t	valsize;
2338 
2339 	/*
2340 	 * If the user gives us more than we wanted, we ignore it, but if we
2341 	 * don't get the minimum length the caller wants, we return EINVAL.
2342 	 * On success, sopt->sopt_valsize is set to however much we actually
2343 	 * retrieved.
2344 	 */
2345 	if ((valsize = sopt->sopt_valsize) < minlen)
2346 		return EINVAL;
2347 	if (valsize > len)
2348 		sopt->sopt_valsize = valsize = len;
2349 
2350 	if (sopt->sopt_td != NULL)
2351 		return (copyin(sopt->sopt_val, buf, valsize));
2352 
2353 	bcopy(sopt->sopt_val, buf, valsize);
2354 	return (0);
2355 }
2356 
2357 /*
2358  * Kernel version of setsockopt(2).
2359  *
2360  * XXX: optlen is size_t, not socklen_t
2361  */
2362 int
2363 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2364     size_t optlen)
2365 {
2366 	struct sockopt sopt;
2367 
2368 	sopt.sopt_level = level;
2369 	sopt.sopt_name = optname;
2370 	sopt.sopt_dir = SOPT_SET;
2371 	sopt.sopt_val = optval;
2372 	sopt.sopt_valsize = optlen;
2373 	sopt.sopt_td = NULL;
2374 	return (sosetopt(so, &sopt));
2375 }
2376 
2377 int
2378 sosetopt(struct socket *so, struct sockopt *sopt)
2379 {
2380 	int	error, optval;
2381 	struct	linger l;
2382 	struct	timeval tv;
2383 	u_long  val;
2384 #ifdef MAC
2385 	struct mac extmac;
2386 #endif
2387 
2388 	error = 0;
2389 	if (sopt->sopt_level != SOL_SOCKET) {
2390 		if (so->so_proto && so->so_proto->pr_ctloutput)
2391 			return ((*so->so_proto->pr_ctloutput)
2392 				  (so, sopt));
2393 		error = ENOPROTOOPT;
2394 	} else {
2395 		switch (sopt->sopt_name) {
2396 #ifdef INET
2397 		case SO_ACCEPTFILTER:
2398 			error = do_setopt_accept_filter(so, sopt);
2399 			if (error)
2400 				goto bad;
2401 			break;
2402 #endif
2403 		case SO_LINGER:
2404 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2405 			if (error)
2406 				goto bad;
2407 
2408 			SOCK_LOCK(so);
2409 			so->so_linger = l.l_linger;
2410 			if (l.l_onoff)
2411 				so->so_options |= SO_LINGER;
2412 			else
2413 				so->so_options &= ~SO_LINGER;
2414 			SOCK_UNLOCK(so);
2415 			break;
2416 
2417 		case SO_DEBUG:
2418 		case SO_KEEPALIVE:
2419 		case SO_DONTROUTE:
2420 		case SO_USELOOPBACK:
2421 		case SO_BROADCAST:
2422 		case SO_REUSEADDR:
2423 		case SO_REUSEPORT:
2424 		case SO_OOBINLINE:
2425 		case SO_TIMESTAMP:
2426 		case SO_BINTIME:
2427 		case SO_NOSIGPIPE:
2428 		case SO_NO_DDP:
2429 		case SO_NO_OFFLOAD:
2430 			error = sooptcopyin(sopt, &optval, sizeof optval,
2431 					    sizeof optval);
2432 			if (error)
2433 				goto bad;
2434 			SOCK_LOCK(so);
2435 			if (optval)
2436 				so->so_options |= sopt->sopt_name;
2437 			else
2438 				so->so_options &= ~sopt->sopt_name;
2439 			SOCK_UNLOCK(so);
2440 			break;
2441 
2442 		case SO_SETFIB:
2443 			error = sooptcopyin(sopt, &optval, sizeof optval,
2444 					    sizeof optval);
2445 			if (optval < 1 || optval > rt_numfibs) {
2446 				error = EINVAL;
2447 				goto bad;
2448 			}
2449 			if ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2450 			    (so->so_proto->pr_domain->dom_family == PF_ROUTE)) {
2451 				so->so_fibnum = optval;
2452 				/* Note: ignore error */
2453 				if (so->so_proto && so->so_proto->pr_ctloutput)
2454 					(*so->so_proto->pr_ctloutput)(so, sopt);
2455 			} else {
2456 				so->so_fibnum = 0;
2457 			}
2458 			break;
2459 		case SO_SNDBUF:
2460 		case SO_RCVBUF:
2461 		case SO_SNDLOWAT:
2462 		case SO_RCVLOWAT:
2463 			error = sooptcopyin(sopt, &optval, sizeof optval,
2464 					    sizeof optval);
2465 			if (error)
2466 				goto bad;
2467 
2468 			/*
2469 			 * Values < 1 make no sense for any of these options,
2470 			 * so disallow them.
2471 			 */
2472 			if (optval < 1) {
2473 				error = EINVAL;
2474 				goto bad;
2475 			}
2476 
2477 			switch (sopt->sopt_name) {
2478 			case SO_SNDBUF:
2479 			case SO_RCVBUF:
2480 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2481 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2482 				    so, curthread) == 0) {
2483 					error = ENOBUFS;
2484 					goto bad;
2485 				}
2486 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2487 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2488 				break;
2489 
2490 			/*
2491 			 * Make sure the low-water is never greater than the
2492 			 * high-water.
2493 			 */
2494 			case SO_SNDLOWAT:
2495 				SOCKBUF_LOCK(&so->so_snd);
2496 				so->so_snd.sb_lowat =
2497 				    (optval > so->so_snd.sb_hiwat) ?
2498 				    so->so_snd.sb_hiwat : optval;
2499 				SOCKBUF_UNLOCK(&so->so_snd);
2500 				break;
2501 			case SO_RCVLOWAT:
2502 				SOCKBUF_LOCK(&so->so_rcv);
2503 				so->so_rcv.sb_lowat =
2504 				    (optval > so->so_rcv.sb_hiwat) ?
2505 				    so->so_rcv.sb_hiwat : optval;
2506 				SOCKBUF_UNLOCK(&so->so_rcv);
2507 				break;
2508 			}
2509 			break;
2510 
2511 		case SO_SNDTIMEO:
2512 		case SO_RCVTIMEO:
2513 #ifdef COMPAT_FREEBSD32
2514 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2515 				struct timeval32 tv32;
2516 
2517 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2518 				    sizeof tv32);
2519 				CP(tv32, tv, tv_sec);
2520 				CP(tv32, tv, tv_usec);
2521 			} else
2522 #endif
2523 				error = sooptcopyin(sopt, &tv, sizeof tv,
2524 				    sizeof tv);
2525 			if (error)
2526 				goto bad;
2527 
2528 			/* assert(hz > 0); */
2529 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2530 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2531 				error = EDOM;
2532 				goto bad;
2533 			}
2534 			/* assert(tick > 0); */
2535 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2536 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2537 			if (val > INT_MAX) {
2538 				error = EDOM;
2539 				goto bad;
2540 			}
2541 			if (val == 0 && tv.tv_usec != 0)
2542 				val = 1;
2543 
2544 			switch (sopt->sopt_name) {
2545 			case SO_SNDTIMEO:
2546 				so->so_snd.sb_timeo = val;
2547 				break;
2548 			case SO_RCVTIMEO:
2549 				so->so_rcv.sb_timeo = val;
2550 				break;
2551 			}
2552 			break;
2553 
2554 		case SO_LABEL:
2555 #ifdef MAC
2556 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2557 			    sizeof extmac);
2558 			if (error)
2559 				goto bad;
2560 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2561 			    so, &extmac);
2562 #else
2563 			error = EOPNOTSUPP;
2564 #endif
2565 			break;
2566 
2567 		default:
2568 			error = ENOPROTOOPT;
2569 			break;
2570 		}
2571 		if (error == 0 && so->so_proto != NULL &&
2572 		    so->so_proto->pr_ctloutput != NULL) {
2573 			(void) ((*so->so_proto->pr_ctloutput)
2574 				  (so, sopt));
2575 		}
2576 	}
2577 bad:
2578 	return (error);
2579 }
2580 
2581 /*
2582  * Helper routine for getsockopt.
2583  */
2584 int
2585 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2586 {
2587 	int	error;
2588 	size_t	valsize;
2589 
2590 	error = 0;
2591 
2592 	/*
2593 	 * Documented get behavior is that we always return a value, possibly
2594 	 * truncated to fit in the user's buffer.  Traditional behavior is
2595 	 * that we always tell the user precisely how much we copied, rather
2596 	 * than something useful like the total amount we had available for
2597 	 * her.  Note that this interface is not idempotent; the entire
2598 	 * answer must generated ahead of time.
2599 	 */
2600 	valsize = min(len, sopt->sopt_valsize);
2601 	sopt->sopt_valsize = valsize;
2602 	if (sopt->sopt_val != NULL) {
2603 		if (sopt->sopt_td != NULL)
2604 			error = copyout(buf, sopt->sopt_val, valsize);
2605 		else
2606 			bcopy(buf, sopt->sopt_val, valsize);
2607 	}
2608 	return (error);
2609 }
2610 
2611 int
2612 sogetopt(struct socket *so, struct sockopt *sopt)
2613 {
2614 	int	error, optval;
2615 	struct	linger l;
2616 	struct	timeval tv;
2617 #ifdef MAC
2618 	struct mac extmac;
2619 #endif
2620 
2621 	error = 0;
2622 	if (sopt->sopt_level != SOL_SOCKET) {
2623 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2624 			return ((*so->so_proto->pr_ctloutput)
2625 				  (so, sopt));
2626 		} else
2627 			return (ENOPROTOOPT);
2628 	} else {
2629 		switch (sopt->sopt_name) {
2630 #ifdef INET
2631 		case SO_ACCEPTFILTER:
2632 			error = do_getopt_accept_filter(so, sopt);
2633 			break;
2634 #endif
2635 		case SO_LINGER:
2636 			SOCK_LOCK(so);
2637 			l.l_onoff = so->so_options & SO_LINGER;
2638 			l.l_linger = so->so_linger;
2639 			SOCK_UNLOCK(so);
2640 			error = sooptcopyout(sopt, &l, sizeof l);
2641 			break;
2642 
2643 		case SO_USELOOPBACK:
2644 		case SO_DONTROUTE:
2645 		case SO_DEBUG:
2646 		case SO_KEEPALIVE:
2647 		case SO_REUSEADDR:
2648 		case SO_REUSEPORT:
2649 		case SO_BROADCAST:
2650 		case SO_OOBINLINE:
2651 		case SO_ACCEPTCONN:
2652 		case SO_TIMESTAMP:
2653 		case SO_BINTIME:
2654 		case SO_NOSIGPIPE:
2655 			optval = so->so_options & sopt->sopt_name;
2656 integer:
2657 			error = sooptcopyout(sopt, &optval, sizeof optval);
2658 			break;
2659 
2660 		case SO_TYPE:
2661 			optval = so->so_type;
2662 			goto integer;
2663 
2664 		case SO_ERROR:
2665 			SOCK_LOCK(so);
2666 			optval = so->so_error;
2667 			so->so_error = 0;
2668 			SOCK_UNLOCK(so);
2669 			goto integer;
2670 
2671 		case SO_SNDBUF:
2672 			optval = so->so_snd.sb_hiwat;
2673 			goto integer;
2674 
2675 		case SO_RCVBUF:
2676 			optval = so->so_rcv.sb_hiwat;
2677 			goto integer;
2678 
2679 		case SO_SNDLOWAT:
2680 			optval = so->so_snd.sb_lowat;
2681 			goto integer;
2682 
2683 		case SO_RCVLOWAT:
2684 			optval = so->so_rcv.sb_lowat;
2685 			goto integer;
2686 
2687 		case SO_SNDTIMEO:
2688 		case SO_RCVTIMEO:
2689 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2690 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2691 
2692 			tv.tv_sec = optval / hz;
2693 			tv.tv_usec = (optval % hz) * tick;
2694 #ifdef COMPAT_FREEBSD32
2695 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2696 				struct timeval32 tv32;
2697 
2698 				CP(tv, tv32, tv_sec);
2699 				CP(tv, tv32, tv_usec);
2700 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2701 			} else
2702 #endif
2703 				error = sooptcopyout(sopt, &tv, sizeof tv);
2704 			break;
2705 
2706 		case SO_LABEL:
2707 #ifdef MAC
2708 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2709 			    sizeof(extmac));
2710 			if (error)
2711 				return (error);
2712 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2713 			    so, &extmac);
2714 			if (error)
2715 				return (error);
2716 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2717 #else
2718 			error = EOPNOTSUPP;
2719 #endif
2720 			break;
2721 
2722 		case SO_PEERLABEL:
2723 #ifdef MAC
2724 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2725 			    sizeof(extmac));
2726 			if (error)
2727 				return (error);
2728 			error = mac_getsockopt_peerlabel(
2729 			    sopt->sopt_td->td_ucred, so, &extmac);
2730 			if (error)
2731 				return (error);
2732 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2733 #else
2734 			error = EOPNOTSUPP;
2735 #endif
2736 			break;
2737 
2738 		case SO_LISTENQLIMIT:
2739 			optval = so->so_qlimit;
2740 			goto integer;
2741 
2742 		case SO_LISTENQLEN:
2743 			optval = so->so_qlen;
2744 			goto integer;
2745 
2746 		case SO_LISTENINCQLEN:
2747 			optval = so->so_incqlen;
2748 			goto integer;
2749 
2750 		default:
2751 			error = ENOPROTOOPT;
2752 			break;
2753 		}
2754 		return (error);
2755 	}
2756 }
2757 
2758 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2759 int
2760 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2761 {
2762 	struct mbuf *m, *m_prev;
2763 	int sopt_size = sopt->sopt_valsize;
2764 
2765 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2766 	if (m == NULL)
2767 		return ENOBUFS;
2768 	if (sopt_size > MLEN) {
2769 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2770 		if ((m->m_flags & M_EXT) == 0) {
2771 			m_free(m);
2772 			return ENOBUFS;
2773 		}
2774 		m->m_len = min(MCLBYTES, sopt_size);
2775 	} else {
2776 		m->m_len = min(MLEN, sopt_size);
2777 	}
2778 	sopt_size -= m->m_len;
2779 	*mp = m;
2780 	m_prev = m;
2781 
2782 	while (sopt_size) {
2783 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2784 		if (m == NULL) {
2785 			m_freem(*mp);
2786 			return ENOBUFS;
2787 		}
2788 		if (sopt_size > MLEN) {
2789 			MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2790 			    M_DONTWAIT);
2791 			if ((m->m_flags & M_EXT) == 0) {
2792 				m_freem(m);
2793 				m_freem(*mp);
2794 				return ENOBUFS;
2795 			}
2796 			m->m_len = min(MCLBYTES, sopt_size);
2797 		} else {
2798 			m->m_len = min(MLEN, sopt_size);
2799 		}
2800 		sopt_size -= m->m_len;
2801 		m_prev->m_next = m;
2802 		m_prev = m;
2803 	}
2804 	return (0);
2805 }
2806 
2807 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2808 int
2809 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2810 {
2811 	struct mbuf *m0 = m;
2812 
2813 	if (sopt->sopt_val == NULL)
2814 		return (0);
2815 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2816 		if (sopt->sopt_td != NULL) {
2817 			int error;
2818 
2819 			error = copyin(sopt->sopt_val, mtod(m, char *),
2820 				       m->m_len);
2821 			if (error != 0) {
2822 				m_freem(m0);
2823 				return(error);
2824 			}
2825 		} else
2826 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2827 		sopt->sopt_valsize -= m->m_len;
2828 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2829 		m = m->m_next;
2830 	}
2831 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2832 		panic("ip6_sooptmcopyin");
2833 	return (0);
2834 }
2835 
2836 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2837 int
2838 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2839 {
2840 	struct mbuf *m0 = m;
2841 	size_t valsize = 0;
2842 
2843 	if (sopt->sopt_val == NULL)
2844 		return (0);
2845 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2846 		if (sopt->sopt_td != NULL) {
2847 			int error;
2848 
2849 			error = copyout(mtod(m, char *), sopt->sopt_val,
2850 				       m->m_len);
2851 			if (error != 0) {
2852 				m_freem(m0);
2853 				return(error);
2854 			}
2855 		} else
2856 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2857 	       sopt->sopt_valsize -= m->m_len;
2858 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2859 	       valsize += m->m_len;
2860 	       m = m->m_next;
2861 	}
2862 	if (m != NULL) {
2863 		/* enough soopt buffer should be given from user-land */
2864 		m_freem(m0);
2865 		return(EINVAL);
2866 	}
2867 	sopt->sopt_valsize = valsize;
2868 	return (0);
2869 }
2870 
2871 /*
2872  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2873  * out-of-band data, which will then notify socket consumers.
2874  */
2875 void
2876 sohasoutofband(struct socket *so)
2877 {
2878 
2879 	if (so->so_sigio != NULL)
2880 		pgsigio(&so->so_sigio, SIGURG, 0);
2881 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2882 }
2883 
2884 int
2885 sopoll(struct socket *so, int events, struct ucred *active_cred,
2886     struct thread *td)
2887 {
2888 
2889 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2890 	    td));
2891 }
2892 
2893 int
2894 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2895     struct thread *td)
2896 {
2897 	int revents = 0;
2898 
2899 	SOCKBUF_LOCK(&so->so_snd);
2900 	SOCKBUF_LOCK(&so->so_rcv);
2901 	if (events & (POLLIN | POLLRDNORM))
2902 		if (soreadabledata(so))
2903 			revents |= events & (POLLIN | POLLRDNORM);
2904 
2905 	if (events & (POLLOUT | POLLWRNORM))
2906 		if (sowriteable(so))
2907 			revents |= events & (POLLOUT | POLLWRNORM);
2908 
2909 	if (events & (POLLPRI | POLLRDBAND))
2910 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2911 			revents |= events & (POLLPRI | POLLRDBAND);
2912 
2913 	if ((events & POLLINIGNEOF) == 0) {
2914 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2915 			revents |= events & (POLLIN | POLLRDNORM);
2916 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2917 				revents |= POLLHUP;
2918 		}
2919 	}
2920 
2921 	if (revents == 0) {
2922 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2923 			selrecord(td, &so->so_rcv.sb_sel);
2924 			so->so_rcv.sb_flags |= SB_SEL;
2925 		}
2926 
2927 		if (events & (POLLOUT | POLLWRNORM)) {
2928 			selrecord(td, &so->so_snd.sb_sel);
2929 			so->so_snd.sb_flags |= SB_SEL;
2930 		}
2931 	}
2932 
2933 	SOCKBUF_UNLOCK(&so->so_rcv);
2934 	SOCKBUF_UNLOCK(&so->so_snd);
2935 	return (revents);
2936 }
2937 
2938 int
2939 soo_kqfilter(struct file *fp, struct knote *kn)
2940 {
2941 	struct socket *so = kn->kn_fp->f_data;
2942 	struct sockbuf *sb;
2943 
2944 	switch (kn->kn_filter) {
2945 	case EVFILT_READ:
2946 		if (so->so_options & SO_ACCEPTCONN)
2947 			kn->kn_fop = &solisten_filtops;
2948 		else
2949 			kn->kn_fop = &soread_filtops;
2950 		sb = &so->so_rcv;
2951 		break;
2952 	case EVFILT_WRITE:
2953 		kn->kn_fop = &sowrite_filtops;
2954 		sb = &so->so_snd;
2955 		break;
2956 	default:
2957 		return (EINVAL);
2958 	}
2959 
2960 	SOCKBUF_LOCK(sb);
2961 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2962 	sb->sb_flags |= SB_KNOTE;
2963 	SOCKBUF_UNLOCK(sb);
2964 	return (0);
2965 }
2966 
2967 /*
2968  * Some routines that return EOPNOTSUPP for entry points that are not
2969  * supported by a protocol.  Fill in as needed.
2970  */
2971 int
2972 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2973 {
2974 
2975 	return EOPNOTSUPP;
2976 }
2977 
2978 int
2979 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2980 {
2981 
2982 	return EOPNOTSUPP;
2983 }
2984 
2985 int
2986 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2987 {
2988 
2989 	return EOPNOTSUPP;
2990 }
2991 
2992 int
2993 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2994 {
2995 
2996 	return EOPNOTSUPP;
2997 }
2998 
2999 int
3000 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3001 {
3002 
3003 	return EOPNOTSUPP;
3004 }
3005 
3006 int
3007 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3008     struct ifnet *ifp, struct thread *td)
3009 {
3010 
3011 	return EOPNOTSUPP;
3012 }
3013 
3014 int
3015 pru_disconnect_notsupp(struct socket *so)
3016 {
3017 
3018 	return EOPNOTSUPP;
3019 }
3020 
3021 int
3022 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3023 {
3024 
3025 	return EOPNOTSUPP;
3026 }
3027 
3028 int
3029 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3030 {
3031 
3032 	return EOPNOTSUPP;
3033 }
3034 
3035 int
3036 pru_rcvd_notsupp(struct socket *so, int flags)
3037 {
3038 
3039 	return EOPNOTSUPP;
3040 }
3041 
3042 int
3043 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3044 {
3045 
3046 	return EOPNOTSUPP;
3047 }
3048 
3049 int
3050 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3051     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3052 {
3053 
3054 	return EOPNOTSUPP;
3055 }
3056 
3057 /*
3058  * This isn't really a ``null'' operation, but it's the default one and
3059  * doesn't do anything destructive.
3060  */
3061 int
3062 pru_sense_null(struct socket *so, struct stat *sb)
3063 {
3064 
3065 	sb->st_blksize = so->so_snd.sb_hiwat;
3066 	return 0;
3067 }
3068 
3069 int
3070 pru_shutdown_notsupp(struct socket *so)
3071 {
3072 
3073 	return EOPNOTSUPP;
3074 }
3075 
3076 int
3077 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3078 {
3079 
3080 	return EOPNOTSUPP;
3081 }
3082 
3083 int
3084 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3085     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3086 {
3087 
3088 	return EOPNOTSUPP;
3089 }
3090 
3091 int
3092 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3093     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3094 {
3095 
3096 	return EOPNOTSUPP;
3097 }
3098 
3099 int
3100 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3101     struct thread *td)
3102 {
3103 
3104 	return EOPNOTSUPP;
3105 }
3106 
3107 static void
3108 filt_sordetach(struct knote *kn)
3109 {
3110 	struct socket *so = kn->kn_fp->f_data;
3111 
3112 	SOCKBUF_LOCK(&so->so_rcv);
3113 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3114 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3115 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3116 	SOCKBUF_UNLOCK(&so->so_rcv);
3117 }
3118 
3119 /*ARGSUSED*/
3120 static int
3121 filt_soread(struct knote *kn, long hint)
3122 {
3123 	struct socket *so;
3124 
3125 	so = kn->kn_fp->f_data;
3126 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3127 
3128 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3129 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3130 		kn->kn_flags |= EV_EOF;
3131 		kn->kn_fflags = so->so_error;
3132 		return (1);
3133 	} else if (so->so_error)	/* temporary udp error */
3134 		return (1);
3135 	else if (kn->kn_sfflags & NOTE_LOWAT)
3136 		return (kn->kn_data >= kn->kn_sdata);
3137 	else
3138 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3139 }
3140 
3141 static void
3142 filt_sowdetach(struct knote *kn)
3143 {
3144 	struct socket *so = kn->kn_fp->f_data;
3145 
3146 	SOCKBUF_LOCK(&so->so_snd);
3147 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3148 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3149 		so->so_snd.sb_flags &= ~SB_KNOTE;
3150 	SOCKBUF_UNLOCK(&so->so_snd);
3151 }
3152 
3153 /*ARGSUSED*/
3154 static int
3155 filt_sowrite(struct knote *kn, long hint)
3156 {
3157 	struct socket *so;
3158 
3159 	so = kn->kn_fp->f_data;
3160 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3161 	kn->kn_data = sbspace(&so->so_snd);
3162 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3163 		kn->kn_flags |= EV_EOF;
3164 		kn->kn_fflags = so->so_error;
3165 		return (1);
3166 	} else if (so->so_error)	/* temporary udp error */
3167 		return (1);
3168 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3169 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3170 		return (0);
3171 	else if (kn->kn_sfflags & NOTE_LOWAT)
3172 		return (kn->kn_data >= kn->kn_sdata);
3173 	else
3174 		return (kn->kn_data >= so->so_snd.sb_lowat);
3175 }
3176 
3177 /*ARGSUSED*/
3178 static int
3179 filt_solisten(struct knote *kn, long hint)
3180 {
3181 	struct socket *so = kn->kn_fp->f_data;
3182 
3183 	kn->kn_data = so->so_qlen;
3184 	return (! TAILQ_EMPTY(&so->so_comp));
3185 }
3186 
3187 int
3188 socheckuid(struct socket *so, uid_t uid)
3189 {
3190 
3191 	if (so == NULL)
3192 		return (EPERM);
3193 	if (so->so_cred->cr_uid != uid)
3194 		return (EPERM);
3195 	return (0);
3196 }
3197 
3198 static int
3199 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3200 {
3201 	int error;
3202 	int val;
3203 
3204 	val = somaxconn;
3205 	error = sysctl_handle_int(oidp, &val, 0, req);
3206 	if (error || !req->newptr )
3207 		return (error);
3208 
3209 	if (val < 1 || val > USHRT_MAX)
3210 		return (EINVAL);
3211 
3212 	somaxconn = val;
3213 	return (0);
3214 }
3215 
3216 /*
3217  * These functions are used by protocols to notify the socket layer (and its
3218  * consumers) of state changes in the sockets driven by protocol-side events.
3219  */
3220 
3221 /*
3222  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3223  *
3224  * Normal sequence from the active (originating) side is that
3225  * soisconnecting() is called during processing of connect() call, resulting
3226  * in an eventual call to soisconnected() if/when the connection is
3227  * established.  When the connection is torn down soisdisconnecting() is
3228  * called during processing of disconnect() call, and soisdisconnected() is
3229  * called when the connection to the peer is totally severed.  The semantics
3230  * of these routines are such that connectionless protocols can call
3231  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3232  * calls when setting up a ``connection'' takes no time.
3233  *
3234  * From the passive side, a socket is created with two queues of sockets:
3235  * so_incomp for connections in progress and so_comp for connections already
3236  * made and awaiting user acceptance.  As a protocol is preparing incoming
3237  * connections, it creates a socket structure queued on so_incomp by calling
3238  * sonewconn().  When the connection is established, soisconnected() is
3239  * called, and transfers the socket structure to so_comp, making it available
3240  * to accept().
3241  *
3242  * If a socket is closed with sockets on either so_incomp or so_comp, these
3243  * sockets are dropped.
3244  *
3245  * If higher-level protocols are implemented in the kernel, the wakeups done
3246  * here will sometimes cause software-interrupt process scheduling.
3247  */
3248 void
3249 soisconnecting(struct socket *so)
3250 {
3251 
3252 	SOCK_LOCK(so);
3253 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3254 	so->so_state |= SS_ISCONNECTING;
3255 	SOCK_UNLOCK(so);
3256 }
3257 
3258 void
3259 soisconnected(struct socket *so)
3260 {
3261 	struct socket *head;
3262 	int ret;
3263 
3264 restart:
3265 	ACCEPT_LOCK();
3266 	SOCK_LOCK(so);
3267 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3268 	so->so_state |= SS_ISCONNECTED;
3269 	head = so->so_head;
3270 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3271 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3272 			SOCK_UNLOCK(so);
3273 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3274 			head->so_incqlen--;
3275 			so->so_qstate &= ~SQ_INCOMP;
3276 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3277 			head->so_qlen++;
3278 			so->so_qstate |= SQ_COMP;
3279 			ACCEPT_UNLOCK();
3280 			sorwakeup(head);
3281 			wakeup_one(&head->so_timeo);
3282 		} else {
3283 			ACCEPT_UNLOCK();
3284 			soupcall_set(so, SO_RCV,
3285 			    head->so_accf->so_accept_filter->accf_callback,
3286 			    head->so_accf->so_accept_filter_arg);
3287 			so->so_options &= ~SO_ACCEPTFILTER;
3288 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3289 			    head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3290 			if (ret == SU_ISCONNECTED)
3291 				soupcall_clear(so, SO_RCV);
3292 			SOCK_UNLOCK(so);
3293 			if (ret == SU_ISCONNECTED)
3294 				goto restart;
3295 		}
3296 		return;
3297 	}
3298 	SOCK_UNLOCK(so);
3299 	ACCEPT_UNLOCK();
3300 	wakeup(&so->so_timeo);
3301 	sorwakeup(so);
3302 	sowwakeup(so);
3303 }
3304 
3305 void
3306 soisdisconnecting(struct socket *so)
3307 {
3308 
3309 	/*
3310 	 * Note: This code assumes that SOCK_LOCK(so) and
3311 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3312 	 */
3313 	SOCKBUF_LOCK(&so->so_rcv);
3314 	so->so_state &= ~SS_ISCONNECTING;
3315 	so->so_state |= SS_ISDISCONNECTING;
3316 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3317 	sorwakeup_locked(so);
3318 	SOCKBUF_LOCK(&so->so_snd);
3319 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3320 	sowwakeup_locked(so);
3321 	wakeup(&so->so_timeo);
3322 }
3323 
3324 void
3325 soisdisconnected(struct socket *so)
3326 {
3327 
3328 	/*
3329 	 * Note: This code assumes that SOCK_LOCK(so) and
3330 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3331 	 */
3332 	SOCKBUF_LOCK(&so->so_rcv);
3333 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3334 	so->so_state |= SS_ISDISCONNECTED;
3335 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3336 	sorwakeup_locked(so);
3337 	SOCKBUF_LOCK(&so->so_snd);
3338 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3339 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3340 	sowwakeup_locked(so);
3341 	wakeup(&so->so_timeo);
3342 }
3343 
3344 /*
3345  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3346  */
3347 struct sockaddr *
3348 sodupsockaddr(const struct sockaddr *sa, int mflags)
3349 {
3350 	struct sockaddr *sa2;
3351 
3352 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3353 	if (sa2)
3354 		bcopy(sa, sa2, sa->sa_len);
3355 	return sa2;
3356 }
3357 
3358 /*
3359  * Register per-socket buffer upcalls.
3360  */
3361 void
3362 soupcall_set(struct socket *so, int which,
3363     int (*func)(struct socket *, void *, int), void *arg)
3364 {
3365 	struct sockbuf *sb;
3366 
3367 	switch (which) {
3368 	case SO_RCV:
3369 		sb = &so->so_rcv;
3370 		break;
3371 	case SO_SND:
3372 		sb = &so->so_snd;
3373 		break;
3374 	default:
3375 		panic("soupcall_set: bad which");
3376 	}
3377 	SOCKBUF_LOCK_ASSERT(sb);
3378 #if 0
3379 	/* XXX: accf_http actually wants to do this on purpose. */
3380 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3381 #endif
3382 	sb->sb_upcall = func;
3383 	sb->sb_upcallarg = arg;
3384 	sb->sb_flags |= SB_UPCALL;
3385 }
3386 
3387 void
3388 soupcall_clear(struct socket *so, int which)
3389 {
3390 	struct sockbuf *sb;
3391 
3392 	switch (which) {
3393 	case SO_RCV:
3394 		sb = &so->so_rcv;
3395 		break;
3396 	case SO_SND:
3397 		sb = &so->so_snd;
3398 		break;
3399 	default:
3400 		panic("soupcall_clear: bad which");
3401 	}
3402 	SOCKBUF_LOCK_ASSERT(sb);
3403 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3404 	sb->sb_upcall = NULL;
3405 	sb->sb_upcallarg = NULL;
3406 	sb->sb_flags &= ~SB_UPCALL;
3407 }
3408 
3409 /*
3410  * Create an external-format (``xsocket'') structure using the information in
3411  * the kernel-format socket structure pointed to by so.  This is done to
3412  * reduce the spew of irrelevant information over this interface, to isolate
3413  * user code from changes in the kernel structure, and potentially to provide
3414  * information-hiding if we decide that some of this information should be
3415  * hidden from users.
3416  */
3417 void
3418 sotoxsocket(struct socket *so, struct xsocket *xso)
3419 {
3420 
3421 	xso->xso_len = sizeof *xso;
3422 	xso->xso_so = so;
3423 	xso->so_type = so->so_type;
3424 	xso->so_options = so->so_options;
3425 	xso->so_linger = so->so_linger;
3426 	xso->so_state = so->so_state;
3427 	xso->so_pcb = so->so_pcb;
3428 	xso->xso_protocol = so->so_proto->pr_protocol;
3429 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3430 	xso->so_qlen = so->so_qlen;
3431 	xso->so_incqlen = so->so_incqlen;
3432 	xso->so_qlimit = so->so_qlimit;
3433 	xso->so_timeo = so->so_timeo;
3434 	xso->so_error = so->so_error;
3435 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3436 	xso->so_oobmark = so->so_oobmark;
3437 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3438 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3439 	xso->so_uid = so->so_cred->cr_uid;
3440 }
3441 
3442 
3443 /*
3444  * Socket accessor functions to provide external consumers with
3445  * a safe interface to socket state
3446  *
3447  */
3448 
3449 void
3450 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3451 {
3452 
3453 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3454 		func(so, arg);
3455 }
3456 
3457 struct sockbuf *
3458 so_sockbuf_rcv(struct socket *so)
3459 {
3460 
3461 	return (&so->so_rcv);
3462 }
3463 
3464 struct sockbuf *
3465 so_sockbuf_snd(struct socket *so)
3466 {
3467 
3468 	return (&so->so_snd);
3469 }
3470 
3471 int
3472 so_state_get(const struct socket *so)
3473 {
3474 
3475 	return (so->so_state);
3476 }
3477 
3478 void
3479 so_state_set(struct socket *so, int val)
3480 {
3481 
3482 	so->so_state = val;
3483 }
3484 
3485 int
3486 so_options_get(const struct socket *so)
3487 {
3488 
3489 	return (so->so_options);
3490 }
3491 
3492 void
3493 so_options_set(struct socket *so, int val)
3494 {
3495 
3496 	so->so_options = val;
3497 }
3498 
3499 int
3500 so_error_get(const struct socket *so)
3501 {
3502 
3503 	return (so->so_error);
3504 }
3505 
3506 void
3507 so_error_set(struct socket *so, int val)
3508 {
3509 
3510 	so->so_error = val;
3511 }
3512 
3513 int
3514 so_linger_get(const struct socket *so)
3515 {
3516 
3517 	return (so->so_linger);
3518 }
3519 
3520 void
3521 so_linger_set(struct socket *so, int val)
3522 {
3523 
3524 	so->so_linger = val;
3525 }
3526 
3527 struct protosw *
3528 so_protosw_get(const struct socket *so)
3529 {
3530 
3531 	return (so->so_proto);
3532 }
3533 
3534 void
3535 so_protosw_set(struct socket *so, struct protosw *val)
3536 {
3537 
3538 	so->so_proto = val;
3539 }
3540 
3541 void
3542 so_sorwakeup(struct socket *so)
3543 {
3544 
3545 	sorwakeup(so);
3546 }
3547 
3548 void
3549 so_sowwakeup(struct socket *so)
3550 {
3551 
3552 	sowwakeup(so);
3553 }
3554 
3555 void
3556 so_sorwakeup_locked(struct socket *so)
3557 {
3558 
3559 	sorwakeup_locked(so);
3560 }
3561 
3562 void
3563 so_sowwakeup_locked(struct socket *so)
3564 {
3565 
3566 	sowwakeup_locked(so);
3567 }
3568 
3569 void
3570 so_lock(struct socket *so)
3571 {
3572 	SOCK_LOCK(so);
3573 }
3574 
3575 void
3576 so_unlock(struct socket *so)
3577 {
3578 	SOCK_UNLOCK(so);
3579 }
3580