xref: /freebsd/sys/kern/uipc_socket.c (revision 641a6cfb86023499caafe26a4d821a0b885cf00b)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004 The FreeBSD Foundation
5  * Copyright (c) 2004-2008 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
33  */
34 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  *
96  * NOTE: With regard to VNETs the general rule is that callers do not set
97  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99  * and sorflush(), which are usually called from a pre-set VNET context.
100  * sopoll() currently does not need a VNET context to be set.
101  */
102 
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
105 
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_zero.h"
109 #include "opt_compat.h"
110 
111 #include <sys/param.h>
112 #include <sys/systm.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
116 #include <sys/mac.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h>			/* for struct knote */
122 #include <sys/kernel.h>
123 #include <sys/event.h>
124 #include <sys/eventhandler.h>
125 #include <sys/poll.h>
126 #include <sys/proc.h>
127 #include <sys/protosw.h>
128 #include <sys/socket.h>
129 #include <sys/socketvar.h>
130 #include <sys/resourcevar.h>
131 #include <net/route.h>
132 #include <sys/signalvar.h>
133 #include <sys/stat.h>
134 #include <sys/sx.h>
135 #include <sys/sysctl.h>
136 #include <sys/uio.h>
137 #include <sys/jail.h>
138 
139 #include <net/vnet.h>
140 
141 #include <security/mac/mac_framework.h>
142 
143 #include <vm/uma.h>
144 
145 #ifdef COMPAT_FREEBSD32
146 #include <sys/mount.h>
147 #include <sys/sysent.h>
148 #include <compat/freebsd32/freebsd32.h>
149 #endif
150 
151 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
152 		    int flags);
153 
154 static void	filt_sordetach(struct knote *kn);
155 static int	filt_soread(struct knote *kn, long hint);
156 static void	filt_sowdetach(struct knote *kn);
157 static int	filt_sowrite(struct knote *kn, long hint);
158 static int	filt_solisten(struct knote *kn, long hint);
159 
160 static struct filterops solisten_filtops = {
161 	.f_isfd = 1,
162 	.f_detach = filt_sordetach,
163 	.f_event = filt_solisten,
164 };
165 static struct filterops soread_filtops = {
166 	.f_isfd = 1,
167 	.f_detach = filt_sordetach,
168 	.f_event = filt_soread,
169 };
170 static struct filterops sowrite_filtops = {
171 	.f_isfd = 1,
172 	.f_detach = filt_sowdetach,
173 	.f_event = filt_sowrite,
174 };
175 
176 uma_zone_t socket_zone;
177 so_gen_t	so_gencnt;	/* generation count for sockets */
178 
179 int	maxsockets;
180 
181 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
182 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
183 
184 #define	VNET_SO_ASSERT(so)						\
185 	VNET_ASSERT(curvnet != NULL,					\
186 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
187 
188 static int somaxconn = SOMAXCONN;
189 static int sysctl_somaxconn(SYSCTL_HANDLER_ARGS);
190 /* XXX: we dont have SYSCTL_USHORT */
191 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn, CTLTYPE_UINT | CTLFLAG_RW,
192     0, sizeof(int), sysctl_somaxconn, "I", "Maximum pending socket connection "
193     "queue size");
194 static int numopensockets;
195 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
196     &numopensockets, 0, "Number of open sockets");
197 #ifdef ZERO_COPY_SOCKETS
198 /* These aren't static because they're used in other files. */
199 int so_zero_copy_send = 1;
200 int so_zero_copy_receive = 1;
201 SYSCTL_NODE(_kern_ipc, OID_AUTO, zero_copy, CTLFLAG_RD, 0,
202     "Zero copy controls");
203 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, receive, CTLFLAG_RW,
204     &so_zero_copy_receive, 0, "Enable zero copy receive");
205 SYSCTL_INT(_kern_ipc_zero_copy, OID_AUTO, send, CTLFLAG_RW,
206     &so_zero_copy_send, 0, "Enable zero copy send");
207 #endif /* ZERO_COPY_SOCKETS */
208 
209 /*
210  * accept_mtx locks down per-socket fields relating to accept queues.  See
211  * socketvar.h for an annotation of the protected fields of struct socket.
212  */
213 struct mtx accept_mtx;
214 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
215 
216 /*
217  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
218  * so_gencnt field.
219  */
220 static struct mtx so_global_mtx;
221 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
222 
223 /*
224  * General IPC sysctl name space, used by sockets and a variety of other IPC
225  * types.
226  */
227 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
228 
229 /*
230  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
231  * of the change so that they can update their dependent limits as required.
232  */
233 static int
234 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
235 {
236 	int error, newmaxsockets;
237 
238 	newmaxsockets = maxsockets;
239 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
240 	if (error == 0 && req->newptr) {
241 		if (newmaxsockets > maxsockets) {
242 			maxsockets = newmaxsockets;
243 			if (maxsockets > ((maxfiles / 4) * 3)) {
244 				maxfiles = (maxsockets * 5) / 4;
245 				maxfilesperproc = (maxfiles * 9) / 10;
246 			}
247 			EVENTHANDLER_INVOKE(maxsockets_change);
248 		} else
249 			error = EINVAL;
250 	}
251 	return (error);
252 }
253 
254 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
255     &maxsockets, 0, sysctl_maxsockets, "IU",
256     "Maximum number of sockets avaliable");
257 
258 /*
259  * Initialise maxsockets.  This SYSINIT must be run after
260  * tunable_mbinit().
261  */
262 static void
263 init_maxsockets(void *ignored)
264 {
265 
266 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
267 	maxsockets = imax(maxsockets, imax(maxfiles, nmbclusters));
268 }
269 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
270 
271 /*
272  * Socket operation routines.  These routines are called by the routines in
273  * sys_socket.c or from a system process, and implement the semantics of
274  * socket operations by switching out to the protocol specific routines.
275  */
276 
277 /*
278  * Get a socket structure from our zone, and initialize it.  Note that it
279  * would probably be better to allocate socket and PCB at the same time, but
280  * I'm not convinced that all the protocols can be easily modified to do
281  * this.
282  *
283  * soalloc() returns a socket with a ref count of 0.
284  */
285 static struct socket *
286 soalloc(struct vnet *vnet)
287 {
288 	struct socket *so;
289 
290 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
291 	if (so == NULL)
292 		return (NULL);
293 #ifdef MAC
294 	if (mac_socket_init(so, M_NOWAIT) != 0) {
295 		uma_zfree(socket_zone, so);
296 		return (NULL);
297 	}
298 #endif
299 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
300 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
301 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
302 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
303 	TAILQ_INIT(&so->so_aiojobq);
304 	mtx_lock(&so_global_mtx);
305 	so->so_gencnt = ++so_gencnt;
306 	++numopensockets;
307 #ifdef VIMAGE
308 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
309 	    __func__, __LINE__, so));
310 	vnet->vnet_sockcnt++;
311 	so->so_vnet = vnet;
312 #endif
313 	mtx_unlock(&so_global_mtx);
314 	return (so);
315 }
316 
317 /*
318  * Free the storage associated with a socket at the socket layer, tear down
319  * locks, labels, etc.  All protocol state is assumed already to have been
320  * torn down (and possibly never set up) by the caller.
321  */
322 static void
323 sodealloc(struct socket *so)
324 {
325 
326 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
327 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
328 
329 	mtx_lock(&so_global_mtx);
330 	so->so_gencnt = ++so_gencnt;
331 	--numopensockets;	/* Could be below, but faster here. */
332 #ifdef VIMAGE
333 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
334 	    __func__, __LINE__, so));
335 	so->so_vnet->vnet_sockcnt--;
336 #endif
337 	mtx_unlock(&so_global_mtx);
338 	if (so->so_rcv.sb_hiwat)
339 		(void)chgsbsize(so->so_cred->cr_uidinfo,
340 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
341 	if (so->so_snd.sb_hiwat)
342 		(void)chgsbsize(so->so_cred->cr_uidinfo,
343 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
344 #ifdef INET
345 	/* remove acccept filter if one is present. */
346 	if (so->so_accf != NULL)
347 		do_setopt_accept_filter(so, NULL);
348 #endif
349 #ifdef MAC
350 	mac_socket_destroy(so);
351 #endif
352 	crfree(so->so_cred);
353 	sx_destroy(&so->so_snd.sb_sx);
354 	sx_destroy(&so->so_rcv.sb_sx);
355 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
356 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
357 	uma_zfree(socket_zone, so);
358 }
359 
360 /*
361  * socreate returns a socket with a ref count of 1.  The socket should be
362  * closed with soclose().
363  */
364 int
365 socreate(int dom, struct socket **aso, int type, int proto,
366     struct ucred *cred, struct thread *td)
367 {
368 	struct protosw *prp;
369 	struct socket *so;
370 	int error;
371 
372 	if (proto)
373 		prp = pffindproto(dom, proto, type);
374 	else
375 		prp = pffindtype(dom, type);
376 
377 	if (prp == NULL || prp->pr_usrreqs->pru_attach == NULL ||
378 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
379 		return (EPROTONOSUPPORT);
380 
381 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
382 		return (EPROTONOSUPPORT);
383 
384 	if (prp->pr_type != type)
385 		return (EPROTOTYPE);
386 	so = soalloc(CRED_TO_VNET(cred));
387 	if (so == NULL)
388 		return (ENOBUFS);
389 
390 	TAILQ_INIT(&so->so_incomp);
391 	TAILQ_INIT(&so->so_comp);
392 	so->so_type = type;
393 	so->so_cred = crhold(cred);
394 	if ((prp->pr_domain->dom_family == PF_INET) ||
395 	    (prp->pr_domain->dom_family == PF_INET6) ||
396 	    (prp->pr_domain->dom_family == PF_ROUTE))
397 		so->so_fibnum = td->td_proc->p_fibnum;
398 	else
399 		so->so_fibnum = 0;
400 	so->so_proto = prp;
401 #ifdef MAC
402 	mac_socket_create(cred, so);
403 #endif
404 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
405 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
406 	so->so_count = 1;
407 	/*
408 	 * Auto-sizing of socket buffers is managed by the protocols and
409 	 * the appropriate flags must be set in the pru_attach function.
410 	 */
411 	CURVNET_SET(so->so_vnet);
412 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
413 	CURVNET_RESTORE();
414 	if (error) {
415 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
416 		    so->so_count));
417 		so->so_count = 0;
418 		sodealloc(so);
419 		return (error);
420 	}
421 	*aso = so;
422 	return (0);
423 }
424 
425 #ifdef REGRESSION
426 static int regression_sonewconn_earlytest = 1;
427 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
428     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
429 #endif
430 
431 /*
432  * When an attempt at a new connection is noted on a socket which accepts
433  * connections, sonewconn is called.  If the connection is possible (subject
434  * to space constraints, etc.) then we allocate a new structure, propoerly
435  * linked into the data structure of the original socket, and return this.
436  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
437  *
438  * Note: the ref count on the socket is 0 on return.
439  */
440 struct socket *
441 sonewconn(struct socket *head, int connstatus)
442 {
443 	struct socket *so;
444 	int over;
445 
446 	ACCEPT_LOCK();
447 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
448 	ACCEPT_UNLOCK();
449 #ifdef REGRESSION
450 	if (regression_sonewconn_earlytest && over)
451 #else
452 	if (over)
453 #endif
454 		return (NULL);
455 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
456 	    __func__, __LINE__, head));
457 	so = soalloc(head->so_vnet);
458 	if (so == NULL)
459 		return (NULL);
460 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
461 		connstatus = 0;
462 	so->so_head = head;
463 	so->so_type = head->so_type;
464 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
465 	so->so_linger = head->so_linger;
466 	so->so_state = head->so_state | SS_NOFDREF;
467 	so->so_fibnum = head->so_fibnum;
468 	so->so_proto = head->so_proto;
469 	so->so_cred = crhold(head->so_cred);
470 #ifdef MAC
471 	mac_socket_newconn(head, so);
472 #endif
473 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
474 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
475 	VNET_SO_ASSERT(head);
476 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
477 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
478 		sodealloc(so);
479 		return (NULL);
480 	}
481 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
482 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
483 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
484 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
485 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
486 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
487 	so->so_state |= connstatus;
488 	ACCEPT_LOCK();
489 	if (connstatus) {
490 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
491 		so->so_qstate |= SQ_COMP;
492 		head->so_qlen++;
493 	} else {
494 		/*
495 		 * Keep removing sockets from the head until there's room for
496 		 * us to insert on the tail.  In pre-locking revisions, this
497 		 * was a simple if(), but as we could be racing with other
498 		 * threads and soabort() requires dropping locks, we must
499 		 * loop waiting for the condition to be true.
500 		 */
501 		while (head->so_incqlen > head->so_qlimit) {
502 			struct socket *sp;
503 			sp = TAILQ_FIRST(&head->so_incomp);
504 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
505 			head->so_incqlen--;
506 			sp->so_qstate &= ~SQ_INCOMP;
507 			sp->so_head = NULL;
508 			ACCEPT_UNLOCK();
509 			soabort(sp);
510 			ACCEPT_LOCK();
511 		}
512 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
513 		so->so_qstate |= SQ_INCOMP;
514 		head->so_incqlen++;
515 	}
516 	ACCEPT_UNLOCK();
517 	if (connstatus) {
518 		sorwakeup(head);
519 		wakeup_one(&head->so_timeo);
520 	}
521 	return (so);
522 }
523 
524 int
525 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
526 {
527 	int error;
528 
529 	CURVNET_SET(so->so_vnet);
530 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
531 	CURVNET_RESTORE();
532 	return error;
533 }
534 
535 /*
536  * solisten() transitions a socket from a non-listening state to a listening
537  * state, but can also be used to update the listen queue depth on an
538  * existing listen socket.  The protocol will call back into the sockets
539  * layer using solisten_proto_check() and solisten_proto() to check and set
540  * socket-layer listen state.  Call backs are used so that the protocol can
541  * acquire both protocol and socket layer locks in whatever order is required
542  * by the protocol.
543  *
544  * Protocol implementors are advised to hold the socket lock across the
545  * socket-layer test and set to avoid races at the socket layer.
546  */
547 int
548 solisten(struct socket *so, int backlog, struct thread *td)
549 {
550 	int error;
551 
552 	CURVNET_SET(so->so_vnet);
553 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
554 	CURVNET_RESTORE();
555 	return error;
556 }
557 
558 int
559 solisten_proto_check(struct socket *so)
560 {
561 
562 	SOCK_LOCK_ASSERT(so);
563 
564 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
565 	    SS_ISDISCONNECTING))
566 		return (EINVAL);
567 	return (0);
568 }
569 
570 void
571 solisten_proto(struct socket *so, int backlog)
572 {
573 
574 	SOCK_LOCK_ASSERT(so);
575 
576 	if (backlog < 0 || backlog > somaxconn)
577 		backlog = somaxconn;
578 	so->so_qlimit = backlog;
579 	so->so_options |= SO_ACCEPTCONN;
580 }
581 
582 /*
583  * Evaluate the reference count and named references on a socket; if no
584  * references remain, free it.  This should be called whenever a reference is
585  * released, such as in sorele(), but also when named reference flags are
586  * cleared in socket or protocol code.
587  *
588  * sofree() will free the socket if:
589  *
590  * - There are no outstanding file descriptor references or related consumers
591  *   (so_count == 0).
592  *
593  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
594  *
595  * - The protocol does not have an outstanding strong reference on the socket
596  *   (SS_PROTOREF).
597  *
598  * - The socket is not in a completed connection queue, so a process has been
599  *   notified that it is present.  If it is removed, the user process may
600  *   block in accept() despite select() saying the socket was ready.
601  */
602 void
603 sofree(struct socket *so)
604 {
605 	struct protosw *pr = so->so_proto;
606 	struct socket *head;
607 
608 	ACCEPT_LOCK_ASSERT();
609 	SOCK_LOCK_ASSERT(so);
610 
611 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
612 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
613 		SOCK_UNLOCK(so);
614 		ACCEPT_UNLOCK();
615 		return;
616 	}
617 
618 	head = so->so_head;
619 	if (head != NULL) {
620 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
621 		    (so->so_qstate & SQ_INCOMP) != 0,
622 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
623 		    "SQ_INCOMP"));
624 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
625 		    (so->so_qstate & SQ_INCOMP) == 0,
626 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
627 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
628 		head->so_incqlen--;
629 		so->so_qstate &= ~SQ_INCOMP;
630 		so->so_head = NULL;
631 	}
632 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
633 	    (so->so_qstate & SQ_INCOMP) == 0,
634 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
635 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
636 	if (so->so_options & SO_ACCEPTCONN) {
637 		KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
638 		KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_incomp populated"));
639 	}
640 	SOCK_UNLOCK(so);
641 	ACCEPT_UNLOCK();
642 
643 	VNET_SO_ASSERT(so);
644 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
645 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
646 	if (pr->pr_usrreqs->pru_detach != NULL)
647 		(*pr->pr_usrreqs->pru_detach)(so);
648 
649 	/*
650 	 * From this point on, we assume that no other references to this
651 	 * socket exist anywhere else in the stack.  Therefore, no locks need
652 	 * to be acquired or held.
653 	 *
654 	 * We used to do a lot of socket buffer and socket locking here, as
655 	 * well as invoke sorflush() and perform wakeups.  The direct call to
656 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
657 	 * necessary from sorflush().
658 	 *
659 	 * Notice that the socket buffer and kqueue state are torn down
660 	 * before calling pru_detach.  This means that protocols shold not
661 	 * assume they can perform socket wakeups, etc, in their detach code.
662 	 */
663 	sbdestroy(&so->so_snd, so);
664 	sbdestroy(&so->so_rcv, so);
665 	seldrain(&so->so_snd.sb_sel);
666 	seldrain(&so->so_rcv.sb_sel);
667 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
668 	knlist_destroy(&so->so_snd.sb_sel.si_note);
669 	sodealloc(so);
670 }
671 
672 /*
673  * Close a socket on last file table reference removal.  Initiate disconnect
674  * if connected.  Free socket when disconnect complete.
675  *
676  * This function will sorele() the socket.  Note that soclose() may be called
677  * prior to the ref count reaching zero.  The actual socket structure will
678  * not be freed until the ref count reaches zero.
679  */
680 int
681 soclose(struct socket *so)
682 {
683 	int error = 0;
684 
685 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
686 
687 	CURVNET_SET(so->so_vnet);
688 	funsetown(&so->so_sigio);
689 	if (so->so_state & SS_ISCONNECTED) {
690 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
691 			error = sodisconnect(so);
692 			if (error) {
693 				if (error == ENOTCONN)
694 					error = 0;
695 				goto drop;
696 			}
697 		}
698 		if (so->so_options & SO_LINGER) {
699 			if ((so->so_state & SS_ISDISCONNECTING) &&
700 			    (so->so_state & SS_NBIO))
701 				goto drop;
702 			while (so->so_state & SS_ISCONNECTED) {
703 				error = tsleep(&so->so_timeo,
704 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
705 				if (error)
706 					break;
707 			}
708 		}
709 	}
710 
711 drop:
712 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
713 		(*so->so_proto->pr_usrreqs->pru_close)(so);
714 	if (so->so_options & SO_ACCEPTCONN) {
715 		struct socket *sp;
716 		ACCEPT_LOCK();
717 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
718 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
719 			so->so_incqlen--;
720 			sp->so_qstate &= ~SQ_INCOMP;
721 			sp->so_head = NULL;
722 			ACCEPT_UNLOCK();
723 			soabort(sp);
724 			ACCEPT_LOCK();
725 		}
726 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
727 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
728 			so->so_qlen--;
729 			sp->so_qstate &= ~SQ_COMP;
730 			sp->so_head = NULL;
731 			ACCEPT_UNLOCK();
732 			soabort(sp);
733 			ACCEPT_LOCK();
734 		}
735 		ACCEPT_UNLOCK();
736 	}
737 	ACCEPT_LOCK();
738 	SOCK_LOCK(so);
739 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
740 	so->so_state |= SS_NOFDREF;
741 	sorele(so);
742 	CURVNET_RESTORE();
743 	return (error);
744 }
745 
746 /*
747  * soabort() is used to abruptly tear down a connection, such as when a
748  * resource limit is reached (listen queue depth exceeded), or if a listen
749  * socket is closed while there are sockets waiting to be accepted.
750  *
751  * This interface is tricky, because it is called on an unreferenced socket,
752  * and must be called only by a thread that has actually removed the socket
753  * from the listen queue it was on, or races with other threads are risked.
754  *
755  * This interface will call into the protocol code, so must not be called
756  * with any socket locks held.  Protocols do call it while holding their own
757  * recursible protocol mutexes, but this is something that should be subject
758  * to review in the future.
759  */
760 void
761 soabort(struct socket *so)
762 {
763 
764 	/*
765 	 * In as much as is possible, assert that no references to this
766 	 * socket are held.  This is not quite the same as asserting that the
767 	 * current thread is responsible for arranging for no references, but
768 	 * is as close as we can get for now.
769 	 */
770 	KASSERT(so->so_count == 0, ("soabort: so_count"));
771 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
772 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
773 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
774 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
775 	VNET_SO_ASSERT(so);
776 
777 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
778 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
779 	ACCEPT_LOCK();
780 	SOCK_LOCK(so);
781 	sofree(so);
782 }
783 
784 int
785 soaccept(struct socket *so, struct sockaddr **nam)
786 {
787 	int error;
788 
789 	SOCK_LOCK(so);
790 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
791 	so->so_state &= ~SS_NOFDREF;
792 	SOCK_UNLOCK(so);
793 
794 	CURVNET_SET(so->so_vnet);
795 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
796 	CURVNET_RESTORE();
797 	return (error);
798 }
799 
800 int
801 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
802 {
803 	int error;
804 
805 	if (so->so_options & SO_ACCEPTCONN)
806 		return (EOPNOTSUPP);
807 
808 	CURVNET_SET(so->so_vnet);
809 	/*
810 	 * If protocol is connection-based, can only connect once.
811 	 * Otherwise, if connected, try to disconnect first.  This allows
812 	 * user to disconnect by connecting to, e.g., a null address.
813 	 */
814 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
815 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
816 	    (error = sodisconnect(so)))) {
817 		error = EISCONN;
818 	} else {
819 		/*
820 		 * Prevent accumulated error from previous connection from
821 		 * biting us.
822 		 */
823 		so->so_error = 0;
824 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
825 	}
826 	CURVNET_RESTORE();
827 
828 	return (error);
829 }
830 
831 int
832 soconnect2(struct socket *so1, struct socket *so2)
833 {
834 	int error;
835 
836 	CURVNET_SET(so1->so_vnet);
837 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
838 	CURVNET_RESTORE();
839 	return (error);
840 }
841 
842 int
843 sodisconnect(struct socket *so)
844 {
845 	int error;
846 
847 	if ((so->so_state & SS_ISCONNECTED) == 0)
848 		return (ENOTCONN);
849 	if (so->so_state & SS_ISDISCONNECTING)
850 		return (EALREADY);
851 	VNET_SO_ASSERT(so);
852 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
853 	return (error);
854 }
855 
856 #ifdef ZERO_COPY_SOCKETS
857 struct so_zerocopy_stats{
858 	int size_ok;
859 	int align_ok;
860 	int found_ifp;
861 };
862 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
863 #include <netinet/in.h>
864 #include <net/route.h>
865 #include <netinet/in_pcb.h>
866 #include <vm/vm.h>
867 #include <vm/vm_page.h>
868 #include <vm/vm_object.h>
869 
870 /*
871  * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
872  * sosend_dgram() and sosend_generic() use m_uiotombuf().
873  *
874  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
875  * all of the data referenced by the uio.  If desired, it uses zero-copy.
876  * *space will be updated to reflect data copied in.
877  *
878  * NB: If atomic I/O is requested, the caller must already have checked that
879  * space can hold resid bytes.
880  *
881  * NB: In the event of an error, the caller may need to free the partial
882  * chain pointed to by *mpp.  The contents of both *uio and *space may be
883  * modified even in the case of an error.
884  */
885 static int
886 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
887     int flags)
888 {
889 	struct mbuf *m, **mp, *top;
890 	long len;
891 	ssize_t resid;
892 	int error;
893 #ifdef ZERO_COPY_SOCKETS
894 	int cow_send;
895 #endif
896 
897 	*retmp = top = NULL;
898 	mp = &top;
899 	len = 0;
900 	resid = uio->uio_resid;
901 	error = 0;
902 	do {
903 #ifdef ZERO_COPY_SOCKETS
904 		cow_send = 0;
905 #endif /* ZERO_COPY_SOCKETS */
906 		if (resid >= MINCLSIZE) {
907 #ifdef ZERO_COPY_SOCKETS
908 			if (top == NULL) {
909 				m = m_gethdr(M_WAITOK, MT_DATA);
910 				m->m_pkthdr.len = 0;
911 				m->m_pkthdr.rcvif = NULL;
912 			} else
913 				m = m_get(M_WAITOK, MT_DATA);
914 			if (so_zero_copy_send &&
915 			    resid>=PAGE_SIZE &&
916 			    *space>=PAGE_SIZE &&
917 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
918 				so_zerocp_stats.size_ok++;
919 				so_zerocp_stats.align_ok++;
920 				cow_send = socow_setup(m, uio);
921 				len = cow_send;
922 			}
923 			if (!cow_send) {
924 				m_clget(m, M_WAITOK);
925 				len = min(min(MCLBYTES, resid), *space);
926 			}
927 #else /* ZERO_COPY_SOCKETS */
928 			if (top == NULL) {
929 				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
930 				m->m_pkthdr.len = 0;
931 				m->m_pkthdr.rcvif = NULL;
932 			} else
933 				m = m_getcl(M_WAIT, MT_DATA, 0);
934 			len = min(min(MCLBYTES, resid), *space);
935 #endif /* ZERO_COPY_SOCKETS */
936 		} else {
937 			if (top == NULL) {
938 				m = m_gethdr(M_WAIT, MT_DATA);
939 				m->m_pkthdr.len = 0;
940 				m->m_pkthdr.rcvif = NULL;
941 
942 				len = min(min(MHLEN, resid), *space);
943 				/*
944 				 * For datagram protocols, leave room
945 				 * for protocol headers in first mbuf.
946 				 */
947 				if (atomic && m && len < MHLEN)
948 					MH_ALIGN(m, len);
949 			} else {
950 				m = m_get(M_WAIT, MT_DATA);
951 				len = min(min(MLEN, resid), *space);
952 			}
953 		}
954 		if (m == NULL) {
955 			error = ENOBUFS;
956 			goto out;
957 		}
958 
959 		*space -= len;
960 #ifdef ZERO_COPY_SOCKETS
961 		if (cow_send)
962 			error = 0;
963 		else
964 #endif /* ZERO_COPY_SOCKETS */
965 		error = uiomove(mtod(m, void *), (int)len, uio);
966 		resid = uio->uio_resid;
967 		m->m_len = len;
968 		*mp = m;
969 		top->m_pkthdr.len += len;
970 		if (error)
971 			goto out;
972 		mp = &m->m_next;
973 		if (resid <= 0) {
974 			if (flags & MSG_EOR)
975 				top->m_flags |= M_EOR;
976 			break;
977 		}
978 	} while (*space > 0 && atomic);
979 out:
980 	*retmp = top;
981 	return (error);
982 }
983 #endif /*ZERO_COPY_SOCKETS*/
984 
985 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
986 
987 int
988 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
989     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
990 {
991 	long space;
992 	ssize_t resid;
993 	int clen = 0, error, dontroute;
994 #ifdef ZERO_COPY_SOCKETS
995 	int atomic = sosendallatonce(so) || top;
996 #endif
997 
998 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
999 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1000 	    ("sodgram_send: !PR_ATOMIC"));
1001 
1002 	if (uio != NULL)
1003 		resid = uio->uio_resid;
1004 	else
1005 		resid = top->m_pkthdr.len;
1006 	/*
1007 	 * In theory resid should be unsigned.  However, space must be
1008 	 * signed, as it might be less than 0 if we over-committed, and we
1009 	 * must use a signed comparison of space and resid.  On the other
1010 	 * hand, a negative resid causes us to loop sending 0-length
1011 	 * segments to the protocol.
1012 	 */
1013 	if (resid < 0) {
1014 		error = EINVAL;
1015 		goto out;
1016 	}
1017 
1018 	dontroute =
1019 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1020 	if (td != NULL)
1021 		td->td_ru.ru_msgsnd++;
1022 	if (control != NULL)
1023 		clen = control->m_len;
1024 
1025 	SOCKBUF_LOCK(&so->so_snd);
1026 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1027 		SOCKBUF_UNLOCK(&so->so_snd);
1028 		error = EPIPE;
1029 		goto out;
1030 	}
1031 	if (so->so_error) {
1032 		error = so->so_error;
1033 		so->so_error = 0;
1034 		SOCKBUF_UNLOCK(&so->so_snd);
1035 		goto out;
1036 	}
1037 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1038 		/*
1039 		 * `sendto' and `sendmsg' is allowed on a connection-based
1040 		 * socket if it supports implied connect.  Return ENOTCONN if
1041 		 * not connected and no address is supplied.
1042 		 */
1043 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1044 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1045 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1046 			    !(resid == 0 && clen != 0)) {
1047 				SOCKBUF_UNLOCK(&so->so_snd);
1048 				error = ENOTCONN;
1049 				goto out;
1050 			}
1051 		} else if (addr == NULL) {
1052 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1053 				error = ENOTCONN;
1054 			else
1055 				error = EDESTADDRREQ;
1056 			SOCKBUF_UNLOCK(&so->so_snd);
1057 			goto out;
1058 		}
1059 	}
1060 
1061 	/*
1062 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1063 	 * problem and need fixing.
1064 	 */
1065 	space = sbspace(&so->so_snd);
1066 	if (flags & MSG_OOB)
1067 		space += 1024;
1068 	space -= clen;
1069 	SOCKBUF_UNLOCK(&so->so_snd);
1070 	if (resid > space) {
1071 		error = EMSGSIZE;
1072 		goto out;
1073 	}
1074 	if (uio == NULL) {
1075 		resid = 0;
1076 		if (flags & MSG_EOR)
1077 			top->m_flags |= M_EOR;
1078 	} else {
1079 #ifdef ZERO_COPY_SOCKETS
1080 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1081 		if (error)
1082 			goto out;
1083 #else
1084 		/*
1085 		 * Copy the data from userland into a mbuf chain.
1086 		 * If no data is to be copied in, a single empty mbuf
1087 		 * is returned.
1088 		 */
1089 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1090 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1091 		if (top == NULL) {
1092 			error = EFAULT;	/* only possible error */
1093 			goto out;
1094 		}
1095 		space -= resid - uio->uio_resid;
1096 #endif
1097 		resid = uio->uio_resid;
1098 	}
1099 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1100 	/*
1101 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1102 	 * than with.
1103 	 */
1104 	if (dontroute) {
1105 		SOCK_LOCK(so);
1106 		so->so_options |= SO_DONTROUTE;
1107 		SOCK_UNLOCK(so);
1108 	}
1109 	/*
1110 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1111 	 * of date.  We could have recieved a reset packet in an interrupt or
1112 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1113 	 * probably recheck again inside the locking protection here, but
1114 	 * there are probably other places that this also happens.  We must
1115 	 * rethink this.
1116 	 */
1117 	VNET_SO_ASSERT(so);
1118 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1119 	    (flags & MSG_OOB) ? PRUS_OOB :
1120 	/*
1121 	 * If the user set MSG_EOF, the protocol understands this flag and
1122 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1123 	 */
1124 	    ((flags & MSG_EOF) &&
1125 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1126 	     (resid <= 0)) ?
1127 		PRUS_EOF :
1128 		/* If there is more to send set PRUS_MORETOCOME */
1129 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1130 		top, addr, control, td);
1131 	if (dontroute) {
1132 		SOCK_LOCK(so);
1133 		so->so_options &= ~SO_DONTROUTE;
1134 		SOCK_UNLOCK(so);
1135 	}
1136 	clen = 0;
1137 	control = NULL;
1138 	top = NULL;
1139 out:
1140 	if (top != NULL)
1141 		m_freem(top);
1142 	if (control != NULL)
1143 		m_freem(control);
1144 	return (error);
1145 }
1146 
1147 /*
1148  * Send on a socket.  If send must go all at once and message is larger than
1149  * send buffering, then hard error.  Lock against other senders.  If must go
1150  * all at once and not enough room now, then inform user that this would
1151  * block and do nothing.  Otherwise, if nonblocking, send as much as
1152  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1153  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1154  * in mbuf chain must be small enough to send all at once.
1155  *
1156  * Returns nonzero on error, timeout or signal; callers must check for short
1157  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1158  * on return.
1159  */
1160 int
1161 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1162     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1163 {
1164 	long space;
1165 	ssize_t resid;
1166 	int clen = 0, error, dontroute;
1167 	int atomic = sosendallatonce(so) || top;
1168 
1169 	if (uio != NULL)
1170 		resid = uio->uio_resid;
1171 	else
1172 		resid = top->m_pkthdr.len;
1173 	/*
1174 	 * In theory resid should be unsigned.  However, space must be
1175 	 * signed, as it might be less than 0 if we over-committed, and we
1176 	 * must use a signed comparison of space and resid.  On the other
1177 	 * hand, a negative resid causes us to loop sending 0-length
1178 	 * segments to the protocol.
1179 	 *
1180 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1181 	 * type sockets since that's an error.
1182 	 */
1183 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1184 		error = EINVAL;
1185 		goto out;
1186 	}
1187 
1188 	dontroute =
1189 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1190 	    (so->so_proto->pr_flags & PR_ATOMIC);
1191 	if (td != NULL)
1192 		td->td_ru.ru_msgsnd++;
1193 	if (control != NULL)
1194 		clen = control->m_len;
1195 
1196 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1197 	if (error)
1198 		goto out;
1199 
1200 restart:
1201 	do {
1202 		SOCKBUF_LOCK(&so->so_snd);
1203 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1204 			SOCKBUF_UNLOCK(&so->so_snd);
1205 			error = EPIPE;
1206 			goto release;
1207 		}
1208 		if (so->so_error) {
1209 			error = so->so_error;
1210 			so->so_error = 0;
1211 			SOCKBUF_UNLOCK(&so->so_snd);
1212 			goto release;
1213 		}
1214 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1215 			/*
1216 			 * `sendto' and `sendmsg' is allowed on a connection-
1217 			 * based socket if it supports implied connect.
1218 			 * Return ENOTCONN if not connected and no address is
1219 			 * supplied.
1220 			 */
1221 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1222 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1223 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1224 				    !(resid == 0 && clen != 0)) {
1225 					SOCKBUF_UNLOCK(&so->so_snd);
1226 					error = ENOTCONN;
1227 					goto release;
1228 				}
1229 			} else if (addr == NULL) {
1230 				SOCKBUF_UNLOCK(&so->so_snd);
1231 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1232 					error = ENOTCONN;
1233 				else
1234 					error = EDESTADDRREQ;
1235 				goto release;
1236 			}
1237 		}
1238 		space = sbspace(&so->so_snd);
1239 		if (flags & MSG_OOB)
1240 			space += 1024;
1241 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1242 		    clen > so->so_snd.sb_hiwat) {
1243 			SOCKBUF_UNLOCK(&so->so_snd);
1244 			error = EMSGSIZE;
1245 			goto release;
1246 		}
1247 		if (space < resid + clen &&
1248 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1249 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1250 				SOCKBUF_UNLOCK(&so->so_snd);
1251 				error = EWOULDBLOCK;
1252 				goto release;
1253 			}
1254 			error = sbwait(&so->so_snd);
1255 			SOCKBUF_UNLOCK(&so->so_snd);
1256 			if (error)
1257 				goto release;
1258 			goto restart;
1259 		}
1260 		SOCKBUF_UNLOCK(&so->so_snd);
1261 		space -= clen;
1262 		do {
1263 			if (uio == NULL) {
1264 				resid = 0;
1265 				if (flags & MSG_EOR)
1266 					top->m_flags |= M_EOR;
1267 			} else {
1268 #ifdef ZERO_COPY_SOCKETS
1269 				error = sosend_copyin(uio, &top, atomic,
1270 				    &space, flags);
1271 				if (error != 0)
1272 					goto release;
1273 #else
1274 				/*
1275 				 * Copy the data from userland into a mbuf
1276 				 * chain.  If no data is to be copied in,
1277 				 * a single empty mbuf is returned.
1278 				 */
1279 				top = m_uiotombuf(uio, M_WAITOK, space,
1280 				    (atomic ? max_hdr : 0),
1281 				    (atomic ? M_PKTHDR : 0) |
1282 				    ((flags & MSG_EOR) ? M_EOR : 0));
1283 				if (top == NULL) {
1284 					error = EFAULT; /* only possible error */
1285 					goto release;
1286 				}
1287 				space -= resid - uio->uio_resid;
1288 #endif
1289 				resid = uio->uio_resid;
1290 			}
1291 			if (dontroute) {
1292 				SOCK_LOCK(so);
1293 				so->so_options |= SO_DONTROUTE;
1294 				SOCK_UNLOCK(so);
1295 			}
1296 			/*
1297 			 * XXX all the SBS_CANTSENDMORE checks previously
1298 			 * done could be out of date.  We could have recieved
1299 			 * a reset packet in an interrupt or maybe we slept
1300 			 * while doing page faults in uiomove() etc.  We
1301 			 * could probably recheck again inside the locking
1302 			 * protection here, but there are probably other
1303 			 * places that this also happens.  We must rethink
1304 			 * this.
1305 			 */
1306 			VNET_SO_ASSERT(so);
1307 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1308 			    (flags & MSG_OOB) ? PRUS_OOB :
1309 			/*
1310 			 * If the user set MSG_EOF, the protocol understands
1311 			 * this flag and nothing left to send then use
1312 			 * PRU_SEND_EOF instead of PRU_SEND.
1313 			 */
1314 			    ((flags & MSG_EOF) &&
1315 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1316 			     (resid <= 0)) ?
1317 				PRUS_EOF :
1318 			/* If there is more to send set PRUS_MORETOCOME. */
1319 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1320 			    top, addr, control, td);
1321 			if (dontroute) {
1322 				SOCK_LOCK(so);
1323 				so->so_options &= ~SO_DONTROUTE;
1324 				SOCK_UNLOCK(so);
1325 			}
1326 			clen = 0;
1327 			control = NULL;
1328 			top = NULL;
1329 			if (error)
1330 				goto release;
1331 		} while (resid && space > 0);
1332 	} while (resid);
1333 
1334 release:
1335 	sbunlock(&so->so_snd);
1336 out:
1337 	if (top != NULL)
1338 		m_freem(top);
1339 	if (control != NULL)
1340 		m_freem(control);
1341 	return (error);
1342 }
1343 
1344 int
1345 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1346     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1347 {
1348 	int error;
1349 
1350 	CURVNET_SET(so->so_vnet);
1351 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1352 	    control, flags, td);
1353 	CURVNET_RESTORE();
1354 	return (error);
1355 }
1356 
1357 /*
1358  * The part of soreceive() that implements reading non-inline out-of-band
1359  * data from a socket.  For more complete comments, see soreceive(), from
1360  * which this code originated.
1361  *
1362  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1363  * unable to return an mbuf chain to the caller.
1364  */
1365 static int
1366 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1367 {
1368 	struct protosw *pr = so->so_proto;
1369 	struct mbuf *m;
1370 	int error;
1371 
1372 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1373 	VNET_SO_ASSERT(so);
1374 
1375 	m = m_get(M_WAIT, MT_DATA);
1376 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1377 	if (error)
1378 		goto bad;
1379 	do {
1380 #ifdef ZERO_COPY_SOCKETS
1381 		if (so_zero_copy_receive) {
1382 			int disposable;
1383 
1384 			if ((m->m_flags & M_EXT)
1385 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1386 				disposable = 1;
1387 			else
1388 				disposable = 0;
1389 
1390 			error = uiomoveco(mtod(m, void *),
1391 					  min(uio->uio_resid, m->m_len),
1392 					  uio, disposable);
1393 		} else
1394 #endif /* ZERO_COPY_SOCKETS */
1395 		error = uiomove(mtod(m, void *),
1396 		    (int) min(uio->uio_resid, m->m_len), uio);
1397 		m = m_free(m);
1398 	} while (uio->uio_resid && error == 0 && m);
1399 bad:
1400 	if (m != NULL)
1401 		m_freem(m);
1402 	return (error);
1403 }
1404 
1405 /*
1406  * Following replacement or removal of the first mbuf on the first mbuf chain
1407  * of a socket buffer, push necessary state changes back into the socket
1408  * buffer so that other consumers see the values consistently.  'nextrecord'
1409  * is the callers locally stored value of the original value of
1410  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1411  * NOTE: 'nextrecord' may be NULL.
1412  */
1413 static __inline void
1414 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1415 {
1416 
1417 	SOCKBUF_LOCK_ASSERT(sb);
1418 	/*
1419 	 * First, update for the new value of nextrecord.  If necessary, make
1420 	 * it the first record.
1421 	 */
1422 	if (sb->sb_mb != NULL)
1423 		sb->sb_mb->m_nextpkt = nextrecord;
1424 	else
1425 		sb->sb_mb = nextrecord;
1426 
1427         /*
1428          * Now update any dependent socket buffer fields to reflect the new
1429          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1430 	 * addition of a second clause that takes care of the case where
1431 	 * sb_mb has been updated, but remains the last record.
1432          */
1433         if (sb->sb_mb == NULL) {
1434                 sb->sb_mbtail = NULL;
1435                 sb->sb_lastrecord = NULL;
1436         } else if (sb->sb_mb->m_nextpkt == NULL)
1437                 sb->sb_lastrecord = sb->sb_mb;
1438 }
1439 
1440 
1441 /*
1442  * Implement receive operations on a socket.  We depend on the way that
1443  * records are added to the sockbuf by sbappend.  In particular, each record
1444  * (mbufs linked through m_next) must begin with an address if the protocol
1445  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1446  * data, and then zero or more mbufs of data.  In order to allow parallelism
1447  * between network receive and copying to user space, as well as avoid
1448  * sleeping with a mutex held, we release the socket buffer mutex during the
1449  * user space copy.  Although the sockbuf is locked, new data may still be
1450  * appended, and thus we must maintain consistency of the sockbuf during that
1451  * time.
1452  *
1453  * The caller may receive the data as a single mbuf chain by supplying an
1454  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1455  * the count in uio_resid.
1456  */
1457 int
1458 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1459     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1460 {
1461 	struct mbuf *m, **mp;
1462 	int flags, error, offset;
1463 	ssize_t len;
1464 	struct protosw *pr = so->so_proto;
1465 	struct mbuf *nextrecord;
1466 	int moff, type = 0;
1467 	ssize_t orig_resid = uio->uio_resid;
1468 
1469 	mp = mp0;
1470 	if (psa != NULL)
1471 		*psa = NULL;
1472 	if (controlp != NULL)
1473 		*controlp = NULL;
1474 	if (flagsp != NULL)
1475 		flags = *flagsp &~ MSG_EOR;
1476 	else
1477 		flags = 0;
1478 	if (flags & MSG_OOB)
1479 		return (soreceive_rcvoob(so, uio, flags));
1480 	if (mp != NULL)
1481 		*mp = NULL;
1482 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1483 	    && uio->uio_resid) {
1484 		VNET_SO_ASSERT(so);
1485 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1486 	}
1487 
1488 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1489 	if (error)
1490 		return (error);
1491 
1492 restart:
1493 	SOCKBUF_LOCK(&so->so_rcv);
1494 	m = so->so_rcv.sb_mb;
1495 	/*
1496 	 * If we have less data than requested, block awaiting more (subject
1497 	 * to any timeout) if:
1498 	 *   1. the current count is less than the low water mark, or
1499 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1500 	 *	receive operation at once if we block (resid <= hiwat).
1501 	 *   3. MSG_DONTWAIT is not set
1502 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1503 	 * we have to do the receive in sections, and thus risk returning a
1504 	 * short count if a timeout or signal occurs after we start.
1505 	 */
1506 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1507 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1508 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1509 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1510 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1511 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1512 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1513 		    m, so->so_rcv.sb_cc));
1514 		if (so->so_error) {
1515 			if (m != NULL)
1516 				goto dontblock;
1517 			error = so->so_error;
1518 			if ((flags & MSG_PEEK) == 0)
1519 				so->so_error = 0;
1520 			SOCKBUF_UNLOCK(&so->so_rcv);
1521 			goto release;
1522 		}
1523 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1524 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1525 			if (m == NULL) {
1526 				SOCKBUF_UNLOCK(&so->so_rcv);
1527 				goto release;
1528 			} else
1529 				goto dontblock;
1530 		}
1531 		for (; m != NULL; m = m->m_next)
1532 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1533 				m = so->so_rcv.sb_mb;
1534 				goto dontblock;
1535 			}
1536 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1537 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1538 			SOCKBUF_UNLOCK(&so->so_rcv);
1539 			error = ENOTCONN;
1540 			goto release;
1541 		}
1542 		if (uio->uio_resid == 0) {
1543 			SOCKBUF_UNLOCK(&so->so_rcv);
1544 			goto release;
1545 		}
1546 		if ((so->so_state & SS_NBIO) ||
1547 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1548 			SOCKBUF_UNLOCK(&so->so_rcv);
1549 			error = EWOULDBLOCK;
1550 			goto release;
1551 		}
1552 		SBLASTRECORDCHK(&so->so_rcv);
1553 		SBLASTMBUFCHK(&so->so_rcv);
1554 		error = sbwait(&so->so_rcv);
1555 		SOCKBUF_UNLOCK(&so->so_rcv);
1556 		if (error)
1557 			goto release;
1558 		goto restart;
1559 	}
1560 dontblock:
1561 	/*
1562 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1563 	 * pointer to the next record in the socket buffer.  We must keep the
1564 	 * various socket buffer pointers and local stack versions of the
1565 	 * pointers in sync, pushing out modifications before dropping the
1566 	 * socket buffer mutex, and re-reading them when picking it up.
1567 	 *
1568 	 * Otherwise, we will race with the network stack appending new data
1569 	 * or records onto the socket buffer by using inconsistent/stale
1570 	 * versions of the field, possibly resulting in socket buffer
1571 	 * corruption.
1572 	 *
1573 	 * By holding the high-level sblock(), we prevent simultaneous
1574 	 * readers from pulling off the front of the socket buffer.
1575 	 */
1576 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1577 	if (uio->uio_td)
1578 		uio->uio_td->td_ru.ru_msgrcv++;
1579 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1580 	SBLASTRECORDCHK(&so->so_rcv);
1581 	SBLASTMBUFCHK(&so->so_rcv);
1582 	nextrecord = m->m_nextpkt;
1583 	if (pr->pr_flags & PR_ADDR) {
1584 		KASSERT(m->m_type == MT_SONAME,
1585 		    ("m->m_type == %d", m->m_type));
1586 		orig_resid = 0;
1587 		if (psa != NULL)
1588 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1589 			    M_NOWAIT);
1590 		if (flags & MSG_PEEK) {
1591 			m = m->m_next;
1592 		} else {
1593 			sbfree(&so->so_rcv, m);
1594 			so->so_rcv.sb_mb = m_free(m);
1595 			m = so->so_rcv.sb_mb;
1596 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1597 		}
1598 	}
1599 
1600 	/*
1601 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1602 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1603 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1604 	 * perform externalization (or freeing if controlp == NULL).
1605 	 */
1606 	if (m != NULL && m->m_type == MT_CONTROL) {
1607 		struct mbuf *cm = NULL, *cmn;
1608 		struct mbuf **cme = &cm;
1609 
1610 		do {
1611 			if (flags & MSG_PEEK) {
1612 				if (controlp != NULL) {
1613 					*controlp = m_copy(m, 0, m->m_len);
1614 					controlp = &(*controlp)->m_next;
1615 				}
1616 				m = m->m_next;
1617 			} else {
1618 				sbfree(&so->so_rcv, m);
1619 				so->so_rcv.sb_mb = m->m_next;
1620 				m->m_next = NULL;
1621 				*cme = m;
1622 				cme = &(*cme)->m_next;
1623 				m = so->so_rcv.sb_mb;
1624 			}
1625 		} while (m != NULL && m->m_type == MT_CONTROL);
1626 		if ((flags & MSG_PEEK) == 0)
1627 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1628 		while (cm != NULL) {
1629 			cmn = cm->m_next;
1630 			cm->m_next = NULL;
1631 			if (pr->pr_domain->dom_externalize != NULL) {
1632 				SOCKBUF_UNLOCK(&so->so_rcv);
1633 				VNET_SO_ASSERT(so);
1634 				error = (*pr->pr_domain->dom_externalize)
1635 				    (cm, controlp);
1636 				SOCKBUF_LOCK(&so->so_rcv);
1637 			} else if (controlp != NULL)
1638 				*controlp = cm;
1639 			else
1640 				m_freem(cm);
1641 			if (controlp != NULL) {
1642 				orig_resid = 0;
1643 				while (*controlp != NULL)
1644 					controlp = &(*controlp)->m_next;
1645 			}
1646 			cm = cmn;
1647 		}
1648 		if (m != NULL)
1649 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1650 		else
1651 			nextrecord = so->so_rcv.sb_mb;
1652 		orig_resid = 0;
1653 	}
1654 	if (m != NULL) {
1655 		if ((flags & MSG_PEEK) == 0) {
1656 			KASSERT(m->m_nextpkt == nextrecord,
1657 			    ("soreceive: post-control, nextrecord !sync"));
1658 			if (nextrecord == NULL) {
1659 				KASSERT(so->so_rcv.sb_mb == m,
1660 				    ("soreceive: post-control, sb_mb!=m"));
1661 				KASSERT(so->so_rcv.sb_lastrecord == m,
1662 				    ("soreceive: post-control, lastrecord!=m"));
1663 			}
1664 		}
1665 		type = m->m_type;
1666 		if (type == MT_OOBDATA)
1667 			flags |= MSG_OOB;
1668 	} else {
1669 		if ((flags & MSG_PEEK) == 0) {
1670 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1671 			    ("soreceive: sb_mb != nextrecord"));
1672 			if (so->so_rcv.sb_mb == NULL) {
1673 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1674 				    ("soreceive: sb_lastercord != NULL"));
1675 			}
1676 		}
1677 	}
1678 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1679 	SBLASTRECORDCHK(&so->so_rcv);
1680 	SBLASTMBUFCHK(&so->so_rcv);
1681 
1682 	/*
1683 	 * Now continue to read any data mbufs off of the head of the socket
1684 	 * buffer until the read request is satisfied.  Note that 'type' is
1685 	 * used to store the type of any mbuf reads that have happened so far
1686 	 * such that soreceive() can stop reading if the type changes, which
1687 	 * causes soreceive() to return only one of regular data and inline
1688 	 * out-of-band data in a single socket receive operation.
1689 	 */
1690 	moff = 0;
1691 	offset = 0;
1692 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1693 		/*
1694 		 * If the type of mbuf has changed since the last mbuf
1695 		 * examined ('type'), end the receive operation.
1696 	 	 */
1697 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1698 		if (m->m_type == MT_OOBDATA) {
1699 			if (type != MT_OOBDATA)
1700 				break;
1701 		} else if (type == MT_OOBDATA)
1702 			break;
1703 		else
1704 		    KASSERT(m->m_type == MT_DATA,
1705 			("m->m_type == %d", m->m_type));
1706 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1707 		len = uio->uio_resid;
1708 		if (so->so_oobmark && len > so->so_oobmark - offset)
1709 			len = so->so_oobmark - offset;
1710 		if (len > m->m_len - moff)
1711 			len = m->m_len - moff;
1712 		/*
1713 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1714 		 * them out via the uio, then free.  Sockbuf must be
1715 		 * consistent here (points to current mbuf, it points to next
1716 		 * record) when we drop priority; we must note any additions
1717 		 * to the sockbuf when we block interrupts again.
1718 		 */
1719 		if (mp == NULL) {
1720 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1721 			SBLASTRECORDCHK(&so->so_rcv);
1722 			SBLASTMBUFCHK(&so->so_rcv);
1723 			SOCKBUF_UNLOCK(&so->so_rcv);
1724 #ifdef ZERO_COPY_SOCKETS
1725 			if (so_zero_copy_receive) {
1726 				int disposable;
1727 
1728 				if ((m->m_flags & M_EXT)
1729 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1730 					disposable = 1;
1731 				else
1732 					disposable = 0;
1733 
1734 				error = uiomoveco(mtod(m, char *) + moff,
1735 						  (int)len, uio,
1736 						  disposable);
1737 			} else
1738 #endif /* ZERO_COPY_SOCKETS */
1739 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1740 			SOCKBUF_LOCK(&so->so_rcv);
1741 			if (error) {
1742 				/*
1743 				 * The MT_SONAME mbuf has already been removed
1744 				 * from the record, so it is necessary to
1745 				 * remove the data mbufs, if any, to preserve
1746 				 * the invariant in the case of PR_ADDR that
1747 				 * requires MT_SONAME mbufs at the head of
1748 				 * each record.
1749 				 */
1750 				if (m && pr->pr_flags & PR_ATOMIC &&
1751 				    ((flags & MSG_PEEK) == 0))
1752 					(void)sbdroprecord_locked(&so->so_rcv);
1753 				SOCKBUF_UNLOCK(&so->so_rcv);
1754 				goto release;
1755 			}
1756 		} else
1757 			uio->uio_resid -= len;
1758 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1759 		if (len == m->m_len - moff) {
1760 			if (m->m_flags & M_EOR)
1761 				flags |= MSG_EOR;
1762 			if (flags & MSG_PEEK) {
1763 				m = m->m_next;
1764 				moff = 0;
1765 			} else {
1766 				nextrecord = m->m_nextpkt;
1767 				sbfree(&so->so_rcv, m);
1768 				if (mp != NULL) {
1769 					*mp = m;
1770 					mp = &m->m_next;
1771 					so->so_rcv.sb_mb = m = m->m_next;
1772 					*mp = NULL;
1773 				} else {
1774 					so->so_rcv.sb_mb = m_free(m);
1775 					m = so->so_rcv.sb_mb;
1776 				}
1777 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1778 				SBLASTRECORDCHK(&so->so_rcv);
1779 				SBLASTMBUFCHK(&so->so_rcv);
1780 			}
1781 		} else {
1782 			if (flags & MSG_PEEK)
1783 				moff += len;
1784 			else {
1785 				if (mp != NULL) {
1786 					int copy_flag;
1787 
1788 					if (flags & MSG_DONTWAIT)
1789 						copy_flag = M_DONTWAIT;
1790 					else
1791 						copy_flag = M_WAIT;
1792 					if (copy_flag == M_WAIT)
1793 						SOCKBUF_UNLOCK(&so->so_rcv);
1794 					*mp = m_copym(m, 0, len, copy_flag);
1795 					if (copy_flag == M_WAIT)
1796 						SOCKBUF_LOCK(&so->so_rcv);
1797  					if (*mp == NULL) {
1798  						/*
1799  						 * m_copym() couldn't
1800 						 * allocate an mbuf.  Adjust
1801 						 * uio_resid back (it was
1802 						 * adjusted down by len
1803 						 * bytes, which we didn't end
1804 						 * up "copying" over).
1805  						 */
1806  						uio->uio_resid += len;
1807  						break;
1808  					}
1809 				}
1810 				m->m_data += len;
1811 				m->m_len -= len;
1812 				so->so_rcv.sb_cc -= len;
1813 			}
1814 		}
1815 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1816 		if (so->so_oobmark) {
1817 			if ((flags & MSG_PEEK) == 0) {
1818 				so->so_oobmark -= len;
1819 				if (so->so_oobmark == 0) {
1820 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1821 					break;
1822 				}
1823 			} else {
1824 				offset += len;
1825 				if (offset == so->so_oobmark)
1826 					break;
1827 			}
1828 		}
1829 		if (flags & MSG_EOR)
1830 			break;
1831 		/*
1832 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1833 		 * must not quit until "uio->uio_resid == 0" or an error
1834 		 * termination.  If a signal/timeout occurs, return with a
1835 		 * short count but without error.  Keep sockbuf locked
1836 		 * against other readers.
1837 		 */
1838 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1839 		    !sosendallatonce(so) && nextrecord == NULL) {
1840 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1841 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1842 				break;
1843 			/*
1844 			 * Notify the protocol that some data has been
1845 			 * drained before blocking.
1846 			 */
1847 			if (pr->pr_flags & PR_WANTRCVD) {
1848 				SOCKBUF_UNLOCK(&so->so_rcv);
1849 				VNET_SO_ASSERT(so);
1850 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1851 				SOCKBUF_LOCK(&so->so_rcv);
1852 			}
1853 			SBLASTRECORDCHK(&so->so_rcv);
1854 			SBLASTMBUFCHK(&so->so_rcv);
1855 			/*
1856 			 * We could receive some data while was notifying
1857 			 * the protocol. Skip blocking in this case.
1858 			 */
1859 			if (so->so_rcv.sb_mb == NULL) {
1860 				error = sbwait(&so->so_rcv);
1861 				if (error) {
1862 					SOCKBUF_UNLOCK(&so->so_rcv);
1863 					goto release;
1864 				}
1865 			}
1866 			m = so->so_rcv.sb_mb;
1867 			if (m != NULL)
1868 				nextrecord = m->m_nextpkt;
1869 		}
1870 	}
1871 
1872 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1873 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1874 		flags |= MSG_TRUNC;
1875 		if ((flags & MSG_PEEK) == 0)
1876 			(void) sbdroprecord_locked(&so->so_rcv);
1877 	}
1878 	if ((flags & MSG_PEEK) == 0) {
1879 		if (m == NULL) {
1880 			/*
1881 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1882 			 * part makes sure sb_lastrecord is up-to-date if
1883 			 * there is still data in the socket buffer.
1884 			 */
1885 			so->so_rcv.sb_mb = nextrecord;
1886 			if (so->so_rcv.sb_mb == NULL) {
1887 				so->so_rcv.sb_mbtail = NULL;
1888 				so->so_rcv.sb_lastrecord = NULL;
1889 			} else if (nextrecord->m_nextpkt == NULL)
1890 				so->so_rcv.sb_lastrecord = nextrecord;
1891 		}
1892 		SBLASTRECORDCHK(&so->so_rcv);
1893 		SBLASTMBUFCHK(&so->so_rcv);
1894 		/*
1895 		 * If soreceive() is being done from the socket callback,
1896 		 * then don't need to generate ACK to peer to update window,
1897 		 * since ACK will be generated on return to TCP.
1898 		 */
1899 		if (!(flags & MSG_SOCALLBCK) &&
1900 		    (pr->pr_flags & PR_WANTRCVD)) {
1901 			SOCKBUF_UNLOCK(&so->so_rcv);
1902 			VNET_SO_ASSERT(so);
1903 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1904 			SOCKBUF_LOCK(&so->so_rcv);
1905 		}
1906 	}
1907 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1908 	if (orig_resid == uio->uio_resid && orig_resid &&
1909 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1910 		SOCKBUF_UNLOCK(&so->so_rcv);
1911 		goto restart;
1912 	}
1913 	SOCKBUF_UNLOCK(&so->so_rcv);
1914 
1915 	if (flagsp != NULL)
1916 		*flagsp |= flags;
1917 release:
1918 	sbunlock(&so->so_rcv);
1919 	return (error);
1920 }
1921 
1922 /*
1923  * Optimized version of soreceive() for stream (TCP) sockets.
1924  */
1925 int
1926 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1927     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1928 {
1929 	int len = 0, error = 0, flags, oresid;
1930 	struct sockbuf *sb;
1931 	struct mbuf *m, *n = NULL;
1932 
1933 	/* We only do stream sockets. */
1934 	if (so->so_type != SOCK_STREAM)
1935 		return (EINVAL);
1936 	if (psa != NULL)
1937 		*psa = NULL;
1938 	if (controlp != NULL)
1939 		return (EINVAL);
1940 	if (flagsp != NULL)
1941 		flags = *flagsp &~ MSG_EOR;
1942 	else
1943 		flags = 0;
1944 	if (flags & MSG_OOB)
1945 		return (soreceive_rcvoob(so, uio, flags));
1946 	if (mp0 != NULL)
1947 		*mp0 = NULL;
1948 
1949 	sb = &so->so_rcv;
1950 
1951 	/* Prevent other readers from entering the socket. */
1952 	error = sblock(sb, SBLOCKWAIT(flags));
1953 	if (error)
1954 		goto out;
1955 	SOCKBUF_LOCK(sb);
1956 
1957 	/* Easy one, no space to copyout anything. */
1958 	if (uio->uio_resid == 0) {
1959 		error = EINVAL;
1960 		goto out;
1961 	}
1962 	oresid = uio->uio_resid;
1963 
1964 	/* We will never ever get anything unless we are or were connected. */
1965 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1966 		error = ENOTCONN;
1967 		goto out;
1968 	}
1969 
1970 restart:
1971 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1972 
1973 	/* Abort if socket has reported problems. */
1974 	if (so->so_error) {
1975 		if (sb->sb_cc > 0)
1976 			goto deliver;
1977 		if (oresid > uio->uio_resid)
1978 			goto out;
1979 		error = so->so_error;
1980 		if (!(flags & MSG_PEEK))
1981 			so->so_error = 0;
1982 		goto out;
1983 	}
1984 
1985 	/* Door is closed.  Deliver what is left, if any. */
1986 	if (sb->sb_state & SBS_CANTRCVMORE) {
1987 		if (sb->sb_cc > 0)
1988 			goto deliver;
1989 		else
1990 			goto out;
1991 	}
1992 
1993 	/* Socket buffer is empty and we shall not block. */
1994 	if (sb->sb_cc == 0 &&
1995 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1996 		error = EAGAIN;
1997 		goto out;
1998 	}
1999 
2000 	/* Socket buffer got some data that we shall deliver now. */
2001 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
2002 	    ((sb->sb_flags & SS_NBIO) ||
2003 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2004 	     sb->sb_cc >= sb->sb_lowat ||
2005 	     sb->sb_cc >= uio->uio_resid ||
2006 	     sb->sb_cc >= sb->sb_hiwat) ) {
2007 		goto deliver;
2008 	}
2009 
2010 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2011 	if ((flags & MSG_WAITALL) &&
2012 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
2013 		goto deliver;
2014 
2015 	/*
2016 	 * Wait and block until (more) data comes in.
2017 	 * NB: Drops the sockbuf lock during wait.
2018 	 */
2019 	error = sbwait(sb);
2020 	if (error)
2021 		goto out;
2022 	goto restart;
2023 
2024 deliver:
2025 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2026 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2027 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2028 
2029 	/* Statistics. */
2030 	if (uio->uio_td)
2031 		uio->uio_td->td_ru.ru_msgrcv++;
2032 
2033 	/* Fill uio until full or current end of socket buffer is reached. */
2034 	len = min(uio->uio_resid, sb->sb_cc);
2035 	if (mp0 != NULL) {
2036 		/* Dequeue as many mbufs as possible. */
2037 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2038 			for (*mp0 = m = sb->sb_mb;
2039 			     m != NULL && m->m_len <= len;
2040 			     m = m->m_next) {
2041 				len -= m->m_len;
2042 				uio->uio_resid -= m->m_len;
2043 				sbfree(sb, m);
2044 				n = m;
2045 			}
2046 			sb->sb_mb = m;
2047 			if (sb->sb_mb == NULL)
2048 				SB_EMPTY_FIXUP(sb);
2049 			n->m_next = NULL;
2050 		}
2051 		/* Copy the remainder. */
2052 		if (len > 0) {
2053 			KASSERT(sb->sb_mb != NULL,
2054 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2055 
2056 			m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2057 			if (m == NULL)
2058 				len = 0;	/* Don't flush data from sockbuf. */
2059 			else
2060 				uio->uio_resid -= m->m_len;
2061 			if (*mp0 != NULL)
2062 				n->m_next = m;
2063 			else
2064 				*mp0 = m;
2065 			if (*mp0 == NULL) {
2066 				error = ENOBUFS;
2067 				goto out;
2068 			}
2069 		}
2070 	} else {
2071 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2072 		SOCKBUF_UNLOCK(sb);
2073 		error = m_mbuftouio(uio, sb->sb_mb, len);
2074 		SOCKBUF_LOCK(sb);
2075 		if (error)
2076 			goto out;
2077 	}
2078 	SBLASTRECORDCHK(sb);
2079 	SBLASTMBUFCHK(sb);
2080 
2081 	/*
2082 	 * Remove the delivered data from the socket buffer unless we
2083 	 * were only peeking.
2084 	 */
2085 	if (!(flags & MSG_PEEK)) {
2086 		if (len > 0)
2087 			sbdrop_locked(sb, len);
2088 
2089 		/* Notify protocol that we drained some data. */
2090 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2091 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2092 		     !(flags & MSG_SOCALLBCK))) {
2093 			SOCKBUF_UNLOCK(sb);
2094 			VNET_SO_ASSERT(so);
2095 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2096 			SOCKBUF_LOCK(sb);
2097 		}
2098 	}
2099 
2100 	/*
2101 	 * For MSG_WAITALL we may have to loop again and wait for
2102 	 * more data to come in.
2103 	 */
2104 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2105 		goto restart;
2106 out:
2107 	SOCKBUF_LOCK_ASSERT(sb);
2108 	SBLASTRECORDCHK(sb);
2109 	SBLASTMBUFCHK(sb);
2110 	SOCKBUF_UNLOCK(sb);
2111 	sbunlock(sb);
2112 	return (error);
2113 }
2114 
2115 /*
2116  * Optimized version of soreceive() for simple datagram cases from userspace.
2117  * Unlike in the stream case, we're able to drop a datagram if copyout()
2118  * fails, and because we handle datagrams atomically, we don't need to use a
2119  * sleep lock to prevent I/O interlacing.
2120  */
2121 int
2122 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2123     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2124 {
2125 	struct mbuf *m, *m2;
2126 	int flags, error;
2127 	ssize_t len;
2128 	struct protosw *pr = so->so_proto;
2129 	struct mbuf *nextrecord;
2130 
2131 	if (psa != NULL)
2132 		*psa = NULL;
2133 	if (controlp != NULL)
2134 		*controlp = NULL;
2135 	if (flagsp != NULL)
2136 		flags = *flagsp &~ MSG_EOR;
2137 	else
2138 		flags = 0;
2139 
2140 	/*
2141 	 * For any complicated cases, fall back to the full
2142 	 * soreceive_generic().
2143 	 */
2144 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2145 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2146 		    flagsp));
2147 
2148 	/*
2149 	 * Enforce restrictions on use.
2150 	 */
2151 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2152 	    ("soreceive_dgram: wantrcvd"));
2153 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2154 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2155 	    ("soreceive_dgram: SBS_RCVATMARK"));
2156 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2157 	    ("soreceive_dgram: P_CONNREQUIRED"));
2158 
2159 	/*
2160 	 * Loop blocking while waiting for a datagram.
2161 	 */
2162 	SOCKBUF_LOCK(&so->so_rcv);
2163 	while ((m = so->so_rcv.sb_mb) == NULL) {
2164 		KASSERT(so->so_rcv.sb_cc == 0,
2165 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2166 		    so->so_rcv.sb_cc));
2167 		if (so->so_error) {
2168 			error = so->so_error;
2169 			so->so_error = 0;
2170 			SOCKBUF_UNLOCK(&so->so_rcv);
2171 			return (error);
2172 		}
2173 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2174 		    uio->uio_resid == 0) {
2175 			SOCKBUF_UNLOCK(&so->so_rcv);
2176 			return (0);
2177 		}
2178 		if ((so->so_state & SS_NBIO) ||
2179 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2180 			SOCKBUF_UNLOCK(&so->so_rcv);
2181 			return (EWOULDBLOCK);
2182 		}
2183 		SBLASTRECORDCHK(&so->so_rcv);
2184 		SBLASTMBUFCHK(&so->so_rcv);
2185 		error = sbwait(&so->so_rcv);
2186 		if (error) {
2187 			SOCKBUF_UNLOCK(&so->so_rcv);
2188 			return (error);
2189 		}
2190 	}
2191 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2192 
2193 	if (uio->uio_td)
2194 		uio->uio_td->td_ru.ru_msgrcv++;
2195 	SBLASTRECORDCHK(&so->so_rcv);
2196 	SBLASTMBUFCHK(&so->so_rcv);
2197 	nextrecord = m->m_nextpkt;
2198 	if (nextrecord == NULL) {
2199 		KASSERT(so->so_rcv.sb_lastrecord == m,
2200 		    ("soreceive_dgram: lastrecord != m"));
2201 	}
2202 
2203 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2204 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2205 
2206 	/*
2207 	 * Pull 'm' and its chain off the front of the packet queue.
2208 	 */
2209 	so->so_rcv.sb_mb = NULL;
2210 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2211 
2212 	/*
2213 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2214 	 */
2215 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2216 		sbfree(&so->so_rcv, m2);
2217 
2218 	/*
2219 	 * Do a few last checks before we let go of the lock.
2220 	 */
2221 	SBLASTRECORDCHK(&so->so_rcv);
2222 	SBLASTMBUFCHK(&so->so_rcv);
2223 	SOCKBUF_UNLOCK(&so->so_rcv);
2224 
2225 	if (pr->pr_flags & PR_ADDR) {
2226 		KASSERT(m->m_type == MT_SONAME,
2227 		    ("m->m_type == %d", m->m_type));
2228 		if (psa != NULL)
2229 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2230 			    M_NOWAIT);
2231 		m = m_free(m);
2232 	}
2233 	if (m == NULL) {
2234 		/* XXXRW: Can this happen? */
2235 		return (0);
2236 	}
2237 
2238 	/*
2239 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2240 	 * queue.
2241 	 *
2242 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2243 	 * in the first mbuf chain on the socket buffer.  We call into the
2244 	 * protocol to perform externalization (or freeing if controlp ==
2245 	 * NULL).
2246 	 */
2247 	if (m->m_type == MT_CONTROL) {
2248 		struct mbuf *cm = NULL, *cmn;
2249 		struct mbuf **cme = &cm;
2250 
2251 		do {
2252 			m2 = m->m_next;
2253 			m->m_next = NULL;
2254 			*cme = m;
2255 			cme = &(*cme)->m_next;
2256 			m = m2;
2257 		} while (m != NULL && m->m_type == MT_CONTROL);
2258 		while (cm != NULL) {
2259 			cmn = cm->m_next;
2260 			cm->m_next = NULL;
2261 			if (pr->pr_domain->dom_externalize != NULL) {
2262 				error = (*pr->pr_domain->dom_externalize)
2263 				    (cm, controlp);
2264 			} else if (controlp != NULL)
2265 				*controlp = cm;
2266 			else
2267 				m_freem(cm);
2268 			if (controlp != NULL) {
2269 				while (*controlp != NULL)
2270 					controlp = &(*controlp)->m_next;
2271 			}
2272 			cm = cmn;
2273 		}
2274 	}
2275 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2276 
2277 	while (m != NULL && uio->uio_resid > 0) {
2278 		len = uio->uio_resid;
2279 		if (len > m->m_len)
2280 			len = m->m_len;
2281 		error = uiomove(mtod(m, char *), (int)len, uio);
2282 		if (error) {
2283 			m_freem(m);
2284 			return (error);
2285 		}
2286 		if (len == m->m_len)
2287 			m = m_free(m);
2288 		else {
2289 			m->m_data += len;
2290 			m->m_len -= len;
2291 		}
2292 	}
2293 	if (m != NULL)
2294 		flags |= MSG_TRUNC;
2295 	m_freem(m);
2296 	if (flagsp != NULL)
2297 		*flagsp |= flags;
2298 	return (0);
2299 }
2300 
2301 int
2302 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2303     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2304 {
2305 	int error;
2306 
2307 	CURVNET_SET(so->so_vnet);
2308 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2309 	    controlp, flagsp));
2310 	CURVNET_RESTORE();
2311 	return (error);
2312 }
2313 
2314 int
2315 soshutdown(struct socket *so, int how)
2316 {
2317 	struct protosw *pr = so->so_proto;
2318 	int error;
2319 
2320 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2321 		return (EINVAL);
2322 
2323 	CURVNET_SET(so->so_vnet);
2324 	if (pr->pr_usrreqs->pru_flush != NULL) {
2325 	        (*pr->pr_usrreqs->pru_flush)(so, how);
2326 	}
2327 	if (how != SHUT_WR)
2328 		sorflush(so);
2329 	if (how != SHUT_RD) {
2330 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2331 		CURVNET_RESTORE();
2332 		return (error);
2333 	}
2334 	CURVNET_RESTORE();
2335 	return (0);
2336 }
2337 
2338 void
2339 sorflush(struct socket *so)
2340 {
2341 	struct sockbuf *sb = &so->so_rcv;
2342 	struct protosw *pr = so->so_proto;
2343 	struct sockbuf asb;
2344 
2345 	VNET_SO_ASSERT(so);
2346 
2347 	/*
2348 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2349 	 * held, and in order to generally avoid holding the lock for a long
2350 	 * time, we make a copy of the socket buffer and clear the original
2351 	 * (except locks, state).  The new socket buffer copy won't have
2352 	 * initialized locks so we can only call routines that won't use or
2353 	 * assert those locks.
2354 	 *
2355 	 * Dislodge threads currently blocked in receive and wait to acquire
2356 	 * a lock against other simultaneous readers before clearing the
2357 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2358 	 * despite any existing socket disposition on interruptable waiting.
2359 	 */
2360 	socantrcvmore(so);
2361 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2362 
2363 	/*
2364 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2365 	 * and mutex data unchanged.
2366 	 */
2367 	SOCKBUF_LOCK(sb);
2368 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2369 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2370 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2371 	bzero(&sb->sb_startzero,
2372 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2373 	SOCKBUF_UNLOCK(sb);
2374 	sbunlock(sb);
2375 
2376 	/*
2377 	 * Dispose of special rights and flush the socket buffer.  Don't call
2378 	 * any unsafe routines (that rely on locks being initialized) on asb.
2379 	 */
2380 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2381 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2382 	sbrelease_internal(&asb, so);
2383 }
2384 
2385 /*
2386  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2387  * additional variant to handle the case where the option value needs to be
2388  * some kind of integer, but not a specific size.  In addition to their use
2389  * here, these functions are also called by the protocol-level pr_ctloutput()
2390  * routines.
2391  */
2392 int
2393 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2394 {
2395 	size_t	valsize;
2396 
2397 	/*
2398 	 * If the user gives us more than we wanted, we ignore it, but if we
2399 	 * don't get the minimum length the caller wants, we return EINVAL.
2400 	 * On success, sopt->sopt_valsize is set to however much we actually
2401 	 * retrieved.
2402 	 */
2403 	if ((valsize = sopt->sopt_valsize) < minlen)
2404 		return EINVAL;
2405 	if (valsize > len)
2406 		sopt->sopt_valsize = valsize = len;
2407 
2408 	if (sopt->sopt_td != NULL)
2409 		return (copyin(sopt->sopt_val, buf, valsize));
2410 
2411 	bcopy(sopt->sopt_val, buf, valsize);
2412 	return (0);
2413 }
2414 
2415 /*
2416  * Kernel version of setsockopt(2).
2417  *
2418  * XXX: optlen is size_t, not socklen_t
2419  */
2420 int
2421 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2422     size_t optlen)
2423 {
2424 	struct sockopt sopt;
2425 
2426 	sopt.sopt_level = level;
2427 	sopt.sopt_name = optname;
2428 	sopt.sopt_dir = SOPT_SET;
2429 	sopt.sopt_val = optval;
2430 	sopt.sopt_valsize = optlen;
2431 	sopt.sopt_td = NULL;
2432 	return (sosetopt(so, &sopt));
2433 }
2434 
2435 int
2436 sosetopt(struct socket *so, struct sockopt *sopt)
2437 {
2438 	int	error, optval;
2439 	struct	linger l;
2440 	struct	timeval tv;
2441 	u_long  val;
2442 	uint32_t val32;
2443 #ifdef MAC
2444 	struct mac extmac;
2445 #endif
2446 
2447 	CURVNET_SET(so->so_vnet);
2448 	error = 0;
2449 	if (sopt->sopt_level != SOL_SOCKET) {
2450 		if (so->so_proto->pr_ctloutput != NULL) {
2451 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2452 			CURVNET_RESTORE();
2453 			return (error);
2454 		}
2455 		error = ENOPROTOOPT;
2456 	} else {
2457 		switch (sopt->sopt_name) {
2458 #ifdef INET
2459 		case SO_ACCEPTFILTER:
2460 			error = do_setopt_accept_filter(so, sopt);
2461 			if (error)
2462 				goto bad;
2463 			break;
2464 #endif
2465 		case SO_LINGER:
2466 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2467 			if (error)
2468 				goto bad;
2469 
2470 			SOCK_LOCK(so);
2471 			so->so_linger = l.l_linger;
2472 			if (l.l_onoff)
2473 				so->so_options |= SO_LINGER;
2474 			else
2475 				so->so_options &= ~SO_LINGER;
2476 			SOCK_UNLOCK(so);
2477 			break;
2478 
2479 		case SO_DEBUG:
2480 		case SO_KEEPALIVE:
2481 		case SO_DONTROUTE:
2482 		case SO_USELOOPBACK:
2483 		case SO_BROADCAST:
2484 		case SO_REUSEADDR:
2485 		case SO_REUSEPORT:
2486 		case SO_OOBINLINE:
2487 		case SO_TIMESTAMP:
2488 		case SO_BINTIME:
2489 		case SO_NOSIGPIPE:
2490 		case SO_NO_DDP:
2491 		case SO_NO_OFFLOAD:
2492 			error = sooptcopyin(sopt, &optval, sizeof optval,
2493 					    sizeof optval);
2494 			if (error)
2495 				goto bad;
2496 			SOCK_LOCK(so);
2497 			if (optval)
2498 				so->so_options |= sopt->sopt_name;
2499 			else
2500 				so->so_options &= ~sopt->sopt_name;
2501 			SOCK_UNLOCK(so);
2502 			break;
2503 
2504 		case SO_SETFIB:
2505 			error = sooptcopyin(sopt, &optval, sizeof optval,
2506 					    sizeof optval);
2507 			if (error)
2508 				goto bad;
2509 
2510 			if (optval < 0 || optval >= rt_numfibs) {
2511 				error = EINVAL;
2512 				goto bad;
2513 			}
2514 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2515 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2516 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2517 				so->so_fibnum = optval;
2518 			else
2519 				so->so_fibnum = 0;
2520 			break;
2521 
2522 		case SO_USER_COOKIE:
2523 			error = sooptcopyin(sopt, &val32, sizeof val32,
2524 					    sizeof val32);
2525 			if (error)
2526 				goto bad;
2527 			so->so_user_cookie = val32;
2528 			break;
2529 
2530 		case SO_SNDBUF:
2531 		case SO_RCVBUF:
2532 		case SO_SNDLOWAT:
2533 		case SO_RCVLOWAT:
2534 			error = sooptcopyin(sopt, &optval, sizeof optval,
2535 					    sizeof optval);
2536 			if (error)
2537 				goto bad;
2538 
2539 			/*
2540 			 * Values < 1 make no sense for any of these options,
2541 			 * so disallow them.
2542 			 */
2543 			if (optval < 1) {
2544 				error = EINVAL;
2545 				goto bad;
2546 			}
2547 
2548 			switch (sopt->sopt_name) {
2549 			case SO_SNDBUF:
2550 			case SO_RCVBUF:
2551 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2552 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2553 				    so, curthread) == 0) {
2554 					error = ENOBUFS;
2555 					goto bad;
2556 				}
2557 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2558 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2559 				break;
2560 
2561 			/*
2562 			 * Make sure the low-water is never greater than the
2563 			 * high-water.
2564 			 */
2565 			case SO_SNDLOWAT:
2566 				SOCKBUF_LOCK(&so->so_snd);
2567 				so->so_snd.sb_lowat =
2568 				    (optval > so->so_snd.sb_hiwat) ?
2569 				    so->so_snd.sb_hiwat : optval;
2570 				SOCKBUF_UNLOCK(&so->so_snd);
2571 				break;
2572 			case SO_RCVLOWAT:
2573 				SOCKBUF_LOCK(&so->so_rcv);
2574 				so->so_rcv.sb_lowat =
2575 				    (optval > so->so_rcv.sb_hiwat) ?
2576 				    so->so_rcv.sb_hiwat : optval;
2577 				SOCKBUF_UNLOCK(&so->so_rcv);
2578 				break;
2579 			}
2580 			break;
2581 
2582 		case SO_SNDTIMEO:
2583 		case SO_RCVTIMEO:
2584 #ifdef COMPAT_FREEBSD32
2585 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2586 				struct timeval32 tv32;
2587 
2588 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2589 				    sizeof tv32);
2590 				CP(tv32, tv, tv_sec);
2591 				CP(tv32, tv, tv_usec);
2592 			} else
2593 #endif
2594 				error = sooptcopyin(sopt, &tv, sizeof tv,
2595 				    sizeof tv);
2596 			if (error)
2597 				goto bad;
2598 
2599 			/* assert(hz > 0); */
2600 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2601 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2602 				error = EDOM;
2603 				goto bad;
2604 			}
2605 			/* assert(tick > 0); */
2606 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2607 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2608 			if (val > INT_MAX) {
2609 				error = EDOM;
2610 				goto bad;
2611 			}
2612 			if (val == 0 && tv.tv_usec != 0)
2613 				val = 1;
2614 
2615 			switch (sopt->sopt_name) {
2616 			case SO_SNDTIMEO:
2617 				so->so_snd.sb_timeo = val;
2618 				break;
2619 			case SO_RCVTIMEO:
2620 				so->so_rcv.sb_timeo = val;
2621 				break;
2622 			}
2623 			break;
2624 
2625 		case SO_LABEL:
2626 #ifdef MAC
2627 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2628 			    sizeof extmac);
2629 			if (error)
2630 				goto bad;
2631 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2632 			    so, &extmac);
2633 #else
2634 			error = EOPNOTSUPP;
2635 #endif
2636 			break;
2637 
2638 		default:
2639 			error = ENOPROTOOPT;
2640 			break;
2641 		}
2642 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2643 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2644 	}
2645 bad:
2646 	CURVNET_RESTORE();
2647 	return (error);
2648 }
2649 
2650 /*
2651  * Helper routine for getsockopt.
2652  */
2653 int
2654 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2655 {
2656 	int	error;
2657 	size_t	valsize;
2658 
2659 	error = 0;
2660 
2661 	/*
2662 	 * Documented get behavior is that we always return a value, possibly
2663 	 * truncated to fit in the user's buffer.  Traditional behavior is
2664 	 * that we always tell the user precisely how much we copied, rather
2665 	 * than something useful like the total amount we had available for
2666 	 * her.  Note that this interface is not idempotent; the entire
2667 	 * answer must generated ahead of time.
2668 	 */
2669 	valsize = min(len, sopt->sopt_valsize);
2670 	sopt->sopt_valsize = valsize;
2671 	if (sopt->sopt_val != NULL) {
2672 		if (sopt->sopt_td != NULL)
2673 			error = copyout(buf, sopt->sopt_val, valsize);
2674 		else
2675 			bcopy(buf, sopt->sopt_val, valsize);
2676 	}
2677 	return (error);
2678 }
2679 
2680 int
2681 sogetopt(struct socket *so, struct sockopt *sopt)
2682 {
2683 	int	error, optval;
2684 	struct	linger l;
2685 	struct	timeval tv;
2686 #ifdef MAC
2687 	struct mac extmac;
2688 #endif
2689 
2690 	CURVNET_SET(so->so_vnet);
2691 	error = 0;
2692 	if (sopt->sopt_level != SOL_SOCKET) {
2693 		if (so->so_proto->pr_ctloutput != NULL)
2694 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2695 		else
2696 			error = ENOPROTOOPT;
2697 		CURVNET_RESTORE();
2698 		return (error);
2699 	} else {
2700 		switch (sopt->sopt_name) {
2701 #ifdef INET
2702 		case SO_ACCEPTFILTER:
2703 			error = do_getopt_accept_filter(so, sopt);
2704 			break;
2705 #endif
2706 		case SO_LINGER:
2707 			SOCK_LOCK(so);
2708 			l.l_onoff = so->so_options & SO_LINGER;
2709 			l.l_linger = so->so_linger;
2710 			SOCK_UNLOCK(so);
2711 			error = sooptcopyout(sopt, &l, sizeof l);
2712 			break;
2713 
2714 		case SO_USELOOPBACK:
2715 		case SO_DONTROUTE:
2716 		case SO_DEBUG:
2717 		case SO_KEEPALIVE:
2718 		case SO_REUSEADDR:
2719 		case SO_REUSEPORT:
2720 		case SO_BROADCAST:
2721 		case SO_OOBINLINE:
2722 		case SO_ACCEPTCONN:
2723 		case SO_TIMESTAMP:
2724 		case SO_BINTIME:
2725 		case SO_NOSIGPIPE:
2726 			optval = so->so_options & sopt->sopt_name;
2727 integer:
2728 			error = sooptcopyout(sopt, &optval, sizeof optval);
2729 			break;
2730 
2731 		case SO_TYPE:
2732 			optval = so->so_type;
2733 			goto integer;
2734 
2735 		case SO_PROTOCOL:
2736 			optval = so->so_proto->pr_protocol;
2737 			goto integer;
2738 
2739 		case SO_ERROR:
2740 			SOCK_LOCK(so);
2741 			optval = so->so_error;
2742 			so->so_error = 0;
2743 			SOCK_UNLOCK(so);
2744 			goto integer;
2745 
2746 		case SO_SNDBUF:
2747 			optval = so->so_snd.sb_hiwat;
2748 			goto integer;
2749 
2750 		case SO_RCVBUF:
2751 			optval = so->so_rcv.sb_hiwat;
2752 			goto integer;
2753 
2754 		case SO_SNDLOWAT:
2755 			optval = so->so_snd.sb_lowat;
2756 			goto integer;
2757 
2758 		case SO_RCVLOWAT:
2759 			optval = so->so_rcv.sb_lowat;
2760 			goto integer;
2761 
2762 		case SO_SNDTIMEO:
2763 		case SO_RCVTIMEO:
2764 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2765 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2766 
2767 			tv.tv_sec = optval / hz;
2768 			tv.tv_usec = (optval % hz) * tick;
2769 #ifdef COMPAT_FREEBSD32
2770 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2771 				struct timeval32 tv32;
2772 
2773 				CP(tv, tv32, tv_sec);
2774 				CP(tv, tv32, tv_usec);
2775 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2776 			} else
2777 #endif
2778 				error = sooptcopyout(sopt, &tv, sizeof tv);
2779 			break;
2780 
2781 		case SO_LABEL:
2782 #ifdef MAC
2783 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2784 			    sizeof(extmac));
2785 			if (error)
2786 				goto bad;
2787 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2788 			    so, &extmac);
2789 			if (error)
2790 				goto bad;
2791 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2792 #else
2793 			error = EOPNOTSUPP;
2794 #endif
2795 			break;
2796 
2797 		case SO_PEERLABEL:
2798 #ifdef MAC
2799 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2800 			    sizeof(extmac));
2801 			if (error)
2802 				goto bad;
2803 			error = mac_getsockopt_peerlabel(
2804 			    sopt->sopt_td->td_ucred, so, &extmac);
2805 			if (error)
2806 				goto bad;
2807 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2808 #else
2809 			error = EOPNOTSUPP;
2810 #endif
2811 			break;
2812 
2813 		case SO_LISTENQLIMIT:
2814 			optval = so->so_qlimit;
2815 			goto integer;
2816 
2817 		case SO_LISTENQLEN:
2818 			optval = so->so_qlen;
2819 			goto integer;
2820 
2821 		case SO_LISTENINCQLEN:
2822 			optval = so->so_incqlen;
2823 			goto integer;
2824 
2825 		default:
2826 			error = ENOPROTOOPT;
2827 			break;
2828 		}
2829 	}
2830 #ifdef MAC
2831 bad:
2832 #endif
2833 	CURVNET_RESTORE();
2834 	return (error);
2835 }
2836 
2837 int
2838 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2839 {
2840 	struct mbuf *m, *m_prev;
2841 	int sopt_size = sopt->sopt_valsize;
2842 
2843 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2844 	if (m == NULL)
2845 		return ENOBUFS;
2846 	if (sopt_size > MLEN) {
2847 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2848 		if ((m->m_flags & M_EXT) == 0) {
2849 			m_free(m);
2850 			return ENOBUFS;
2851 		}
2852 		m->m_len = min(MCLBYTES, sopt_size);
2853 	} else {
2854 		m->m_len = min(MLEN, sopt_size);
2855 	}
2856 	sopt_size -= m->m_len;
2857 	*mp = m;
2858 	m_prev = m;
2859 
2860 	while (sopt_size) {
2861 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2862 		if (m == NULL) {
2863 			m_freem(*mp);
2864 			return ENOBUFS;
2865 		}
2866 		if (sopt_size > MLEN) {
2867 			MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2868 			    M_DONTWAIT);
2869 			if ((m->m_flags & M_EXT) == 0) {
2870 				m_freem(m);
2871 				m_freem(*mp);
2872 				return ENOBUFS;
2873 			}
2874 			m->m_len = min(MCLBYTES, sopt_size);
2875 		} else {
2876 			m->m_len = min(MLEN, sopt_size);
2877 		}
2878 		sopt_size -= m->m_len;
2879 		m_prev->m_next = m;
2880 		m_prev = m;
2881 	}
2882 	return (0);
2883 }
2884 
2885 int
2886 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2887 {
2888 	struct mbuf *m0 = m;
2889 
2890 	if (sopt->sopt_val == NULL)
2891 		return (0);
2892 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2893 		if (sopt->sopt_td != NULL) {
2894 			int error;
2895 
2896 			error = copyin(sopt->sopt_val, mtod(m, char *),
2897 				       m->m_len);
2898 			if (error != 0) {
2899 				m_freem(m0);
2900 				return(error);
2901 			}
2902 		} else
2903 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2904 		sopt->sopt_valsize -= m->m_len;
2905 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2906 		m = m->m_next;
2907 	}
2908 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2909 		panic("ip6_sooptmcopyin");
2910 	return (0);
2911 }
2912 
2913 int
2914 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2915 {
2916 	struct mbuf *m0 = m;
2917 	size_t valsize = 0;
2918 
2919 	if (sopt->sopt_val == NULL)
2920 		return (0);
2921 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2922 		if (sopt->sopt_td != NULL) {
2923 			int error;
2924 
2925 			error = copyout(mtod(m, char *), sopt->sopt_val,
2926 				       m->m_len);
2927 			if (error != 0) {
2928 				m_freem(m0);
2929 				return(error);
2930 			}
2931 		} else
2932 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2933 	       sopt->sopt_valsize -= m->m_len;
2934 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2935 	       valsize += m->m_len;
2936 	       m = m->m_next;
2937 	}
2938 	if (m != NULL) {
2939 		/* enough soopt buffer should be given from user-land */
2940 		m_freem(m0);
2941 		return(EINVAL);
2942 	}
2943 	sopt->sopt_valsize = valsize;
2944 	return (0);
2945 }
2946 
2947 /*
2948  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2949  * out-of-band data, which will then notify socket consumers.
2950  */
2951 void
2952 sohasoutofband(struct socket *so)
2953 {
2954 
2955 	if (so->so_sigio != NULL)
2956 		pgsigio(&so->so_sigio, SIGURG, 0);
2957 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2958 }
2959 
2960 int
2961 sopoll(struct socket *so, int events, struct ucred *active_cred,
2962     struct thread *td)
2963 {
2964 
2965 	/*
2966 	 * We do not need to set or assert curvnet as long as everyone uses
2967 	 * sopoll_generic().
2968 	 */
2969 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2970 	    td));
2971 }
2972 
2973 int
2974 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2975     struct thread *td)
2976 {
2977 	int revents = 0;
2978 
2979 	SOCKBUF_LOCK(&so->so_snd);
2980 	SOCKBUF_LOCK(&so->so_rcv);
2981 	if (events & (POLLIN | POLLRDNORM))
2982 		if (soreadabledata(so))
2983 			revents |= events & (POLLIN | POLLRDNORM);
2984 
2985 	if (events & (POLLOUT | POLLWRNORM))
2986 		if (sowriteable(so))
2987 			revents |= events & (POLLOUT | POLLWRNORM);
2988 
2989 	if (events & (POLLPRI | POLLRDBAND))
2990 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2991 			revents |= events & (POLLPRI | POLLRDBAND);
2992 
2993 	if ((events & POLLINIGNEOF) == 0) {
2994 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2995 			revents |= events & (POLLIN | POLLRDNORM);
2996 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2997 				revents |= POLLHUP;
2998 		}
2999 	}
3000 
3001 	if (revents == 0) {
3002 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3003 			selrecord(td, &so->so_rcv.sb_sel);
3004 			so->so_rcv.sb_flags |= SB_SEL;
3005 		}
3006 
3007 		if (events & (POLLOUT | POLLWRNORM)) {
3008 			selrecord(td, &so->so_snd.sb_sel);
3009 			so->so_snd.sb_flags |= SB_SEL;
3010 		}
3011 	}
3012 
3013 	SOCKBUF_UNLOCK(&so->so_rcv);
3014 	SOCKBUF_UNLOCK(&so->so_snd);
3015 	return (revents);
3016 }
3017 
3018 int
3019 soo_kqfilter(struct file *fp, struct knote *kn)
3020 {
3021 	struct socket *so = kn->kn_fp->f_data;
3022 	struct sockbuf *sb;
3023 
3024 	switch (kn->kn_filter) {
3025 	case EVFILT_READ:
3026 		if (so->so_options & SO_ACCEPTCONN)
3027 			kn->kn_fop = &solisten_filtops;
3028 		else
3029 			kn->kn_fop = &soread_filtops;
3030 		sb = &so->so_rcv;
3031 		break;
3032 	case EVFILT_WRITE:
3033 		kn->kn_fop = &sowrite_filtops;
3034 		sb = &so->so_snd;
3035 		break;
3036 	default:
3037 		return (EINVAL);
3038 	}
3039 
3040 	SOCKBUF_LOCK(sb);
3041 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3042 	sb->sb_flags |= SB_KNOTE;
3043 	SOCKBUF_UNLOCK(sb);
3044 	return (0);
3045 }
3046 
3047 /*
3048  * Some routines that return EOPNOTSUPP for entry points that are not
3049  * supported by a protocol.  Fill in as needed.
3050  */
3051 int
3052 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3053 {
3054 
3055 	return EOPNOTSUPP;
3056 }
3057 
3058 int
3059 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3060 {
3061 
3062 	return EOPNOTSUPP;
3063 }
3064 
3065 int
3066 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3067 {
3068 
3069 	return EOPNOTSUPP;
3070 }
3071 
3072 int
3073 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3074 {
3075 
3076 	return EOPNOTSUPP;
3077 }
3078 
3079 int
3080 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3081 {
3082 
3083 	return EOPNOTSUPP;
3084 }
3085 
3086 int
3087 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3088     struct ifnet *ifp, struct thread *td)
3089 {
3090 
3091 	return EOPNOTSUPP;
3092 }
3093 
3094 int
3095 pru_disconnect_notsupp(struct socket *so)
3096 {
3097 
3098 	return EOPNOTSUPP;
3099 }
3100 
3101 int
3102 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3103 {
3104 
3105 	return EOPNOTSUPP;
3106 }
3107 
3108 int
3109 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3110 {
3111 
3112 	return EOPNOTSUPP;
3113 }
3114 
3115 int
3116 pru_rcvd_notsupp(struct socket *so, int flags)
3117 {
3118 
3119 	return EOPNOTSUPP;
3120 }
3121 
3122 int
3123 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3124 {
3125 
3126 	return EOPNOTSUPP;
3127 }
3128 
3129 int
3130 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3131     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3132 {
3133 
3134 	return EOPNOTSUPP;
3135 }
3136 
3137 /*
3138  * This isn't really a ``null'' operation, but it's the default one and
3139  * doesn't do anything destructive.
3140  */
3141 int
3142 pru_sense_null(struct socket *so, struct stat *sb)
3143 {
3144 
3145 	sb->st_blksize = so->so_snd.sb_hiwat;
3146 	return 0;
3147 }
3148 
3149 int
3150 pru_shutdown_notsupp(struct socket *so)
3151 {
3152 
3153 	return EOPNOTSUPP;
3154 }
3155 
3156 int
3157 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3158 {
3159 
3160 	return EOPNOTSUPP;
3161 }
3162 
3163 int
3164 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3165     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3166 {
3167 
3168 	return EOPNOTSUPP;
3169 }
3170 
3171 int
3172 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3173     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3174 {
3175 
3176 	return EOPNOTSUPP;
3177 }
3178 
3179 int
3180 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3181     struct thread *td)
3182 {
3183 
3184 	return EOPNOTSUPP;
3185 }
3186 
3187 static void
3188 filt_sordetach(struct knote *kn)
3189 {
3190 	struct socket *so = kn->kn_fp->f_data;
3191 
3192 	SOCKBUF_LOCK(&so->so_rcv);
3193 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3194 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3195 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3196 	SOCKBUF_UNLOCK(&so->so_rcv);
3197 }
3198 
3199 /*ARGSUSED*/
3200 static int
3201 filt_soread(struct knote *kn, long hint)
3202 {
3203 	struct socket *so;
3204 
3205 	so = kn->kn_fp->f_data;
3206 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3207 
3208 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3209 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3210 		kn->kn_flags |= EV_EOF;
3211 		kn->kn_fflags = so->so_error;
3212 		return (1);
3213 	} else if (so->so_error)	/* temporary udp error */
3214 		return (1);
3215 	else if (kn->kn_sfflags & NOTE_LOWAT)
3216 		return (kn->kn_data >= kn->kn_sdata);
3217 	else
3218 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3219 }
3220 
3221 static void
3222 filt_sowdetach(struct knote *kn)
3223 {
3224 	struct socket *so = kn->kn_fp->f_data;
3225 
3226 	SOCKBUF_LOCK(&so->so_snd);
3227 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3228 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3229 		so->so_snd.sb_flags &= ~SB_KNOTE;
3230 	SOCKBUF_UNLOCK(&so->so_snd);
3231 }
3232 
3233 /*ARGSUSED*/
3234 static int
3235 filt_sowrite(struct knote *kn, long hint)
3236 {
3237 	struct socket *so;
3238 
3239 	so = kn->kn_fp->f_data;
3240 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3241 	kn->kn_data = sbspace(&so->so_snd);
3242 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3243 		kn->kn_flags |= EV_EOF;
3244 		kn->kn_fflags = so->so_error;
3245 		return (1);
3246 	} else if (so->so_error)	/* temporary udp error */
3247 		return (1);
3248 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3249 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3250 		return (0);
3251 	else if (kn->kn_sfflags & NOTE_LOWAT)
3252 		return (kn->kn_data >= kn->kn_sdata);
3253 	else
3254 		return (kn->kn_data >= so->so_snd.sb_lowat);
3255 }
3256 
3257 /*ARGSUSED*/
3258 static int
3259 filt_solisten(struct knote *kn, long hint)
3260 {
3261 	struct socket *so = kn->kn_fp->f_data;
3262 
3263 	kn->kn_data = so->so_qlen;
3264 	return (! TAILQ_EMPTY(&so->so_comp));
3265 }
3266 
3267 int
3268 socheckuid(struct socket *so, uid_t uid)
3269 {
3270 
3271 	if (so == NULL)
3272 		return (EPERM);
3273 	if (so->so_cred->cr_uid != uid)
3274 		return (EPERM);
3275 	return (0);
3276 }
3277 
3278 static int
3279 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3280 {
3281 	int error;
3282 	int val;
3283 
3284 	val = somaxconn;
3285 	error = sysctl_handle_int(oidp, &val, 0, req);
3286 	if (error || !req->newptr )
3287 		return (error);
3288 
3289 	if (val < 1 || val > USHRT_MAX)
3290 		return (EINVAL);
3291 
3292 	somaxconn = val;
3293 	return (0);
3294 }
3295 
3296 /*
3297  * These functions are used by protocols to notify the socket layer (and its
3298  * consumers) of state changes in the sockets driven by protocol-side events.
3299  */
3300 
3301 /*
3302  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3303  *
3304  * Normal sequence from the active (originating) side is that
3305  * soisconnecting() is called during processing of connect() call, resulting
3306  * in an eventual call to soisconnected() if/when the connection is
3307  * established.  When the connection is torn down soisdisconnecting() is
3308  * called during processing of disconnect() call, and soisdisconnected() is
3309  * called when the connection to the peer is totally severed.  The semantics
3310  * of these routines are such that connectionless protocols can call
3311  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3312  * calls when setting up a ``connection'' takes no time.
3313  *
3314  * From the passive side, a socket is created with two queues of sockets:
3315  * so_incomp for connections in progress and so_comp for connections already
3316  * made and awaiting user acceptance.  As a protocol is preparing incoming
3317  * connections, it creates a socket structure queued on so_incomp by calling
3318  * sonewconn().  When the connection is established, soisconnected() is
3319  * called, and transfers the socket structure to so_comp, making it available
3320  * to accept().
3321  *
3322  * If a socket is closed with sockets on either so_incomp or so_comp, these
3323  * sockets are dropped.
3324  *
3325  * If higher-level protocols are implemented in the kernel, the wakeups done
3326  * here will sometimes cause software-interrupt process scheduling.
3327  */
3328 void
3329 soisconnecting(struct socket *so)
3330 {
3331 
3332 	SOCK_LOCK(so);
3333 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3334 	so->so_state |= SS_ISCONNECTING;
3335 	SOCK_UNLOCK(so);
3336 }
3337 
3338 void
3339 soisconnected(struct socket *so)
3340 {
3341 	struct socket *head;
3342 	int ret;
3343 
3344 restart:
3345 	ACCEPT_LOCK();
3346 	SOCK_LOCK(so);
3347 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3348 	so->so_state |= SS_ISCONNECTED;
3349 	head = so->so_head;
3350 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3351 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3352 			SOCK_UNLOCK(so);
3353 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3354 			head->so_incqlen--;
3355 			so->so_qstate &= ~SQ_INCOMP;
3356 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3357 			head->so_qlen++;
3358 			so->so_qstate |= SQ_COMP;
3359 			ACCEPT_UNLOCK();
3360 			sorwakeup(head);
3361 			wakeup_one(&head->so_timeo);
3362 		} else {
3363 			ACCEPT_UNLOCK();
3364 			soupcall_set(so, SO_RCV,
3365 			    head->so_accf->so_accept_filter->accf_callback,
3366 			    head->so_accf->so_accept_filter_arg);
3367 			so->so_options &= ~SO_ACCEPTFILTER;
3368 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3369 			    head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3370 			if (ret == SU_ISCONNECTED)
3371 				soupcall_clear(so, SO_RCV);
3372 			SOCK_UNLOCK(so);
3373 			if (ret == SU_ISCONNECTED)
3374 				goto restart;
3375 		}
3376 		return;
3377 	}
3378 	SOCK_UNLOCK(so);
3379 	ACCEPT_UNLOCK();
3380 	wakeup(&so->so_timeo);
3381 	sorwakeup(so);
3382 	sowwakeup(so);
3383 }
3384 
3385 void
3386 soisdisconnecting(struct socket *so)
3387 {
3388 
3389 	/*
3390 	 * Note: This code assumes that SOCK_LOCK(so) and
3391 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3392 	 */
3393 	SOCKBUF_LOCK(&so->so_rcv);
3394 	so->so_state &= ~SS_ISCONNECTING;
3395 	so->so_state |= SS_ISDISCONNECTING;
3396 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3397 	sorwakeup_locked(so);
3398 	SOCKBUF_LOCK(&so->so_snd);
3399 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3400 	sowwakeup_locked(so);
3401 	wakeup(&so->so_timeo);
3402 }
3403 
3404 void
3405 soisdisconnected(struct socket *so)
3406 {
3407 
3408 	/*
3409 	 * Note: This code assumes that SOCK_LOCK(so) and
3410 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3411 	 */
3412 	SOCKBUF_LOCK(&so->so_rcv);
3413 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3414 	so->so_state |= SS_ISDISCONNECTED;
3415 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3416 	sorwakeup_locked(so);
3417 	SOCKBUF_LOCK(&so->so_snd);
3418 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3419 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3420 	sowwakeup_locked(so);
3421 	wakeup(&so->so_timeo);
3422 }
3423 
3424 /*
3425  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3426  */
3427 struct sockaddr *
3428 sodupsockaddr(const struct sockaddr *sa, int mflags)
3429 {
3430 	struct sockaddr *sa2;
3431 
3432 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3433 	if (sa2)
3434 		bcopy(sa, sa2, sa->sa_len);
3435 	return sa2;
3436 }
3437 
3438 /*
3439  * Register per-socket buffer upcalls.
3440  */
3441 void
3442 soupcall_set(struct socket *so, int which,
3443     int (*func)(struct socket *, void *, int), void *arg)
3444 {
3445 	struct sockbuf *sb;
3446 
3447 	switch (which) {
3448 	case SO_RCV:
3449 		sb = &so->so_rcv;
3450 		break;
3451 	case SO_SND:
3452 		sb = &so->so_snd;
3453 		break;
3454 	default:
3455 		panic("soupcall_set: bad which");
3456 	}
3457 	SOCKBUF_LOCK_ASSERT(sb);
3458 #if 0
3459 	/* XXX: accf_http actually wants to do this on purpose. */
3460 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3461 #endif
3462 	sb->sb_upcall = func;
3463 	sb->sb_upcallarg = arg;
3464 	sb->sb_flags |= SB_UPCALL;
3465 }
3466 
3467 void
3468 soupcall_clear(struct socket *so, int which)
3469 {
3470 	struct sockbuf *sb;
3471 
3472 	switch (which) {
3473 	case SO_RCV:
3474 		sb = &so->so_rcv;
3475 		break;
3476 	case SO_SND:
3477 		sb = &so->so_snd;
3478 		break;
3479 	default:
3480 		panic("soupcall_clear: bad which");
3481 	}
3482 	SOCKBUF_LOCK_ASSERT(sb);
3483 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3484 	sb->sb_upcall = NULL;
3485 	sb->sb_upcallarg = NULL;
3486 	sb->sb_flags &= ~SB_UPCALL;
3487 }
3488 
3489 /*
3490  * Create an external-format (``xsocket'') structure using the information in
3491  * the kernel-format socket structure pointed to by so.  This is done to
3492  * reduce the spew of irrelevant information over this interface, to isolate
3493  * user code from changes in the kernel structure, and potentially to provide
3494  * information-hiding if we decide that some of this information should be
3495  * hidden from users.
3496  */
3497 void
3498 sotoxsocket(struct socket *so, struct xsocket *xso)
3499 {
3500 
3501 	xso->xso_len = sizeof *xso;
3502 	xso->xso_so = so;
3503 	xso->so_type = so->so_type;
3504 	xso->so_options = so->so_options;
3505 	xso->so_linger = so->so_linger;
3506 	xso->so_state = so->so_state;
3507 	xso->so_pcb = so->so_pcb;
3508 	xso->xso_protocol = so->so_proto->pr_protocol;
3509 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3510 	xso->so_qlen = so->so_qlen;
3511 	xso->so_incqlen = so->so_incqlen;
3512 	xso->so_qlimit = so->so_qlimit;
3513 	xso->so_timeo = so->so_timeo;
3514 	xso->so_error = so->so_error;
3515 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3516 	xso->so_oobmark = so->so_oobmark;
3517 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3518 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3519 	xso->so_uid = so->so_cred->cr_uid;
3520 }
3521 
3522 
3523 /*
3524  * Socket accessor functions to provide external consumers with
3525  * a safe interface to socket state
3526  *
3527  */
3528 
3529 void
3530 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3531 {
3532 
3533 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3534 		func(so, arg);
3535 }
3536 
3537 struct sockbuf *
3538 so_sockbuf_rcv(struct socket *so)
3539 {
3540 
3541 	return (&so->so_rcv);
3542 }
3543 
3544 struct sockbuf *
3545 so_sockbuf_snd(struct socket *so)
3546 {
3547 
3548 	return (&so->so_snd);
3549 }
3550 
3551 int
3552 so_state_get(const struct socket *so)
3553 {
3554 
3555 	return (so->so_state);
3556 }
3557 
3558 void
3559 so_state_set(struct socket *so, int val)
3560 {
3561 
3562 	so->so_state = val;
3563 }
3564 
3565 int
3566 so_options_get(const struct socket *so)
3567 {
3568 
3569 	return (so->so_options);
3570 }
3571 
3572 void
3573 so_options_set(struct socket *so, int val)
3574 {
3575 
3576 	so->so_options = val;
3577 }
3578 
3579 int
3580 so_error_get(const struct socket *so)
3581 {
3582 
3583 	return (so->so_error);
3584 }
3585 
3586 void
3587 so_error_set(struct socket *so, int val)
3588 {
3589 
3590 	so->so_error = val;
3591 }
3592 
3593 int
3594 so_linger_get(const struct socket *so)
3595 {
3596 
3597 	return (so->so_linger);
3598 }
3599 
3600 void
3601 so_linger_set(struct socket *so, int val)
3602 {
3603 
3604 	so->so_linger = val;
3605 }
3606 
3607 struct protosw *
3608 so_protosw_get(const struct socket *so)
3609 {
3610 
3611 	return (so->so_proto);
3612 }
3613 
3614 void
3615 so_protosw_set(struct socket *so, struct protosw *val)
3616 {
3617 
3618 	so->so_proto = val;
3619 }
3620 
3621 void
3622 so_sorwakeup(struct socket *so)
3623 {
3624 
3625 	sorwakeup(so);
3626 }
3627 
3628 void
3629 so_sowwakeup(struct socket *so)
3630 {
3631 
3632 	sowwakeup(so);
3633 }
3634 
3635 void
3636 so_sorwakeup_locked(struct socket *so)
3637 {
3638 
3639 	sorwakeup_locked(so);
3640 }
3641 
3642 void
3643 so_sowwakeup_locked(struct socket *so)
3644 {
3645 
3646 	sowwakeup_locked(so);
3647 }
3648 
3649 void
3650 so_lock(struct socket *so)
3651 {
3652 	SOCK_LOCK(so);
3653 }
3654 
3655 void
3656 so_unlock(struct socket *so)
3657 {
3658 	SOCK_UNLOCK(so);
3659 }
3660