xref: /freebsd/sys/kern/uipc_socket.c (revision 1669d8afc64812c8d2d1d147ae1fd42ff441e1b1)
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) 2004-2007 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_mac.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 <sys/signalvar.h>
126 #include <sys/stat.h>
127 #include <sys/sx.h>
128 #include <sys/sysctl.h>
129 #include <sys/uio.h>
130 #include <sys/jail.h>
131 
132 #include <security/mac/mac_framework.h>
133 
134 #include <vm/uma.h>
135 
136 #ifdef COMPAT_IA32
137 #include <sys/mount.h>
138 #include <compat/freebsd32/freebsd32.h>
139 
140 extern struct sysentvec ia32_freebsd_sysvec;
141 #endif
142 
143 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
144 		    int flags);
145 
146 static void	filt_sordetach(struct knote *kn);
147 static int	filt_soread(struct knote *kn, long hint);
148 static void	filt_sowdetach(struct knote *kn);
149 static int	filt_sowrite(struct knote *kn, long hint);
150 static int	filt_solisten(struct knote *kn, long hint);
151 
152 static struct filterops solisten_filtops =
153 	{ 1, NULL, filt_sordetach, filt_solisten };
154 static struct filterops soread_filtops =
155 	{ 1, NULL, filt_sordetach, filt_soread };
156 static struct filterops sowrite_filtops =
157 	{ 1, NULL, filt_sowdetach, filt_sowrite };
158 
159 uma_zone_t socket_zone;
160 so_gen_t	so_gencnt;	/* generation count for sockets */
161 
162 int	maxsockets;
163 
164 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
165 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
166 
167 static int somaxconn = SOMAXCONN;
168 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
169 /* XXX: we dont have SYSCTL_USHORT */
170 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
171     0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
172     "queue size");
173 static int numopensockets;
174 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
175     &numopensockets, 0, "Number of open sockets");
176 #ifdef ZERO_COPY_SOCKETS
177 /* These aren't static because they're used in other files. */
178 int so_zero_copy_send = 1;
179 int so_zero_copy_receive = 1;
180 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
181     "Zero copy controls");
182 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
183     &so_zero_copy_receive, 0, "Enable zero copy receive");
184 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
185     &so_zero_copy_send, 0, "Enable zero copy send");
186 #endif /* ZERO_COPY_SOCKETS */
187 
188 /*
189  * accept_mtx locks down per-socket fields relating to accept queues.  See
190  * socketvar.h for an annotation of the protected fields of struct socket.
191  */
192 struct mtx accept_mtx;
193 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
194 
195 /*
196  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
197  * so_gencnt field.
198  */
199 static struct mtx so_global_mtx;
200 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
201 
202 /*
203  * General IPC sysctl name space, used by sockets and a variety of other IPC
204  * types.
205  */
206 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
207 
208 /*
209  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
210  * of the change so that they can update their dependent limits as required.
211  */
212 static int
213 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
214 {
215 	int error, newmaxsockets;
216 
217 	newmaxsockets = maxsockets;
218 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
219 	if (error == 0 && req->newptr) {
220 		if (newmaxsockets > maxsockets) {
221 			maxsockets = newmaxsockets;
222 			if (maxsockets > ((maxfiles / 4) * 3)) {
223 				maxfiles = (maxsockets * 5) / 4;
224 				maxfilesperproc = (maxfiles * 9) / 10;
225 			}
226 			EVENTHANDLER_INVOKE(maxsockets_change);
227 		} else
228 			error = EINVAL;
229 	}
230 	return (error);
231 }
232 
233 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
234     &maxsockets, 0, sysctl_maxsockets, "IU",
235     "Maximum number of sockets avaliable");
236 
237 /*
238  * Initialise maxsockets.
239  */
240 static void init_maxsockets(void *ignored)
241 {
242 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
243 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
244 }
245 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
246 
247 /*
248  * Socket operation routines.  These routines are called by the routines in
249  * sys_socket.c or from a system process, and implement the semantics of
250  * socket operations by switching out to the protocol specific routines.
251  */
252 
253 /*
254  * Get a socket structure from our zone, and initialize it.  Note that it
255  * would probably be better to allocate socket and PCB at the same time, but
256  * I'm not convinced that all the protocols can be easily modified to do
257  * this.
258  *
259  * soalloc() returns a socket with a ref count of 0.
260  */
261 static struct socket *
262 soalloc(void)
263 {
264 	struct socket *so;
265 
266 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
267 	if (so == NULL)
268 		return (NULL);
269 #ifdef MAC
270 	if (mac_socket_init(so, M_NOWAIT) != 0) {
271 		uma_zfree(socket_zone, so);
272 		return (NULL);
273 	}
274 #endif
275 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
276 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
277 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
278 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
279 	TAILQ_INIT(&so->so_aiojobq);
280 	mtx_lock(&so_global_mtx);
281 	so->so_gencnt = ++so_gencnt;
282 	++numopensockets;
283 	mtx_unlock(&so_global_mtx);
284 	return (so);
285 }
286 
287 /*
288  * Free the storage associated with a socket at the socket layer, tear down
289  * locks, labels, etc.  All protocol state is assumed already to have been
290  * torn down (and possibly never set up) by the caller.
291  */
292 static void
293 sodealloc(struct socket *so)
294 {
295 
296 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
297 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
298 
299 	mtx_lock(&so_global_mtx);
300 	so->so_gencnt = ++so_gencnt;
301 	--numopensockets;	/* Could be below, but faster here. */
302 	mtx_unlock(&so_global_mtx);
303 	if (so->so_rcv.sb_hiwat)
304 		(void)chgsbsize(so->so_cred->cr_uidinfo,
305 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
306 	if (so->so_snd.sb_hiwat)
307 		(void)chgsbsize(so->so_cred->cr_uidinfo,
308 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
309 #ifdef INET
310 	/* remove acccept filter if one is present. */
311 	if (so->so_accf != NULL)
312 		do_setopt_accept_filter(so, NULL);
313 #endif
314 #ifdef MAC
315 	mac_socket_destroy(so);
316 #endif
317 	crfree(so->so_cred);
318 	sx_destroy(&so->so_snd.sb_sx);
319 	sx_destroy(&so->so_rcv.sb_sx);
320 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
321 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
322 	uma_zfree(socket_zone, so);
323 }
324 
325 /*
326  * socreate returns a socket with a ref count of 1.  The socket should be
327  * closed with soclose().
328  */
329 int
330 socreate(int dom, struct socket **aso, int type, int proto,
331     struct ucred *cred, struct thread *td)
332 {
333 	struct protosw *prp;
334 	struct socket *so;
335 	int error;
336 
337 	if (proto)
338 		prp = pffindproto(dom, proto, type);
339 	else
340 		prp = pffindtype(dom, type);
341 
342 	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
343 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
344 		return (EPROTONOSUPPORT);
345 
346 	if (jailed(cred) && jail_socket_unixiproute_only &&
347 	    prp->pr_domain->dom_family != PF_LOCAL &&
348 	    prp->pr_domain->dom_family != PF_INET &&
349 	    prp->pr_domain->dom_family != PF_ROUTE) {
350 		return (EPROTONOSUPPORT);
351 	}
352 
353 	if (prp->pr_type != type)
354 		return (EPROTOTYPE);
355 	so = soalloc();
356 	if (so == NULL)
357 		return (ENOBUFS);
358 
359 	TAILQ_INIT(&so->so_incomp);
360 	TAILQ_INIT(&so->so_comp);
361 	so->so_type = type;
362 	so->so_cred = crhold(cred);
363 	so->so_proto = prp;
364 #ifdef MAC
365 	mac_socket_create(cred, so);
366 #endif
367 	knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
368 	    NULL, NULL, NULL);
369 	knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
370 	    NULL, NULL, NULL);
371 	so->so_count = 1;
372 	/*
373 	 * Auto-sizing of socket buffers is managed by the protocols and
374 	 * the appropriate flags must be set in the pru_attach function.
375 	 */
376 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
377 	if (error) {
378 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
379 		    so->so_count));
380 		so->so_count = 0;
381 		sodealloc(so);
382 		return (error);
383 	}
384 	*aso = so;
385 	return (0);
386 }
387 
388 #ifdef REGRESSION
389 static int regression_sonewconn_earlytest = 1;
390 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
391     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
392 #endif
393 
394 /*
395  * When an attempt at a new connection is noted on a socket which accepts
396  * connections, sonewconn is called.  If the connection is possible (subject
397  * to space constraints, etc.) then we allocate a new structure, propoerly
398  * linked into the data structure of the original socket, and return this.
399  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
400  *
401  * Note: the ref count on the socket is 0 on return.
402  */
403 struct socket *
404 sonewconn(struct socket *head, int connstatus)
405 {
406 	struct socket *so;
407 	int over;
408 
409 	ACCEPT_LOCK();
410 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
411 	ACCEPT_UNLOCK();
412 #ifdef REGRESSION
413 	if (regression_sonewconn_earlytest && over)
414 #else
415 	if (over)
416 #endif
417 		return (NULL);
418 	so = soalloc();
419 	if (so == NULL)
420 		return (NULL);
421 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
422 		connstatus = 0;
423 	so->so_head = head;
424 	so->so_type = head->so_type;
425 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
426 	so->so_linger = head->so_linger;
427 	so->so_state = head->so_state | SS_NOFDREF;
428 	so->so_proto = head->so_proto;
429 	so->so_cred = crhold(head->so_cred);
430 #ifdef MAC
431 	SOCK_LOCK(head);
432 	mac_socket_newconn(head, so);
433 	SOCK_UNLOCK(head);
434 #endif
435 	knlist_init(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv),
436 	    NULL, NULL, NULL);
437 	knlist_init(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd),
438 	    NULL, NULL, NULL);
439 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
440 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
441 		sodealloc(so);
442 		return (NULL);
443 	}
444 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
445 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
446 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
447 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
448 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
449 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
450 	so->so_state |= connstatus;
451 	ACCEPT_LOCK();
452 	if (connstatus) {
453 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
454 		so->so_qstate |= SQ_COMP;
455 		head->so_qlen++;
456 	} else {
457 		/*
458 		 * Keep removing sockets from the head until there's room for
459 		 * us to insert on the tail.  In pre-locking revisions, this
460 		 * was a simple if(), but as we could be racing with other
461 		 * threads and soabort() requires dropping locks, we must
462 		 * loop waiting for the condition to be true.
463 		 */
464 		while (head->so_incqlen > head->so_qlimit) {
465 			struct socket *sp;
466 			sp = TAILQ_FIRST(&head->so_incomp);
467 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
468 			head->so_incqlen--;
469 			sp->so_qstate &= ~SQ_INCOMP;
470 			sp->so_head = NULL;
471 			ACCEPT_UNLOCK();
472 			soabort(sp);
473 			ACCEPT_LOCK();
474 		}
475 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
476 		so->so_qstate |= SQ_INCOMP;
477 		head->so_incqlen++;
478 	}
479 	ACCEPT_UNLOCK();
480 	if (connstatus) {
481 		sorwakeup(head);
482 		wakeup_one(&head->so_timeo);
483 	}
484 	return (so);
485 }
486 
487 int
488 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
489 {
490 
491 	return ((*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td));
492 }
493 
494 /*
495  * solisten() transitions a socket from a non-listening state to a listening
496  * state, but can also be used to update the listen queue depth on an
497  * existing listen socket.  The protocol will call back into the sockets
498  * layer using solisten_proto_check() and solisten_proto() to check and set
499  * socket-layer listen state.  Call backs are used so that the protocol can
500  * acquire both protocol and socket layer locks in whatever order is required
501  * by the protocol.
502  *
503  * Protocol implementors are advised to hold the socket lock across the
504  * socket-layer test and set to avoid races at the socket layer.
505  */
506 int
507 solisten(struct socket *so, int backlog, struct thread *td)
508 {
509 
510 	return ((*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td));
511 }
512 
513 int
514 solisten_proto_check(struct socket *so)
515 {
516 
517 	SOCK_LOCK_ASSERT(so);
518 
519 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
520 	    SS_ISDISCONNECTING))
521 		return (EINVAL);
522 	return (0);
523 }
524 
525 void
526 solisten_proto(struct socket *so, int backlog)
527 {
528 
529 	SOCK_LOCK_ASSERT(so);
530 
531 	if (backlog < 0 || backlog > somaxconn)
532 		backlog = somaxconn;
533 	so->so_qlimit = backlog;
534 	so->so_options |= SO_ACCEPTCONN;
535 }
536 
537 /*
538  * Attempt to free a socket.  This should really be sotryfree().
539  *
540  * sofree() will succeed if:
541  *
542  * - There are no outstanding file descriptor references or related consumers
543  *   (so_count == 0).
544  *
545  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
546  *
547  * - The protocol does not have an outstanding strong reference on the socket
548  *   (SS_PROTOREF).
549  *
550  * - The socket is not in a completed connection queue, so a process has been
551  *   notified that it is present.  If it is removed, the user process may
552  *   block in accept() despite select() saying the socket was ready.
553  *
554  * Otherwise, it will quietly abort so that a future call to sofree(), when
555  * conditions are right, can succeed.
556  */
557 void
558 sofree(struct socket *so)
559 {
560 	struct protosw *pr = so->so_proto;
561 	struct socket *head;
562 
563 	ACCEPT_LOCK_ASSERT();
564 	SOCK_LOCK_ASSERT(so);
565 
566 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
567 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
568 		SOCK_UNLOCK(so);
569 		ACCEPT_UNLOCK();
570 		return;
571 	}
572 
573 	head = so->so_head;
574 	if (head != NULL) {
575 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
576 		    (so->so_qstate & SQ_INCOMP) != 0,
577 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
578 		    "SQ_INCOMP"));
579 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
580 		    (so->so_qstate & SQ_INCOMP) == 0,
581 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
582 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
583 		head->so_incqlen--;
584 		so->so_qstate &= ~SQ_INCOMP;
585 		so->so_head = NULL;
586 	}
587 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
588 	    (so->so_qstate & SQ_INCOMP) == 0,
589 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
590 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
591 	if (so->so_options & SO_ACCEPTCONN) {
592 		KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
593 		KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
594 	}
595 	SOCK_UNLOCK(so);
596 	ACCEPT_UNLOCK();
597 
598 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
599 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
600 	if (pr->pr_usrreqs->pru_detach != NULL)
601 		(*pr->pr_usrreqs->pru_detach)(so);
602 
603 	/*
604 	 * From this point on, we assume that no other references to this
605 	 * socket exist anywhere else in the stack.  Therefore, no locks need
606 	 * to be acquired or held.
607 	 *
608 	 * We used to do a lot of socket buffer and socket locking here, as
609 	 * well as invoke sorflush() and perform wakeups.  The direct call to
610 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
611 	 * necessary from sorflush().
612 	 *
613 	 * Notice that the socket buffer and kqueue state are torn down
614 	 * before calling pru_detach.  This means that protocols shold not
615 	 * assume they can perform socket wakeups, etc, in their detach code.
616 	 */
617 	sbdestroy(&so->so_snd, so);
618 	sbdestroy(&so->so_rcv, so);
619 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
620 	knlist_destroy(&so->so_snd.sb_sel.si_note);
621 	sodealloc(so);
622 }
623 
624 /*
625  * Close a socket on last file table reference removal.  Initiate disconnect
626  * if connected.  Free socket when disconnect complete.
627  *
628  * This function will sorele() the socket.  Note that soclose() may be called
629  * prior to the ref count reaching zero.  The actual socket structure will
630  * not be freed until the ref count reaches zero.
631  */
632 int
633 soclose(struct socket *so)
634 {
635 	int error = 0;
636 
637 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
638 
639 	funsetown(&so->so_sigio);
640 	if (so->so_state & SS_ISCONNECTED) {
641 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
642 			error = sodisconnect(so);
643 			if (error)
644 				goto drop;
645 		}
646 		if (so->so_options & SO_LINGER) {
647 			if ((so->so_state & SS_ISDISCONNECTING) &&
648 			    (so->so_state & SS_NBIO))
649 				goto drop;
650 			while (so->so_state & SS_ISCONNECTED) {
651 				error = tsleep(&so->so_timeo,
652 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
653 				if (error)
654 					break;
655 			}
656 		}
657 	}
658 
659 drop:
660 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
661 		(*so->so_proto->pr_usrreqs->pru_close)(so);
662 	if (so->so_options & SO_ACCEPTCONN) {
663 		struct socket *sp;
664 		ACCEPT_LOCK();
665 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
666 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
667 			so->so_incqlen--;
668 			sp->so_qstate &= ~SQ_INCOMP;
669 			sp->so_head = NULL;
670 			ACCEPT_UNLOCK();
671 			soabort(sp);
672 			ACCEPT_LOCK();
673 		}
674 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
675 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
676 			so->so_qlen--;
677 			sp->so_qstate &= ~SQ_COMP;
678 			sp->so_head = NULL;
679 			ACCEPT_UNLOCK();
680 			soabort(sp);
681 			ACCEPT_LOCK();
682 		}
683 		ACCEPT_UNLOCK();
684 	}
685 	ACCEPT_LOCK();
686 	SOCK_LOCK(so);
687 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
688 	so->so_state |= SS_NOFDREF;
689 	sorele(so);
690 	return (error);
691 }
692 
693 /*
694  * soabort() is used to abruptly tear down a connection, such as when a
695  * resource limit is reached (listen queue depth exceeded), or if a listen
696  * socket is closed while there are sockets waiting to be accepted.
697  *
698  * This interface is tricky, because it is called on an unreferenced socket,
699  * and must be called only by a thread that has actually removed the socket
700  * from the listen queue it was on, or races with other threads are risked.
701  *
702  * This interface will call into the protocol code, so must not be called
703  * with any socket locks held.  Protocols do call it while holding their own
704  * recursible protocol mutexes, but this is something that should be subject
705  * to review in the future.
706  */
707 void
708 soabort(struct socket *so)
709 {
710 
711 	/*
712 	 * In as much as is possible, assert that no references to this
713 	 * socket are held.  This is not quite the same as asserting that the
714 	 * current thread is responsible for arranging for no references, but
715 	 * is as close as we can get for now.
716 	 */
717 	KASSERT(so->so_count == 0, ("soabort: so_count"));
718 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
719 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
720 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
721 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
722 
723 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
724 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
725 	ACCEPT_LOCK();
726 	SOCK_LOCK(so);
727 	sofree(so);
728 }
729 
730 int
731 soaccept(struct socket *so, struct sockaddr **nam)
732 {
733 	int error;
734 
735 	SOCK_LOCK(so);
736 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
737 	so->so_state &= ~SS_NOFDREF;
738 	SOCK_UNLOCK(so);
739 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
740 	return (error);
741 }
742 
743 int
744 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
745 {
746 	int error;
747 
748 	if (so->so_options & SO_ACCEPTCONN)
749 		return (EOPNOTSUPP);
750 	/*
751 	 * If protocol is connection-based, can only connect once.
752 	 * Otherwise, if connected, try to disconnect first.  This allows
753 	 * user to disconnect by connecting to, e.g., a null address.
754 	 */
755 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
756 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
757 	    (error = sodisconnect(so)))) {
758 		error = EISCONN;
759 	} else {
760 		/*
761 		 * Prevent accumulated error from previous connection from
762 		 * biting us.
763 		 */
764 		so->so_error = 0;
765 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
766 	}
767 
768 	return (error);
769 }
770 
771 int
772 soconnect2(struct socket *so1, struct socket *so2)
773 {
774 
775 	return ((*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2));
776 }
777 
778 int
779 sodisconnect(struct socket *so)
780 {
781 	int error;
782 
783 	if ((so->so_state & SS_ISCONNECTED) == 0)
784 		return (ENOTCONN);
785 	if (so->so_state & SS_ISDISCONNECTING)
786 		return (EALREADY);
787 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
788 	return (error);
789 }
790 
791 #ifdef ZERO_COPY_SOCKETS
792 struct so_zerocopy_stats{
793 	int size_ok;
794 	int align_ok;
795 	int found_ifp;
796 };
797 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
798 #include <netinet/in.h>
799 #include <net/route.h>
800 #include <netinet/in_pcb.h>
801 #include <vm/vm.h>
802 #include <vm/vm_page.h>
803 #include <vm/vm_object.h>
804 
805 /*
806  * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
807  * sosend_dgram() and sosend_generic() use m_uiotombuf().
808  *
809  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
810  * all of the data referenced by the uio.  If desired, it uses zero-copy.
811  * *space will be updated to reflect data copied in.
812  *
813  * NB: If atomic I/O is requested, the caller must already have checked that
814  * space can hold resid bytes.
815  *
816  * NB: In the event of an error, the caller may need to free the partial
817  * chain pointed to by *mpp.  The contents of both *uio and *space may be
818  * modified even in the case of an error.
819  */
820 static int
821 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
822     int flags)
823 {
824 	struct mbuf *m, **mp, *top;
825 	long len, resid;
826 	int error;
827 #ifdef ZERO_COPY_SOCKETS
828 	int cow_send;
829 #endif
830 
831 	*retmp = top = NULL;
832 	mp = &top;
833 	len = 0;
834 	resid = uio->uio_resid;
835 	error = 0;
836 	do {
837 #ifdef ZERO_COPY_SOCKETS
838 		cow_send = 0;
839 #endif /* ZERO_COPY_SOCKETS */
840 		if (resid >= MINCLSIZE) {
841 #ifdef ZERO_COPY_SOCKETS
842 			if (top == NULL) {
843 				m = m_gethdr(M_WAITOK, MT_DATA);
844 				m->m_pkthdr.len = 0;
845 				m->m_pkthdr.rcvif = NULL;
846 			} else
847 				m = m_get(M_WAITOK, MT_DATA);
848 			if (so_zero_copy_send &&
849 			    resid>=PAGE_SIZE &&
850 			    *space>=PAGE_SIZE &&
851 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
852 				so_zerocp_stats.size_ok++;
853 				so_zerocp_stats.align_ok++;
854 				cow_send = socow_setup(m, uio);
855 				len = cow_send;
856 			}
857 			if (!cow_send) {
858 				m_clget(m, M_WAITOK);
859 				len = min(min(MCLBYTES, resid), *space);
860 			}
861 #else /* ZERO_COPY_SOCKETS */
862 			if (top == NULL) {
863 				m = m_getcl(M_TRYWAIT, MT_DATA, M_PKTHDR);
864 				m->m_pkthdr.len = 0;
865 				m->m_pkthdr.rcvif = NULL;
866 			} else
867 				m = m_getcl(M_TRYWAIT, MT_DATA, 0);
868 			len = min(min(MCLBYTES, resid), *space);
869 #endif /* ZERO_COPY_SOCKETS */
870 		} else {
871 			if (top == NULL) {
872 				m = m_gethdr(M_TRYWAIT, MT_DATA);
873 				m->m_pkthdr.len = 0;
874 				m->m_pkthdr.rcvif = NULL;
875 
876 				len = min(min(MHLEN, resid), *space);
877 				/*
878 				 * For datagram protocols, leave room
879 				 * for protocol headers in first mbuf.
880 				 */
881 				if (atomic && m && len < MHLEN)
882 					MH_ALIGN(m, len);
883 			} else {
884 				m = m_get(M_TRYWAIT, MT_DATA);
885 				len = min(min(MLEN, resid), *space);
886 			}
887 		}
888 		if (m == NULL) {
889 			error = ENOBUFS;
890 			goto out;
891 		}
892 
893 		*space -= len;
894 #ifdef ZERO_COPY_SOCKETS
895 		if (cow_send)
896 			error = 0;
897 		else
898 #endif /* ZERO_COPY_SOCKETS */
899 		error = uiomove(mtod(m, void *), (int)len, uio);
900 		resid = uio->uio_resid;
901 		m->m_len = len;
902 		*mp = m;
903 		top->m_pkthdr.len += len;
904 		if (error)
905 			goto out;
906 		mp = &m->m_next;
907 		if (resid <= 0) {
908 			if (flags & MSG_EOR)
909 				top->m_flags |= M_EOR;
910 			break;
911 		}
912 	} while (*space > 0 && atomic);
913 out:
914 	*retmp = top;
915 	return (error);
916 }
917 #endif /*ZERO_COPY_SOCKETS*/
918 
919 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
920 
921 int
922 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
923     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
924 {
925 	long space, resid;
926 	int clen = 0, error, dontroute;
927 #ifdef ZERO_COPY_SOCKETS
928 	int atomic = sosendallatonce(so) || top;
929 #endif
930 
931 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
932 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
933 	    ("sodgram_send: !PR_ATOMIC"));
934 
935 	if (uio != NULL)
936 		resid = uio->uio_resid;
937 	else
938 		resid = top->m_pkthdr.len;
939 	/*
940 	 * In theory resid should be unsigned.  However, space must be
941 	 * signed, as it might be less than 0 if we over-committed, and we
942 	 * must use a signed comparison of space and resid.  On the other
943 	 * hand, a negative resid causes us to loop sending 0-length
944 	 * segments to the protocol.
945 	 *
946 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
947 	 * type sockets since that's an error.
948 	 */
949 	if (resid < 0) {
950 		error = EINVAL;
951 		goto out;
952 	}
953 
954 	dontroute =
955 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
956 	if (td != NULL)
957 		td->td_ru.ru_msgsnd++;
958 	if (control != NULL)
959 		clen = control->m_len;
960 
961 	SOCKBUF_LOCK(&so->so_snd);
962 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
963 		SOCKBUF_UNLOCK(&so->so_snd);
964 		error = EPIPE;
965 		goto out;
966 	}
967 	if (so->so_error) {
968 		error = so->so_error;
969 		so->so_error = 0;
970 		SOCKBUF_UNLOCK(&so->so_snd);
971 		goto out;
972 	}
973 	if ((so->so_state & SS_ISCONNECTED) == 0) {
974 		/*
975 		 * `sendto' and `sendmsg' is allowed on a connection-based
976 		 * socket if it supports implied connect.  Return ENOTCONN if
977 		 * not connected and no address is supplied.
978 		 */
979 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
980 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
981 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
982 			    !(resid == 0 && clen != 0)) {
983 				SOCKBUF_UNLOCK(&so->so_snd);
984 				error = ENOTCONN;
985 				goto out;
986 			}
987 		} else if (addr == NULL) {
988 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
989 				error = ENOTCONN;
990 			else
991 				error = EDESTADDRREQ;
992 			SOCKBUF_UNLOCK(&so->so_snd);
993 			goto out;
994 		}
995 	}
996 
997 	/*
998 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
999 	 * problem and need fixing.
1000 	 */
1001 	space = sbspace(&so->so_snd);
1002 	if (flags & MSG_OOB)
1003 		space += 1024;
1004 	space -= clen;
1005 	SOCKBUF_UNLOCK(&so->so_snd);
1006 	if (resid > space) {
1007 		error = EMSGSIZE;
1008 		goto out;
1009 	}
1010 	if (uio == NULL) {
1011 		resid = 0;
1012 		if (flags & MSG_EOR)
1013 			top->m_flags |= M_EOR;
1014 	} else {
1015 #ifdef ZERO_COPY_SOCKETS
1016 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1017 		if (error)
1018 			goto out;
1019 #else
1020 		/*
1021 		 * Copy the data from userland into a mbuf chain.
1022 		 * If no data is to be copied in, a single empty mbuf
1023 		 * is returned.
1024 		 */
1025 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1026 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1027 		if (top == NULL) {
1028 			error = EFAULT;	/* only possible error */
1029 			goto out;
1030 		}
1031 		space -= resid - uio->uio_resid;
1032 #endif
1033 		resid = uio->uio_resid;
1034 	}
1035 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1036 	/*
1037 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1038 	 * than with.
1039 	 */
1040 	if (dontroute) {
1041 		SOCK_LOCK(so);
1042 		so->so_options |= SO_DONTROUTE;
1043 		SOCK_UNLOCK(so);
1044 	}
1045 	/*
1046 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1047 	 * of date.  We could have recieved a reset packet in an interrupt or
1048 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1049 	 * probably recheck again inside the locking protection here, but
1050 	 * there are probably other places that this also happens.  We must
1051 	 * rethink this.
1052 	 */
1053 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1054 	    (flags & MSG_OOB) ? PRUS_OOB :
1055 	/*
1056 	 * If the user set MSG_EOF, the protocol understands this flag and
1057 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1058 	 */
1059 	    ((flags & MSG_EOF) &&
1060 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1061 	     (resid <= 0)) ?
1062 		PRUS_EOF :
1063 		/* If there is more to send set PRUS_MORETOCOME */
1064 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1065 		top, addr, control, td);
1066 	if (dontroute) {
1067 		SOCK_LOCK(so);
1068 		so->so_options &= ~SO_DONTROUTE;
1069 		SOCK_UNLOCK(so);
1070 	}
1071 	clen = 0;
1072 	control = NULL;
1073 	top = NULL;
1074 out:
1075 	if (top != NULL)
1076 		m_freem(top);
1077 	if (control != NULL)
1078 		m_freem(control);
1079 	return (error);
1080 }
1081 
1082 /*
1083  * Send on a socket.  If send must go all at once and message is larger than
1084  * send buffering, then hard error.  Lock against other senders.  If must go
1085  * all at once and not enough room now, then inform user that this would
1086  * block and do nothing.  Otherwise, if nonblocking, send as much as
1087  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1088  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1089  * in mbuf chain must be small enough to send all at once.
1090  *
1091  * Returns nonzero on error, timeout or signal; callers must check for short
1092  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1093  * on return.
1094  */
1095 int
1096 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1097     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1098 {
1099 	long space, resid;
1100 	int clen = 0, error, dontroute;
1101 	int atomic = sosendallatonce(so) || top;
1102 
1103 	if (uio != NULL)
1104 		resid = uio->uio_resid;
1105 	else
1106 		resid = top->m_pkthdr.len;
1107 	/*
1108 	 * In theory resid should be unsigned.  However, space must be
1109 	 * signed, as it might be less than 0 if we over-committed, and we
1110 	 * must use a signed comparison of space and resid.  On the other
1111 	 * hand, a negative resid causes us to loop sending 0-length
1112 	 * segments to the protocol.
1113 	 *
1114 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1115 	 * type sockets since that's an error.
1116 	 */
1117 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1118 		error = EINVAL;
1119 		goto out;
1120 	}
1121 
1122 	dontroute =
1123 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1124 	    (so->so_proto->pr_flags & PR_ATOMIC);
1125 	if (td != NULL)
1126 		td->td_ru.ru_msgsnd++;
1127 	if (control != NULL)
1128 		clen = control->m_len;
1129 
1130 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1131 	if (error)
1132 		goto out;
1133 
1134 restart:
1135 	do {
1136 		SOCKBUF_LOCK(&so->so_snd);
1137 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1138 			SOCKBUF_UNLOCK(&so->so_snd);
1139 			error = EPIPE;
1140 			goto release;
1141 		}
1142 		if (so->so_error) {
1143 			error = so->so_error;
1144 			so->so_error = 0;
1145 			SOCKBUF_UNLOCK(&so->so_snd);
1146 			goto release;
1147 		}
1148 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1149 			/*
1150 			 * `sendto' and `sendmsg' is allowed on a connection-
1151 			 * based socket if it supports implied connect.
1152 			 * Return ENOTCONN if not connected and no address is
1153 			 * supplied.
1154 			 */
1155 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1156 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1157 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1158 				    !(resid == 0 && clen != 0)) {
1159 					SOCKBUF_UNLOCK(&so->so_snd);
1160 					error = ENOTCONN;
1161 					goto release;
1162 				}
1163 			} else if (addr == NULL) {
1164 				SOCKBUF_UNLOCK(&so->so_snd);
1165 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1166 					error = ENOTCONN;
1167 				else
1168 					error = EDESTADDRREQ;
1169 				goto release;
1170 			}
1171 		}
1172 		space = sbspace(&so->so_snd);
1173 		if (flags & MSG_OOB)
1174 			space += 1024;
1175 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1176 		    clen > so->so_snd.sb_hiwat) {
1177 			SOCKBUF_UNLOCK(&so->so_snd);
1178 			error = EMSGSIZE;
1179 			goto release;
1180 		}
1181 		if (space < resid + clen &&
1182 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1183 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1184 				SOCKBUF_UNLOCK(&so->so_snd);
1185 				error = EWOULDBLOCK;
1186 				goto release;
1187 			}
1188 			error = sbwait(&so->so_snd);
1189 			SOCKBUF_UNLOCK(&so->so_snd);
1190 			if (error)
1191 				goto release;
1192 			goto restart;
1193 		}
1194 		SOCKBUF_UNLOCK(&so->so_snd);
1195 		space -= clen;
1196 		do {
1197 			if (uio == NULL) {
1198 				resid = 0;
1199 				if (flags & MSG_EOR)
1200 					top->m_flags |= M_EOR;
1201 			} else {
1202 #ifdef ZERO_COPY_SOCKETS
1203 				error = sosend_copyin(uio, &top, atomic,
1204 				    &space, flags);
1205 				if (error != 0)
1206 					goto release;
1207 #else
1208 				/*
1209 				 * Copy the data from userland into a mbuf
1210 				 * chain.  If no data is to be copied in,
1211 				 * a single empty mbuf is returned.
1212 				 */
1213 				top = m_uiotombuf(uio, M_WAITOK, space,
1214 				    (atomic ? max_hdr : 0),
1215 				    (atomic ? M_PKTHDR : 0) |
1216 				    ((flags & MSG_EOR) ? M_EOR : 0));
1217 				if (top == NULL) {
1218 					error = EFAULT; /* only possible error */
1219 					goto release;
1220 				}
1221 				space -= resid - uio->uio_resid;
1222 #endif
1223 				resid = uio->uio_resid;
1224 			}
1225 			if (dontroute) {
1226 				SOCK_LOCK(so);
1227 				so->so_options |= SO_DONTROUTE;
1228 				SOCK_UNLOCK(so);
1229 			}
1230 			/*
1231 			 * XXX all the SBS_CANTSENDMORE checks previously
1232 			 * done could be out of date.  We could have recieved
1233 			 * a reset packet in an interrupt or maybe we slept
1234 			 * while doing page faults in uiomove() etc.  We
1235 			 * could probably recheck again inside the locking
1236 			 * protection here, but there are probably other
1237 			 * places that this also happens.  We must rethink
1238 			 * this.
1239 			 */
1240 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1241 			    (flags & MSG_OOB) ? PRUS_OOB :
1242 			/*
1243 			 * If the user set MSG_EOF, the protocol understands
1244 			 * this flag and nothing left to send then use
1245 			 * PRU_SEND_EOF instead of PRU_SEND.
1246 			 */
1247 			    ((flags & MSG_EOF) &&
1248 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1249 			     (resid <= 0)) ?
1250 				PRUS_EOF :
1251 			/* If there is more to send set PRUS_MORETOCOME. */
1252 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1253 			    top, addr, control, td);
1254 			if (dontroute) {
1255 				SOCK_LOCK(so);
1256 				so->so_options &= ~SO_DONTROUTE;
1257 				SOCK_UNLOCK(so);
1258 			}
1259 			clen = 0;
1260 			control = NULL;
1261 			top = NULL;
1262 			if (error)
1263 				goto release;
1264 		} while (resid && space > 0);
1265 	} while (resid);
1266 
1267 release:
1268 	sbunlock(&so->so_snd);
1269 out:
1270 	if (top != NULL)
1271 		m_freem(top);
1272 	if (control != NULL)
1273 		m_freem(control);
1274 	return (error);
1275 }
1276 
1277 int
1278 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1279     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1280 {
1281 
1282 	/* XXXRW: Temporary debugging. */
1283 	KASSERT(so->so_proto->pr_usrreqs->pru_sosend != sosend,
1284 	    ("sosend: protocol calls sosend"));
1285 
1286 	return (so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1287 	    control, flags, td));
1288 }
1289 
1290 /*
1291  * The part of soreceive() that implements reading non-inline out-of-band
1292  * data from a socket.  For more complete comments, see soreceive(), from
1293  * which this code originated.
1294  *
1295  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1296  * unable to return an mbuf chain to the caller.
1297  */
1298 static int
1299 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1300 {
1301 	struct protosw *pr = so->so_proto;
1302 	struct mbuf *m;
1303 	int error;
1304 
1305 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1306 
1307 	m = m_get(M_TRYWAIT, MT_DATA);
1308 	if (m == NULL)
1309 		return (ENOBUFS);
1310 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1311 	if (error)
1312 		goto bad;
1313 	do {
1314 #ifdef ZERO_COPY_SOCKETS
1315 		if (so_zero_copy_receive) {
1316 			int disposable;
1317 
1318 			if ((m->m_flags & M_EXT)
1319 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1320 				disposable = 1;
1321 			else
1322 				disposable = 0;
1323 
1324 			error = uiomoveco(mtod(m, void *),
1325 					  min(uio->uio_resid, m->m_len),
1326 					  uio, disposable);
1327 		} else
1328 #endif /* ZERO_COPY_SOCKETS */
1329 		error = uiomove(mtod(m, void *),
1330 		    (int) min(uio->uio_resid, m->m_len), uio);
1331 		m = m_free(m);
1332 	} while (uio->uio_resid && error == 0 && m);
1333 bad:
1334 	if (m != NULL)
1335 		m_freem(m);
1336 	return (error);
1337 }
1338 
1339 /*
1340  * Following replacement or removal of the first mbuf on the first mbuf chain
1341  * of a socket buffer, push necessary state changes back into the socket
1342  * buffer so that other consumers see the values consistently.  'nextrecord'
1343  * is the callers locally stored value of the original value of
1344  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1345  * NOTE: 'nextrecord' may be NULL.
1346  */
1347 static __inline void
1348 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1349 {
1350 
1351 	SOCKBUF_LOCK_ASSERT(sb);
1352 	/*
1353 	 * First, update for the new value of nextrecord.  If necessary, make
1354 	 * it the first record.
1355 	 */
1356 	if (sb->sb_mb != NULL)
1357 		sb->sb_mb->m_nextpkt = nextrecord;
1358 	else
1359 		sb->sb_mb = nextrecord;
1360 
1361         /*
1362          * Now update any dependent socket buffer fields to reflect the new
1363          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1364 	 * addition of a second clause that takes care of the case where
1365 	 * sb_mb has been updated, but remains the last record.
1366          */
1367         if (sb->sb_mb == NULL) {
1368                 sb->sb_mbtail = NULL;
1369                 sb->sb_lastrecord = NULL;
1370         } else if (sb->sb_mb->m_nextpkt == NULL)
1371                 sb->sb_lastrecord = sb->sb_mb;
1372 }
1373 
1374 
1375 /*
1376  * Implement receive operations on a socket.  We depend on the way that
1377  * records are added to the sockbuf by sbappend.  In particular, each record
1378  * (mbufs linked through m_next) must begin with an address if the protocol
1379  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1380  * data, and then zero or more mbufs of data.  In order to allow parallelism
1381  * between network receive and copying to user space, as well as avoid
1382  * sleeping with a mutex held, we release the socket buffer mutex during the
1383  * user space copy.  Although the sockbuf is locked, new data may still be
1384  * appended, and thus we must maintain consistency of the sockbuf during that
1385  * time.
1386  *
1387  * The caller may receive the data as a single mbuf chain by supplying an
1388  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1389  * the count in uio_resid.
1390  */
1391 int
1392 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1393     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1394 {
1395 	struct mbuf *m, **mp;
1396 	int flags, len, error, offset;
1397 	struct protosw *pr = so->so_proto;
1398 	struct mbuf *nextrecord;
1399 	int moff, type = 0;
1400 	int orig_resid = uio->uio_resid;
1401 
1402 	mp = mp0;
1403 	if (psa != NULL)
1404 		*psa = NULL;
1405 	if (controlp != NULL)
1406 		*controlp = NULL;
1407 	if (flagsp != NULL)
1408 		flags = *flagsp &~ MSG_EOR;
1409 	else
1410 		flags = 0;
1411 	if (flags & MSG_OOB)
1412 		return (soreceive_rcvoob(so, uio, flags));
1413 	if (mp != NULL)
1414 		*mp = NULL;
1415 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1416 	    && uio->uio_resid)
1417 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1418 
1419 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1420 	if (error)
1421 		return (error);
1422 
1423 restart:
1424 	SOCKBUF_LOCK(&so->so_rcv);
1425 	m = so->so_rcv.sb_mb;
1426 	/*
1427 	 * If we have less data than requested, block awaiting more (subject
1428 	 * to any timeout) if:
1429 	 *   1. the current count is less than the low water mark, or
1430 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1431 	 *	receive operation at once if we block (resid <= hiwat).
1432 	 *   3. MSG_DONTWAIT is not set
1433 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1434 	 * we have to do the receive in sections, and thus risk returning a
1435 	 * short count if a timeout or signal occurs after we start.
1436 	 */
1437 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1438 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1439 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1440 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1441 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1442 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1443 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1444 		    m, so->so_rcv.sb_cc));
1445 		if (so->so_error) {
1446 			if (m != NULL)
1447 				goto dontblock;
1448 			error = so->so_error;
1449 			if ((flags & MSG_PEEK) == 0)
1450 				so->so_error = 0;
1451 			SOCKBUF_UNLOCK(&so->so_rcv);
1452 			goto release;
1453 		}
1454 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1455 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1456 			if (m == NULL) {
1457 				SOCKBUF_UNLOCK(&so->so_rcv);
1458 				goto release;
1459 			} else
1460 				goto dontblock;
1461 		}
1462 		for (; m != NULL; m = m->m_next)
1463 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1464 				m = so->so_rcv.sb_mb;
1465 				goto dontblock;
1466 			}
1467 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1468 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1469 			SOCKBUF_UNLOCK(&so->so_rcv);
1470 			error = ENOTCONN;
1471 			goto release;
1472 		}
1473 		if (uio->uio_resid == 0) {
1474 			SOCKBUF_UNLOCK(&so->so_rcv);
1475 			goto release;
1476 		}
1477 		if ((so->so_state & SS_NBIO) ||
1478 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1479 			SOCKBUF_UNLOCK(&so->so_rcv);
1480 			error = EWOULDBLOCK;
1481 			goto release;
1482 		}
1483 		SBLASTRECORDCHK(&so->so_rcv);
1484 		SBLASTMBUFCHK(&so->so_rcv);
1485 		error = sbwait(&so->so_rcv);
1486 		SOCKBUF_UNLOCK(&so->so_rcv);
1487 		if (error)
1488 			goto release;
1489 		goto restart;
1490 	}
1491 dontblock:
1492 	/*
1493 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1494 	 * pointer to the next record in the socket buffer.  We must keep the
1495 	 * various socket buffer pointers and local stack versions of the
1496 	 * pointers in sync, pushing out modifications before dropping the
1497 	 * socket buffer mutex, and re-reading them when picking it up.
1498 	 *
1499 	 * Otherwise, we will race with the network stack appending new data
1500 	 * or records onto the socket buffer by using inconsistent/stale
1501 	 * versions of the field, possibly resulting in socket buffer
1502 	 * corruption.
1503 	 *
1504 	 * By holding the high-level sblock(), we prevent simultaneous
1505 	 * readers from pulling off the front of the socket buffer.
1506 	 */
1507 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1508 	if (uio->uio_td)
1509 		uio->uio_td->td_ru.ru_msgrcv++;
1510 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1511 	SBLASTRECORDCHK(&so->so_rcv);
1512 	SBLASTMBUFCHK(&so->so_rcv);
1513 	nextrecord = m->m_nextpkt;
1514 	if (pr->pr_flags & PR_ADDR) {
1515 		KASSERT(m->m_type == MT_SONAME,
1516 		    ("m->m_type == %d", m->m_type));
1517 		orig_resid = 0;
1518 		if (psa != NULL)
1519 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1520 			    M_NOWAIT);
1521 		if (flags & MSG_PEEK) {
1522 			m = m->m_next;
1523 		} else {
1524 			sbfree(&so->so_rcv, m);
1525 			so->so_rcv.sb_mb = m_free(m);
1526 			m = so->so_rcv.sb_mb;
1527 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1528 		}
1529 	}
1530 
1531 	/*
1532 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1533 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1534 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1535 	 * perform externalization (or freeing if controlp == NULL).
1536 	 */
1537 	if (m != NULL && m->m_type == MT_CONTROL) {
1538 		struct mbuf *cm = NULL, *cmn;
1539 		struct mbuf **cme = &cm;
1540 
1541 		do {
1542 			if (flags & MSG_PEEK) {
1543 				if (controlp != NULL) {
1544 					*controlp = m_copy(m, 0, m->m_len);
1545 					controlp = &(*controlp)->m_next;
1546 				}
1547 				m = m->m_next;
1548 			} else {
1549 				sbfree(&so->so_rcv, m);
1550 				so->so_rcv.sb_mb = m->m_next;
1551 				m->m_next = NULL;
1552 				*cme = m;
1553 				cme = &(*cme)->m_next;
1554 				m = so->so_rcv.sb_mb;
1555 			}
1556 		} while (m != NULL && m->m_type == MT_CONTROL);
1557 		if ((flags & MSG_PEEK) == 0)
1558 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1559 		while (cm != NULL) {
1560 			cmn = cm->m_next;
1561 			cm->m_next = NULL;
1562 			if (pr->pr_domain->dom_externalize != NULL) {
1563 				SOCKBUF_UNLOCK(&so->so_rcv);
1564 				error = (*pr->pr_domain->dom_externalize)
1565 				    (cm, controlp);
1566 				SOCKBUF_LOCK(&so->so_rcv);
1567 			} else if (controlp != NULL)
1568 				*controlp = cm;
1569 			else
1570 				m_freem(cm);
1571 			if (controlp != NULL) {
1572 				orig_resid = 0;
1573 				while (*controlp != NULL)
1574 					controlp = &(*controlp)->m_next;
1575 			}
1576 			cm = cmn;
1577 		}
1578 		if (m != NULL)
1579 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1580 		else
1581 			nextrecord = so->so_rcv.sb_mb;
1582 		orig_resid = 0;
1583 	}
1584 	if (m != NULL) {
1585 		if ((flags & MSG_PEEK) == 0) {
1586 			KASSERT(m->m_nextpkt == nextrecord,
1587 			    ("soreceive: post-control, nextrecord !sync"));
1588 			if (nextrecord == NULL) {
1589 				KASSERT(so->so_rcv.sb_mb == m,
1590 				    ("soreceive: post-control, sb_mb!=m"));
1591 				KASSERT(so->so_rcv.sb_lastrecord == m,
1592 				    ("soreceive: post-control, lastrecord!=m"));
1593 			}
1594 		}
1595 		type = m->m_type;
1596 		if (type == MT_OOBDATA)
1597 			flags |= MSG_OOB;
1598 	} else {
1599 		if ((flags & MSG_PEEK) == 0) {
1600 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1601 			    ("soreceive: sb_mb != nextrecord"));
1602 			if (so->so_rcv.sb_mb == NULL) {
1603 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1604 				    ("soreceive: sb_lastercord != NULL"));
1605 			}
1606 		}
1607 	}
1608 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1609 	SBLASTRECORDCHK(&so->so_rcv);
1610 	SBLASTMBUFCHK(&so->so_rcv);
1611 
1612 	/*
1613 	 * Now continue to read any data mbufs off of the head of the socket
1614 	 * buffer until the read request is satisfied.  Note that 'type' is
1615 	 * used to store the type of any mbuf reads that have happened so far
1616 	 * such that soreceive() can stop reading if the type changes, which
1617 	 * causes soreceive() to return only one of regular data and inline
1618 	 * out-of-band data in a single socket receive operation.
1619 	 */
1620 	moff = 0;
1621 	offset = 0;
1622 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1623 		/*
1624 		 * If the type of mbuf has changed since the last mbuf
1625 		 * examined ('type'), end the receive operation.
1626 	 	 */
1627 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1628 		if (m->m_type == MT_OOBDATA) {
1629 			if (type != MT_OOBDATA)
1630 				break;
1631 		} else if (type == MT_OOBDATA)
1632 			break;
1633 		else
1634 		    KASSERT(m->m_type == MT_DATA,
1635 			("m->m_type == %d", m->m_type));
1636 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1637 		len = uio->uio_resid;
1638 		if (so->so_oobmark && len > so->so_oobmark - offset)
1639 			len = so->so_oobmark - offset;
1640 		if (len > m->m_len - moff)
1641 			len = m->m_len - moff;
1642 		/*
1643 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1644 		 * them out via the uio, then free.  Sockbuf must be
1645 		 * consistent here (points to current mbuf, it points to next
1646 		 * record) when we drop priority; we must note any additions
1647 		 * to the sockbuf when we block interrupts again.
1648 		 */
1649 		if (mp == NULL) {
1650 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1651 			SBLASTRECORDCHK(&so->so_rcv);
1652 			SBLASTMBUFCHK(&so->so_rcv);
1653 			SOCKBUF_UNLOCK(&so->so_rcv);
1654 #ifdef ZERO_COPY_SOCKETS
1655 			if (so_zero_copy_receive) {
1656 				int disposable;
1657 
1658 				if ((m->m_flags & M_EXT)
1659 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1660 					disposable = 1;
1661 				else
1662 					disposable = 0;
1663 
1664 				error = uiomoveco(mtod(m, char *) + moff,
1665 						  (int)len, uio,
1666 						  disposable);
1667 			} else
1668 #endif /* ZERO_COPY_SOCKETS */
1669 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1670 			SOCKBUF_LOCK(&so->so_rcv);
1671 			if (error) {
1672 				/*
1673 				 * The MT_SONAME mbuf has already been removed
1674 				 * from the record, so it is necessary to
1675 				 * remove the data mbufs, if any, to preserve
1676 				 * the invariant in the case of PR_ADDR that
1677 				 * requires MT_SONAME mbufs at the head of
1678 				 * each record.
1679 				 */
1680 				if (m && pr->pr_flags & PR_ATOMIC &&
1681 				    ((flags & MSG_PEEK) == 0))
1682 					(void)sbdroprecord_locked(&so->so_rcv);
1683 				SOCKBUF_UNLOCK(&so->so_rcv);
1684 				goto release;
1685 			}
1686 		} else
1687 			uio->uio_resid -= len;
1688 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1689 		if (len == m->m_len - moff) {
1690 			if (m->m_flags & M_EOR)
1691 				flags |= MSG_EOR;
1692 			if (flags & MSG_PEEK) {
1693 				m = m->m_next;
1694 				moff = 0;
1695 			} else {
1696 				nextrecord = m->m_nextpkt;
1697 				sbfree(&so->so_rcv, m);
1698 				if (mp != NULL) {
1699 					*mp = m;
1700 					mp = &m->m_next;
1701 					so->so_rcv.sb_mb = m = m->m_next;
1702 					*mp = NULL;
1703 				} else {
1704 					so->so_rcv.sb_mb = m_free(m);
1705 					m = so->so_rcv.sb_mb;
1706 				}
1707 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1708 				SBLASTRECORDCHK(&so->so_rcv);
1709 				SBLASTMBUFCHK(&so->so_rcv);
1710 			}
1711 		} else {
1712 			if (flags & MSG_PEEK)
1713 				moff += len;
1714 			else {
1715 				if (mp != NULL) {
1716 					int copy_flag;
1717 
1718 					if (flags & MSG_DONTWAIT)
1719 						copy_flag = M_DONTWAIT;
1720 					else
1721 						copy_flag = M_TRYWAIT;
1722 					if (copy_flag == M_TRYWAIT)
1723 						SOCKBUF_UNLOCK(&so->so_rcv);
1724 					*mp = m_copym(m, 0, len, copy_flag);
1725 					if (copy_flag == M_TRYWAIT)
1726 						SOCKBUF_LOCK(&so->so_rcv);
1727  					if (*mp == NULL) {
1728  						/*
1729  						 * m_copym() couldn't
1730 						 * allocate an mbuf.  Adjust
1731 						 * uio_resid back (it was
1732 						 * adjusted down by len
1733 						 * bytes, which we didn't end
1734 						 * up "copying" over).
1735  						 */
1736  						uio->uio_resid += len;
1737  						break;
1738  					}
1739 				}
1740 				m->m_data += len;
1741 				m->m_len -= len;
1742 				so->so_rcv.sb_cc -= len;
1743 			}
1744 		}
1745 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1746 		if (so->so_oobmark) {
1747 			if ((flags & MSG_PEEK) == 0) {
1748 				so->so_oobmark -= len;
1749 				if (so->so_oobmark == 0) {
1750 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1751 					break;
1752 				}
1753 			} else {
1754 				offset += len;
1755 				if (offset == so->so_oobmark)
1756 					break;
1757 			}
1758 		}
1759 		if (flags & MSG_EOR)
1760 			break;
1761 		/*
1762 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1763 		 * must not quit until "uio->uio_resid == 0" or an error
1764 		 * termination.  If a signal/timeout occurs, return with a
1765 		 * short count but without error.  Keep sockbuf locked
1766 		 * against other readers.
1767 		 */
1768 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1769 		    !sosendallatonce(so) && nextrecord == NULL) {
1770 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1771 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1772 				break;
1773 			/*
1774 			 * Notify the protocol that some data has been
1775 			 * drained before blocking.
1776 			 */
1777 			if (pr->pr_flags & PR_WANTRCVD) {
1778 				SOCKBUF_UNLOCK(&so->so_rcv);
1779 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1780 				SOCKBUF_LOCK(&so->so_rcv);
1781 			}
1782 			SBLASTRECORDCHK(&so->so_rcv);
1783 			SBLASTMBUFCHK(&so->so_rcv);
1784 			error = sbwait(&so->so_rcv);
1785 			if (error) {
1786 				SOCKBUF_UNLOCK(&so->so_rcv);
1787 				goto release;
1788 			}
1789 			m = so->so_rcv.sb_mb;
1790 			if (m != NULL)
1791 				nextrecord = m->m_nextpkt;
1792 		}
1793 	}
1794 
1795 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1796 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1797 		flags |= MSG_TRUNC;
1798 		if ((flags & MSG_PEEK) == 0)
1799 			(void) sbdroprecord_locked(&so->so_rcv);
1800 	}
1801 	if ((flags & MSG_PEEK) == 0) {
1802 		if (m == NULL) {
1803 			/*
1804 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1805 			 * part makes sure sb_lastrecord is up-to-date if
1806 			 * there is still data in the socket buffer.
1807 			 */
1808 			so->so_rcv.sb_mb = nextrecord;
1809 			if (so->so_rcv.sb_mb == NULL) {
1810 				so->so_rcv.sb_mbtail = NULL;
1811 				so->so_rcv.sb_lastrecord = NULL;
1812 			} else if (nextrecord->m_nextpkt == NULL)
1813 				so->so_rcv.sb_lastrecord = nextrecord;
1814 		}
1815 		SBLASTRECORDCHK(&so->so_rcv);
1816 		SBLASTMBUFCHK(&so->so_rcv);
1817 		/*
1818 		 * If soreceive() is being done from the socket callback,
1819 		 * then don't need to generate ACK to peer to update window,
1820 		 * since ACK will be generated on return to TCP.
1821 		 */
1822 		if (!(flags & MSG_SOCALLBCK) &&
1823 		    (pr->pr_flags & PR_WANTRCVD)) {
1824 			SOCKBUF_UNLOCK(&so->so_rcv);
1825 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1826 			SOCKBUF_LOCK(&so->so_rcv);
1827 		}
1828 	}
1829 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1830 	if (orig_resid == uio->uio_resid && orig_resid &&
1831 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1832 		SOCKBUF_UNLOCK(&so->so_rcv);
1833 		goto restart;
1834 	}
1835 	SOCKBUF_UNLOCK(&so->so_rcv);
1836 
1837 	if (flagsp != NULL)
1838 		*flagsp |= flags;
1839 release:
1840 	sbunlock(&so->so_rcv);
1841 	return (error);
1842 }
1843 
1844 int
1845 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
1846     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1847 {
1848 
1849 	/* XXXRW: Temporary debugging. */
1850 	KASSERT(so->so_proto->pr_usrreqs->pru_soreceive != soreceive,
1851 	    ("soreceive: protocol calls soreceive"));
1852 
1853 	return (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
1854 	    controlp, flagsp));
1855 }
1856 
1857 int
1858 soshutdown(struct socket *so, int how)
1859 {
1860 	struct protosw *pr = so->so_proto;
1861 
1862 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
1863 		return (EINVAL);
1864 
1865 	if (how != SHUT_WR)
1866 		sorflush(so);
1867 	if (how != SHUT_RD)
1868 		return ((*pr->pr_usrreqs->pru_shutdown)(so));
1869 	return (0);
1870 }
1871 
1872 void
1873 sorflush(struct socket *so)
1874 {
1875 	struct sockbuf *sb = &so->so_rcv;
1876 	struct protosw *pr = so->so_proto;
1877 	struct sockbuf asb;
1878 
1879 	/*
1880 	 * In order to avoid calling dom_dispose with the socket buffer mutex
1881 	 * held, and in order to generally avoid holding the lock for a long
1882 	 * time, we make a copy of the socket buffer and clear the original
1883 	 * (except locks, state).  The new socket buffer copy won't have
1884 	 * initialized locks so we can only call routines that won't use or
1885 	 * assert those locks.
1886 	 *
1887 	 * Dislodge threads currently blocked in receive and wait to acquire
1888 	 * a lock against other simultaneous readers before clearing the
1889 	 * socket buffer.  Don't let our acquire be interrupted by a signal
1890 	 * despite any existing socket disposition on interruptable waiting.
1891 	 */
1892 	socantrcvmore(so);
1893 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
1894 
1895 	/*
1896 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
1897 	 * and mutex data unchanged.
1898 	 */
1899 	SOCKBUF_LOCK(sb);
1900 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
1901 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
1902 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1903 	bzero(&sb->sb_startzero,
1904 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
1905 	SOCKBUF_UNLOCK(sb);
1906 	sbunlock(sb);
1907 
1908 	/*
1909 	 * Dispose of special rights and flush the socket buffer.  Don't call
1910 	 * any unsafe routines (that rely on locks being initialized) on asb.
1911 	 */
1912 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1913 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
1914 	sbrelease_internal(&asb, so);
1915 }
1916 
1917 /*
1918  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
1919  * additional variant to handle the case where the option value needs to be
1920  * some kind of integer, but not a specific size.  In addition to their use
1921  * here, these functions are also called by the protocol-level pr_ctloutput()
1922  * routines.
1923  */
1924 int
1925 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
1926 {
1927 	size_t	valsize;
1928 
1929 	/*
1930 	 * If the user gives us more than we wanted, we ignore it, but if we
1931 	 * don't get the minimum length the caller wants, we return EINVAL.
1932 	 * On success, sopt->sopt_valsize is set to however much we actually
1933 	 * retrieved.
1934 	 */
1935 	if ((valsize = sopt->sopt_valsize) < minlen)
1936 		return EINVAL;
1937 	if (valsize > len)
1938 		sopt->sopt_valsize = valsize = len;
1939 
1940 	if (sopt->sopt_td != NULL)
1941 		return (copyin(sopt->sopt_val, buf, valsize));
1942 
1943 	bcopy(sopt->sopt_val, buf, valsize);
1944 	return (0);
1945 }
1946 
1947 /*
1948  * Kernel version of setsockopt(2).
1949  *
1950  * XXX: optlen is size_t, not socklen_t
1951  */
1952 int
1953 so_setsockopt(struct socket *so, int level, int optname, void *optval,
1954     size_t optlen)
1955 {
1956 	struct sockopt sopt;
1957 
1958 	sopt.sopt_level = level;
1959 	sopt.sopt_name = optname;
1960 	sopt.sopt_dir = SOPT_SET;
1961 	sopt.sopt_val = optval;
1962 	sopt.sopt_valsize = optlen;
1963 	sopt.sopt_td = NULL;
1964 	return (sosetopt(so, &sopt));
1965 }
1966 
1967 int
1968 sosetopt(struct socket *so, struct sockopt *sopt)
1969 {
1970 	int	error, optval;
1971 	struct	linger l;
1972 	struct	timeval tv;
1973 	u_long  val;
1974 #ifdef MAC
1975 	struct mac extmac;
1976 #endif
1977 
1978 	error = 0;
1979 	if (sopt->sopt_level != SOL_SOCKET) {
1980 		if (so->so_proto && so->so_proto->pr_ctloutput)
1981 			return ((*so->so_proto->pr_ctloutput)
1982 				  (so, sopt));
1983 		error = ENOPROTOOPT;
1984 	} else {
1985 		switch (sopt->sopt_name) {
1986 #ifdef INET
1987 		case SO_ACCEPTFILTER:
1988 			error = do_setopt_accept_filter(so, sopt);
1989 			if (error)
1990 				goto bad;
1991 			break;
1992 #endif
1993 		case SO_LINGER:
1994 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
1995 			if (error)
1996 				goto bad;
1997 
1998 			SOCK_LOCK(so);
1999 			so->so_linger = l.l_linger;
2000 			if (l.l_onoff)
2001 				so->so_options |= SO_LINGER;
2002 			else
2003 				so->so_options &= ~SO_LINGER;
2004 			SOCK_UNLOCK(so);
2005 			break;
2006 
2007 		case SO_DEBUG:
2008 		case SO_KEEPALIVE:
2009 		case SO_DONTROUTE:
2010 		case SO_USELOOPBACK:
2011 		case SO_BROADCAST:
2012 		case SO_REUSEADDR:
2013 		case SO_REUSEPORT:
2014 		case SO_OOBINLINE:
2015 		case SO_TIMESTAMP:
2016 		case SO_BINTIME:
2017 		case SO_NOSIGPIPE:
2018 			error = sooptcopyin(sopt, &optval, sizeof optval,
2019 					    sizeof optval);
2020 			if (error)
2021 				goto bad;
2022 			SOCK_LOCK(so);
2023 			if (optval)
2024 				so->so_options |= sopt->sopt_name;
2025 			else
2026 				so->so_options &= ~sopt->sopt_name;
2027 			SOCK_UNLOCK(so);
2028 			break;
2029 
2030 		case SO_SNDBUF:
2031 		case SO_RCVBUF:
2032 		case SO_SNDLOWAT:
2033 		case SO_RCVLOWAT:
2034 			error = sooptcopyin(sopt, &optval, sizeof optval,
2035 					    sizeof optval);
2036 			if (error)
2037 				goto bad;
2038 
2039 			/*
2040 			 * Values < 1 make no sense for any of these options,
2041 			 * so disallow them.
2042 			 */
2043 			if (optval < 1) {
2044 				error = EINVAL;
2045 				goto bad;
2046 			}
2047 
2048 			switch (sopt->sopt_name) {
2049 			case SO_SNDBUF:
2050 			case SO_RCVBUF:
2051 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2052 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2053 				    so, curthread) == 0) {
2054 					error = ENOBUFS;
2055 					goto bad;
2056 				}
2057 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2058 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2059 				break;
2060 
2061 			/*
2062 			 * Make sure the low-water is never greater than the
2063 			 * high-water.
2064 			 */
2065 			case SO_SNDLOWAT:
2066 				SOCKBUF_LOCK(&so->so_snd);
2067 				so->so_snd.sb_lowat =
2068 				    (optval > so->so_snd.sb_hiwat) ?
2069 				    so->so_snd.sb_hiwat : optval;
2070 				SOCKBUF_UNLOCK(&so->so_snd);
2071 				break;
2072 			case SO_RCVLOWAT:
2073 				SOCKBUF_LOCK(&so->so_rcv);
2074 				so->so_rcv.sb_lowat =
2075 				    (optval > so->so_rcv.sb_hiwat) ?
2076 				    so->so_rcv.sb_hiwat : optval;
2077 				SOCKBUF_UNLOCK(&so->so_rcv);
2078 				break;
2079 			}
2080 			break;
2081 
2082 		case SO_SNDTIMEO:
2083 		case SO_RCVTIMEO:
2084 #ifdef COMPAT_IA32
2085 			if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2086 				struct timeval32 tv32;
2087 
2088 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2089 				    sizeof tv32);
2090 				CP(tv32, tv, tv_sec);
2091 				CP(tv32, tv, tv_usec);
2092 			} else
2093 #endif
2094 				error = sooptcopyin(sopt, &tv, sizeof tv,
2095 				    sizeof tv);
2096 			if (error)
2097 				goto bad;
2098 
2099 			/* assert(hz > 0); */
2100 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2101 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2102 				error = EDOM;
2103 				goto bad;
2104 			}
2105 			/* assert(tick > 0); */
2106 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2107 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2108 			if (val > INT_MAX) {
2109 				error = EDOM;
2110 				goto bad;
2111 			}
2112 			if (val == 0 && tv.tv_usec != 0)
2113 				val = 1;
2114 
2115 			switch (sopt->sopt_name) {
2116 			case SO_SNDTIMEO:
2117 				so->so_snd.sb_timeo = val;
2118 				break;
2119 			case SO_RCVTIMEO:
2120 				so->so_rcv.sb_timeo = val;
2121 				break;
2122 			}
2123 			break;
2124 
2125 		case SO_LABEL:
2126 #ifdef MAC
2127 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2128 			    sizeof extmac);
2129 			if (error)
2130 				goto bad;
2131 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2132 			    so, &extmac);
2133 #else
2134 			error = EOPNOTSUPP;
2135 #endif
2136 			break;
2137 
2138 		default:
2139 			error = ENOPROTOOPT;
2140 			break;
2141 		}
2142 		if (error == 0 && so->so_proto != NULL &&
2143 		    so->so_proto->pr_ctloutput != NULL) {
2144 			(void) ((*so->so_proto->pr_ctloutput)
2145 				  (so, sopt));
2146 		}
2147 	}
2148 bad:
2149 	return (error);
2150 }
2151 
2152 /*
2153  * Helper routine for getsockopt.
2154  */
2155 int
2156 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2157 {
2158 	int	error;
2159 	size_t	valsize;
2160 
2161 	error = 0;
2162 
2163 	/*
2164 	 * Documented get behavior is that we always return a value, possibly
2165 	 * truncated to fit in the user's buffer.  Traditional behavior is
2166 	 * that we always tell the user precisely how much we copied, rather
2167 	 * than something useful like the total amount we had available for
2168 	 * her.  Note that this interface is not idempotent; the entire
2169 	 * answer must generated ahead of time.
2170 	 */
2171 	valsize = min(len, sopt->sopt_valsize);
2172 	sopt->sopt_valsize = valsize;
2173 	if (sopt->sopt_val != NULL) {
2174 		if (sopt->sopt_td != NULL)
2175 			error = copyout(buf, sopt->sopt_val, valsize);
2176 		else
2177 			bcopy(buf, sopt->sopt_val, valsize);
2178 	}
2179 	return (error);
2180 }
2181 
2182 int
2183 sogetopt(struct socket *so, struct sockopt *sopt)
2184 {
2185 	int	error, optval;
2186 	struct	linger l;
2187 	struct	timeval tv;
2188 #ifdef MAC
2189 	struct mac extmac;
2190 #endif
2191 
2192 	error = 0;
2193 	if (sopt->sopt_level != SOL_SOCKET) {
2194 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2195 			return ((*so->so_proto->pr_ctloutput)
2196 				  (so, sopt));
2197 		} else
2198 			return (ENOPROTOOPT);
2199 	} else {
2200 		switch (sopt->sopt_name) {
2201 #ifdef INET
2202 		case SO_ACCEPTFILTER:
2203 			error = do_getopt_accept_filter(so, sopt);
2204 			break;
2205 #endif
2206 		case SO_LINGER:
2207 			SOCK_LOCK(so);
2208 			l.l_onoff = so->so_options & SO_LINGER;
2209 			l.l_linger = so->so_linger;
2210 			SOCK_UNLOCK(so);
2211 			error = sooptcopyout(sopt, &l, sizeof l);
2212 			break;
2213 
2214 		case SO_USELOOPBACK:
2215 		case SO_DONTROUTE:
2216 		case SO_DEBUG:
2217 		case SO_KEEPALIVE:
2218 		case SO_REUSEADDR:
2219 		case SO_REUSEPORT:
2220 		case SO_BROADCAST:
2221 		case SO_OOBINLINE:
2222 		case SO_ACCEPTCONN:
2223 		case SO_TIMESTAMP:
2224 		case SO_BINTIME:
2225 		case SO_NOSIGPIPE:
2226 			optval = so->so_options & sopt->sopt_name;
2227 integer:
2228 			error = sooptcopyout(sopt, &optval, sizeof optval);
2229 			break;
2230 
2231 		case SO_TYPE:
2232 			optval = so->so_type;
2233 			goto integer;
2234 
2235 		case SO_ERROR:
2236 			SOCK_LOCK(so);
2237 			optval = so->so_error;
2238 			so->so_error = 0;
2239 			SOCK_UNLOCK(so);
2240 			goto integer;
2241 
2242 		case SO_SNDBUF:
2243 			optval = so->so_snd.sb_hiwat;
2244 			goto integer;
2245 
2246 		case SO_RCVBUF:
2247 			optval = so->so_rcv.sb_hiwat;
2248 			goto integer;
2249 
2250 		case SO_SNDLOWAT:
2251 			optval = so->so_snd.sb_lowat;
2252 			goto integer;
2253 
2254 		case SO_RCVLOWAT:
2255 			optval = so->so_rcv.sb_lowat;
2256 			goto integer;
2257 
2258 		case SO_SNDTIMEO:
2259 		case SO_RCVTIMEO:
2260 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2261 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2262 
2263 			tv.tv_sec = optval / hz;
2264 			tv.tv_usec = (optval % hz) * tick;
2265 #ifdef COMPAT_IA32
2266 			if (curthread->td_proc->p_sysent == &ia32_freebsd_sysvec) {
2267 				struct timeval32 tv32;
2268 
2269 				CP(tv, tv32, tv_sec);
2270 				CP(tv, tv32, tv_usec);
2271 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2272 			} else
2273 #endif
2274 				error = sooptcopyout(sopt, &tv, sizeof tv);
2275 			break;
2276 
2277 		case SO_LABEL:
2278 #ifdef MAC
2279 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2280 			    sizeof(extmac));
2281 			if (error)
2282 				return (error);
2283 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2284 			    so, &extmac);
2285 			if (error)
2286 				return (error);
2287 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2288 #else
2289 			error = EOPNOTSUPP;
2290 #endif
2291 			break;
2292 
2293 		case SO_PEERLABEL:
2294 #ifdef MAC
2295 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2296 			    sizeof(extmac));
2297 			if (error)
2298 				return (error);
2299 			error = mac_getsockopt_peerlabel(
2300 			    sopt->sopt_td->td_ucred, so, &extmac);
2301 			if (error)
2302 				return (error);
2303 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2304 #else
2305 			error = EOPNOTSUPP;
2306 #endif
2307 			break;
2308 
2309 		case SO_LISTENQLIMIT:
2310 			optval = so->so_qlimit;
2311 			goto integer;
2312 
2313 		case SO_LISTENQLEN:
2314 			optval = so->so_qlen;
2315 			goto integer;
2316 
2317 		case SO_LISTENINCQLEN:
2318 			optval = so->so_incqlen;
2319 			goto integer;
2320 
2321 		default:
2322 			error = ENOPROTOOPT;
2323 			break;
2324 		}
2325 		return (error);
2326 	}
2327 }
2328 
2329 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2330 int
2331 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2332 {
2333 	struct mbuf *m, *m_prev;
2334 	int sopt_size = sopt->sopt_valsize;
2335 
2336 	MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2337 	if (m == NULL)
2338 		return ENOBUFS;
2339 	if (sopt_size > MLEN) {
2340 		MCLGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT);
2341 		if ((m->m_flags & M_EXT) == 0) {
2342 			m_free(m);
2343 			return ENOBUFS;
2344 		}
2345 		m->m_len = min(MCLBYTES, sopt_size);
2346 	} else {
2347 		m->m_len = min(MLEN, sopt_size);
2348 	}
2349 	sopt_size -= m->m_len;
2350 	*mp = m;
2351 	m_prev = m;
2352 
2353 	while (sopt_size) {
2354 		MGET(m, sopt->sopt_td ? M_TRYWAIT : M_DONTWAIT, MT_DATA);
2355 		if (m == NULL) {
2356 			m_freem(*mp);
2357 			return ENOBUFS;
2358 		}
2359 		if (sopt_size > MLEN) {
2360 			MCLGET(m, sopt->sopt_td != NULL ? M_TRYWAIT :
2361 			    M_DONTWAIT);
2362 			if ((m->m_flags & M_EXT) == 0) {
2363 				m_freem(m);
2364 				m_freem(*mp);
2365 				return ENOBUFS;
2366 			}
2367 			m->m_len = min(MCLBYTES, sopt_size);
2368 		} else {
2369 			m->m_len = min(MLEN, sopt_size);
2370 		}
2371 		sopt_size -= m->m_len;
2372 		m_prev->m_next = m;
2373 		m_prev = m;
2374 	}
2375 	return (0);
2376 }
2377 
2378 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2379 int
2380 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2381 {
2382 	struct mbuf *m0 = m;
2383 
2384 	if (sopt->sopt_val == NULL)
2385 		return (0);
2386 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2387 		if (sopt->sopt_td != NULL) {
2388 			int error;
2389 
2390 			error = copyin(sopt->sopt_val, mtod(m, char *),
2391 				       m->m_len);
2392 			if (error != 0) {
2393 				m_freem(m0);
2394 				return(error);
2395 			}
2396 		} else
2397 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2398 		sopt->sopt_valsize -= m->m_len;
2399 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2400 		m = m->m_next;
2401 	}
2402 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2403 		panic("ip6_sooptmcopyin");
2404 	return (0);
2405 }
2406 
2407 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2408 int
2409 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2410 {
2411 	struct mbuf *m0 = m;
2412 	size_t valsize = 0;
2413 
2414 	if (sopt->sopt_val == NULL)
2415 		return (0);
2416 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2417 		if (sopt->sopt_td != NULL) {
2418 			int error;
2419 
2420 			error = copyout(mtod(m, char *), sopt->sopt_val,
2421 				       m->m_len);
2422 			if (error != 0) {
2423 				m_freem(m0);
2424 				return(error);
2425 			}
2426 		} else
2427 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2428 	       sopt->sopt_valsize -= m->m_len;
2429 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2430 	       valsize += m->m_len;
2431 	       m = m->m_next;
2432 	}
2433 	if (m != NULL) {
2434 		/* enough soopt buffer should be given from user-land */
2435 		m_freem(m0);
2436 		return(EINVAL);
2437 	}
2438 	sopt->sopt_valsize = valsize;
2439 	return (0);
2440 }
2441 
2442 /*
2443  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2444  * out-of-band data, which will then notify socket consumers.
2445  */
2446 void
2447 sohasoutofband(struct socket *so)
2448 {
2449 
2450 	if (so->so_sigio != NULL)
2451 		pgsigio(&so->so_sigio, SIGURG, 0);
2452 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2453 }
2454 
2455 int
2456 sopoll(struct socket *so, int events, struct ucred *active_cred,
2457     struct thread *td)
2458 {
2459 
2460 	/* XXXRW: Temporary debugging. */
2461 	KASSERT(so->so_proto->pr_usrreqs->pru_sopoll != sopoll,
2462 	    ("sopoll: protocol calls sopoll"));
2463 
2464 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2465 	    td));
2466 }
2467 
2468 int
2469 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2470     struct thread *td)
2471 {
2472 	int revents = 0;
2473 
2474 	SOCKBUF_LOCK(&so->so_snd);
2475 	SOCKBUF_LOCK(&so->so_rcv);
2476 	if (events & (POLLIN | POLLRDNORM))
2477 		if (soreadable(so))
2478 			revents |= events & (POLLIN | POLLRDNORM);
2479 
2480 	if (events & POLLINIGNEOF)
2481 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat ||
2482 		    !TAILQ_EMPTY(&so->so_comp) || so->so_error)
2483 			revents |= POLLINIGNEOF;
2484 
2485 	if (events & (POLLOUT | POLLWRNORM))
2486 		if (sowriteable(so))
2487 			revents |= events & (POLLOUT | POLLWRNORM);
2488 
2489 	if (events & (POLLPRI | POLLRDBAND))
2490 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2491 			revents |= events & (POLLPRI | POLLRDBAND);
2492 
2493 	if (revents == 0) {
2494 		if (events &
2495 		    (POLLIN | POLLINIGNEOF | POLLPRI | POLLRDNORM |
2496 		     POLLRDBAND)) {
2497 			selrecord(td, &so->so_rcv.sb_sel);
2498 			so->so_rcv.sb_flags |= SB_SEL;
2499 		}
2500 
2501 		if (events & (POLLOUT | POLLWRNORM)) {
2502 			selrecord(td, &so->so_snd.sb_sel);
2503 			so->so_snd.sb_flags |= SB_SEL;
2504 		}
2505 	}
2506 
2507 	SOCKBUF_UNLOCK(&so->so_rcv);
2508 	SOCKBUF_UNLOCK(&so->so_snd);
2509 	return (revents);
2510 }
2511 
2512 int
2513 soo_kqfilter(struct file *fp, struct knote *kn)
2514 {
2515 	struct socket *so = kn->kn_fp->f_data;
2516 	struct sockbuf *sb;
2517 
2518 	switch (kn->kn_filter) {
2519 	case EVFILT_READ:
2520 		if (so->so_options & SO_ACCEPTCONN)
2521 			kn->kn_fop = &solisten_filtops;
2522 		else
2523 			kn->kn_fop = &soread_filtops;
2524 		sb = &so->so_rcv;
2525 		break;
2526 	case EVFILT_WRITE:
2527 		kn->kn_fop = &sowrite_filtops;
2528 		sb = &so->so_snd;
2529 		break;
2530 	default:
2531 		return (EINVAL);
2532 	}
2533 
2534 	SOCKBUF_LOCK(sb);
2535 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2536 	sb->sb_flags |= SB_KNOTE;
2537 	SOCKBUF_UNLOCK(sb);
2538 	return (0);
2539 }
2540 
2541 /*
2542  * Some routines that return EOPNOTSUPP for entry points that are not
2543  * supported by a protocol.  Fill in as needed.
2544  */
2545 int
2546 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2547 {
2548 
2549 	return EOPNOTSUPP;
2550 }
2551 
2552 int
2553 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2554 {
2555 
2556 	return EOPNOTSUPP;
2557 }
2558 
2559 int
2560 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2561 {
2562 
2563 	return EOPNOTSUPP;
2564 }
2565 
2566 int
2567 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2568 {
2569 
2570 	return EOPNOTSUPP;
2571 }
2572 
2573 int
2574 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
2575 {
2576 
2577 	return EOPNOTSUPP;
2578 }
2579 
2580 int
2581 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
2582     struct ifnet *ifp, struct thread *td)
2583 {
2584 
2585 	return EOPNOTSUPP;
2586 }
2587 
2588 int
2589 pru_disconnect_notsupp(struct socket *so)
2590 {
2591 
2592 	return EOPNOTSUPP;
2593 }
2594 
2595 int
2596 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
2597 {
2598 
2599 	return EOPNOTSUPP;
2600 }
2601 
2602 int
2603 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
2604 {
2605 
2606 	return EOPNOTSUPP;
2607 }
2608 
2609 int
2610 pru_rcvd_notsupp(struct socket *so, int flags)
2611 {
2612 
2613 	return EOPNOTSUPP;
2614 }
2615 
2616 int
2617 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
2618 {
2619 
2620 	return EOPNOTSUPP;
2621 }
2622 
2623 int
2624 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
2625     struct sockaddr *addr, struct mbuf *control, struct thread *td)
2626 {
2627 
2628 	return EOPNOTSUPP;
2629 }
2630 
2631 /*
2632  * This isn't really a ``null'' operation, but it's the default one and
2633  * doesn't do anything destructive.
2634  */
2635 int
2636 pru_sense_null(struct socket *so, struct stat *sb)
2637 {
2638 
2639 	sb->st_blksize = so->so_snd.sb_hiwat;
2640 	return 0;
2641 }
2642 
2643 int
2644 pru_shutdown_notsupp(struct socket *so)
2645 {
2646 
2647 	return EOPNOTSUPP;
2648 }
2649 
2650 int
2651 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
2652 {
2653 
2654 	return EOPNOTSUPP;
2655 }
2656 
2657 int
2658 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
2659     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
2660 {
2661 
2662 	return EOPNOTSUPP;
2663 }
2664 
2665 int
2666 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
2667     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2668 {
2669 
2670 	return EOPNOTSUPP;
2671 }
2672 
2673 int
2674 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
2675     struct thread *td)
2676 {
2677 
2678 	return EOPNOTSUPP;
2679 }
2680 
2681 static void
2682 filt_sordetach(struct knote *kn)
2683 {
2684 	struct socket *so = kn->kn_fp->f_data;
2685 
2686 	SOCKBUF_LOCK(&so->so_rcv);
2687 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
2688 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
2689 		so->so_rcv.sb_flags &= ~SB_KNOTE;
2690 	SOCKBUF_UNLOCK(&so->so_rcv);
2691 }
2692 
2693 /*ARGSUSED*/
2694 static int
2695 filt_soread(struct knote *kn, long hint)
2696 {
2697 	struct socket *so;
2698 
2699 	so = kn->kn_fp->f_data;
2700 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2701 
2702 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
2703 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2704 		kn->kn_flags |= EV_EOF;
2705 		kn->kn_fflags = so->so_error;
2706 		return (1);
2707 	} else if (so->so_error)	/* temporary udp error */
2708 		return (1);
2709 	else if (kn->kn_sfflags & NOTE_LOWAT)
2710 		return (kn->kn_data >= kn->kn_sdata);
2711 	else
2712 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
2713 }
2714 
2715 static void
2716 filt_sowdetach(struct knote *kn)
2717 {
2718 	struct socket *so = kn->kn_fp->f_data;
2719 
2720 	SOCKBUF_LOCK(&so->so_snd);
2721 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
2722 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
2723 		so->so_snd.sb_flags &= ~SB_KNOTE;
2724 	SOCKBUF_UNLOCK(&so->so_snd);
2725 }
2726 
2727 /*ARGSUSED*/
2728 static int
2729 filt_sowrite(struct knote *kn, long hint)
2730 {
2731 	struct socket *so;
2732 
2733 	so = kn->kn_fp->f_data;
2734 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
2735 	kn->kn_data = sbspace(&so->so_snd);
2736 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
2737 		kn->kn_flags |= EV_EOF;
2738 		kn->kn_fflags = so->so_error;
2739 		return (1);
2740 	} else if (so->so_error)	/* temporary udp error */
2741 		return (1);
2742 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
2743 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
2744 		return (0);
2745 	else if (kn->kn_sfflags & NOTE_LOWAT)
2746 		return (kn->kn_data >= kn->kn_sdata);
2747 	else
2748 		return (kn->kn_data >= so->so_snd.sb_lowat);
2749 }
2750 
2751 /*ARGSUSED*/
2752 static int
2753 filt_solisten(struct knote *kn, long hint)
2754 {
2755 	struct socket *so = kn->kn_fp->f_data;
2756 
2757 	kn->kn_data = so->so_qlen;
2758 	return (! TAILQ_EMPTY(&so->so_comp));
2759 }
2760 
2761 int
2762 socheckuid(struct socket *so, uid_t uid)
2763 {
2764 
2765 	if (so == NULL)
2766 		return (EPERM);
2767 	if (so->so_cred->cr_uid != uid)
2768 		return (EPERM);
2769 	return (0);
2770 }
2771 
2772 static int
2773 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
2774 {
2775 	int error;
2776 	int val;
2777 
2778 	val = somaxconn;
2779 	error = sysctl_handle_int(oidp, &val, 0, req);
2780 	if (error || !req->newptr )
2781 		return (error);
2782 
2783 	if (val < 1 || val > USHRT_MAX)
2784 		return (EINVAL);
2785 
2786 	somaxconn = val;
2787 	return (0);
2788 }
2789 
2790 /*
2791  * These functions are used by protocols to notify the socket layer (and its
2792  * consumers) of state changes in the sockets driven by protocol-side events.
2793  */
2794 
2795 /*
2796  * Procedures to manipulate state flags of socket and do appropriate wakeups.
2797  *
2798  * Normal sequence from the active (originating) side is that
2799  * soisconnecting() is called during processing of connect() call, resulting
2800  * in an eventual call to soisconnected() if/when the connection is
2801  * established.  When the connection is torn down soisdisconnecting() is
2802  * called during processing of disconnect() call, and soisdisconnected() is
2803  * called when the connection to the peer is totally severed.  The semantics
2804  * of these routines are such that connectionless protocols can call
2805  * soisconnected() and soisdisconnected() only, bypassing the in-progress
2806  * calls when setting up a ``connection'' takes no time.
2807  *
2808  * From the passive side, a socket is created with two queues of sockets:
2809  * so_incomp for connections in progress and so_comp for connections already
2810  * made and awaiting user acceptance.  As a protocol is preparing incoming
2811  * connections, it creates a socket structure queued on so_incomp by calling
2812  * sonewconn().  When the connection is established, soisconnected() is
2813  * called, and transfers the socket structure to so_comp, making it available
2814  * to accept().
2815  *
2816  * If a socket is closed with sockets on either so_incomp or so_comp, these
2817  * sockets are dropped.
2818  *
2819  * If higher-level protocols are implemented in the kernel, the wakeups done
2820  * here will sometimes cause software-interrupt process scheduling.
2821  */
2822 void
2823 soisconnecting(struct socket *so)
2824 {
2825 
2826 	SOCK_LOCK(so);
2827 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
2828 	so->so_state |= SS_ISCONNECTING;
2829 	SOCK_UNLOCK(so);
2830 }
2831 
2832 void
2833 soisconnected(struct socket *so)
2834 {
2835 	struct socket *head;
2836 
2837 	ACCEPT_LOCK();
2838 	SOCK_LOCK(so);
2839 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
2840 	so->so_state |= SS_ISCONNECTED;
2841 	head = so->so_head;
2842 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
2843 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
2844 			SOCK_UNLOCK(so);
2845 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
2846 			head->so_incqlen--;
2847 			so->so_qstate &= ~SQ_INCOMP;
2848 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
2849 			head->so_qlen++;
2850 			so->so_qstate |= SQ_COMP;
2851 			ACCEPT_UNLOCK();
2852 			sorwakeup(head);
2853 			wakeup_one(&head->so_timeo);
2854 		} else {
2855 			ACCEPT_UNLOCK();
2856 			so->so_upcall =
2857 			    head->so_accf->so_accept_filter->accf_callback;
2858 			so->so_upcallarg = head->so_accf->so_accept_filter_arg;
2859 			so->so_rcv.sb_flags |= SB_UPCALL;
2860 			so->so_options &= ~SO_ACCEPTFILTER;
2861 			SOCK_UNLOCK(so);
2862 			so->so_upcall(so, so->so_upcallarg, M_DONTWAIT);
2863 		}
2864 		return;
2865 	}
2866 	SOCK_UNLOCK(so);
2867 	ACCEPT_UNLOCK();
2868 	wakeup(&so->so_timeo);
2869 	sorwakeup(so);
2870 	sowwakeup(so);
2871 }
2872 
2873 void
2874 soisdisconnecting(struct socket *so)
2875 {
2876 
2877 	/*
2878 	 * Note: This code assumes that SOCK_LOCK(so) and
2879 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
2880 	 */
2881 	SOCKBUF_LOCK(&so->so_rcv);
2882 	so->so_state &= ~SS_ISCONNECTING;
2883 	so->so_state |= SS_ISDISCONNECTING;
2884 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
2885 	sorwakeup_locked(so);
2886 	SOCKBUF_LOCK(&so->so_snd);
2887 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
2888 	sowwakeup_locked(so);
2889 	wakeup(&so->so_timeo);
2890 }
2891 
2892 void
2893 soisdisconnected(struct socket *so)
2894 {
2895 
2896 	/*
2897 	 * Note: This code assumes that SOCK_LOCK(so) and
2898 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
2899 	 */
2900 	SOCKBUF_LOCK(&so->so_rcv);
2901 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
2902 	so->so_state |= SS_ISDISCONNECTED;
2903 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
2904 	sorwakeup_locked(so);
2905 	SOCKBUF_LOCK(&so->so_snd);
2906 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
2907 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
2908 	sowwakeup_locked(so);
2909 	wakeup(&so->so_timeo);
2910 }
2911 
2912 /*
2913  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
2914  */
2915 struct sockaddr *
2916 sodupsockaddr(const struct sockaddr *sa, int mflags)
2917 {
2918 	struct sockaddr *sa2;
2919 
2920 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
2921 	if (sa2)
2922 		bcopy(sa, sa2, sa->sa_len);
2923 	return sa2;
2924 }
2925 
2926 /*
2927  * Create an external-format (``xsocket'') structure using the information in
2928  * the kernel-format socket structure pointed to by so.  This is done to
2929  * reduce the spew of irrelevant information over this interface, to isolate
2930  * user code from changes in the kernel structure, and potentially to provide
2931  * information-hiding if we decide that some of this information should be
2932  * hidden from users.
2933  */
2934 void
2935 sotoxsocket(struct socket *so, struct xsocket *xso)
2936 {
2937 
2938 	xso->xso_len = sizeof *xso;
2939 	xso->xso_so = so;
2940 	xso->so_type = so->so_type;
2941 	xso->so_options = so->so_options;
2942 	xso->so_linger = so->so_linger;
2943 	xso->so_state = so->so_state;
2944 	xso->so_pcb = so->so_pcb;
2945 	xso->xso_protocol = so->so_proto->pr_protocol;
2946 	xso->xso_family = so->so_proto->pr_domain->dom_family;
2947 	xso->so_qlen = so->so_qlen;
2948 	xso->so_incqlen = so->so_incqlen;
2949 	xso->so_qlimit = so->so_qlimit;
2950 	xso->so_timeo = so->so_timeo;
2951 	xso->so_error = so->so_error;
2952 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
2953 	xso->so_oobmark = so->so_oobmark;
2954 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
2955 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
2956 	xso->so_uid = so->so_cred->cr_uid;
2957 }
2958