xref: /freebsd/sys/kern/uipc_socket.c (revision aaf04b7cb637af9400e6b9ae1bd531ef828a7c82)
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_comp 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, resid;
891 	int error;
892 #ifdef ZERO_COPY_SOCKETS
893 	int cow_send;
894 #endif
895 
896 	*retmp = top = NULL;
897 	mp = &top;
898 	len = 0;
899 	resid = uio->uio_resid;
900 	error = 0;
901 	do {
902 #ifdef ZERO_COPY_SOCKETS
903 		cow_send = 0;
904 #endif /* ZERO_COPY_SOCKETS */
905 		if (resid >= MINCLSIZE) {
906 #ifdef ZERO_COPY_SOCKETS
907 			if (top == NULL) {
908 				m = m_gethdr(M_WAITOK, MT_DATA);
909 				m->m_pkthdr.len = 0;
910 				m->m_pkthdr.rcvif = NULL;
911 			} else
912 				m = m_get(M_WAITOK, MT_DATA);
913 			if (so_zero_copy_send &&
914 			    resid>=PAGE_SIZE &&
915 			    *space>=PAGE_SIZE &&
916 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
917 				so_zerocp_stats.size_ok++;
918 				so_zerocp_stats.align_ok++;
919 				cow_send = socow_setup(m, uio);
920 				len = cow_send;
921 			}
922 			if (!cow_send) {
923 				m_clget(m, M_WAITOK);
924 				len = min(min(MCLBYTES, resid), *space);
925 			}
926 #else /* ZERO_COPY_SOCKETS */
927 			if (top == NULL) {
928 				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
929 				m->m_pkthdr.len = 0;
930 				m->m_pkthdr.rcvif = NULL;
931 			} else
932 				m = m_getcl(M_WAIT, MT_DATA, 0);
933 			len = min(min(MCLBYTES, resid), *space);
934 #endif /* ZERO_COPY_SOCKETS */
935 		} else {
936 			if (top == NULL) {
937 				m = m_gethdr(M_WAIT, MT_DATA);
938 				m->m_pkthdr.len = 0;
939 				m->m_pkthdr.rcvif = NULL;
940 
941 				len = min(min(MHLEN, resid), *space);
942 				/*
943 				 * For datagram protocols, leave room
944 				 * for protocol headers in first mbuf.
945 				 */
946 				if (atomic && m && len < MHLEN)
947 					MH_ALIGN(m, len);
948 			} else {
949 				m = m_get(M_WAIT, MT_DATA);
950 				len = min(min(MLEN, resid), *space);
951 			}
952 		}
953 		if (m == NULL) {
954 			error = ENOBUFS;
955 			goto out;
956 		}
957 
958 		*space -= len;
959 #ifdef ZERO_COPY_SOCKETS
960 		if (cow_send)
961 			error = 0;
962 		else
963 #endif /* ZERO_COPY_SOCKETS */
964 		error = uiomove(mtod(m, void *), (int)len, uio);
965 		resid = uio->uio_resid;
966 		m->m_len = len;
967 		*mp = m;
968 		top->m_pkthdr.len += len;
969 		if (error)
970 			goto out;
971 		mp = &m->m_next;
972 		if (resid <= 0) {
973 			if (flags & MSG_EOR)
974 				top->m_flags |= M_EOR;
975 			break;
976 		}
977 	} while (*space > 0 && atomic);
978 out:
979 	*retmp = top;
980 	return (error);
981 }
982 #endif /*ZERO_COPY_SOCKETS*/
983 
984 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
985 
986 int
987 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
988     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
989 {
990 	long space, resid;
991 	int clen = 0, error, dontroute;
992 #ifdef ZERO_COPY_SOCKETS
993 	int atomic = sosendallatonce(so) || top;
994 #endif
995 
996 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
997 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
998 	    ("sodgram_send: !PR_ATOMIC"));
999 
1000 	if (uio != NULL)
1001 		resid = uio->uio_resid;
1002 	else
1003 		resid = top->m_pkthdr.len;
1004 	/*
1005 	 * In theory resid should be unsigned.  However, space must be
1006 	 * signed, as it might be less than 0 if we over-committed, and we
1007 	 * must use a signed comparison of space and resid.  On the other
1008 	 * hand, a negative resid causes us to loop sending 0-length
1009 	 * segments to the protocol.
1010 	 */
1011 	if (resid < 0) {
1012 		error = EINVAL;
1013 		goto out;
1014 	}
1015 
1016 	dontroute =
1017 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1018 	if (td != NULL)
1019 		td->td_ru.ru_msgsnd++;
1020 	if (control != NULL)
1021 		clen = control->m_len;
1022 
1023 	SOCKBUF_LOCK(&so->so_snd);
1024 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1025 		SOCKBUF_UNLOCK(&so->so_snd);
1026 		error = EPIPE;
1027 		goto out;
1028 	}
1029 	if (so->so_error) {
1030 		error = so->so_error;
1031 		so->so_error = 0;
1032 		SOCKBUF_UNLOCK(&so->so_snd);
1033 		goto out;
1034 	}
1035 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1036 		/*
1037 		 * `sendto' and `sendmsg' is allowed on a connection-based
1038 		 * socket if it supports implied connect.  Return ENOTCONN if
1039 		 * not connected and no address is supplied.
1040 		 */
1041 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1042 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1043 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1044 			    !(resid == 0 && clen != 0)) {
1045 				SOCKBUF_UNLOCK(&so->so_snd);
1046 				error = ENOTCONN;
1047 				goto out;
1048 			}
1049 		} else if (addr == NULL) {
1050 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1051 				error = ENOTCONN;
1052 			else
1053 				error = EDESTADDRREQ;
1054 			SOCKBUF_UNLOCK(&so->so_snd);
1055 			goto out;
1056 		}
1057 	}
1058 
1059 	/*
1060 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1061 	 * problem and need fixing.
1062 	 */
1063 	space = sbspace(&so->so_snd);
1064 	if (flags & MSG_OOB)
1065 		space += 1024;
1066 	space -= clen;
1067 	SOCKBUF_UNLOCK(&so->so_snd);
1068 	if (resid > space) {
1069 		error = EMSGSIZE;
1070 		goto out;
1071 	}
1072 	if (uio == NULL) {
1073 		resid = 0;
1074 		if (flags & MSG_EOR)
1075 			top->m_flags |= M_EOR;
1076 	} else {
1077 #ifdef ZERO_COPY_SOCKETS
1078 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1079 		if (error)
1080 			goto out;
1081 #else
1082 		/*
1083 		 * Copy the data from userland into a mbuf chain.
1084 		 * If no data is to be copied in, a single empty mbuf
1085 		 * is returned.
1086 		 */
1087 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1088 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1089 		if (top == NULL) {
1090 			error = EFAULT;	/* only possible error */
1091 			goto out;
1092 		}
1093 		space -= resid - uio->uio_resid;
1094 #endif
1095 		resid = uio->uio_resid;
1096 	}
1097 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1098 	/*
1099 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1100 	 * than with.
1101 	 */
1102 	if (dontroute) {
1103 		SOCK_LOCK(so);
1104 		so->so_options |= SO_DONTROUTE;
1105 		SOCK_UNLOCK(so);
1106 	}
1107 	/*
1108 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1109 	 * of date.  We could have recieved a reset packet in an interrupt or
1110 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1111 	 * probably recheck again inside the locking protection here, but
1112 	 * there are probably other places that this also happens.  We must
1113 	 * rethink this.
1114 	 */
1115 	VNET_SO_ASSERT(so);
1116 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1117 	    (flags & MSG_OOB) ? PRUS_OOB :
1118 	/*
1119 	 * If the user set MSG_EOF, the protocol understands this flag and
1120 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1121 	 */
1122 	    ((flags & MSG_EOF) &&
1123 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1124 	     (resid <= 0)) ?
1125 		PRUS_EOF :
1126 		/* If there is more to send set PRUS_MORETOCOME */
1127 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1128 		top, addr, control, td);
1129 	if (dontroute) {
1130 		SOCK_LOCK(so);
1131 		so->so_options &= ~SO_DONTROUTE;
1132 		SOCK_UNLOCK(so);
1133 	}
1134 	clen = 0;
1135 	control = NULL;
1136 	top = NULL;
1137 out:
1138 	if (top != NULL)
1139 		m_freem(top);
1140 	if (control != NULL)
1141 		m_freem(control);
1142 	return (error);
1143 }
1144 
1145 /*
1146  * Send on a socket.  If send must go all at once and message is larger than
1147  * send buffering, then hard error.  Lock against other senders.  If must go
1148  * all at once and not enough room now, then inform user that this would
1149  * block and do nothing.  Otherwise, if nonblocking, send as much as
1150  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1151  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1152  * in mbuf chain must be small enough to send all at once.
1153  *
1154  * Returns nonzero on error, timeout or signal; callers must check for short
1155  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1156  * on return.
1157  */
1158 int
1159 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1160     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1161 {
1162 	long space, resid;
1163 	int clen = 0, error, dontroute;
1164 	int atomic = sosendallatonce(so) || top;
1165 
1166 	if (uio != NULL)
1167 		resid = uio->uio_resid;
1168 	else
1169 		resid = top->m_pkthdr.len;
1170 	/*
1171 	 * In theory resid should be unsigned.  However, space must be
1172 	 * signed, as it might be less than 0 if we over-committed, and we
1173 	 * must use a signed comparison of space and resid.  On the other
1174 	 * hand, a negative resid causes us to loop sending 0-length
1175 	 * segments to the protocol.
1176 	 *
1177 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1178 	 * type sockets since that's an error.
1179 	 */
1180 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1181 		error = EINVAL;
1182 		goto out;
1183 	}
1184 
1185 	dontroute =
1186 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1187 	    (so->so_proto->pr_flags & PR_ATOMIC);
1188 	if (td != NULL)
1189 		td->td_ru.ru_msgsnd++;
1190 	if (control != NULL)
1191 		clen = control->m_len;
1192 
1193 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1194 	if (error)
1195 		goto out;
1196 
1197 restart:
1198 	do {
1199 		SOCKBUF_LOCK(&so->so_snd);
1200 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1201 			SOCKBUF_UNLOCK(&so->so_snd);
1202 			error = EPIPE;
1203 			goto release;
1204 		}
1205 		if (so->so_error) {
1206 			error = so->so_error;
1207 			so->so_error = 0;
1208 			SOCKBUF_UNLOCK(&so->so_snd);
1209 			goto release;
1210 		}
1211 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1212 			/*
1213 			 * `sendto' and `sendmsg' is allowed on a connection-
1214 			 * based socket if it supports implied connect.
1215 			 * Return ENOTCONN if not connected and no address is
1216 			 * supplied.
1217 			 */
1218 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1219 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1220 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1221 				    !(resid == 0 && clen != 0)) {
1222 					SOCKBUF_UNLOCK(&so->so_snd);
1223 					error = ENOTCONN;
1224 					goto release;
1225 				}
1226 			} else if (addr == NULL) {
1227 				SOCKBUF_UNLOCK(&so->so_snd);
1228 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1229 					error = ENOTCONN;
1230 				else
1231 					error = EDESTADDRREQ;
1232 				goto release;
1233 			}
1234 		}
1235 		space = sbspace(&so->so_snd);
1236 		if (flags & MSG_OOB)
1237 			space += 1024;
1238 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1239 		    clen > so->so_snd.sb_hiwat) {
1240 			SOCKBUF_UNLOCK(&so->so_snd);
1241 			error = EMSGSIZE;
1242 			goto release;
1243 		}
1244 		if (space < resid + clen &&
1245 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1246 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1247 				SOCKBUF_UNLOCK(&so->so_snd);
1248 				error = EWOULDBLOCK;
1249 				goto release;
1250 			}
1251 			error = sbwait(&so->so_snd);
1252 			SOCKBUF_UNLOCK(&so->so_snd);
1253 			if (error)
1254 				goto release;
1255 			goto restart;
1256 		}
1257 		SOCKBUF_UNLOCK(&so->so_snd);
1258 		space -= clen;
1259 		do {
1260 			if (uio == NULL) {
1261 				resid = 0;
1262 				if (flags & MSG_EOR)
1263 					top->m_flags |= M_EOR;
1264 			} else {
1265 #ifdef ZERO_COPY_SOCKETS
1266 				error = sosend_copyin(uio, &top, atomic,
1267 				    &space, flags);
1268 				if (error != 0)
1269 					goto release;
1270 #else
1271 				/*
1272 				 * Copy the data from userland into a mbuf
1273 				 * chain.  If no data is to be copied in,
1274 				 * a single empty mbuf is returned.
1275 				 */
1276 				top = m_uiotombuf(uio, M_WAITOK, space,
1277 				    (atomic ? max_hdr : 0),
1278 				    (atomic ? M_PKTHDR : 0) |
1279 				    ((flags & MSG_EOR) ? M_EOR : 0));
1280 				if (top == NULL) {
1281 					error = EFAULT; /* only possible error */
1282 					goto release;
1283 				}
1284 				space -= resid - uio->uio_resid;
1285 #endif
1286 				resid = uio->uio_resid;
1287 			}
1288 			if (dontroute) {
1289 				SOCK_LOCK(so);
1290 				so->so_options |= SO_DONTROUTE;
1291 				SOCK_UNLOCK(so);
1292 			}
1293 			/*
1294 			 * XXX all the SBS_CANTSENDMORE checks previously
1295 			 * done could be out of date.  We could have recieved
1296 			 * a reset packet in an interrupt or maybe we slept
1297 			 * while doing page faults in uiomove() etc.  We
1298 			 * could probably recheck again inside the locking
1299 			 * protection here, but there are probably other
1300 			 * places that this also happens.  We must rethink
1301 			 * this.
1302 			 */
1303 			VNET_SO_ASSERT(so);
1304 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1305 			    (flags & MSG_OOB) ? PRUS_OOB :
1306 			/*
1307 			 * If the user set MSG_EOF, the protocol understands
1308 			 * this flag and nothing left to send then use
1309 			 * PRU_SEND_EOF instead of PRU_SEND.
1310 			 */
1311 			    ((flags & MSG_EOF) &&
1312 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1313 			     (resid <= 0)) ?
1314 				PRUS_EOF :
1315 			/* If there is more to send set PRUS_MORETOCOME. */
1316 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1317 			    top, addr, control, td);
1318 			if (dontroute) {
1319 				SOCK_LOCK(so);
1320 				so->so_options &= ~SO_DONTROUTE;
1321 				SOCK_UNLOCK(so);
1322 			}
1323 			clen = 0;
1324 			control = NULL;
1325 			top = NULL;
1326 			if (error)
1327 				goto release;
1328 		} while (resid && space > 0);
1329 	} while (resid);
1330 
1331 release:
1332 	sbunlock(&so->so_snd);
1333 out:
1334 	if (top != NULL)
1335 		m_freem(top);
1336 	if (control != NULL)
1337 		m_freem(control);
1338 	return (error);
1339 }
1340 
1341 int
1342 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1343     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1344 {
1345 	int error;
1346 
1347 	CURVNET_SET(so->so_vnet);
1348 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1349 	    control, flags, td);
1350 	CURVNET_RESTORE();
1351 	return (error);
1352 }
1353 
1354 /*
1355  * The part of soreceive() that implements reading non-inline out-of-band
1356  * data from a socket.  For more complete comments, see soreceive(), from
1357  * which this code originated.
1358  *
1359  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1360  * unable to return an mbuf chain to the caller.
1361  */
1362 static int
1363 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1364 {
1365 	struct protosw *pr = so->so_proto;
1366 	struct mbuf *m;
1367 	int error;
1368 
1369 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1370 	VNET_SO_ASSERT(so);
1371 
1372 	m = m_get(M_WAIT, MT_DATA);
1373 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1374 	if (error)
1375 		goto bad;
1376 	do {
1377 #ifdef ZERO_COPY_SOCKETS
1378 		if (so_zero_copy_receive) {
1379 			int disposable;
1380 
1381 			if ((m->m_flags & M_EXT)
1382 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1383 				disposable = 1;
1384 			else
1385 				disposable = 0;
1386 
1387 			error = uiomoveco(mtod(m, void *),
1388 					  min(uio->uio_resid, m->m_len),
1389 					  uio, disposable);
1390 		} else
1391 #endif /* ZERO_COPY_SOCKETS */
1392 		error = uiomove(mtod(m, void *),
1393 		    (int) min(uio->uio_resid, m->m_len), uio);
1394 		m = m_free(m);
1395 	} while (uio->uio_resid && error == 0 && m);
1396 bad:
1397 	if (m != NULL)
1398 		m_freem(m);
1399 	return (error);
1400 }
1401 
1402 /*
1403  * Following replacement or removal of the first mbuf on the first mbuf chain
1404  * of a socket buffer, push necessary state changes back into the socket
1405  * buffer so that other consumers see the values consistently.  'nextrecord'
1406  * is the callers locally stored value of the original value of
1407  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1408  * NOTE: 'nextrecord' may be NULL.
1409  */
1410 static __inline void
1411 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1412 {
1413 
1414 	SOCKBUF_LOCK_ASSERT(sb);
1415 	/*
1416 	 * First, update for the new value of nextrecord.  If necessary, make
1417 	 * it the first record.
1418 	 */
1419 	if (sb->sb_mb != NULL)
1420 		sb->sb_mb->m_nextpkt = nextrecord;
1421 	else
1422 		sb->sb_mb = nextrecord;
1423 
1424         /*
1425          * Now update any dependent socket buffer fields to reflect the new
1426          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1427 	 * addition of a second clause that takes care of the case where
1428 	 * sb_mb has been updated, but remains the last record.
1429          */
1430         if (sb->sb_mb == NULL) {
1431                 sb->sb_mbtail = NULL;
1432                 sb->sb_lastrecord = NULL;
1433         } else if (sb->sb_mb->m_nextpkt == NULL)
1434                 sb->sb_lastrecord = sb->sb_mb;
1435 }
1436 
1437 
1438 /*
1439  * Implement receive operations on a socket.  We depend on the way that
1440  * records are added to the sockbuf by sbappend.  In particular, each record
1441  * (mbufs linked through m_next) must begin with an address if the protocol
1442  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1443  * data, and then zero or more mbufs of data.  In order to allow parallelism
1444  * between network receive and copying to user space, as well as avoid
1445  * sleeping with a mutex held, we release the socket buffer mutex during the
1446  * user space copy.  Although the sockbuf is locked, new data may still be
1447  * appended, and thus we must maintain consistency of the sockbuf during that
1448  * time.
1449  *
1450  * The caller may receive the data as a single mbuf chain by supplying an
1451  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1452  * the count in uio_resid.
1453  */
1454 int
1455 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1456     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1457 {
1458 	struct mbuf *m, **mp;
1459 	int flags, len, error, offset;
1460 	struct protosw *pr = so->so_proto;
1461 	struct mbuf *nextrecord;
1462 	int moff, type = 0;
1463 	int orig_resid = uio->uio_resid;
1464 
1465 	mp = mp0;
1466 	if (psa != NULL)
1467 		*psa = NULL;
1468 	if (controlp != NULL)
1469 		*controlp = NULL;
1470 	if (flagsp != NULL)
1471 		flags = *flagsp &~ MSG_EOR;
1472 	else
1473 		flags = 0;
1474 	if (flags & MSG_OOB)
1475 		return (soreceive_rcvoob(so, uio, flags));
1476 	if (mp != NULL)
1477 		*mp = NULL;
1478 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1479 	    && uio->uio_resid) {
1480 		VNET_SO_ASSERT(so);
1481 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1482 	}
1483 
1484 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1485 	if (error)
1486 		return (error);
1487 
1488 restart:
1489 	SOCKBUF_LOCK(&so->so_rcv);
1490 	m = so->so_rcv.sb_mb;
1491 	/*
1492 	 * If we have less data than requested, block awaiting more (subject
1493 	 * to any timeout) if:
1494 	 *   1. the current count is less than the low water mark, or
1495 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1496 	 *	receive operation at once if we block (resid <= hiwat).
1497 	 *   3. MSG_DONTWAIT is not set
1498 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1499 	 * we have to do the receive in sections, and thus risk returning a
1500 	 * short count if a timeout or signal occurs after we start.
1501 	 */
1502 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1503 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1504 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1505 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1506 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1507 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1508 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1509 		    m, so->so_rcv.sb_cc));
1510 		if (so->so_error) {
1511 			if (m != NULL)
1512 				goto dontblock;
1513 			error = so->so_error;
1514 			if ((flags & MSG_PEEK) == 0)
1515 				so->so_error = 0;
1516 			SOCKBUF_UNLOCK(&so->so_rcv);
1517 			goto release;
1518 		}
1519 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1520 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1521 			if (m == NULL) {
1522 				SOCKBUF_UNLOCK(&so->so_rcv);
1523 				goto release;
1524 			} else
1525 				goto dontblock;
1526 		}
1527 		for (; m != NULL; m = m->m_next)
1528 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1529 				m = so->so_rcv.sb_mb;
1530 				goto dontblock;
1531 			}
1532 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1533 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1534 			SOCKBUF_UNLOCK(&so->so_rcv);
1535 			error = ENOTCONN;
1536 			goto release;
1537 		}
1538 		if (uio->uio_resid == 0) {
1539 			SOCKBUF_UNLOCK(&so->so_rcv);
1540 			goto release;
1541 		}
1542 		if ((so->so_state & SS_NBIO) ||
1543 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1544 			SOCKBUF_UNLOCK(&so->so_rcv);
1545 			error = EWOULDBLOCK;
1546 			goto release;
1547 		}
1548 		SBLASTRECORDCHK(&so->so_rcv);
1549 		SBLASTMBUFCHK(&so->so_rcv);
1550 		error = sbwait(&so->so_rcv);
1551 		SOCKBUF_UNLOCK(&so->so_rcv);
1552 		if (error)
1553 			goto release;
1554 		goto restart;
1555 	}
1556 dontblock:
1557 	/*
1558 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1559 	 * pointer to the next record in the socket buffer.  We must keep the
1560 	 * various socket buffer pointers and local stack versions of the
1561 	 * pointers in sync, pushing out modifications before dropping the
1562 	 * socket buffer mutex, and re-reading them when picking it up.
1563 	 *
1564 	 * Otherwise, we will race with the network stack appending new data
1565 	 * or records onto the socket buffer by using inconsistent/stale
1566 	 * versions of the field, possibly resulting in socket buffer
1567 	 * corruption.
1568 	 *
1569 	 * By holding the high-level sblock(), we prevent simultaneous
1570 	 * readers from pulling off the front of the socket buffer.
1571 	 */
1572 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1573 	if (uio->uio_td)
1574 		uio->uio_td->td_ru.ru_msgrcv++;
1575 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1576 	SBLASTRECORDCHK(&so->so_rcv);
1577 	SBLASTMBUFCHK(&so->so_rcv);
1578 	nextrecord = m->m_nextpkt;
1579 	if (pr->pr_flags & PR_ADDR) {
1580 		KASSERT(m->m_type == MT_SONAME,
1581 		    ("m->m_type == %d", m->m_type));
1582 		orig_resid = 0;
1583 		if (psa != NULL)
1584 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1585 			    M_NOWAIT);
1586 		if (flags & MSG_PEEK) {
1587 			m = m->m_next;
1588 		} else {
1589 			sbfree(&so->so_rcv, m);
1590 			so->so_rcv.sb_mb = m_free(m);
1591 			m = so->so_rcv.sb_mb;
1592 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1593 		}
1594 	}
1595 
1596 	/*
1597 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1598 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1599 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1600 	 * perform externalization (or freeing if controlp == NULL).
1601 	 */
1602 	if (m != NULL && m->m_type == MT_CONTROL) {
1603 		struct mbuf *cm = NULL, *cmn;
1604 		struct mbuf **cme = &cm;
1605 
1606 		do {
1607 			if (flags & MSG_PEEK) {
1608 				if (controlp != NULL) {
1609 					*controlp = m_copy(m, 0, m->m_len);
1610 					controlp = &(*controlp)->m_next;
1611 				}
1612 				m = m->m_next;
1613 			} else {
1614 				sbfree(&so->so_rcv, m);
1615 				so->so_rcv.sb_mb = m->m_next;
1616 				m->m_next = NULL;
1617 				*cme = m;
1618 				cme = &(*cme)->m_next;
1619 				m = so->so_rcv.sb_mb;
1620 			}
1621 		} while (m != NULL && m->m_type == MT_CONTROL);
1622 		if ((flags & MSG_PEEK) == 0)
1623 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1624 		while (cm != NULL) {
1625 			cmn = cm->m_next;
1626 			cm->m_next = NULL;
1627 			if (pr->pr_domain->dom_externalize != NULL) {
1628 				SOCKBUF_UNLOCK(&so->so_rcv);
1629 				VNET_SO_ASSERT(so);
1630 				error = (*pr->pr_domain->dom_externalize)
1631 				    (cm, controlp);
1632 				SOCKBUF_LOCK(&so->so_rcv);
1633 			} else if (controlp != NULL)
1634 				*controlp = cm;
1635 			else
1636 				m_freem(cm);
1637 			if (controlp != NULL) {
1638 				orig_resid = 0;
1639 				while (*controlp != NULL)
1640 					controlp = &(*controlp)->m_next;
1641 			}
1642 			cm = cmn;
1643 		}
1644 		if (m != NULL)
1645 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1646 		else
1647 			nextrecord = so->so_rcv.sb_mb;
1648 		orig_resid = 0;
1649 	}
1650 	if (m != NULL) {
1651 		if ((flags & MSG_PEEK) == 0) {
1652 			KASSERT(m->m_nextpkt == nextrecord,
1653 			    ("soreceive: post-control, nextrecord !sync"));
1654 			if (nextrecord == NULL) {
1655 				KASSERT(so->so_rcv.sb_mb == m,
1656 				    ("soreceive: post-control, sb_mb!=m"));
1657 				KASSERT(so->so_rcv.sb_lastrecord == m,
1658 				    ("soreceive: post-control, lastrecord!=m"));
1659 			}
1660 		}
1661 		type = m->m_type;
1662 		if (type == MT_OOBDATA)
1663 			flags |= MSG_OOB;
1664 	} else {
1665 		if ((flags & MSG_PEEK) == 0) {
1666 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1667 			    ("soreceive: sb_mb != nextrecord"));
1668 			if (so->so_rcv.sb_mb == NULL) {
1669 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1670 				    ("soreceive: sb_lastercord != NULL"));
1671 			}
1672 		}
1673 	}
1674 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1675 	SBLASTRECORDCHK(&so->so_rcv);
1676 	SBLASTMBUFCHK(&so->so_rcv);
1677 
1678 	/*
1679 	 * Now continue to read any data mbufs off of the head of the socket
1680 	 * buffer until the read request is satisfied.  Note that 'type' is
1681 	 * used to store the type of any mbuf reads that have happened so far
1682 	 * such that soreceive() can stop reading if the type changes, which
1683 	 * causes soreceive() to return only one of regular data and inline
1684 	 * out-of-band data in a single socket receive operation.
1685 	 */
1686 	moff = 0;
1687 	offset = 0;
1688 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1689 		/*
1690 		 * If the type of mbuf has changed since the last mbuf
1691 		 * examined ('type'), end the receive operation.
1692 	 	 */
1693 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1694 		if (m->m_type == MT_OOBDATA) {
1695 			if (type != MT_OOBDATA)
1696 				break;
1697 		} else if (type == MT_OOBDATA)
1698 			break;
1699 		else
1700 		    KASSERT(m->m_type == MT_DATA,
1701 			("m->m_type == %d", m->m_type));
1702 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1703 		len = uio->uio_resid;
1704 		if (so->so_oobmark && len > so->so_oobmark - offset)
1705 			len = so->so_oobmark - offset;
1706 		if (len > m->m_len - moff)
1707 			len = m->m_len - moff;
1708 		/*
1709 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1710 		 * them out via the uio, then free.  Sockbuf must be
1711 		 * consistent here (points to current mbuf, it points to next
1712 		 * record) when we drop priority; we must note any additions
1713 		 * to the sockbuf when we block interrupts again.
1714 		 */
1715 		if (mp == NULL) {
1716 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1717 			SBLASTRECORDCHK(&so->so_rcv);
1718 			SBLASTMBUFCHK(&so->so_rcv);
1719 			SOCKBUF_UNLOCK(&so->so_rcv);
1720 #ifdef ZERO_COPY_SOCKETS
1721 			if (so_zero_copy_receive) {
1722 				int disposable;
1723 
1724 				if ((m->m_flags & M_EXT)
1725 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1726 					disposable = 1;
1727 				else
1728 					disposable = 0;
1729 
1730 				error = uiomoveco(mtod(m, char *) + moff,
1731 						  (int)len, uio,
1732 						  disposable);
1733 			} else
1734 #endif /* ZERO_COPY_SOCKETS */
1735 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1736 			SOCKBUF_LOCK(&so->so_rcv);
1737 			if (error) {
1738 				/*
1739 				 * The MT_SONAME mbuf has already been removed
1740 				 * from the record, so it is necessary to
1741 				 * remove the data mbufs, if any, to preserve
1742 				 * the invariant in the case of PR_ADDR that
1743 				 * requires MT_SONAME mbufs at the head of
1744 				 * each record.
1745 				 */
1746 				if (m && pr->pr_flags & PR_ATOMIC &&
1747 				    ((flags & MSG_PEEK) == 0))
1748 					(void)sbdroprecord_locked(&so->so_rcv);
1749 				SOCKBUF_UNLOCK(&so->so_rcv);
1750 				goto release;
1751 			}
1752 		} else
1753 			uio->uio_resid -= len;
1754 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1755 		if (len == m->m_len - moff) {
1756 			if (m->m_flags & M_EOR)
1757 				flags |= MSG_EOR;
1758 			if (flags & MSG_PEEK) {
1759 				m = m->m_next;
1760 				moff = 0;
1761 			} else {
1762 				nextrecord = m->m_nextpkt;
1763 				sbfree(&so->so_rcv, m);
1764 				if (mp != NULL) {
1765 					*mp = m;
1766 					mp = &m->m_next;
1767 					so->so_rcv.sb_mb = m = m->m_next;
1768 					*mp = NULL;
1769 				} else {
1770 					so->so_rcv.sb_mb = m_free(m);
1771 					m = so->so_rcv.sb_mb;
1772 				}
1773 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1774 				SBLASTRECORDCHK(&so->so_rcv);
1775 				SBLASTMBUFCHK(&so->so_rcv);
1776 			}
1777 		} else {
1778 			if (flags & MSG_PEEK)
1779 				moff += len;
1780 			else {
1781 				if (mp != NULL) {
1782 					int copy_flag;
1783 
1784 					if (flags & MSG_DONTWAIT)
1785 						copy_flag = M_DONTWAIT;
1786 					else
1787 						copy_flag = M_WAIT;
1788 					if (copy_flag == M_WAIT)
1789 						SOCKBUF_UNLOCK(&so->so_rcv);
1790 					*mp = m_copym(m, 0, len, copy_flag);
1791 					if (copy_flag == M_WAIT)
1792 						SOCKBUF_LOCK(&so->so_rcv);
1793  					if (*mp == NULL) {
1794  						/*
1795  						 * m_copym() couldn't
1796 						 * allocate an mbuf.  Adjust
1797 						 * uio_resid back (it was
1798 						 * adjusted down by len
1799 						 * bytes, which we didn't end
1800 						 * up "copying" over).
1801  						 */
1802  						uio->uio_resid += len;
1803  						break;
1804  					}
1805 				}
1806 				m->m_data += len;
1807 				m->m_len -= len;
1808 				so->so_rcv.sb_cc -= len;
1809 			}
1810 		}
1811 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1812 		if (so->so_oobmark) {
1813 			if ((flags & MSG_PEEK) == 0) {
1814 				so->so_oobmark -= len;
1815 				if (so->so_oobmark == 0) {
1816 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1817 					break;
1818 				}
1819 			} else {
1820 				offset += len;
1821 				if (offset == so->so_oobmark)
1822 					break;
1823 			}
1824 		}
1825 		if (flags & MSG_EOR)
1826 			break;
1827 		/*
1828 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1829 		 * must not quit until "uio->uio_resid == 0" or an error
1830 		 * termination.  If a signal/timeout occurs, return with a
1831 		 * short count but without error.  Keep sockbuf locked
1832 		 * against other readers.
1833 		 */
1834 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1835 		    !sosendallatonce(so) && nextrecord == NULL) {
1836 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1837 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1838 				break;
1839 			/*
1840 			 * Notify the protocol that some data has been
1841 			 * drained before blocking.
1842 			 */
1843 			if (pr->pr_flags & PR_WANTRCVD) {
1844 				SOCKBUF_UNLOCK(&so->so_rcv);
1845 				VNET_SO_ASSERT(so);
1846 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1847 				SOCKBUF_LOCK(&so->so_rcv);
1848 			}
1849 			SBLASTRECORDCHK(&so->so_rcv);
1850 			SBLASTMBUFCHK(&so->so_rcv);
1851 			/*
1852 			 * We could receive some data while was notifying
1853 			 * the protocol. Skip blocking in this case.
1854 			 */
1855 			if (so->so_rcv.sb_mb == NULL) {
1856 				error = sbwait(&so->so_rcv);
1857 				if (error) {
1858 					SOCKBUF_UNLOCK(&so->so_rcv);
1859 					goto release;
1860 				}
1861 			}
1862 			m = so->so_rcv.sb_mb;
1863 			if (m != NULL)
1864 				nextrecord = m->m_nextpkt;
1865 		}
1866 	}
1867 
1868 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1869 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1870 		flags |= MSG_TRUNC;
1871 		if ((flags & MSG_PEEK) == 0)
1872 			(void) sbdroprecord_locked(&so->so_rcv);
1873 	}
1874 	if ((flags & MSG_PEEK) == 0) {
1875 		if (m == NULL) {
1876 			/*
1877 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1878 			 * part makes sure sb_lastrecord is up-to-date if
1879 			 * there is still data in the socket buffer.
1880 			 */
1881 			so->so_rcv.sb_mb = nextrecord;
1882 			if (so->so_rcv.sb_mb == NULL) {
1883 				so->so_rcv.sb_mbtail = NULL;
1884 				so->so_rcv.sb_lastrecord = NULL;
1885 			} else if (nextrecord->m_nextpkt == NULL)
1886 				so->so_rcv.sb_lastrecord = nextrecord;
1887 		}
1888 		SBLASTRECORDCHK(&so->so_rcv);
1889 		SBLASTMBUFCHK(&so->so_rcv);
1890 		/*
1891 		 * If soreceive() is being done from the socket callback,
1892 		 * then don't need to generate ACK to peer to update window,
1893 		 * since ACK will be generated on return to TCP.
1894 		 */
1895 		if (!(flags & MSG_SOCALLBCK) &&
1896 		    (pr->pr_flags & PR_WANTRCVD)) {
1897 			SOCKBUF_UNLOCK(&so->so_rcv);
1898 			VNET_SO_ASSERT(so);
1899 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1900 			SOCKBUF_LOCK(&so->so_rcv);
1901 		}
1902 	}
1903 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1904 	if (orig_resid == uio->uio_resid && orig_resid &&
1905 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1906 		SOCKBUF_UNLOCK(&so->so_rcv);
1907 		goto restart;
1908 	}
1909 	SOCKBUF_UNLOCK(&so->so_rcv);
1910 
1911 	if (flagsp != NULL)
1912 		*flagsp |= flags;
1913 release:
1914 	sbunlock(&so->so_rcv);
1915 	return (error);
1916 }
1917 
1918 /*
1919  * Optimized version of soreceive() for stream (TCP) sockets.
1920  */
1921 int
1922 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1923     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1924 {
1925 	int len = 0, error = 0, flags, oresid;
1926 	struct sockbuf *sb;
1927 	struct mbuf *m, *n = NULL;
1928 
1929 	/* We only do stream sockets. */
1930 	if (so->so_type != SOCK_STREAM)
1931 		return (EINVAL);
1932 	if (psa != NULL)
1933 		*psa = NULL;
1934 	if (controlp != NULL)
1935 		return (EINVAL);
1936 	if (flagsp != NULL)
1937 		flags = *flagsp &~ MSG_EOR;
1938 	else
1939 		flags = 0;
1940 	if (flags & MSG_OOB)
1941 		return (soreceive_rcvoob(so, uio, flags));
1942 	if (mp0 != NULL)
1943 		*mp0 = NULL;
1944 
1945 	sb = &so->so_rcv;
1946 
1947 	/* Prevent other readers from entering the socket. */
1948 	error = sblock(sb, SBLOCKWAIT(flags));
1949 	if (error)
1950 		goto out;
1951 	SOCKBUF_LOCK(sb);
1952 
1953 	/* Easy one, no space to copyout anything. */
1954 	if (uio->uio_resid == 0) {
1955 		error = EINVAL;
1956 		goto out;
1957 	}
1958 	oresid = uio->uio_resid;
1959 
1960 	/* We will never ever get anything unless we are or were connected. */
1961 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1962 		error = ENOTCONN;
1963 		goto out;
1964 	}
1965 
1966 restart:
1967 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1968 
1969 	/* Abort if socket has reported problems. */
1970 	if (so->so_error) {
1971 		if (sb->sb_cc > 0)
1972 			goto deliver;
1973 		if (oresid > uio->uio_resid)
1974 			goto out;
1975 		error = so->so_error;
1976 		if (!(flags & MSG_PEEK))
1977 			so->so_error = 0;
1978 		goto out;
1979 	}
1980 
1981 	/* Door is closed.  Deliver what is left, if any. */
1982 	if (sb->sb_state & SBS_CANTRCVMORE) {
1983 		if (sb->sb_cc > 0)
1984 			goto deliver;
1985 		else
1986 			goto out;
1987 	}
1988 
1989 	/* Socket buffer is empty and we shall not block. */
1990 	if (sb->sb_cc == 0 &&
1991 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1992 		error = EAGAIN;
1993 		goto out;
1994 	}
1995 
1996 	/* Socket buffer got some data that we shall deliver now. */
1997 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1998 	    ((sb->sb_flags & SS_NBIO) ||
1999 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2000 	     sb->sb_cc >= sb->sb_lowat ||
2001 	     sb->sb_cc >= uio->uio_resid ||
2002 	     sb->sb_cc >= sb->sb_hiwat) ) {
2003 		goto deliver;
2004 	}
2005 
2006 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2007 	if ((flags & MSG_WAITALL) &&
2008 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
2009 		goto deliver;
2010 
2011 	/*
2012 	 * Wait and block until (more) data comes in.
2013 	 * NB: Drops the sockbuf lock during wait.
2014 	 */
2015 	error = sbwait(sb);
2016 	if (error)
2017 		goto out;
2018 	goto restart;
2019 
2020 deliver:
2021 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2022 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2023 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2024 
2025 	/* Statistics. */
2026 	if (uio->uio_td)
2027 		uio->uio_td->td_ru.ru_msgrcv++;
2028 
2029 	/* Fill uio until full or current end of socket buffer is reached. */
2030 	len = min(uio->uio_resid, sb->sb_cc);
2031 	if (mp0 != NULL) {
2032 		/* Dequeue as many mbufs as possible. */
2033 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2034 			for (*mp0 = m = sb->sb_mb;
2035 			     m != NULL && m->m_len <= len;
2036 			     m = m->m_next) {
2037 				len -= m->m_len;
2038 				uio->uio_resid -= m->m_len;
2039 				sbfree(sb, m);
2040 				n = m;
2041 			}
2042 			sb->sb_mb = m;
2043 			if (sb->sb_mb == NULL)
2044 				SB_EMPTY_FIXUP(sb);
2045 			n->m_next = NULL;
2046 		}
2047 		/* Copy the remainder. */
2048 		if (len > 0) {
2049 			KASSERT(sb->sb_mb != NULL,
2050 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2051 
2052 			m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2053 			if (m == NULL)
2054 				len = 0;	/* Don't flush data from sockbuf. */
2055 			else
2056 				uio->uio_resid -= m->m_len;
2057 			if (*mp0 != NULL)
2058 				n->m_next = m;
2059 			else
2060 				*mp0 = m;
2061 			if (*mp0 == NULL) {
2062 				error = ENOBUFS;
2063 				goto out;
2064 			}
2065 		}
2066 	} else {
2067 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2068 		SOCKBUF_UNLOCK(sb);
2069 		error = m_mbuftouio(uio, sb->sb_mb, len);
2070 		SOCKBUF_LOCK(sb);
2071 		if (error)
2072 			goto out;
2073 	}
2074 	SBLASTRECORDCHK(sb);
2075 	SBLASTMBUFCHK(sb);
2076 
2077 	/*
2078 	 * Remove the delivered data from the socket buffer unless we
2079 	 * were only peeking.
2080 	 */
2081 	if (!(flags & MSG_PEEK)) {
2082 		if (len > 0)
2083 			sbdrop_locked(sb, len);
2084 
2085 		/* Notify protocol that we drained some data. */
2086 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2087 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2088 		     !(flags & MSG_SOCALLBCK))) {
2089 			SOCKBUF_UNLOCK(sb);
2090 			VNET_SO_ASSERT(so);
2091 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2092 			SOCKBUF_LOCK(sb);
2093 		}
2094 	}
2095 
2096 	/*
2097 	 * For MSG_WAITALL we may have to loop again and wait for
2098 	 * more data to come in.
2099 	 */
2100 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2101 		goto restart;
2102 out:
2103 	SOCKBUF_LOCK_ASSERT(sb);
2104 	SBLASTRECORDCHK(sb);
2105 	SBLASTMBUFCHK(sb);
2106 	SOCKBUF_UNLOCK(sb);
2107 	sbunlock(sb);
2108 	return (error);
2109 }
2110 
2111 /*
2112  * Optimized version of soreceive() for simple datagram cases from userspace.
2113  * Unlike in the stream case, we're able to drop a datagram if copyout()
2114  * fails, and because we handle datagrams atomically, we don't need to use a
2115  * sleep lock to prevent I/O interlacing.
2116  */
2117 int
2118 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2119     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2120 {
2121 	struct mbuf *m, *m2;
2122 	int flags, len, error;
2123 	struct protosw *pr = so->so_proto;
2124 	struct mbuf *nextrecord;
2125 
2126 	if (psa != NULL)
2127 		*psa = NULL;
2128 	if (controlp != NULL)
2129 		*controlp = NULL;
2130 	if (flagsp != NULL)
2131 		flags = *flagsp &~ MSG_EOR;
2132 	else
2133 		flags = 0;
2134 
2135 	/*
2136 	 * For any complicated cases, fall back to the full
2137 	 * soreceive_generic().
2138 	 */
2139 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2140 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2141 		    flagsp));
2142 
2143 	/*
2144 	 * Enforce restrictions on use.
2145 	 */
2146 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2147 	    ("soreceive_dgram: wantrcvd"));
2148 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2149 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2150 	    ("soreceive_dgram: SBS_RCVATMARK"));
2151 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2152 	    ("soreceive_dgram: P_CONNREQUIRED"));
2153 
2154 	/*
2155 	 * Loop blocking while waiting for a datagram.
2156 	 */
2157 	SOCKBUF_LOCK(&so->so_rcv);
2158 	while ((m = so->so_rcv.sb_mb) == NULL) {
2159 		KASSERT(so->so_rcv.sb_cc == 0,
2160 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2161 		    so->so_rcv.sb_cc));
2162 		if (so->so_error) {
2163 			error = so->so_error;
2164 			so->so_error = 0;
2165 			SOCKBUF_UNLOCK(&so->so_rcv);
2166 			return (error);
2167 		}
2168 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2169 		    uio->uio_resid == 0) {
2170 			SOCKBUF_UNLOCK(&so->so_rcv);
2171 			return (0);
2172 		}
2173 		if ((so->so_state & SS_NBIO) ||
2174 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2175 			SOCKBUF_UNLOCK(&so->so_rcv);
2176 			return (EWOULDBLOCK);
2177 		}
2178 		SBLASTRECORDCHK(&so->so_rcv);
2179 		SBLASTMBUFCHK(&so->so_rcv);
2180 		error = sbwait(&so->so_rcv);
2181 		if (error) {
2182 			SOCKBUF_UNLOCK(&so->so_rcv);
2183 			return (error);
2184 		}
2185 	}
2186 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2187 
2188 	if (uio->uio_td)
2189 		uio->uio_td->td_ru.ru_msgrcv++;
2190 	SBLASTRECORDCHK(&so->so_rcv);
2191 	SBLASTMBUFCHK(&so->so_rcv);
2192 	nextrecord = m->m_nextpkt;
2193 	if (nextrecord == NULL) {
2194 		KASSERT(so->so_rcv.sb_lastrecord == m,
2195 		    ("soreceive_dgram: lastrecord != m"));
2196 	}
2197 
2198 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2199 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2200 
2201 	/*
2202 	 * Pull 'm' and its chain off the front of the packet queue.
2203 	 */
2204 	so->so_rcv.sb_mb = NULL;
2205 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2206 
2207 	/*
2208 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2209 	 */
2210 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2211 		sbfree(&so->so_rcv, m2);
2212 
2213 	/*
2214 	 * Do a few last checks before we let go of the lock.
2215 	 */
2216 	SBLASTRECORDCHK(&so->so_rcv);
2217 	SBLASTMBUFCHK(&so->so_rcv);
2218 	SOCKBUF_UNLOCK(&so->so_rcv);
2219 
2220 	if (pr->pr_flags & PR_ADDR) {
2221 		KASSERT(m->m_type == MT_SONAME,
2222 		    ("m->m_type == %d", m->m_type));
2223 		if (psa != NULL)
2224 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2225 			    M_NOWAIT);
2226 		m = m_free(m);
2227 	}
2228 	if (m == NULL) {
2229 		/* XXXRW: Can this happen? */
2230 		return (0);
2231 	}
2232 
2233 	/*
2234 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2235 	 * queue.
2236 	 *
2237 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2238 	 * in the first mbuf chain on the socket buffer.  We call into the
2239 	 * protocol to perform externalization (or freeing if controlp ==
2240 	 * NULL).
2241 	 */
2242 	if (m->m_type == MT_CONTROL) {
2243 		struct mbuf *cm = NULL, *cmn;
2244 		struct mbuf **cme = &cm;
2245 
2246 		do {
2247 			m2 = m->m_next;
2248 			m->m_next = NULL;
2249 			*cme = m;
2250 			cme = &(*cme)->m_next;
2251 			m = m2;
2252 		} while (m != NULL && m->m_type == MT_CONTROL);
2253 		while (cm != NULL) {
2254 			cmn = cm->m_next;
2255 			cm->m_next = NULL;
2256 			if (pr->pr_domain->dom_externalize != NULL) {
2257 				error = (*pr->pr_domain->dom_externalize)
2258 				    (cm, controlp);
2259 			} else if (controlp != NULL)
2260 				*controlp = cm;
2261 			else
2262 				m_freem(cm);
2263 			if (controlp != NULL) {
2264 				while (*controlp != NULL)
2265 					controlp = &(*controlp)->m_next;
2266 			}
2267 			cm = cmn;
2268 		}
2269 	}
2270 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2271 
2272 	while (m != NULL && uio->uio_resid > 0) {
2273 		len = uio->uio_resid;
2274 		if (len > m->m_len)
2275 			len = m->m_len;
2276 		error = uiomove(mtod(m, char *), (int)len, uio);
2277 		if (error) {
2278 			m_freem(m);
2279 			return (error);
2280 		}
2281 		if (len == m->m_len)
2282 			m = m_free(m);
2283 		else {
2284 			m->m_data += len;
2285 			m->m_len -= len;
2286 		}
2287 	}
2288 	if (m != NULL)
2289 		flags |= MSG_TRUNC;
2290 	m_freem(m);
2291 	if (flagsp != NULL)
2292 		*flagsp |= flags;
2293 	return (0);
2294 }
2295 
2296 int
2297 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2298     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2299 {
2300 	int error;
2301 
2302 	CURVNET_SET(so->so_vnet);
2303 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2304 	    controlp, flagsp));
2305 	CURVNET_RESTORE();
2306 	return (error);
2307 }
2308 
2309 int
2310 soshutdown(struct socket *so, int how)
2311 {
2312 	struct protosw *pr = so->so_proto;
2313 	int error;
2314 
2315 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2316 		return (EINVAL);
2317 
2318 	CURVNET_SET(so->so_vnet);
2319 	if (pr->pr_usrreqs->pru_flush != NULL) {
2320 	        (*pr->pr_usrreqs->pru_flush)(so, how);
2321 	}
2322 	if (how != SHUT_WR)
2323 		sorflush(so);
2324 	if (how != SHUT_RD) {
2325 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2326 		CURVNET_RESTORE();
2327 		return (error);
2328 	}
2329 	CURVNET_RESTORE();
2330 	return (0);
2331 }
2332 
2333 void
2334 sorflush(struct socket *so)
2335 {
2336 	struct sockbuf *sb = &so->so_rcv;
2337 	struct protosw *pr = so->so_proto;
2338 	struct sockbuf asb;
2339 
2340 	VNET_SO_ASSERT(so);
2341 
2342 	/*
2343 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2344 	 * held, and in order to generally avoid holding the lock for a long
2345 	 * time, we make a copy of the socket buffer and clear the original
2346 	 * (except locks, state).  The new socket buffer copy won't have
2347 	 * initialized locks so we can only call routines that won't use or
2348 	 * assert those locks.
2349 	 *
2350 	 * Dislodge threads currently blocked in receive and wait to acquire
2351 	 * a lock against other simultaneous readers before clearing the
2352 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2353 	 * despite any existing socket disposition on interruptable waiting.
2354 	 */
2355 	socantrcvmore(so);
2356 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2357 
2358 	/*
2359 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2360 	 * and mutex data unchanged.
2361 	 */
2362 	SOCKBUF_LOCK(sb);
2363 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2364 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2365 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2366 	bzero(&sb->sb_startzero,
2367 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2368 	SOCKBUF_UNLOCK(sb);
2369 	sbunlock(sb);
2370 
2371 	/*
2372 	 * Dispose of special rights and flush the socket buffer.  Don't call
2373 	 * any unsafe routines (that rely on locks being initialized) on asb.
2374 	 */
2375 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2376 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2377 	sbrelease_internal(&asb, so);
2378 }
2379 
2380 /*
2381  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2382  * additional variant to handle the case where the option value needs to be
2383  * some kind of integer, but not a specific size.  In addition to their use
2384  * here, these functions are also called by the protocol-level pr_ctloutput()
2385  * routines.
2386  */
2387 int
2388 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2389 {
2390 	size_t	valsize;
2391 
2392 	/*
2393 	 * If the user gives us more than we wanted, we ignore it, but if we
2394 	 * don't get the minimum length the caller wants, we return EINVAL.
2395 	 * On success, sopt->sopt_valsize is set to however much we actually
2396 	 * retrieved.
2397 	 */
2398 	if ((valsize = sopt->sopt_valsize) < minlen)
2399 		return EINVAL;
2400 	if (valsize > len)
2401 		sopt->sopt_valsize = valsize = len;
2402 
2403 	if (sopt->sopt_td != NULL)
2404 		return (copyin(sopt->sopt_val, buf, valsize));
2405 
2406 	bcopy(sopt->sopt_val, buf, valsize);
2407 	return (0);
2408 }
2409 
2410 /*
2411  * Kernel version of setsockopt(2).
2412  *
2413  * XXX: optlen is size_t, not socklen_t
2414  */
2415 int
2416 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2417     size_t optlen)
2418 {
2419 	struct sockopt sopt;
2420 
2421 	sopt.sopt_level = level;
2422 	sopt.sopt_name = optname;
2423 	sopt.sopt_dir = SOPT_SET;
2424 	sopt.sopt_val = optval;
2425 	sopt.sopt_valsize = optlen;
2426 	sopt.sopt_td = NULL;
2427 	return (sosetopt(so, &sopt));
2428 }
2429 
2430 int
2431 sosetopt(struct socket *so, struct sockopt *sopt)
2432 {
2433 	int	error, optval;
2434 	struct	linger l;
2435 	struct	timeval tv;
2436 	u_long  val;
2437 	uint32_t val32;
2438 #ifdef MAC
2439 	struct mac extmac;
2440 #endif
2441 
2442 	CURVNET_SET(so->so_vnet);
2443 	error = 0;
2444 	if (sopt->sopt_level != SOL_SOCKET) {
2445 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2446 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2447 			CURVNET_RESTORE();
2448 			return (error);
2449 		}
2450 		error = ENOPROTOOPT;
2451 	} else {
2452 		switch (sopt->sopt_name) {
2453 #ifdef INET
2454 		case SO_ACCEPTFILTER:
2455 			error = do_setopt_accept_filter(so, sopt);
2456 			if (error)
2457 				goto bad;
2458 			break;
2459 #endif
2460 		case SO_LINGER:
2461 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2462 			if (error)
2463 				goto bad;
2464 
2465 			SOCK_LOCK(so);
2466 			so->so_linger = l.l_linger;
2467 			if (l.l_onoff)
2468 				so->so_options |= SO_LINGER;
2469 			else
2470 				so->so_options &= ~SO_LINGER;
2471 			SOCK_UNLOCK(so);
2472 			break;
2473 
2474 		case SO_DEBUG:
2475 		case SO_KEEPALIVE:
2476 		case SO_DONTROUTE:
2477 		case SO_USELOOPBACK:
2478 		case SO_BROADCAST:
2479 		case SO_REUSEADDR:
2480 		case SO_REUSEPORT:
2481 		case SO_OOBINLINE:
2482 		case SO_TIMESTAMP:
2483 		case SO_BINTIME:
2484 		case SO_NOSIGPIPE:
2485 		case SO_NO_DDP:
2486 		case SO_NO_OFFLOAD:
2487 			error = sooptcopyin(sopt, &optval, sizeof optval,
2488 					    sizeof optval);
2489 			if (error)
2490 				goto bad;
2491 			SOCK_LOCK(so);
2492 			if (optval)
2493 				so->so_options |= sopt->sopt_name;
2494 			else
2495 				so->so_options &= ~sopt->sopt_name;
2496 			SOCK_UNLOCK(so);
2497 			break;
2498 
2499 		case SO_SETFIB:
2500 			error = sooptcopyin(sopt, &optval, sizeof optval,
2501 					    sizeof optval);
2502 			if (optval < 0 || optval >= rt_numfibs) {
2503 				error = EINVAL;
2504 				goto bad;
2505 			}
2506 			if (so->so_proto != NULL &&
2507 			   ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2508 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2509 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE))) {
2510 				so->so_fibnum = optval;
2511 				/* Note: ignore error */
2512 				if (so->so_proto->pr_ctloutput)
2513 					(*so->so_proto->pr_ctloutput)(so, sopt);
2514 			} else {
2515 				so->so_fibnum = 0;
2516 			}
2517 			break;
2518 
2519 		case SO_USER_COOKIE:
2520 			error = sooptcopyin(sopt, &val32, sizeof val32,
2521 					    sizeof val32);
2522 			if (error)
2523 				goto bad;
2524 			so->so_user_cookie = val32;
2525 			break;
2526 
2527 		case SO_SNDBUF:
2528 		case SO_RCVBUF:
2529 		case SO_SNDLOWAT:
2530 		case SO_RCVLOWAT:
2531 			error = sooptcopyin(sopt, &optval, sizeof optval,
2532 					    sizeof optval);
2533 			if (error)
2534 				goto bad;
2535 
2536 			/*
2537 			 * Values < 1 make no sense for any of these options,
2538 			 * so disallow them.
2539 			 */
2540 			if (optval < 1) {
2541 				error = EINVAL;
2542 				goto bad;
2543 			}
2544 
2545 			switch (sopt->sopt_name) {
2546 			case SO_SNDBUF:
2547 			case SO_RCVBUF:
2548 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2549 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2550 				    so, curthread) == 0) {
2551 					error = ENOBUFS;
2552 					goto bad;
2553 				}
2554 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2555 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2556 				break;
2557 
2558 			/*
2559 			 * Make sure the low-water is never greater than the
2560 			 * high-water.
2561 			 */
2562 			case SO_SNDLOWAT:
2563 				SOCKBUF_LOCK(&so->so_snd);
2564 				so->so_snd.sb_lowat =
2565 				    (optval > so->so_snd.sb_hiwat) ?
2566 				    so->so_snd.sb_hiwat : optval;
2567 				SOCKBUF_UNLOCK(&so->so_snd);
2568 				break;
2569 			case SO_RCVLOWAT:
2570 				SOCKBUF_LOCK(&so->so_rcv);
2571 				so->so_rcv.sb_lowat =
2572 				    (optval > so->so_rcv.sb_hiwat) ?
2573 				    so->so_rcv.sb_hiwat : optval;
2574 				SOCKBUF_UNLOCK(&so->so_rcv);
2575 				break;
2576 			}
2577 			break;
2578 
2579 		case SO_SNDTIMEO:
2580 		case SO_RCVTIMEO:
2581 #ifdef COMPAT_FREEBSD32
2582 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2583 				struct timeval32 tv32;
2584 
2585 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2586 				    sizeof tv32);
2587 				CP(tv32, tv, tv_sec);
2588 				CP(tv32, tv, tv_usec);
2589 			} else
2590 #endif
2591 				error = sooptcopyin(sopt, &tv, sizeof tv,
2592 				    sizeof tv);
2593 			if (error)
2594 				goto bad;
2595 
2596 			/* assert(hz > 0); */
2597 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2598 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2599 				error = EDOM;
2600 				goto bad;
2601 			}
2602 			/* assert(tick > 0); */
2603 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2604 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2605 			if (val > INT_MAX) {
2606 				error = EDOM;
2607 				goto bad;
2608 			}
2609 			if (val == 0 && tv.tv_usec != 0)
2610 				val = 1;
2611 
2612 			switch (sopt->sopt_name) {
2613 			case SO_SNDTIMEO:
2614 				so->so_snd.sb_timeo = val;
2615 				break;
2616 			case SO_RCVTIMEO:
2617 				so->so_rcv.sb_timeo = val;
2618 				break;
2619 			}
2620 			break;
2621 
2622 		case SO_LABEL:
2623 #ifdef MAC
2624 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2625 			    sizeof extmac);
2626 			if (error)
2627 				goto bad;
2628 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2629 			    so, &extmac);
2630 #else
2631 			error = EOPNOTSUPP;
2632 #endif
2633 			break;
2634 
2635 		default:
2636 			error = ENOPROTOOPT;
2637 			break;
2638 		}
2639 		if (error == 0 && so->so_proto != NULL &&
2640 		    so->so_proto->pr_ctloutput != NULL) {
2641 			(void) ((*so->so_proto->pr_ctloutput)
2642 				  (so, sopt));
2643 		}
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 && so->so_proto->pr_ctloutput)
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_ERROR:
2736 			SOCK_LOCK(so);
2737 			optval = so->so_error;
2738 			so->so_error = 0;
2739 			SOCK_UNLOCK(so);
2740 			goto integer;
2741 
2742 		case SO_SNDBUF:
2743 			optval = so->so_snd.sb_hiwat;
2744 			goto integer;
2745 
2746 		case SO_RCVBUF:
2747 			optval = so->so_rcv.sb_hiwat;
2748 			goto integer;
2749 
2750 		case SO_SNDLOWAT:
2751 			optval = so->so_snd.sb_lowat;
2752 			goto integer;
2753 
2754 		case SO_RCVLOWAT:
2755 			optval = so->so_rcv.sb_lowat;
2756 			goto integer;
2757 
2758 		case SO_SNDTIMEO:
2759 		case SO_RCVTIMEO:
2760 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2761 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2762 
2763 			tv.tv_sec = optval / hz;
2764 			tv.tv_usec = (optval % hz) * tick;
2765 #ifdef COMPAT_FREEBSD32
2766 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2767 				struct timeval32 tv32;
2768 
2769 				CP(tv, tv32, tv_sec);
2770 				CP(tv, tv32, tv_usec);
2771 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2772 			} else
2773 #endif
2774 				error = sooptcopyout(sopt, &tv, sizeof tv);
2775 			break;
2776 
2777 		case SO_LABEL:
2778 #ifdef MAC
2779 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2780 			    sizeof(extmac));
2781 			if (error)
2782 				goto bad;
2783 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2784 			    so, &extmac);
2785 			if (error)
2786 				goto bad;
2787 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2788 #else
2789 			error = EOPNOTSUPP;
2790 #endif
2791 			break;
2792 
2793 		case SO_PEERLABEL:
2794 #ifdef MAC
2795 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2796 			    sizeof(extmac));
2797 			if (error)
2798 				goto bad;
2799 			error = mac_getsockopt_peerlabel(
2800 			    sopt->sopt_td->td_ucred, so, &extmac);
2801 			if (error)
2802 				goto bad;
2803 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2804 #else
2805 			error = EOPNOTSUPP;
2806 #endif
2807 			break;
2808 
2809 		case SO_LISTENQLIMIT:
2810 			optval = so->so_qlimit;
2811 			goto integer;
2812 
2813 		case SO_LISTENQLEN:
2814 			optval = so->so_qlen;
2815 			goto integer;
2816 
2817 		case SO_LISTENINCQLEN:
2818 			optval = so->so_incqlen;
2819 			goto integer;
2820 
2821 		default:
2822 			error = ENOPROTOOPT;
2823 			break;
2824 		}
2825 	}
2826 #ifdef MAC
2827 bad:
2828 #endif
2829 	CURVNET_RESTORE();
2830 	return (error);
2831 }
2832 
2833 int
2834 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2835 {
2836 	struct mbuf *m, *m_prev;
2837 	int sopt_size = sopt->sopt_valsize;
2838 
2839 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2840 	if (m == NULL)
2841 		return ENOBUFS;
2842 	if (sopt_size > MLEN) {
2843 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2844 		if ((m->m_flags & M_EXT) == 0) {
2845 			m_free(m);
2846 			return ENOBUFS;
2847 		}
2848 		m->m_len = min(MCLBYTES, sopt_size);
2849 	} else {
2850 		m->m_len = min(MLEN, sopt_size);
2851 	}
2852 	sopt_size -= m->m_len;
2853 	*mp = m;
2854 	m_prev = m;
2855 
2856 	while (sopt_size) {
2857 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2858 		if (m == NULL) {
2859 			m_freem(*mp);
2860 			return ENOBUFS;
2861 		}
2862 		if (sopt_size > MLEN) {
2863 			MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2864 			    M_DONTWAIT);
2865 			if ((m->m_flags & M_EXT) == 0) {
2866 				m_freem(m);
2867 				m_freem(*mp);
2868 				return ENOBUFS;
2869 			}
2870 			m->m_len = min(MCLBYTES, sopt_size);
2871 		} else {
2872 			m->m_len = min(MLEN, sopt_size);
2873 		}
2874 		sopt_size -= m->m_len;
2875 		m_prev->m_next = m;
2876 		m_prev = m;
2877 	}
2878 	return (0);
2879 }
2880 
2881 int
2882 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2883 {
2884 	struct mbuf *m0 = m;
2885 
2886 	if (sopt->sopt_val == NULL)
2887 		return (0);
2888 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2889 		if (sopt->sopt_td != NULL) {
2890 			int error;
2891 
2892 			error = copyin(sopt->sopt_val, mtod(m, char *),
2893 				       m->m_len);
2894 			if (error != 0) {
2895 				m_freem(m0);
2896 				return(error);
2897 			}
2898 		} else
2899 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2900 		sopt->sopt_valsize -= m->m_len;
2901 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2902 		m = m->m_next;
2903 	}
2904 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2905 		panic("ip6_sooptmcopyin");
2906 	return (0);
2907 }
2908 
2909 int
2910 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2911 {
2912 	struct mbuf *m0 = m;
2913 	size_t valsize = 0;
2914 
2915 	if (sopt->sopt_val == NULL)
2916 		return (0);
2917 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2918 		if (sopt->sopt_td != NULL) {
2919 			int error;
2920 
2921 			error = copyout(mtod(m, char *), sopt->sopt_val,
2922 				       m->m_len);
2923 			if (error != 0) {
2924 				m_freem(m0);
2925 				return(error);
2926 			}
2927 		} else
2928 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2929 	       sopt->sopt_valsize -= m->m_len;
2930 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2931 	       valsize += m->m_len;
2932 	       m = m->m_next;
2933 	}
2934 	if (m != NULL) {
2935 		/* enough soopt buffer should be given from user-land */
2936 		m_freem(m0);
2937 		return(EINVAL);
2938 	}
2939 	sopt->sopt_valsize = valsize;
2940 	return (0);
2941 }
2942 
2943 /*
2944  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2945  * out-of-band data, which will then notify socket consumers.
2946  */
2947 void
2948 sohasoutofband(struct socket *so)
2949 {
2950 
2951 	if (so->so_sigio != NULL)
2952 		pgsigio(&so->so_sigio, SIGURG, 0);
2953 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2954 }
2955 
2956 int
2957 sopoll(struct socket *so, int events, struct ucred *active_cred,
2958     struct thread *td)
2959 {
2960 
2961 	/*
2962 	 * We do not need to set or assert curvnet as long as everyone uses
2963 	 * sopoll_generic().
2964 	 */
2965 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2966 	    td));
2967 }
2968 
2969 int
2970 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2971     struct thread *td)
2972 {
2973 	int revents = 0;
2974 
2975 	SOCKBUF_LOCK(&so->so_snd);
2976 	SOCKBUF_LOCK(&so->so_rcv);
2977 	if (events & (POLLIN | POLLRDNORM))
2978 		if (soreadabledata(so))
2979 			revents |= events & (POLLIN | POLLRDNORM);
2980 
2981 	if (events & (POLLOUT | POLLWRNORM))
2982 		if (sowriteable(so))
2983 			revents |= events & (POLLOUT | POLLWRNORM);
2984 
2985 	if (events & (POLLPRI | POLLRDBAND))
2986 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2987 			revents |= events & (POLLPRI | POLLRDBAND);
2988 
2989 	if ((events & POLLINIGNEOF) == 0) {
2990 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2991 			revents |= events & (POLLIN | POLLRDNORM);
2992 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2993 				revents |= POLLHUP;
2994 		}
2995 	}
2996 
2997 	if (revents == 0) {
2998 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2999 			selrecord(td, &so->so_rcv.sb_sel);
3000 			so->so_rcv.sb_flags |= SB_SEL;
3001 		}
3002 
3003 		if (events & (POLLOUT | POLLWRNORM)) {
3004 			selrecord(td, &so->so_snd.sb_sel);
3005 			so->so_snd.sb_flags |= SB_SEL;
3006 		}
3007 	}
3008 
3009 	SOCKBUF_UNLOCK(&so->so_rcv);
3010 	SOCKBUF_UNLOCK(&so->so_snd);
3011 	return (revents);
3012 }
3013 
3014 int
3015 soo_kqfilter(struct file *fp, struct knote *kn)
3016 {
3017 	struct socket *so = kn->kn_fp->f_data;
3018 	struct sockbuf *sb;
3019 
3020 	switch (kn->kn_filter) {
3021 	case EVFILT_READ:
3022 		if (so->so_options & SO_ACCEPTCONN)
3023 			kn->kn_fop = &solisten_filtops;
3024 		else
3025 			kn->kn_fop = &soread_filtops;
3026 		sb = &so->so_rcv;
3027 		break;
3028 	case EVFILT_WRITE:
3029 		kn->kn_fop = &sowrite_filtops;
3030 		sb = &so->so_snd;
3031 		break;
3032 	default:
3033 		return (EINVAL);
3034 	}
3035 
3036 	SOCKBUF_LOCK(sb);
3037 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3038 	sb->sb_flags |= SB_KNOTE;
3039 	SOCKBUF_UNLOCK(sb);
3040 	return (0);
3041 }
3042 
3043 /*
3044  * Some routines that return EOPNOTSUPP for entry points that are not
3045  * supported by a protocol.  Fill in as needed.
3046  */
3047 int
3048 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3049 {
3050 
3051 	return EOPNOTSUPP;
3052 }
3053 
3054 int
3055 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3056 {
3057 
3058 	return EOPNOTSUPP;
3059 }
3060 
3061 int
3062 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3063 {
3064 
3065 	return EOPNOTSUPP;
3066 }
3067 
3068 int
3069 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3070 {
3071 
3072 	return EOPNOTSUPP;
3073 }
3074 
3075 int
3076 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3077 {
3078 
3079 	return EOPNOTSUPP;
3080 }
3081 
3082 int
3083 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3084     struct ifnet *ifp, struct thread *td)
3085 {
3086 
3087 	return EOPNOTSUPP;
3088 }
3089 
3090 int
3091 pru_disconnect_notsupp(struct socket *so)
3092 {
3093 
3094 	return EOPNOTSUPP;
3095 }
3096 
3097 int
3098 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3099 {
3100 
3101 	return EOPNOTSUPP;
3102 }
3103 
3104 int
3105 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3106 {
3107 
3108 	return EOPNOTSUPP;
3109 }
3110 
3111 int
3112 pru_rcvd_notsupp(struct socket *so, int flags)
3113 {
3114 
3115 	return EOPNOTSUPP;
3116 }
3117 
3118 int
3119 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3120 {
3121 
3122 	return EOPNOTSUPP;
3123 }
3124 
3125 int
3126 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3127     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3128 {
3129 
3130 	return EOPNOTSUPP;
3131 }
3132 
3133 /*
3134  * This isn't really a ``null'' operation, but it's the default one and
3135  * doesn't do anything destructive.
3136  */
3137 int
3138 pru_sense_null(struct socket *so, struct stat *sb)
3139 {
3140 
3141 	sb->st_blksize = so->so_snd.sb_hiwat;
3142 	return 0;
3143 }
3144 
3145 int
3146 pru_shutdown_notsupp(struct socket *so)
3147 {
3148 
3149 	return EOPNOTSUPP;
3150 }
3151 
3152 int
3153 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3154 {
3155 
3156 	return EOPNOTSUPP;
3157 }
3158 
3159 int
3160 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3161     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3162 {
3163 
3164 	return EOPNOTSUPP;
3165 }
3166 
3167 int
3168 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3169     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3170 {
3171 
3172 	return EOPNOTSUPP;
3173 }
3174 
3175 int
3176 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3177     struct thread *td)
3178 {
3179 
3180 	return EOPNOTSUPP;
3181 }
3182 
3183 static void
3184 filt_sordetach(struct knote *kn)
3185 {
3186 	struct socket *so = kn->kn_fp->f_data;
3187 
3188 	SOCKBUF_LOCK(&so->so_rcv);
3189 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3190 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3191 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3192 	SOCKBUF_UNLOCK(&so->so_rcv);
3193 }
3194 
3195 /*ARGSUSED*/
3196 static int
3197 filt_soread(struct knote *kn, long hint)
3198 {
3199 	struct socket *so;
3200 
3201 	so = kn->kn_fp->f_data;
3202 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3203 
3204 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3205 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3206 		kn->kn_flags |= EV_EOF;
3207 		kn->kn_fflags = so->so_error;
3208 		return (1);
3209 	} else if (so->so_error)	/* temporary udp error */
3210 		return (1);
3211 	else if (kn->kn_sfflags & NOTE_LOWAT)
3212 		return (kn->kn_data >= kn->kn_sdata);
3213 	else
3214 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3215 }
3216 
3217 static void
3218 filt_sowdetach(struct knote *kn)
3219 {
3220 	struct socket *so = kn->kn_fp->f_data;
3221 
3222 	SOCKBUF_LOCK(&so->so_snd);
3223 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3224 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3225 		so->so_snd.sb_flags &= ~SB_KNOTE;
3226 	SOCKBUF_UNLOCK(&so->so_snd);
3227 }
3228 
3229 /*ARGSUSED*/
3230 static int
3231 filt_sowrite(struct knote *kn, long hint)
3232 {
3233 	struct socket *so;
3234 
3235 	so = kn->kn_fp->f_data;
3236 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3237 	kn->kn_data = sbspace(&so->so_snd);
3238 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3239 		kn->kn_flags |= EV_EOF;
3240 		kn->kn_fflags = so->so_error;
3241 		return (1);
3242 	} else if (so->so_error)	/* temporary udp error */
3243 		return (1);
3244 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3245 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3246 		return (0);
3247 	else if (kn->kn_sfflags & NOTE_LOWAT)
3248 		return (kn->kn_data >= kn->kn_sdata);
3249 	else
3250 		return (kn->kn_data >= so->so_snd.sb_lowat);
3251 }
3252 
3253 /*ARGSUSED*/
3254 static int
3255 filt_solisten(struct knote *kn, long hint)
3256 {
3257 	struct socket *so = kn->kn_fp->f_data;
3258 
3259 	kn->kn_data = so->so_qlen;
3260 	return (! TAILQ_EMPTY(&so->so_comp));
3261 }
3262 
3263 int
3264 socheckuid(struct socket *so, uid_t uid)
3265 {
3266 
3267 	if (so == NULL)
3268 		return (EPERM);
3269 	if (so->so_cred->cr_uid != uid)
3270 		return (EPERM);
3271 	return (0);
3272 }
3273 
3274 static int
3275 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3276 {
3277 	int error;
3278 	int val;
3279 
3280 	val = somaxconn;
3281 	error = sysctl_handle_int(oidp, &val, 0, req);
3282 	if (error || !req->newptr )
3283 		return (error);
3284 
3285 	if (val < 1 || val > USHRT_MAX)
3286 		return (EINVAL);
3287 
3288 	somaxconn = val;
3289 	return (0);
3290 }
3291 
3292 /*
3293  * These functions are used by protocols to notify the socket layer (and its
3294  * consumers) of state changes in the sockets driven by protocol-side events.
3295  */
3296 
3297 /*
3298  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3299  *
3300  * Normal sequence from the active (originating) side is that
3301  * soisconnecting() is called during processing of connect() call, resulting
3302  * in an eventual call to soisconnected() if/when the connection is
3303  * established.  When the connection is torn down soisdisconnecting() is
3304  * called during processing of disconnect() call, and soisdisconnected() is
3305  * called when the connection to the peer is totally severed.  The semantics
3306  * of these routines are such that connectionless protocols can call
3307  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3308  * calls when setting up a ``connection'' takes no time.
3309  *
3310  * From the passive side, a socket is created with two queues of sockets:
3311  * so_incomp for connections in progress and so_comp for connections already
3312  * made and awaiting user acceptance.  As a protocol is preparing incoming
3313  * connections, it creates a socket structure queued on so_incomp by calling
3314  * sonewconn().  When the connection is established, soisconnected() is
3315  * called, and transfers the socket structure to so_comp, making it available
3316  * to accept().
3317  *
3318  * If a socket is closed with sockets on either so_incomp or so_comp, these
3319  * sockets are dropped.
3320  *
3321  * If higher-level protocols are implemented in the kernel, the wakeups done
3322  * here will sometimes cause software-interrupt process scheduling.
3323  */
3324 void
3325 soisconnecting(struct socket *so)
3326 {
3327 
3328 	SOCK_LOCK(so);
3329 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3330 	so->so_state |= SS_ISCONNECTING;
3331 	SOCK_UNLOCK(so);
3332 }
3333 
3334 void
3335 soisconnected(struct socket *so)
3336 {
3337 	struct socket *head;
3338 	int ret;
3339 
3340 restart:
3341 	ACCEPT_LOCK();
3342 	SOCK_LOCK(so);
3343 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3344 	so->so_state |= SS_ISCONNECTED;
3345 	head = so->so_head;
3346 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3347 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3348 			SOCK_UNLOCK(so);
3349 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3350 			head->so_incqlen--;
3351 			so->so_qstate &= ~SQ_INCOMP;
3352 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3353 			head->so_qlen++;
3354 			so->so_qstate |= SQ_COMP;
3355 			ACCEPT_UNLOCK();
3356 			sorwakeup(head);
3357 			wakeup_one(&head->so_timeo);
3358 		} else {
3359 			ACCEPT_UNLOCK();
3360 			soupcall_set(so, SO_RCV,
3361 			    head->so_accf->so_accept_filter->accf_callback,
3362 			    head->so_accf->so_accept_filter_arg);
3363 			so->so_options &= ~SO_ACCEPTFILTER;
3364 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3365 			    head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3366 			if (ret == SU_ISCONNECTED)
3367 				soupcall_clear(so, SO_RCV);
3368 			SOCK_UNLOCK(so);
3369 			if (ret == SU_ISCONNECTED)
3370 				goto restart;
3371 		}
3372 		return;
3373 	}
3374 	SOCK_UNLOCK(so);
3375 	ACCEPT_UNLOCK();
3376 	wakeup(&so->so_timeo);
3377 	sorwakeup(so);
3378 	sowwakeup(so);
3379 }
3380 
3381 void
3382 soisdisconnecting(struct socket *so)
3383 {
3384 
3385 	/*
3386 	 * Note: This code assumes that SOCK_LOCK(so) and
3387 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3388 	 */
3389 	SOCKBUF_LOCK(&so->so_rcv);
3390 	so->so_state &= ~SS_ISCONNECTING;
3391 	so->so_state |= SS_ISDISCONNECTING;
3392 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3393 	sorwakeup_locked(so);
3394 	SOCKBUF_LOCK(&so->so_snd);
3395 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3396 	sowwakeup_locked(so);
3397 	wakeup(&so->so_timeo);
3398 }
3399 
3400 void
3401 soisdisconnected(struct socket *so)
3402 {
3403 
3404 	/*
3405 	 * Note: This code assumes that SOCK_LOCK(so) and
3406 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3407 	 */
3408 	SOCKBUF_LOCK(&so->so_rcv);
3409 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3410 	so->so_state |= SS_ISDISCONNECTED;
3411 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3412 	sorwakeup_locked(so);
3413 	SOCKBUF_LOCK(&so->so_snd);
3414 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3415 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3416 	sowwakeup_locked(so);
3417 	wakeup(&so->so_timeo);
3418 }
3419 
3420 /*
3421  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3422  */
3423 struct sockaddr *
3424 sodupsockaddr(const struct sockaddr *sa, int mflags)
3425 {
3426 	struct sockaddr *sa2;
3427 
3428 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3429 	if (sa2)
3430 		bcopy(sa, sa2, sa->sa_len);
3431 	return sa2;
3432 }
3433 
3434 /*
3435  * Register per-socket buffer upcalls.
3436  */
3437 void
3438 soupcall_set(struct socket *so, int which,
3439     int (*func)(struct socket *, void *, int), void *arg)
3440 {
3441 	struct sockbuf *sb;
3442 
3443 	switch (which) {
3444 	case SO_RCV:
3445 		sb = &so->so_rcv;
3446 		break;
3447 	case SO_SND:
3448 		sb = &so->so_snd;
3449 		break;
3450 	default:
3451 		panic("soupcall_set: bad which");
3452 	}
3453 	SOCKBUF_LOCK_ASSERT(sb);
3454 #if 0
3455 	/* XXX: accf_http actually wants to do this on purpose. */
3456 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3457 #endif
3458 	sb->sb_upcall = func;
3459 	sb->sb_upcallarg = arg;
3460 	sb->sb_flags |= SB_UPCALL;
3461 }
3462 
3463 void
3464 soupcall_clear(struct socket *so, int which)
3465 {
3466 	struct sockbuf *sb;
3467 
3468 	switch (which) {
3469 	case SO_RCV:
3470 		sb = &so->so_rcv;
3471 		break;
3472 	case SO_SND:
3473 		sb = &so->so_snd;
3474 		break;
3475 	default:
3476 		panic("soupcall_clear: bad which");
3477 	}
3478 	SOCKBUF_LOCK_ASSERT(sb);
3479 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3480 	sb->sb_upcall = NULL;
3481 	sb->sb_upcallarg = NULL;
3482 	sb->sb_flags &= ~SB_UPCALL;
3483 }
3484 
3485 /*
3486  * Create an external-format (``xsocket'') structure using the information in
3487  * the kernel-format socket structure pointed to by so.  This is done to
3488  * reduce the spew of irrelevant information over this interface, to isolate
3489  * user code from changes in the kernel structure, and potentially to provide
3490  * information-hiding if we decide that some of this information should be
3491  * hidden from users.
3492  */
3493 void
3494 sotoxsocket(struct socket *so, struct xsocket *xso)
3495 {
3496 
3497 	xso->xso_len = sizeof *xso;
3498 	xso->xso_so = so;
3499 	xso->so_type = so->so_type;
3500 	xso->so_options = so->so_options;
3501 	xso->so_linger = so->so_linger;
3502 	xso->so_state = so->so_state;
3503 	xso->so_pcb = so->so_pcb;
3504 	xso->xso_protocol = so->so_proto->pr_protocol;
3505 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3506 	xso->so_qlen = so->so_qlen;
3507 	xso->so_incqlen = so->so_incqlen;
3508 	xso->so_qlimit = so->so_qlimit;
3509 	xso->so_timeo = so->so_timeo;
3510 	xso->so_error = so->so_error;
3511 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3512 	xso->so_oobmark = so->so_oobmark;
3513 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3514 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3515 	xso->so_uid = so->so_cred->cr_uid;
3516 }
3517 
3518 
3519 /*
3520  * Socket accessor functions to provide external consumers with
3521  * a safe interface to socket state
3522  *
3523  */
3524 
3525 void
3526 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3527 {
3528 
3529 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3530 		func(so, arg);
3531 }
3532 
3533 struct sockbuf *
3534 so_sockbuf_rcv(struct socket *so)
3535 {
3536 
3537 	return (&so->so_rcv);
3538 }
3539 
3540 struct sockbuf *
3541 so_sockbuf_snd(struct socket *so)
3542 {
3543 
3544 	return (&so->so_snd);
3545 }
3546 
3547 int
3548 so_state_get(const struct socket *so)
3549 {
3550 
3551 	return (so->so_state);
3552 }
3553 
3554 void
3555 so_state_set(struct socket *so, int val)
3556 {
3557 
3558 	so->so_state = val;
3559 }
3560 
3561 int
3562 so_options_get(const struct socket *so)
3563 {
3564 
3565 	return (so->so_options);
3566 }
3567 
3568 void
3569 so_options_set(struct socket *so, int val)
3570 {
3571 
3572 	so->so_options = val;
3573 }
3574 
3575 int
3576 so_error_get(const struct socket *so)
3577 {
3578 
3579 	return (so->so_error);
3580 }
3581 
3582 void
3583 so_error_set(struct socket *so, int val)
3584 {
3585 
3586 	so->so_error = val;
3587 }
3588 
3589 int
3590 so_linger_get(const struct socket *so)
3591 {
3592 
3593 	return (so->so_linger);
3594 }
3595 
3596 void
3597 so_linger_set(struct socket *so, int val)
3598 {
3599 
3600 	so->so_linger = val;
3601 }
3602 
3603 struct protosw *
3604 so_protosw_get(const struct socket *so)
3605 {
3606 
3607 	return (so->so_proto);
3608 }
3609 
3610 void
3611 so_protosw_set(struct socket *so, struct protosw *val)
3612 {
3613 
3614 	so->so_proto = val;
3615 }
3616 
3617 void
3618 so_sorwakeup(struct socket *so)
3619 {
3620 
3621 	sorwakeup(so);
3622 }
3623 
3624 void
3625 so_sowwakeup(struct socket *so)
3626 {
3627 
3628 	sowwakeup(so);
3629 }
3630 
3631 void
3632 so_sorwakeup_locked(struct socket *so)
3633 {
3634 
3635 	sorwakeup_locked(so);
3636 }
3637 
3638 void
3639 so_sowwakeup_locked(struct socket *so)
3640 {
3641 
3642 	sowwakeup_locked(so);
3643 }
3644 
3645 void
3646 so_lock(struct socket *so)
3647 {
3648 	SOCK_LOCK(so);
3649 }
3650 
3651 void
3652 so_unlock(struct socket *so)
3653 {
3654 	SOCK_UNLOCK(so);
3655 }
3656