xref: /freebsd/sys/kern/uipc_socket.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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_ROUTE))
396 		so->so_fibnum = td->td_proc->p_fibnum;
397 	else
398 		so->so_fibnum = 0;
399 	so->so_proto = prp;
400 #ifdef MAC
401 	mac_socket_create(cred, so);
402 #endif
403 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
404 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
405 	so->so_count = 1;
406 	/*
407 	 * Auto-sizing of socket buffers is managed by the protocols and
408 	 * the appropriate flags must be set in the pru_attach function.
409 	 */
410 	CURVNET_SET(so->so_vnet);
411 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
412 	CURVNET_RESTORE();
413 	if (error) {
414 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
415 		    so->so_count));
416 		so->so_count = 0;
417 		sodealloc(so);
418 		return (error);
419 	}
420 	*aso = so;
421 	return (0);
422 }
423 
424 #ifdef REGRESSION
425 static int regression_sonewconn_earlytest = 1;
426 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
427     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
428 #endif
429 
430 /*
431  * When an attempt at a new connection is noted on a socket which accepts
432  * connections, sonewconn is called.  If the connection is possible (subject
433  * to space constraints, etc.) then we allocate a new structure, propoerly
434  * linked into the data structure of the original socket, and return this.
435  * Connstatus may be 0, or SO_ISCONFIRMING, or SO_ISCONNECTED.
436  *
437  * Note: the ref count on the socket is 0 on return.
438  */
439 struct socket *
440 sonewconn(struct socket *head, int connstatus)
441 {
442 	struct socket *so;
443 	int over;
444 
445 	ACCEPT_LOCK();
446 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
447 	ACCEPT_UNLOCK();
448 #ifdef REGRESSION
449 	if (regression_sonewconn_earlytest && over)
450 #else
451 	if (over)
452 #endif
453 		return (NULL);
454 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
455 	    __func__, __LINE__, head));
456 	so = soalloc(head->so_vnet);
457 	if (so == NULL)
458 		return (NULL);
459 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
460 		connstatus = 0;
461 	so->so_head = head;
462 	so->so_type = head->so_type;
463 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
464 	so->so_linger = head->so_linger;
465 	so->so_state = head->so_state | SS_NOFDREF;
466 	so->so_fibnum = head->so_fibnum;
467 	so->so_proto = head->so_proto;
468 	so->so_cred = crhold(head->so_cred);
469 #ifdef MAC
470 	mac_socket_newconn(head, so);
471 #endif
472 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
473 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
474 	VNET_SO_ASSERT(head);
475 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat) ||
476 	    (*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
477 		sodealloc(so);
478 		return (NULL);
479 	}
480 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
481 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
482 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
483 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
484 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
485 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
486 	so->so_state |= connstatus;
487 	ACCEPT_LOCK();
488 	if (connstatus) {
489 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
490 		so->so_qstate |= SQ_COMP;
491 		head->so_qlen++;
492 	} else {
493 		/*
494 		 * Keep removing sockets from the head until there's room for
495 		 * us to insert on the tail.  In pre-locking revisions, this
496 		 * was a simple if(), but as we could be racing with other
497 		 * threads and soabort() requires dropping locks, we must
498 		 * loop waiting for the condition to be true.
499 		 */
500 		while (head->so_incqlen > head->so_qlimit) {
501 			struct socket *sp;
502 			sp = TAILQ_FIRST(&head->so_incomp);
503 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
504 			head->so_incqlen--;
505 			sp->so_qstate &= ~SQ_INCOMP;
506 			sp->so_head = NULL;
507 			ACCEPT_UNLOCK();
508 			soabort(sp);
509 			ACCEPT_LOCK();
510 		}
511 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
512 		so->so_qstate |= SQ_INCOMP;
513 		head->so_incqlen++;
514 	}
515 	ACCEPT_UNLOCK();
516 	if (connstatus) {
517 		sorwakeup(head);
518 		wakeup_one(&head->so_timeo);
519 	}
520 	return (so);
521 }
522 
523 int
524 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
525 {
526 	int error;
527 
528 	CURVNET_SET(so->so_vnet);
529 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
530 	CURVNET_RESTORE();
531 	return error;
532 }
533 
534 /*
535  * solisten() transitions a socket from a non-listening state to a listening
536  * state, but can also be used to update the listen queue depth on an
537  * existing listen socket.  The protocol will call back into the sockets
538  * layer using solisten_proto_check() and solisten_proto() to check and set
539  * socket-layer listen state.  Call backs are used so that the protocol can
540  * acquire both protocol and socket layer locks in whatever order is required
541  * by the protocol.
542  *
543  * Protocol implementors are advised to hold the socket lock across the
544  * socket-layer test and set to avoid races at the socket layer.
545  */
546 int
547 solisten(struct socket *so, int backlog, struct thread *td)
548 {
549 	int error;
550 
551 	CURVNET_SET(so->so_vnet);
552 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
553 	CURVNET_RESTORE();
554 	return error;
555 }
556 
557 int
558 solisten_proto_check(struct socket *so)
559 {
560 
561 	SOCK_LOCK_ASSERT(so);
562 
563 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
564 	    SS_ISDISCONNECTING))
565 		return (EINVAL);
566 	return (0);
567 }
568 
569 void
570 solisten_proto(struct socket *so, int backlog)
571 {
572 
573 	SOCK_LOCK_ASSERT(so);
574 
575 	if (backlog < 0 || backlog > somaxconn)
576 		backlog = somaxconn;
577 	so->so_qlimit = backlog;
578 	so->so_options |= SO_ACCEPTCONN;
579 }
580 
581 /*
582  * Evaluate the reference count and named references on a socket; if no
583  * references remain, free it.  This should be called whenever a reference is
584  * released, such as in sorele(), but also when named reference flags are
585  * cleared in socket or protocol code.
586  *
587  * sofree() will free the socket if:
588  *
589  * - There are no outstanding file descriptor references or related consumers
590  *   (so_count == 0).
591  *
592  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
593  *
594  * - The protocol does not have an outstanding strong reference on the socket
595  *   (SS_PROTOREF).
596  *
597  * - The socket is not in a completed connection queue, so a process has been
598  *   notified that it is present.  If it is removed, the user process may
599  *   block in accept() despite select() saying the socket was ready.
600  */
601 void
602 sofree(struct socket *so)
603 {
604 	struct protosw *pr = so->so_proto;
605 	struct socket *head;
606 
607 	ACCEPT_LOCK_ASSERT();
608 	SOCK_LOCK_ASSERT(so);
609 
610 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
611 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
612 		SOCK_UNLOCK(so);
613 		ACCEPT_UNLOCK();
614 		return;
615 	}
616 
617 	head = so->so_head;
618 	if (head != NULL) {
619 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
620 		    (so->so_qstate & SQ_INCOMP) != 0,
621 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
622 		    "SQ_INCOMP"));
623 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
624 		    (so->so_qstate & SQ_INCOMP) == 0,
625 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
626 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
627 		head->so_incqlen--;
628 		so->so_qstate &= ~SQ_INCOMP;
629 		so->so_head = NULL;
630 	}
631 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
632 	    (so->so_qstate & SQ_INCOMP) == 0,
633 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
634 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
635 	if (so->so_options & SO_ACCEPTCONN) {
636 		KASSERT((TAILQ_EMPTY(&so->so_comp)), ("sofree: so_comp populated"));
637 		KASSERT((TAILQ_EMPTY(&so->so_incomp)), ("sofree: so_comp populated"));
638 	}
639 	SOCK_UNLOCK(so);
640 	ACCEPT_UNLOCK();
641 
642 	VNET_SO_ASSERT(so);
643 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
644 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
645 	if (pr->pr_usrreqs->pru_detach != NULL)
646 		(*pr->pr_usrreqs->pru_detach)(so);
647 
648 	/*
649 	 * From this point on, we assume that no other references to this
650 	 * socket exist anywhere else in the stack.  Therefore, no locks need
651 	 * to be acquired or held.
652 	 *
653 	 * We used to do a lot of socket buffer and socket locking here, as
654 	 * well as invoke sorflush() and perform wakeups.  The direct call to
655 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
656 	 * necessary from sorflush().
657 	 *
658 	 * Notice that the socket buffer and kqueue state are torn down
659 	 * before calling pru_detach.  This means that protocols shold not
660 	 * assume they can perform socket wakeups, etc, in their detach code.
661 	 */
662 	sbdestroy(&so->so_snd, so);
663 	sbdestroy(&so->so_rcv, so);
664 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
665 	knlist_destroy(&so->so_snd.sb_sel.si_note);
666 	sodealloc(so);
667 }
668 
669 /*
670  * Close a socket on last file table reference removal.  Initiate disconnect
671  * if connected.  Free socket when disconnect complete.
672  *
673  * This function will sorele() the socket.  Note that soclose() may be called
674  * prior to the ref count reaching zero.  The actual socket structure will
675  * not be freed until the ref count reaches zero.
676  */
677 int
678 soclose(struct socket *so)
679 {
680 	int error = 0;
681 
682 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
683 
684 	CURVNET_SET(so->so_vnet);
685 	funsetown(&so->so_sigio);
686 	if (so->so_state & SS_ISCONNECTED) {
687 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
688 			error = sodisconnect(so);
689 			if (error) {
690 				if (error == ENOTCONN)
691 					error = 0;
692 				goto drop;
693 			}
694 		}
695 		if (so->so_options & SO_LINGER) {
696 			if ((so->so_state & SS_ISDISCONNECTING) &&
697 			    (so->so_state & SS_NBIO))
698 				goto drop;
699 			while (so->so_state & SS_ISCONNECTED) {
700 				error = tsleep(&so->so_timeo,
701 				    PSOCK | PCATCH, "soclos", so->so_linger * hz);
702 				if (error)
703 					break;
704 			}
705 		}
706 	}
707 
708 drop:
709 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
710 		(*so->so_proto->pr_usrreqs->pru_close)(so);
711 	if (so->so_options & SO_ACCEPTCONN) {
712 		struct socket *sp;
713 		ACCEPT_LOCK();
714 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
715 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
716 			so->so_incqlen--;
717 			sp->so_qstate &= ~SQ_INCOMP;
718 			sp->so_head = NULL;
719 			ACCEPT_UNLOCK();
720 			soabort(sp);
721 			ACCEPT_LOCK();
722 		}
723 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
724 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
725 			so->so_qlen--;
726 			sp->so_qstate &= ~SQ_COMP;
727 			sp->so_head = NULL;
728 			ACCEPT_UNLOCK();
729 			soabort(sp);
730 			ACCEPT_LOCK();
731 		}
732 		ACCEPT_UNLOCK();
733 	}
734 	ACCEPT_LOCK();
735 	SOCK_LOCK(so);
736 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
737 	so->so_state |= SS_NOFDREF;
738 	sorele(so);
739 	CURVNET_RESTORE();
740 	return (error);
741 }
742 
743 /*
744  * soabort() is used to abruptly tear down a connection, such as when a
745  * resource limit is reached (listen queue depth exceeded), or if a listen
746  * socket is closed while there are sockets waiting to be accepted.
747  *
748  * This interface is tricky, because it is called on an unreferenced socket,
749  * and must be called only by a thread that has actually removed the socket
750  * from the listen queue it was on, or races with other threads are risked.
751  *
752  * This interface will call into the protocol code, so must not be called
753  * with any socket locks held.  Protocols do call it while holding their own
754  * recursible protocol mutexes, but this is something that should be subject
755  * to review in the future.
756  */
757 void
758 soabort(struct socket *so)
759 {
760 
761 	/*
762 	 * In as much as is possible, assert that no references to this
763 	 * socket are held.  This is not quite the same as asserting that the
764 	 * current thread is responsible for arranging for no references, but
765 	 * is as close as we can get for now.
766 	 */
767 	KASSERT(so->so_count == 0, ("soabort: so_count"));
768 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
769 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
770 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
771 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
772 	VNET_SO_ASSERT(so);
773 
774 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
775 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
776 	ACCEPT_LOCK();
777 	SOCK_LOCK(so);
778 	sofree(so);
779 }
780 
781 int
782 soaccept(struct socket *so, struct sockaddr **nam)
783 {
784 	int error;
785 
786 	SOCK_LOCK(so);
787 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
788 	so->so_state &= ~SS_NOFDREF;
789 	SOCK_UNLOCK(so);
790 
791 	CURVNET_SET(so->so_vnet);
792 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
793 	CURVNET_RESTORE();
794 	return (error);
795 }
796 
797 int
798 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
799 {
800 	int error;
801 
802 	if (so->so_options & SO_ACCEPTCONN)
803 		return (EOPNOTSUPP);
804 
805 	CURVNET_SET(so->so_vnet);
806 	/*
807 	 * If protocol is connection-based, can only connect once.
808 	 * Otherwise, if connected, try to disconnect first.  This allows
809 	 * user to disconnect by connecting to, e.g., a null address.
810 	 */
811 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
812 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
813 	    (error = sodisconnect(so)))) {
814 		error = EISCONN;
815 	} else {
816 		/*
817 		 * Prevent accumulated error from previous connection from
818 		 * biting us.
819 		 */
820 		so->so_error = 0;
821 		error = (*so->so_proto->pr_usrreqs->pru_connect)(so, nam, td);
822 	}
823 	CURVNET_RESTORE();
824 
825 	return (error);
826 }
827 
828 int
829 soconnect2(struct socket *so1, struct socket *so2)
830 {
831 	int error;
832 
833 	CURVNET_SET(so1->so_vnet);
834 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
835 	CURVNET_RESTORE();
836 	return (error);
837 }
838 
839 int
840 sodisconnect(struct socket *so)
841 {
842 	int error;
843 
844 	if ((so->so_state & SS_ISCONNECTED) == 0)
845 		return (ENOTCONN);
846 	if (so->so_state & SS_ISDISCONNECTING)
847 		return (EALREADY);
848 	VNET_SO_ASSERT(so);
849 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
850 	return (error);
851 }
852 
853 #ifdef ZERO_COPY_SOCKETS
854 struct so_zerocopy_stats{
855 	int size_ok;
856 	int align_ok;
857 	int found_ifp;
858 };
859 struct so_zerocopy_stats so_zerocp_stats = {0,0,0};
860 #include <netinet/in.h>
861 #include <net/route.h>
862 #include <netinet/in_pcb.h>
863 #include <vm/vm.h>
864 #include <vm/vm_page.h>
865 #include <vm/vm_object.h>
866 
867 /*
868  * sosend_copyin() is only used if zero copy sockets are enabled.  Otherwise
869  * sosend_dgram() and sosend_generic() use m_uiotombuf().
870  *
871  * sosend_copyin() accepts a uio and prepares an mbuf chain holding part or
872  * all of the data referenced by the uio.  If desired, it uses zero-copy.
873  * *space will be updated to reflect data copied in.
874  *
875  * NB: If atomic I/O is requested, the caller must already have checked that
876  * space can hold resid bytes.
877  *
878  * NB: In the event of an error, the caller may need to free the partial
879  * chain pointed to by *mpp.  The contents of both *uio and *space may be
880  * modified even in the case of an error.
881  */
882 static int
883 sosend_copyin(struct uio *uio, struct mbuf **retmp, int atomic, long *space,
884     int flags)
885 {
886 	struct mbuf *m, **mp, *top;
887 	long len, resid;
888 	int error;
889 #ifdef ZERO_COPY_SOCKETS
890 	int cow_send;
891 #endif
892 
893 	*retmp = top = NULL;
894 	mp = &top;
895 	len = 0;
896 	resid = uio->uio_resid;
897 	error = 0;
898 	do {
899 #ifdef ZERO_COPY_SOCKETS
900 		cow_send = 0;
901 #endif /* ZERO_COPY_SOCKETS */
902 		if (resid >= MINCLSIZE) {
903 #ifdef ZERO_COPY_SOCKETS
904 			if (top == NULL) {
905 				m = m_gethdr(M_WAITOK, MT_DATA);
906 				m->m_pkthdr.len = 0;
907 				m->m_pkthdr.rcvif = NULL;
908 			} else
909 				m = m_get(M_WAITOK, MT_DATA);
910 			if (so_zero_copy_send &&
911 			    resid>=PAGE_SIZE &&
912 			    *space>=PAGE_SIZE &&
913 			    uio->uio_iov->iov_len>=PAGE_SIZE) {
914 				so_zerocp_stats.size_ok++;
915 				so_zerocp_stats.align_ok++;
916 				cow_send = socow_setup(m, uio);
917 				len = cow_send;
918 			}
919 			if (!cow_send) {
920 				m_clget(m, M_WAITOK);
921 				len = min(min(MCLBYTES, resid), *space);
922 			}
923 #else /* ZERO_COPY_SOCKETS */
924 			if (top == NULL) {
925 				m = m_getcl(M_WAIT, MT_DATA, M_PKTHDR);
926 				m->m_pkthdr.len = 0;
927 				m->m_pkthdr.rcvif = NULL;
928 			} else
929 				m = m_getcl(M_WAIT, MT_DATA, 0);
930 			len = min(min(MCLBYTES, resid), *space);
931 #endif /* ZERO_COPY_SOCKETS */
932 		} else {
933 			if (top == NULL) {
934 				m = m_gethdr(M_WAIT, MT_DATA);
935 				m->m_pkthdr.len = 0;
936 				m->m_pkthdr.rcvif = NULL;
937 
938 				len = min(min(MHLEN, resid), *space);
939 				/*
940 				 * For datagram protocols, leave room
941 				 * for protocol headers in first mbuf.
942 				 */
943 				if (atomic && m && len < MHLEN)
944 					MH_ALIGN(m, len);
945 			} else {
946 				m = m_get(M_WAIT, MT_DATA);
947 				len = min(min(MLEN, resid), *space);
948 			}
949 		}
950 		if (m == NULL) {
951 			error = ENOBUFS;
952 			goto out;
953 		}
954 
955 		*space -= len;
956 #ifdef ZERO_COPY_SOCKETS
957 		if (cow_send)
958 			error = 0;
959 		else
960 #endif /* ZERO_COPY_SOCKETS */
961 		error = uiomove(mtod(m, void *), (int)len, uio);
962 		resid = uio->uio_resid;
963 		m->m_len = len;
964 		*mp = m;
965 		top->m_pkthdr.len += len;
966 		if (error)
967 			goto out;
968 		mp = &m->m_next;
969 		if (resid <= 0) {
970 			if (flags & MSG_EOR)
971 				top->m_flags |= M_EOR;
972 			break;
973 		}
974 	} while (*space > 0 && atomic);
975 out:
976 	*retmp = top;
977 	return (error);
978 }
979 #endif /*ZERO_COPY_SOCKETS*/
980 
981 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
982 
983 int
984 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
985     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
986 {
987 	long space, resid;
988 	int clen = 0, error, dontroute;
989 #ifdef ZERO_COPY_SOCKETS
990 	int atomic = sosendallatonce(so) || top;
991 #endif
992 
993 	KASSERT(so->so_type == SOCK_DGRAM, ("sodgram_send: !SOCK_DGRAM"));
994 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
995 	    ("sodgram_send: !PR_ATOMIC"));
996 
997 	if (uio != NULL)
998 		resid = uio->uio_resid;
999 	else
1000 		resid = top->m_pkthdr.len;
1001 	/*
1002 	 * In theory resid should be unsigned.  However, space must be
1003 	 * signed, as it might be less than 0 if we over-committed, and we
1004 	 * must use a signed comparison of space and resid.  On the other
1005 	 * hand, a negative resid causes us to loop sending 0-length
1006 	 * segments to the protocol.
1007 	 */
1008 	if (resid < 0) {
1009 		error = EINVAL;
1010 		goto out;
1011 	}
1012 
1013 	dontroute =
1014 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1015 	if (td != NULL)
1016 		td->td_ru.ru_msgsnd++;
1017 	if (control != NULL)
1018 		clen = control->m_len;
1019 
1020 	SOCKBUF_LOCK(&so->so_snd);
1021 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1022 		SOCKBUF_UNLOCK(&so->so_snd);
1023 		error = EPIPE;
1024 		goto out;
1025 	}
1026 	if (so->so_error) {
1027 		error = so->so_error;
1028 		so->so_error = 0;
1029 		SOCKBUF_UNLOCK(&so->so_snd);
1030 		goto out;
1031 	}
1032 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1033 		/*
1034 		 * `sendto' and `sendmsg' is allowed on a connection-based
1035 		 * socket if it supports implied connect.  Return ENOTCONN if
1036 		 * not connected and no address is supplied.
1037 		 */
1038 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1039 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1040 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1041 			    !(resid == 0 && clen != 0)) {
1042 				SOCKBUF_UNLOCK(&so->so_snd);
1043 				error = ENOTCONN;
1044 				goto out;
1045 			}
1046 		} else if (addr == NULL) {
1047 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1048 				error = ENOTCONN;
1049 			else
1050 				error = EDESTADDRREQ;
1051 			SOCKBUF_UNLOCK(&so->so_snd);
1052 			goto out;
1053 		}
1054 	}
1055 
1056 	/*
1057 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1058 	 * problem and need fixing.
1059 	 */
1060 	space = sbspace(&so->so_snd);
1061 	if (flags & MSG_OOB)
1062 		space += 1024;
1063 	space -= clen;
1064 	SOCKBUF_UNLOCK(&so->so_snd);
1065 	if (resid > space) {
1066 		error = EMSGSIZE;
1067 		goto out;
1068 	}
1069 	if (uio == NULL) {
1070 		resid = 0;
1071 		if (flags & MSG_EOR)
1072 			top->m_flags |= M_EOR;
1073 	} else {
1074 #ifdef ZERO_COPY_SOCKETS
1075 		error = sosend_copyin(uio, &top, atomic, &space, flags);
1076 		if (error)
1077 			goto out;
1078 #else
1079 		/*
1080 		 * Copy the data from userland into a mbuf chain.
1081 		 * If no data is to be copied in, a single empty mbuf
1082 		 * is returned.
1083 		 */
1084 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1085 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1086 		if (top == NULL) {
1087 			error = EFAULT;	/* only possible error */
1088 			goto out;
1089 		}
1090 		space -= resid - uio->uio_resid;
1091 #endif
1092 		resid = uio->uio_resid;
1093 	}
1094 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1095 	/*
1096 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1097 	 * than with.
1098 	 */
1099 	if (dontroute) {
1100 		SOCK_LOCK(so);
1101 		so->so_options |= SO_DONTROUTE;
1102 		SOCK_UNLOCK(so);
1103 	}
1104 	/*
1105 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1106 	 * of date.  We could have recieved a reset packet in an interrupt or
1107 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1108 	 * probably recheck again inside the locking protection here, but
1109 	 * there are probably other places that this also happens.  We must
1110 	 * rethink this.
1111 	 */
1112 	VNET_SO_ASSERT(so);
1113 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1114 	    (flags & MSG_OOB) ? PRUS_OOB :
1115 	/*
1116 	 * If the user set MSG_EOF, the protocol understands this flag and
1117 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1118 	 */
1119 	    ((flags & MSG_EOF) &&
1120 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1121 	     (resid <= 0)) ?
1122 		PRUS_EOF :
1123 		/* If there is more to send set PRUS_MORETOCOME */
1124 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1125 		top, addr, control, td);
1126 	if (dontroute) {
1127 		SOCK_LOCK(so);
1128 		so->so_options &= ~SO_DONTROUTE;
1129 		SOCK_UNLOCK(so);
1130 	}
1131 	clen = 0;
1132 	control = NULL;
1133 	top = NULL;
1134 out:
1135 	if (top != NULL)
1136 		m_freem(top);
1137 	if (control != NULL)
1138 		m_freem(control);
1139 	return (error);
1140 }
1141 
1142 /*
1143  * Send on a socket.  If send must go all at once and message is larger than
1144  * send buffering, then hard error.  Lock against other senders.  If must go
1145  * all at once and not enough room now, then inform user that this would
1146  * block and do nothing.  Otherwise, if nonblocking, send as much as
1147  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1148  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1149  * in mbuf chain must be small enough to send all at once.
1150  *
1151  * Returns nonzero on error, timeout or signal; callers must check for short
1152  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1153  * on return.
1154  */
1155 int
1156 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1157     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1158 {
1159 	long space, resid;
1160 	int clen = 0, error, dontroute;
1161 	int atomic = sosendallatonce(so) || top;
1162 
1163 	if (uio != NULL)
1164 		resid = uio->uio_resid;
1165 	else
1166 		resid = top->m_pkthdr.len;
1167 	/*
1168 	 * In theory resid should be unsigned.  However, space must be
1169 	 * signed, as it might be less than 0 if we over-committed, and we
1170 	 * must use a signed comparison of space and resid.  On the other
1171 	 * hand, a negative resid causes us to loop sending 0-length
1172 	 * segments to the protocol.
1173 	 *
1174 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1175 	 * type sockets since that's an error.
1176 	 */
1177 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1178 		error = EINVAL;
1179 		goto out;
1180 	}
1181 
1182 	dontroute =
1183 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1184 	    (so->so_proto->pr_flags & PR_ATOMIC);
1185 	if (td != NULL)
1186 		td->td_ru.ru_msgsnd++;
1187 	if (control != NULL)
1188 		clen = control->m_len;
1189 
1190 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1191 	if (error)
1192 		goto out;
1193 
1194 restart:
1195 	do {
1196 		SOCKBUF_LOCK(&so->so_snd);
1197 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1198 			SOCKBUF_UNLOCK(&so->so_snd);
1199 			error = EPIPE;
1200 			goto release;
1201 		}
1202 		if (so->so_error) {
1203 			error = so->so_error;
1204 			so->so_error = 0;
1205 			SOCKBUF_UNLOCK(&so->so_snd);
1206 			goto release;
1207 		}
1208 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1209 			/*
1210 			 * `sendto' and `sendmsg' is allowed on a connection-
1211 			 * based socket if it supports implied connect.
1212 			 * Return ENOTCONN if not connected and no address is
1213 			 * supplied.
1214 			 */
1215 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1216 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1217 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1218 				    !(resid == 0 && clen != 0)) {
1219 					SOCKBUF_UNLOCK(&so->so_snd);
1220 					error = ENOTCONN;
1221 					goto release;
1222 				}
1223 			} else if (addr == NULL) {
1224 				SOCKBUF_UNLOCK(&so->so_snd);
1225 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1226 					error = ENOTCONN;
1227 				else
1228 					error = EDESTADDRREQ;
1229 				goto release;
1230 			}
1231 		}
1232 		space = sbspace(&so->so_snd);
1233 		if (flags & MSG_OOB)
1234 			space += 1024;
1235 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1236 		    clen > so->so_snd.sb_hiwat) {
1237 			SOCKBUF_UNLOCK(&so->so_snd);
1238 			error = EMSGSIZE;
1239 			goto release;
1240 		}
1241 		if (space < resid + clen &&
1242 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1243 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1244 				SOCKBUF_UNLOCK(&so->so_snd);
1245 				error = EWOULDBLOCK;
1246 				goto release;
1247 			}
1248 			error = sbwait(&so->so_snd);
1249 			SOCKBUF_UNLOCK(&so->so_snd);
1250 			if (error)
1251 				goto release;
1252 			goto restart;
1253 		}
1254 		SOCKBUF_UNLOCK(&so->so_snd);
1255 		space -= clen;
1256 		do {
1257 			if (uio == NULL) {
1258 				resid = 0;
1259 				if (flags & MSG_EOR)
1260 					top->m_flags |= M_EOR;
1261 			} else {
1262 #ifdef ZERO_COPY_SOCKETS
1263 				error = sosend_copyin(uio, &top, atomic,
1264 				    &space, flags);
1265 				if (error != 0)
1266 					goto release;
1267 #else
1268 				/*
1269 				 * Copy the data from userland into a mbuf
1270 				 * chain.  If no data is to be copied in,
1271 				 * a single empty mbuf is returned.
1272 				 */
1273 				top = m_uiotombuf(uio, M_WAITOK, space,
1274 				    (atomic ? max_hdr : 0),
1275 				    (atomic ? M_PKTHDR : 0) |
1276 				    ((flags & MSG_EOR) ? M_EOR : 0));
1277 				if (top == NULL) {
1278 					error = EFAULT; /* only possible error */
1279 					goto release;
1280 				}
1281 				space -= resid - uio->uio_resid;
1282 #endif
1283 				resid = uio->uio_resid;
1284 			}
1285 			if (dontroute) {
1286 				SOCK_LOCK(so);
1287 				so->so_options |= SO_DONTROUTE;
1288 				SOCK_UNLOCK(so);
1289 			}
1290 			/*
1291 			 * XXX all the SBS_CANTSENDMORE checks previously
1292 			 * done could be out of date.  We could have recieved
1293 			 * a reset packet in an interrupt or maybe we slept
1294 			 * while doing page faults in uiomove() etc.  We
1295 			 * could probably recheck again inside the locking
1296 			 * protection here, but there are probably other
1297 			 * places that this also happens.  We must rethink
1298 			 * this.
1299 			 */
1300 			VNET_SO_ASSERT(so);
1301 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1302 			    (flags & MSG_OOB) ? PRUS_OOB :
1303 			/*
1304 			 * If the user set MSG_EOF, the protocol understands
1305 			 * this flag and nothing left to send then use
1306 			 * PRU_SEND_EOF instead of PRU_SEND.
1307 			 */
1308 			    ((flags & MSG_EOF) &&
1309 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1310 			     (resid <= 0)) ?
1311 				PRUS_EOF :
1312 			/* If there is more to send set PRUS_MORETOCOME. */
1313 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1314 			    top, addr, control, td);
1315 			if (dontroute) {
1316 				SOCK_LOCK(so);
1317 				so->so_options &= ~SO_DONTROUTE;
1318 				SOCK_UNLOCK(so);
1319 			}
1320 			clen = 0;
1321 			control = NULL;
1322 			top = NULL;
1323 			if (error)
1324 				goto release;
1325 		} while (resid && space > 0);
1326 	} while (resid);
1327 
1328 release:
1329 	sbunlock(&so->so_snd);
1330 out:
1331 	if (top != NULL)
1332 		m_freem(top);
1333 	if (control != NULL)
1334 		m_freem(control);
1335 	return (error);
1336 }
1337 
1338 int
1339 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1340     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1341 {
1342 	int error;
1343 
1344 	CURVNET_SET(so->so_vnet);
1345 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1346 	    control, flags, td);
1347 	CURVNET_RESTORE();
1348 	return (error);
1349 }
1350 
1351 /*
1352  * The part of soreceive() that implements reading non-inline out-of-band
1353  * data from a socket.  For more complete comments, see soreceive(), from
1354  * which this code originated.
1355  *
1356  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1357  * unable to return an mbuf chain to the caller.
1358  */
1359 static int
1360 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1361 {
1362 	struct protosw *pr = so->so_proto;
1363 	struct mbuf *m;
1364 	int error;
1365 
1366 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1367 	VNET_SO_ASSERT(so);
1368 
1369 	m = m_get(M_WAIT, MT_DATA);
1370 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1371 	if (error)
1372 		goto bad;
1373 	do {
1374 #ifdef ZERO_COPY_SOCKETS
1375 		if (so_zero_copy_receive) {
1376 			int disposable;
1377 
1378 			if ((m->m_flags & M_EXT)
1379 			 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1380 				disposable = 1;
1381 			else
1382 				disposable = 0;
1383 
1384 			error = uiomoveco(mtod(m, void *),
1385 					  min(uio->uio_resid, m->m_len),
1386 					  uio, disposable);
1387 		} else
1388 #endif /* ZERO_COPY_SOCKETS */
1389 		error = uiomove(mtod(m, void *),
1390 		    (int) min(uio->uio_resid, m->m_len), uio);
1391 		m = m_free(m);
1392 	} while (uio->uio_resid && error == 0 && m);
1393 bad:
1394 	if (m != NULL)
1395 		m_freem(m);
1396 	return (error);
1397 }
1398 
1399 /*
1400  * Following replacement or removal of the first mbuf on the first mbuf chain
1401  * of a socket buffer, push necessary state changes back into the socket
1402  * buffer so that other consumers see the values consistently.  'nextrecord'
1403  * is the callers locally stored value of the original value of
1404  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1405  * NOTE: 'nextrecord' may be NULL.
1406  */
1407 static __inline void
1408 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1409 {
1410 
1411 	SOCKBUF_LOCK_ASSERT(sb);
1412 	/*
1413 	 * First, update for the new value of nextrecord.  If necessary, make
1414 	 * it the first record.
1415 	 */
1416 	if (sb->sb_mb != NULL)
1417 		sb->sb_mb->m_nextpkt = nextrecord;
1418 	else
1419 		sb->sb_mb = nextrecord;
1420 
1421         /*
1422          * Now update any dependent socket buffer fields to reflect the new
1423          * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1424 	 * addition of a second clause that takes care of the case where
1425 	 * sb_mb has been updated, but remains the last record.
1426          */
1427         if (sb->sb_mb == NULL) {
1428                 sb->sb_mbtail = NULL;
1429                 sb->sb_lastrecord = NULL;
1430         } else if (sb->sb_mb->m_nextpkt == NULL)
1431                 sb->sb_lastrecord = sb->sb_mb;
1432 }
1433 
1434 
1435 /*
1436  * Implement receive operations on a socket.  We depend on the way that
1437  * records are added to the sockbuf by sbappend.  In particular, each record
1438  * (mbufs linked through m_next) must begin with an address if the protocol
1439  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1440  * data, and then zero or more mbufs of data.  In order to allow parallelism
1441  * between network receive and copying to user space, as well as avoid
1442  * sleeping with a mutex held, we release the socket buffer mutex during the
1443  * user space copy.  Although the sockbuf is locked, new data may still be
1444  * appended, and thus we must maintain consistency of the sockbuf during that
1445  * time.
1446  *
1447  * The caller may receive the data as a single mbuf chain by supplying an
1448  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1449  * the count in uio_resid.
1450  */
1451 int
1452 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1453     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1454 {
1455 	struct mbuf *m, **mp;
1456 	int flags, len, error, offset;
1457 	struct protosw *pr = so->so_proto;
1458 	struct mbuf *nextrecord;
1459 	int moff, type = 0;
1460 	int orig_resid = uio->uio_resid;
1461 
1462 	mp = mp0;
1463 	if (psa != NULL)
1464 		*psa = NULL;
1465 	if (controlp != NULL)
1466 		*controlp = NULL;
1467 	if (flagsp != NULL)
1468 		flags = *flagsp &~ MSG_EOR;
1469 	else
1470 		flags = 0;
1471 	if (flags & MSG_OOB)
1472 		return (soreceive_rcvoob(so, uio, flags));
1473 	if (mp != NULL)
1474 		*mp = NULL;
1475 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1476 	    && uio->uio_resid) {
1477 		VNET_SO_ASSERT(so);
1478 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1479 	}
1480 
1481 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1482 	if (error)
1483 		return (error);
1484 
1485 restart:
1486 	SOCKBUF_LOCK(&so->so_rcv);
1487 	m = so->so_rcv.sb_mb;
1488 	/*
1489 	 * If we have less data than requested, block awaiting more (subject
1490 	 * to any timeout) if:
1491 	 *   1. the current count is less than the low water mark, or
1492 	 *   2. MSG_WAITALL is set, and it is possible to do the entire
1493 	 *	receive operation at once if we block (resid <= hiwat).
1494 	 *   3. MSG_DONTWAIT is not set
1495 	 * If MSG_WAITALL is set but resid is larger than the receive buffer,
1496 	 * we have to do the receive in sections, and thus risk returning a
1497 	 * short count if a timeout or signal occurs after we start.
1498 	 */
1499 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1500 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1501 	    (so->so_rcv.sb_cc < so->so_rcv.sb_lowat ||
1502 	    ((flags & MSG_WAITALL) && uio->uio_resid <= so->so_rcv.sb_hiwat)) &&
1503 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1504 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1505 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1506 		    m, so->so_rcv.sb_cc));
1507 		if (so->so_error) {
1508 			if (m != NULL)
1509 				goto dontblock;
1510 			error = so->so_error;
1511 			if ((flags & MSG_PEEK) == 0)
1512 				so->so_error = 0;
1513 			SOCKBUF_UNLOCK(&so->so_rcv);
1514 			goto release;
1515 		}
1516 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1517 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1518 			if (m == NULL) {
1519 				SOCKBUF_UNLOCK(&so->so_rcv);
1520 				goto release;
1521 			} else
1522 				goto dontblock;
1523 		}
1524 		for (; m != NULL; m = m->m_next)
1525 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1526 				m = so->so_rcv.sb_mb;
1527 				goto dontblock;
1528 			}
1529 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1530 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1531 			SOCKBUF_UNLOCK(&so->so_rcv);
1532 			error = ENOTCONN;
1533 			goto release;
1534 		}
1535 		if (uio->uio_resid == 0) {
1536 			SOCKBUF_UNLOCK(&so->so_rcv);
1537 			goto release;
1538 		}
1539 		if ((so->so_state & SS_NBIO) ||
1540 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1541 			SOCKBUF_UNLOCK(&so->so_rcv);
1542 			error = EWOULDBLOCK;
1543 			goto release;
1544 		}
1545 		SBLASTRECORDCHK(&so->so_rcv);
1546 		SBLASTMBUFCHK(&so->so_rcv);
1547 		error = sbwait(&so->so_rcv);
1548 		SOCKBUF_UNLOCK(&so->so_rcv);
1549 		if (error)
1550 			goto release;
1551 		goto restart;
1552 	}
1553 dontblock:
1554 	/*
1555 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1556 	 * pointer to the next record in the socket buffer.  We must keep the
1557 	 * various socket buffer pointers and local stack versions of the
1558 	 * pointers in sync, pushing out modifications before dropping the
1559 	 * socket buffer mutex, and re-reading them when picking it up.
1560 	 *
1561 	 * Otherwise, we will race with the network stack appending new data
1562 	 * or records onto the socket buffer by using inconsistent/stale
1563 	 * versions of the field, possibly resulting in socket buffer
1564 	 * corruption.
1565 	 *
1566 	 * By holding the high-level sblock(), we prevent simultaneous
1567 	 * readers from pulling off the front of the socket buffer.
1568 	 */
1569 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1570 	if (uio->uio_td)
1571 		uio->uio_td->td_ru.ru_msgrcv++;
1572 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1573 	SBLASTRECORDCHK(&so->so_rcv);
1574 	SBLASTMBUFCHK(&so->so_rcv);
1575 	nextrecord = m->m_nextpkt;
1576 	if (pr->pr_flags & PR_ADDR) {
1577 		KASSERT(m->m_type == MT_SONAME,
1578 		    ("m->m_type == %d", m->m_type));
1579 		orig_resid = 0;
1580 		if (psa != NULL)
1581 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1582 			    M_NOWAIT);
1583 		if (flags & MSG_PEEK) {
1584 			m = m->m_next;
1585 		} else {
1586 			sbfree(&so->so_rcv, m);
1587 			so->so_rcv.sb_mb = m_free(m);
1588 			m = so->so_rcv.sb_mb;
1589 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1590 		}
1591 	}
1592 
1593 	/*
1594 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1595 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1596 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1597 	 * perform externalization (or freeing if controlp == NULL).
1598 	 */
1599 	if (m != NULL && m->m_type == MT_CONTROL) {
1600 		struct mbuf *cm = NULL, *cmn;
1601 		struct mbuf **cme = &cm;
1602 
1603 		do {
1604 			if (flags & MSG_PEEK) {
1605 				if (controlp != NULL) {
1606 					*controlp = m_copy(m, 0, m->m_len);
1607 					controlp = &(*controlp)->m_next;
1608 				}
1609 				m = m->m_next;
1610 			} else {
1611 				sbfree(&so->so_rcv, m);
1612 				so->so_rcv.sb_mb = m->m_next;
1613 				m->m_next = NULL;
1614 				*cme = m;
1615 				cme = &(*cme)->m_next;
1616 				m = so->so_rcv.sb_mb;
1617 			}
1618 		} while (m != NULL && m->m_type == MT_CONTROL);
1619 		if ((flags & MSG_PEEK) == 0)
1620 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1621 		while (cm != NULL) {
1622 			cmn = cm->m_next;
1623 			cm->m_next = NULL;
1624 			if (pr->pr_domain->dom_externalize != NULL) {
1625 				SOCKBUF_UNLOCK(&so->so_rcv);
1626 				VNET_SO_ASSERT(so);
1627 				error = (*pr->pr_domain->dom_externalize)
1628 				    (cm, controlp);
1629 				SOCKBUF_LOCK(&so->so_rcv);
1630 			} else if (controlp != NULL)
1631 				*controlp = cm;
1632 			else
1633 				m_freem(cm);
1634 			if (controlp != NULL) {
1635 				orig_resid = 0;
1636 				while (*controlp != NULL)
1637 					controlp = &(*controlp)->m_next;
1638 			}
1639 			cm = cmn;
1640 		}
1641 		if (m != NULL)
1642 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1643 		else
1644 			nextrecord = so->so_rcv.sb_mb;
1645 		orig_resid = 0;
1646 	}
1647 	if (m != NULL) {
1648 		if ((flags & MSG_PEEK) == 0) {
1649 			KASSERT(m->m_nextpkt == nextrecord,
1650 			    ("soreceive: post-control, nextrecord !sync"));
1651 			if (nextrecord == NULL) {
1652 				KASSERT(so->so_rcv.sb_mb == m,
1653 				    ("soreceive: post-control, sb_mb!=m"));
1654 				KASSERT(so->so_rcv.sb_lastrecord == m,
1655 				    ("soreceive: post-control, lastrecord!=m"));
1656 			}
1657 		}
1658 		type = m->m_type;
1659 		if (type == MT_OOBDATA)
1660 			flags |= MSG_OOB;
1661 	} else {
1662 		if ((flags & MSG_PEEK) == 0) {
1663 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1664 			    ("soreceive: sb_mb != nextrecord"));
1665 			if (so->so_rcv.sb_mb == NULL) {
1666 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1667 				    ("soreceive: sb_lastercord != NULL"));
1668 			}
1669 		}
1670 	}
1671 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1672 	SBLASTRECORDCHK(&so->so_rcv);
1673 	SBLASTMBUFCHK(&so->so_rcv);
1674 
1675 	/*
1676 	 * Now continue to read any data mbufs off of the head of the socket
1677 	 * buffer until the read request is satisfied.  Note that 'type' is
1678 	 * used to store the type of any mbuf reads that have happened so far
1679 	 * such that soreceive() can stop reading if the type changes, which
1680 	 * causes soreceive() to return only one of regular data and inline
1681 	 * out-of-band data in a single socket receive operation.
1682 	 */
1683 	moff = 0;
1684 	offset = 0;
1685 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1686 		/*
1687 		 * If the type of mbuf has changed since the last mbuf
1688 		 * examined ('type'), end the receive operation.
1689 	 	 */
1690 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1691 		if (m->m_type == MT_OOBDATA) {
1692 			if (type != MT_OOBDATA)
1693 				break;
1694 		} else if (type == MT_OOBDATA)
1695 			break;
1696 		else
1697 		    KASSERT(m->m_type == MT_DATA,
1698 			("m->m_type == %d", m->m_type));
1699 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1700 		len = uio->uio_resid;
1701 		if (so->so_oobmark && len > so->so_oobmark - offset)
1702 			len = so->so_oobmark - offset;
1703 		if (len > m->m_len - moff)
1704 			len = m->m_len - moff;
1705 		/*
1706 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1707 		 * them out via the uio, then free.  Sockbuf must be
1708 		 * consistent here (points to current mbuf, it points to next
1709 		 * record) when we drop priority; we must note any additions
1710 		 * to the sockbuf when we block interrupts again.
1711 		 */
1712 		if (mp == NULL) {
1713 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1714 			SBLASTRECORDCHK(&so->so_rcv);
1715 			SBLASTMBUFCHK(&so->so_rcv);
1716 			SOCKBUF_UNLOCK(&so->so_rcv);
1717 #ifdef ZERO_COPY_SOCKETS
1718 			if (so_zero_copy_receive) {
1719 				int disposable;
1720 
1721 				if ((m->m_flags & M_EXT)
1722 				 && (m->m_ext.ext_type == EXT_DISPOSABLE))
1723 					disposable = 1;
1724 				else
1725 					disposable = 0;
1726 
1727 				error = uiomoveco(mtod(m, char *) + moff,
1728 						  (int)len, uio,
1729 						  disposable);
1730 			} else
1731 #endif /* ZERO_COPY_SOCKETS */
1732 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1733 			SOCKBUF_LOCK(&so->so_rcv);
1734 			if (error) {
1735 				/*
1736 				 * The MT_SONAME mbuf has already been removed
1737 				 * from the record, so it is necessary to
1738 				 * remove the data mbufs, if any, to preserve
1739 				 * the invariant in the case of PR_ADDR that
1740 				 * requires MT_SONAME mbufs at the head of
1741 				 * each record.
1742 				 */
1743 				if (m && pr->pr_flags & PR_ATOMIC &&
1744 				    ((flags & MSG_PEEK) == 0))
1745 					(void)sbdroprecord_locked(&so->so_rcv);
1746 				SOCKBUF_UNLOCK(&so->so_rcv);
1747 				goto release;
1748 			}
1749 		} else
1750 			uio->uio_resid -= len;
1751 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1752 		if (len == m->m_len - moff) {
1753 			if (m->m_flags & M_EOR)
1754 				flags |= MSG_EOR;
1755 			if (flags & MSG_PEEK) {
1756 				m = m->m_next;
1757 				moff = 0;
1758 			} else {
1759 				nextrecord = m->m_nextpkt;
1760 				sbfree(&so->so_rcv, m);
1761 				if (mp != NULL) {
1762 					*mp = m;
1763 					mp = &m->m_next;
1764 					so->so_rcv.sb_mb = m = m->m_next;
1765 					*mp = NULL;
1766 				} else {
1767 					so->so_rcv.sb_mb = m_free(m);
1768 					m = so->so_rcv.sb_mb;
1769 				}
1770 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1771 				SBLASTRECORDCHK(&so->so_rcv);
1772 				SBLASTMBUFCHK(&so->so_rcv);
1773 			}
1774 		} else {
1775 			if (flags & MSG_PEEK)
1776 				moff += len;
1777 			else {
1778 				if (mp != NULL) {
1779 					int copy_flag;
1780 
1781 					if (flags & MSG_DONTWAIT)
1782 						copy_flag = M_DONTWAIT;
1783 					else
1784 						copy_flag = M_WAIT;
1785 					if (copy_flag == M_WAIT)
1786 						SOCKBUF_UNLOCK(&so->so_rcv);
1787 					*mp = m_copym(m, 0, len, copy_flag);
1788 					if (copy_flag == M_WAIT)
1789 						SOCKBUF_LOCK(&so->so_rcv);
1790  					if (*mp == NULL) {
1791  						/*
1792  						 * m_copym() couldn't
1793 						 * allocate an mbuf.  Adjust
1794 						 * uio_resid back (it was
1795 						 * adjusted down by len
1796 						 * bytes, which we didn't end
1797 						 * up "copying" over).
1798  						 */
1799  						uio->uio_resid += len;
1800  						break;
1801  					}
1802 				}
1803 				m->m_data += len;
1804 				m->m_len -= len;
1805 				so->so_rcv.sb_cc -= len;
1806 			}
1807 		}
1808 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1809 		if (so->so_oobmark) {
1810 			if ((flags & MSG_PEEK) == 0) {
1811 				so->so_oobmark -= len;
1812 				if (so->so_oobmark == 0) {
1813 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1814 					break;
1815 				}
1816 			} else {
1817 				offset += len;
1818 				if (offset == so->so_oobmark)
1819 					break;
1820 			}
1821 		}
1822 		if (flags & MSG_EOR)
1823 			break;
1824 		/*
1825 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1826 		 * must not quit until "uio->uio_resid == 0" or an error
1827 		 * termination.  If a signal/timeout occurs, return with a
1828 		 * short count but without error.  Keep sockbuf locked
1829 		 * against other readers.
1830 		 */
1831 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1832 		    !sosendallatonce(so) && nextrecord == NULL) {
1833 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1834 			if (so->so_error || so->so_rcv.sb_state & SBS_CANTRCVMORE)
1835 				break;
1836 			/*
1837 			 * Notify the protocol that some data has been
1838 			 * drained before blocking.
1839 			 */
1840 			if (pr->pr_flags & PR_WANTRCVD) {
1841 				SOCKBUF_UNLOCK(&so->so_rcv);
1842 				VNET_SO_ASSERT(so);
1843 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1844 				SOCKBUF_LOCK(&so->so_rcv);
1845 			}
1846 			SBLASTRECORDCHK(&so->so_rcv);
1847 			SBLASTMBUFCHK(&so->so_rcv);
1848 			/*
1849 			 * We could receive some data while was notifying
1850 			 * the protocol. Skip blocking in this case.
1851 			 */
1852 			if (so->so_rcv.sb_mb == NULL) {
1853 				error = sbwait(&so->so_rcv);
1854 				if (error) {
1855 					SOCKBUF_UNLOCK(&so->so_rcv);
1856 					goto release;
1857 				}
1858 			}
1859 			m = so->so_rcv.sb_mb;
1860 			if (m != NULL)
1861 				nextrecord = m->m_nextpkt;
1862 		}
1863 	}
1864 
1865 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1866 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1867 		flags |= MSG_TRUNC;
1868 		if ((flags & MSG_PEEK) == 0)
1869 			(void) sbdroprecord_locked(&so->so_rcv);
1870 	}
1871 	if ((flags & MSG_PEEK) == 0) {
1872 		if (m == NULL) {
1873 			/*
1874 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1875 			 * part makes sure sb_lastrecord is up-to-date if
1876 			 * there is still data in the socket buffer.
1877 			 */
1878 			so->so_rcv.sb_mb = nextrecord;
1879 			if (so->so_rcv.sb_mb == NULL) {
1880 				so->so_rcv.sb_mbtail = NULL;
1881 				so->so_rcv.sb_lastrecord = NULL;
1882 			} else if (nextrecord->m_nextpkt == NULL)
1883 				so->so_rcv.sb_lastrecord = nextrecord;
1884 		}
1885 		SBLASTRECORDCHK(&so->so_rcv);
1886 		SBLASTMBUFCHK(&so->so_rcv);
1887 		/*
1888 		 * If soreceive() is being done from the socket callback,
1889 		 * then don't need to generate ACK to peer to update window,
1890 		 * since ACK will be generated on return to TCP.
1891 		 */
1892 		if (!(flags & MSG_SOCALLBCK) &&
1893 		    (pr->pr_flags & PR_WANTRCVD)) {
1894 			SOCKBUF_UNLOCK(&so->so_rcv);
1895 			VNET_SO_ASSERT(so);
1896 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1897 			SOCKBUF_LOCK(&so->so_rcv);
1898 		}
1899 	}
1900 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1901 	if (orig_resid == uio->uio_resid && orig_resid &&
1902 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1903 		SOCKBUF_UNLOCK(&so->so_rcv);
1904 		goto restart;
1905 	}
1906 	SOCKBUF_UNLOCK(&so->so_rcv);
1907 
1908 	if (flagsp != NULL)
1909 		*flagsp |= flags;
1910 release:
1911 	sbunlock(&so->so_rcv);
1912 	return (error);
1913 }
1914 
1915 /*
1916  * Optimized version of soreceive() for stream (TCP) sockets.
1917  */
1918 #ifdef TCP_SORECEIVE_STREAM
1919 int
1920 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1921     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1922 {
1923 	int len = 0, error = 0, flags, oresid;
1924 	struct sockbuf *sb;
1925 	struct mbuf *m, *n = NULL;
1926 
1927 	/* We only do stream sockets. */
1928 	if (so->so_type != SOCK_STREAM)
1929 		return (EINVAL);
1930 	if (psa != NULL)
1931 		*psa = NULL;
1932 	if (controlp != NULL)
1933 		return (EINVAL);
1934 	if (flagsp != NULL)
1935 		flags = *flagsp &~ MSG_EOR;
1936 	else
1937 		flags = 0;
1938 	if (flags & MSG_OOB)
1939 		return (soreceive_rcvoob(so, uio, flags));
1940 	if (mp0 != NULL)
1941 		*mp0 = NULL;
1942 
1943 	sb = &so->so_rcv;
1944 
1945 	/* Prevent other readers from entering the socket. */
1946 	error = sblock(sb, SBLOCKWAIT(flags));
1947 	if (error)
1948 		goto out;
1949 	SOCKBUF_LOCK(sb);
1950 
1951 	/* Easy one, no space to copyout anything. */
1952 	if (uio->uio_resid == 0) {
1953 		error = EINVAL;
1954 		goto out;
1955 	}
1956 	oresid = uio->uio_resid;
1957 
1958 	/* We will never ever get anything unless we are connected. */
1959 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1960 		/* When disconnecting there may be still some data left. */
1961 		if (sb->sb_cc > 0)
1962 			goto deliver;
1963 		if (!(so->so_state & SS_ISDISCONNECTED))
1964 			error = ENOTCONN;
1965 		goto out;
1966 	}
1967 
1968 	/* Socket buffer is empty and we shall not block. */
1969 	if (sb->sb_cc == 0 &&
1970 	    ((sb->sb_flags & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1971 		error = EAGAIN;
1972 		goto out;
1973 	}
1974 
1975 restart:
1976 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1977 
1978 	/* Abort if socket has reported problems. */
1979 	if (so->so_error) {
1980 		if (sb->sb_cc > 0)
1981 			goto deliver;
1982 		if (oresid > uio->uio_resid)
1983 			goto out;
1984 		error = so->so_error;
1985 		if (!(flags & MSG_PEEK))
1986 			so->so_error = 0;
1987 		goto out;
1988 	}
1989 
1990 	/* Door is closed.  Deliver what is left, if any. */
1991 	if (sb->sb_state & SBS_CANTRCVMORE) {
1992 		if (sb->sb_cc > 0)
1993 			goto deliver;
1994 		else
1995 			goto out;
1996 	}
1997 
1998 	/* Socket buffer got some data that we shall deliver now. */
1999 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
2000 	    ((sb->sb_flags & SS_NBIO) ||
2001 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2002 	     sb->sb_cc >= sb->sb_lowat ||
2003 	     sb->sb_cc >= uio->uio_resid ||
2004 	     sb->sb_cc >= sb->sb_hiwat) ) {
2005 		goto deliver;
2006 	}
2007 
2008 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2009 	if ((flags & MSG_WAITALL) &&
2010 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_lowat))
2011 		goto deliver;
2012 
2013 	/*
2014 	 * Wait and block until (more) data comes in.
2015 	 * NB: Drops the sockbuf lock during wait.
2016 	 */
2017 	error = sbwait(sb);
2018 	if (error)
2019 		goto out;
2020 	goto restart;
2021 
2022 deliver:
2023 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2024 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2025 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2026 
2027 	/* Statistics. */
2028 	if (uio->uio_td)
2029 		uio->uio_td->td_ru.ru_msgrcv++;
2030 
2031 	/* Fill uio until full or current end of socket buffer is reached. */
2032 	len = min(uio->uio_resid, sb->sb_cc);
2033 	if (mp0 != NULL) {
2034 		/* Dequeue as many mbufs as possible. */
2035 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2036 			for (*mp0 = m = sb->sb_mb;
2037 			     m != NULL && m->m_len <= len;
2038 			     m = m->m_next) {
2039 				len -= m->m_len;
2040 				uio->uio_resid -= m->m_len;
2041 				sbfree(sb, m);
2042 				n = m;
2043 			}
2044 			sb->sb_mb = m;
2045 			if (sb->sb_mb == NULL)
2046 				SB_EMPTY_FIXUP(sb);
2047 			n->m_next = NULL;
2048 		}
2049 		/* Copy the remainder. */
2050 		if (len > 0) {
2051 			KASSERT(sb->sb_mb != NULL,
2052 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2053 
2054 			m = m_copym(sb->sb_mb, 0, len, M_DONTWAIT);
2055 			if (m == NULL)
2056 				len = 0;	/* Don't flush data from sockbuf. */
2057 			else
2058 				uio->uio_resid -= m->m_len;
2059 			if (*mp0 != NULL)
2060 				n->m_next = m;
2061 			else
2062 				*mp0 = m;
2063 			if (*mp0 == NULL) {
2064 				error = ENOBUFS;
2065 				goto out;
2066 			}
2067 		}
2068 	} else {
2069 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2070 		SOCKBUF_UNLOCK(sb);
2071 		error = m_mbuftouio(uio, sb->sb_mb, len);
2072 		SOCKBUF_LOCK(sb);
2073 		if (error)
2074 			goto out;
2075 	}
2076 	SBLASTRECORDCHK(sb);
2077 	SBLASTMBUFCHK(sb);
2078 
2079 	/*
2080 	 * Remove the delivered data from the socket buffer unless we
2081 	 * were only peeking.
2082 	 */
2083 	if (!(flags & MSG_PEEK)) {
2084 		if (len > 0)
2085 			sbdrop_locked(sb, len);
2086 
2087 		/* Notify protocol that we drained some data. */
2088 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2089 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2090 		     !(flags & MSG_SOCALLBCK))) {
2091 			SOCKBUF_UNLOCK(sb);
2092 			VNET_SO_ASSERT(so);
2093 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2094 			SOCKBUF_LOCK(sb);
2095 		}
2096 	}
2097 
2098 	/*
2099 	 * For MSG_WAITALL we may have to loop again and wait for
2100 	 * more data to come in.
2101 	 */
2102 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2103 		goto restart;
2104 out:
2105 	SOCKBUF_LOCK_ASSERT(sb);
2106 	SBLASTRECORDCHK(sb);
2107 	SBLASTMBUFCHK(sb);
2108 	SOCKBUF_UNLOCK(sb);
2109 	sbunlock(sb);
2110 	return (error);
2111 }
2112 #endif /* TCP_SORECEIVE_STREAM */
2113 
2114 /*
2115  * Optimized version of soreceive() for simple datagram cases from userspace.
2116  * Unlike in the stream case, we're able to drop a datagram if copyout()
2117  * fails, and because we handle datagrams atomically, we don't need to use a
2118  * sleep lock to prevent I/O interlacing.
2119  */
2120 int
2121 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2122     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2123 {
2124 	struct mbuf *m, *m2;
2125 	int flags, len, error;
2126 	struct protosw *pr = so->so_proto;
2127 	struct mbuf *nextrecord;
2128 
2129 	if (psa != NULL)
2130 		*psa = NULL;
2131 	if (controlp != NULL)
2132 		*controlp = NULL;
2133 	if (flagsp != NULL)
2134 		flags = *flagsp &~ MSG_EOR;
2135 	else
2136 		flags = 0;
2137 
2138 	/*
2139 	 * For any complicated cases, fall back to the full
2140 	 * soreceive_generic().
2141 	 */
2142 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2143 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2144 		    flagsp));
2145 
2146 	/*
2147 	 * Enforce restrictions on use.
2148 	 */
2149 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2150 	    ("soreceive_dgram: wantrcvd"));
2151 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2152 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2153 	    ("soreceive_dgram: SBS_RCVATMARK"));
2154 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2155 	    ("soreceive_dgram: P_CONNREQUIRED"));
2156 
2157 	/*
2158 	 * Loop blocking while waiting for a datagram.
2159 	 */
2160 	SOCKBUF_LOCK(&so->so_rcv);
2161 	while ((m = so->so_rcv.sb_mb) == NULL) {
2162 		KASSERT(so->so_rcv.sb_cc == 0,
2163 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2164 		    so->so_rcv.sb_cc));
2165 		if (so->so_error) {
2166 			error = so->so_error;
2167 			so->so_error = 0;
2168 			SOCKBUF_UNLOCK(&so->so_rcv);
2169 			return (error);
2170 		}
2171 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2172 		    uio->uio_resid == 0) {
2173 			SOCKBUF_UNLOCK(&so->so_rcv);
2174 			return (0);
2175 		}
2176 		if ((so->so_state & SS_NBIO) ||
2177 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2178 			SOCKBUF_UNLOCK(&so->so_rcv);
2179 			return (EWOULDBLOCK);
2180 		}
2181 		SBLASTRECORDCHK(&so->so_rcv);
2182 		SBLASTMBUFCHK(&so->so_rcv);
2183 		error = sbwait(&so->so_rcv);
2184 		if (error) {
2185 			SOCKBUF_UNLOCK(&so->so_rcv);
2186 			return (error);
2187 		}
2188 	}
2189 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2190 
2191 	if (uio->uio_td)
2192 		uio->uio_td->td_ru.ru_msgrcv++;
2193 	SBLASTRECORDCHK(&so->so_rcv);
2194 	SBLASTMBUFCHK(&so->so_rcv);
2195 	nextrecord = m->m_nextpkt;
2196 	if (nextrecord == NULL) {
2197 		KASSERT(so->so_rcv.sb_lastrecord == m,
2198 		    ("soreceive_dgram: lastrecord != m"));
2199 	}
2200 
2201 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2202 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2203 
2204 	/*
2205 	 * Pull 'm' and its chain off the front of the packet queue.
2206 	 */
2207 	so->so_rcv.sb_mb = NULL;
2208 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2209 
2210 	/*
2211 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2212 	 */
2213 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2214 		sbfree(&so->so_rcv, m2);
2215 
2216 	/*
2217 	 * Do a few last checks before we let go of the lock.
2218 	 */
2219 	SBLASTRECORDCHK(&so->so_rcv);
2220 	SBLASTMBUFCHK(&so->so_rcv);
2221 	SOCKBUF_UNLOCK(&so->so_rcv);
2222 
2223 	if (pr->pr_flags & PR_ADDR) {
2224 		KASSERT(m->m_type == MT_SONAME,
2225 		    ("m->m_type == %d", m->m_type));
2226 		if (psa != NULL)
2227 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2228 			    M_NOWAIT);
2229 		m = m_free(m);
2230 	}
2231 	if (m == NULL) {
2232 		/* XXXRW: Can this happen? */
2233 		return (0);
2234 	}
2235 
2236 	/*
2237 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2238 	 * queue.
2239 	 *
2240 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2241 	 * in the first mbuf chain on the socket buffer.  We call into the
2242 	 * protocol to perform externalization (or freeing if controlp ==
2243 	 * NULL).
2244 	 */
2245 	if (m->m_type == MT_CONTROL) {
2246 		struct mbuf *cm = NULL, *cmn;
2247 		struct mbuf **cme = &cm;
2248 
2249 		do {
2250 			m2 = m->m_next;
2251 			m->m_next = NULL;
2252 			*cme = m;
2253 			cme = &(*cme)->m_next;
2254 			m = m2;
2255 		} while (m != NULL && m->m_type == MT_CONTROL);
2256 		while (cm != NULL) {
2257 			cmn = cm->m_next;
2258 			cm->m_next = NULL;
2259 			if (pr->pr_domain->dom_externalize != NULL) {
2260 				error = (*pr->pr_domain->dom_externalize)
2261 				    (cm, controlp);
2262 			} else if (controlp != NULL)
2263 				*controlp = cm;
2264 			else
2265 				m_freem(cm);
2266 			if (controlp != NULL) {
2267 				while (*controlp != NULL)
2268 					controlp = &(*controlp)->m_next;
2269 			}
2270 			cm = cmn;
2271 		}
2272 	}
2273 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2274 
2275 	while (m != NULL && uio->uio_resid > 0) {
2276 		len = uio->uio_resid;
2277 		if (len > m->m_len)
2278 			len = m->m_len;
2279 		error = uiomove(mtod(m, char *), (int)len, uio);
2280 		if (error) {
2281 			m_freem(m);
2282 			return (error);
2283 		}
2284 		if (len == m->m_len)
2285 			m = m_free(m);
2286 		else {
2287 			m->m_data += len;
2288 			m->m_len -= len;
2289 		}
2290 	}
2291 	if (m != NULL)
2292 		flags |= MSG_TRUNC;
2293 	m_freem(m);
2294 	if (flagsp != NULL)
2295 		*flagsp |= flags;
2296 	return (0);
2297 }
2298 
2299 int
2300 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2301     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2302 {
2303 	int error;
2304 
2305 	CURVNET_SET(so->so_vnet);
2306 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2307 	    controlp, flagsp));
2308 	CURVNET_RESTORE();
2309 	return (error);
2310 }
2311 
2312 int
2313 soshutdown(struct socket *so, int how)
2314 {
2315 	struct protosw *pr = so->so_proto;
2316 	int error;
2317 
2318 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2319 		return (EINVAL);
2320 
2321 	CURVNET_SET(so->so_vnet);
2322 	if (pr->pr_usrreqs->pru_flush != NULL) {
2323 	        (*pr->pr_usrreqs->pru_flush)(so, how);
2324 	}
2325 	if (how != SHUT_WR)
2326 		sorflush(so);
2327 	if (how != SHUT_RD) {
2328 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2329 		CURVNET_RESTORE();
2330 		return (error);
2331 	}
2332 	CURVNET_RESTORE();
2333 	return (0);
2334 }
2335 
2336 void
2337 sorflush(struct socket *so)
2338 {
2339 	struct sockbuf *sb = &so->so_rcv;
2340 	struct protosw *pr = so->so_proto;
2341 	struct sockbuf asb;
2342 
2343 	VNET_SO_ASSERT(so);
2344 
2345 	/*
2346 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2347 	 * held, and in order to generally avoid holding the lock for a long
2348 	 * time, we make a copy of the socket buffer and clear the original
2349 	 * (except locks, state).  The new socket buffer copy won't have
2350 	 * initialized locks so we can only call routines that won't use or
2351 	 * assert those locks.
2352 	 *
2353 	 * Dislodge threads currently blocked in receive and wait to acquire
2354 	 * a lock against other simultaneous readers before clearing the
2355 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2356 	 * despite any existing socket disposition on interruptable waiting.
2357 	 */
2358 	socantrcvmore(so);
2359 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2360 
2361 	/*
2362 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2363 	 * and mutex data unchanged.
2364 	 */
2365 	SOCKBUF_LOCK(sb);
2366 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2367 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2368 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2369 	bzero(&sb->sb_startzero,
2370 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2371 	SOCKBUF_UNLOCK(sb);
2372 	sbunlock(sb);
2373 
2374 	/*
2375 	 * Dispose of special rights and flush the socket buffer.  Don't call
2376 	 * any unsafe routines (that rely on locks being initialized) on asb.
2377 	 */
2378 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2379 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2380 	sbrelease_internal(&asb, so);
2381 }
2382 
2383 /*
2384  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2385  * additional variant to handle the case where the option value needs to be
2386  * some kind of integer, but not a specific size.  In addition to their use
2387  * here, these functions are also called by the protocol-level pr_ctloutput()
2388  * routines.
2389  */
2390 int
2391 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2392 {
2393 	size_t	valsize;
2394 
2395 	/*
2396 	 * If the user gives us more than we wanted, we ignore it, but if we
2397 	 * don't get the minimum length the caller wants, we return EINVAL.
2398 	 * On success, sopt->sopt_valsize is set to however much we actually
2399 	 * retrieved.
2400 	 */
2401 	if ((valsize = sopt->sopt_valsize) < minlen)
2402 		return EINVAL;
2403 	if (valsize > len)
2404 		sopt->sopt_valsize = valsize = len;
2405 
2406 	if (sopt->sopt_td != NULL)
2407 		return (copyin(sopt->sopt_val, buf, valsize));
2408 
2409 	bcopy(sopt->sopt_val, buf, valsize);
2410 	return (0);
2411 }
2412 
2413 /*
2414  * Kernel version of setsockopt(2).
2415  *
2416  * XXX: optlen is size_t, not socklen_t
2417  */
2418 int
2419 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2420     size_t optlen)
2421 {
2422 	struct sockopt sopt;
2423 
2424 	sopt.sopt_level = level;
2425 	sopt.sopt_name = optname;
2426 	sopt.sopt_dir = SOPT_SET;
2427 	sopt.sopt_val = optval;
2428 	sopt.sopt_valsize = optlen;
2429 	sopt.sopt_td = NULL;
2430 	return (sosetopt(so, &sopt));
2431 }
2432 
2433 int
2434 sosetopt(struct socket *so, struct sockopt *sopt)
2435 {
2436 	int	error, optval;
2437 	struct	linger l;
2438 	struct	timeval tv;
2439 	u_long  val;
2440 	uint32_t val32;
2441 #ifdef MAC
2442 	struct mac extmac;
2443 #endif
2444 
2445 	CURVNET_SET(so->so_vnet);
2446 	error = 0;
2447 	if (sopt->sopt_level != SOL_SOCKET) {
2448 		if (so->so_proto && so->so_proto->pr_ctloutput) {
2449 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2450 			CURVNET_RESTORE();
2451 			return (error);
2452 		}
2453 		error = ENOPROTOOPT;
2454 	} else {
2455 		switch (sopt->sopt_name) {
2456 #ifdef INET
2457 		case SO_ACCEPTFILTER:
2458 			error = do_setopt_accept_filter(so, sopt);
2459 			if (error)
2460 				goto bad;
2461 			break;
2462 #endif
2463 		case SO_LINGER:
2464 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2465 			if (error)
2466 				goto bad;
2467 
2468 			SOCK_LOCK(so);
2469 			so->so_linger = l.l_linger;
2470 			if (l.l_onoff)
2471 				so->so_options |= SO_LINGER;
2472 			else
2473 				so->so_options &= ~SO_LINGER;
2474 			SOCK_UNLOCK(so);
2475 			break;
2476 
2477 		case SO_DEBUG:
2478 		case SO_KEEPALIVE:
2479 		case SO_DONTROUTE:
2480 		case SO_USELOOPBACK:
2481 		case SO_BROADCAST:
2482 		case SO_REUSEADDR:
2483 		case SO_REUSEPORT:
2484 		case SO_OOBINLINE:
2485 		case SO_TIMESTAMP:
2486 		case SO_BINTIME:
2487 		case SO_NOSIGPIPE:
2488 		case SO_NO_DDP:
2489 		case SO_NO_OFFLOAD:
2490 			error = sooptcopyin(sopt, &optval, sizeof optval,
2491 					    sizeof optval);
2492 			if (error)
2493 				goto bad;
2494 			SOCK_LOCK(so);
2495 			if (optval)
2496 				so->so_options |= sopt->sopt_name;
2497 			else
2498 				so->so_options &= ~sopt->sopt_name;
2499 			SOCK_UNLOCK(so);
2500 			break;
2501 
2502 		case SO_SETFIB:
2503 			error = sooptcopyin(sopt, &optval, sizeof optval,
2504 					    sizeof optval);
2505 			if (optval < 0 || optval > rt_numfibs) {
2506 				error = EINVAL;
2507 				goto bad;
2508 			}
2509 			if (so->so_proto != NULL &&
2510 			   ((so->so_proto->pr_domain->dom_family == PF_INET) ||
2511 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE))) {
2512 				so->so_fibnum = optval;
2513 				/* Note: ignore error */
2514 				if (so->so_proto->pr_ctloutput)
2515 					(*so->so_proto->pr_ctloutput)(so, sopt);
2516 			} else {
2517 				so->so_fibnum = 0;
2518 			}
2519 			break;
2520 
2521 		case SO_USER_COOKIE:
2522 			error = sooptcopyin(sopt, &val32, sizeof val32,
2523 					    sizeof val32);
2524 			if (error)
2525 				goto bad;
2526 			so->so_user_cookie = val32;
2527 			break;
2528 
2529 		case SO_SNDBUF:
2530 		case SO_RCVBUF:
2531 		case SO_SNDLOWAT:
2532 		case SO_RCVLOWAT:
2533 			error = sooptcopyin(sopt, &optval, sizeof optval,
2534 					    sizeof optval);
2535 			if (error)
2536 				goto bad;
2537 
2538 			/*
2539 			 * Values < 1 make no sense for any of these options,
2540 			 * so disallow them.
2541 			 */
2542 			if (optval < 1) {
2543 				error = EINVAL;
2544 				goto bad;
2545 			}
2546 
2547 			switch (sopt->sopt_name) {
2548 			case SO_SNDBUF:
2549 			case SO_RCVBUF:
2550 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2551 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2552 				    so, curthread) == 0) {
2553 					error = ENOBUFS;
2554 					goto bad;
2555 				}
2556 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2557 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2558 				break;
2559 
2560 			/*
2561 			 * Make sure the low-water is never greater than the
2562 			 * high-water.
2563 			 */
2564 			case SO_SNDLOWAT:
2565 				SOCKBUF_LOCK(&so->so_snd);
2566 				so->so_snd.sb_lowat =
2567 				    (optval > so->so_snd.sb_hiwat) ?
2568 				    so->so_snd.sb_hiwat : optval;
2569 				SOCKBUF_UNLOCK(&so->so_snd);
2570 				break;
2571 			case SO_RCVLOWAT:
2572 				SOCKBUF_LOCK(&so->so_rcv);
2573 				so->so_rcv.sb_lowat =
2574 				    (optval > so->so_rcv.sb_hiwat) ?
2575 				    so->so_rcv.sb_hiwat : optval;
2576 				SOCKBUF_UNLOCK(&so->so_rcv);
2577 				break;
2578 			}
2579 			break;
2580 
2581 		case SO_SNDTIMEO:
2582 		case SO_RCVTIMEO:
2583 #ifdef COMPAT_FREEBSD32
2584 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2585 				struct timeval32 tv32;
2586 
2587 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2588 				    sizeof tv32);
2589 				CP(tv32, tv, tv_sec);
2590 				CP(tv32, tv, tv_usec);
2591 			} else
2592 #endif
2593 				error = sooptcopyin(sopt, &tv, sizeof tv,
2594 				    sizeof tv);
2595 			if (error)
2596 				goto bad;
2597 
2598 			/* assert(hz > 0); */
2599 			if (tv.tv_sec < 0 || tv.tv_sec > INT_MAX / hz ||
2600 			    tv.tv_usec < 0 || tv.tv_usec >= 1000000) {
2601 				error = EDOM;
2602 				goto bad;
2603 			}
2604 			/* assert(tick > 0); */
2605 			/* assert(ULONG_MAX - INT_MAX >= 1000000); */
2606 			val = (u_long)(tv.tv_sec * hz) + tv.tv_usec / tick;
2607 			if (val > INT_MAX) {
2608 				error = EDOM;
2609 				goto bad;
2610 			}
2611 			if (val == 0 && tv.tv_usec != 0)
2612 				val = 1;
2613 
2614 			switch (sopt->sopt_name) {
2615 			case SO_SNDTIMEO:
2616 				so->so_snd.sb_timeo = val;
2617 				break;
2618 			case SO_RCVTIMEO:
2619 				so->so_rcv.sb_timeo = val;
2620 				break;
2621 			}
2622 			break;
2623 
2624 		case SO_LABEL:
2625 #ifdef MAC
2626 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2627 			    sizeof extmac);
2628 			if (error)
2629 				goto bad;
2630 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2631 			    so, &extmac);
2632 #else
2633 			error = EOPNOTSUPP;
2634 #endif
2635 			break;
2636 
2637 		default:
2638 			error = ENOPROTOOPT;
2639 			break;
2640 		}
2641 		if (error == 0 && so->so_proto != NULL &&
2642 		    so->so_proto->pr_ctloutput != NULL) {
2643 			(void) ((*so->so_proto->pr_ctloutput)
2644 				  (so, sopt));
2645 		}
2646 	}
2647 bad:
2648 	CURVNET_RESTORE();
2649 	return (error);
2650 }
2651 
2652 /*
2653  * Helper routine for getsockopt.
2654  */
2655 int
2656 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2657 {
2658 	int	error;
2659 	size_t	valsize;
2660 
2661 	error = 0;
2662 
2663 	/*
2664 	 * Documented get behavior is that we always return a value, possibly
2665 	 * truncated to fit in the user's buffer.  Traditional behavior is
2666 	 * that we always tell the user precisely how much we copied, rather
2667 	 * than something useful like the total amount we had available for
2668 	 * her.  Note that this interface is not idempotent; the entire
2669 	 * answer must generated ahead of time.
2670 	 */
2671 	valsize = min(len, sopt->sopt_valsize);
2672 	sopt->sopt_valsize = valsize;
2673 	if (sopt->sopt_val != NULL) {
2674 		if (sopt->sopt_td != NULL)
2675 			error = copyout(buf, sopt->sopt_val, valsize);
2676 		else
2677 			bcopy(buf, sopt->sopt_val, valsize);
2678 	}
2679 	return (error);
2680 }
2681 
2682 int
2683 sogetopt(struct socket *so, struct sockopt *sopt)
2684 {
2685 	int	error, optval;
2686 	struct	linger l;
2687 	struct	timeval tv;
2688 #ifdef MAC
2689 	struct mac extmac;
2690 #endif
2691 
2692 	CURVNET_SET(so->so_vnet);
2693 	error = 0;
2694 	if (sopt->sopt_level != SOL_SOCKET) {
2695 		if (so->so_proto && so->so_proto->pr_ctloutput)
2696 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2697 		else
2698 			error = ENOPROTOOPT;
2699 		CURVNET_RESTORE();
2700 		return (error);
2701 	} else {
2702 		switch (sopt->sopt_name) {
2703 #ifdef INET
2704 		case SO_ACCEPTFILTER:
2705 			error = do_getopt_accept_filter(so, sopt);
2706 			break;
2707 #endif
2708 		case SO_LINGER:
2709 			SOCK_LOCK(so);
2710 			l.l_onoff = so->so_options & SO_LINGER;
2711 			l.l_linger = so->so_linger;
2712 			SOCK_UNLOCK(so);
2713 			error = sooptcopyout(sopt, &l, sizeof l);
2714 			break;
2715 
2716 		case SO_USELOOPBACK:
2717 		case SO_DONTROUTE:
2718 		case SO_DEBUG:
2719 		case SO_KEEPALIVE:
2720 		case SO_REUSEADDR:
2721 		case SO_REUSEPORT:
2722 		case SO_BROADCAST:
2723 		case SO_OOBINLINE:
2724 		case SO_ACCEPTCONN:
2725 		case SO_TIMESTAMP:
2726 		case SO_BINTIME:
2727 		case SO_NOSIGPIPE:
2728 			optval = so->so_options & sopt->sopt_name;
2729 integer:
2730 			error = sooptcopyout(sopt, &optval, sizeof optval);
2731 			break;
2732 
2733 		case SO_TYPE:
2734 			optval = so->so_type;
2735 			goto integer;
2736 
2737 		case SO_ERROR:
2738 			SOCK_LOCK(so);
2739 			optval = so->so_error;
2740 			so->so_error = 0;
2741 			SOCK_UNLOCK(so);
2742 			goto integer;
2743 
2744 		case SO_SNDBUF:
2745 			optval = so->so_snd.sb_hiwat;
2746 			goto integer;
2747 
2748 		case SO_RCVBUF:
2749 			optval = so->so_rcv.sb_hiwat;
2750 			goto integer;
2751 
2752 		case SO_SNDLOWAT:
2753 			optval = so->so_snd.sb_lowat;
2754 			goto integer;
2755 
2756 		case SO_RCVLOWAT:
2757 			optval = so->so_rcv.sb_lowat;
2758 			goto integer;
2759 
2760 		case SO_SNDTIMEO:
2761 		case SO_RCVTIMEO:
2762 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2763 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2764 
2765 			tv.tv_sec = optval / hz;
2766 			tv.tv_usec = (optval % hz) * tick;
2767 #ifdef COMPAT_FREEBSD32
2768 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2769 				struct timeval32 tv32;
2770 
2771 				CP(tv, tv32, tv_sec);
2772 				CP(tv, tv32, tv_usec);
2773 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2774 			} else
2775 #endif
2776 				error = sooptcopyout(sopt, &tv, sizeof tv);
2777 			break;
2778 
2779 		case SO_LABEL:
2780 #ifdef MAC
2781 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2782 			    sizeof(extmac));
2783 			if (error)
2784 				goto bad;
2785 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2786 			    so, &extmac);
2787 			if (error)
2788 				goto bad;
2789 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2790 #else
2791 			error = EOPNOTSUPP;
2792 #endif
2793 			break;
2794 
2795 		case SO_PEERLABEL:
2796 #ifdef MAC
2797 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2798 			    sizeof(extmac));
2799 			if (error)
2800 				goto bad;
2801 			error = mac_getsockopt_peerlabel(
2802 			    sopt->sopt_td->td_ucred, so, &extmac);
2803 			if (error)
2804 				goto bad;
2805 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2806 #else
2807 			error = EOPNOTSUPP;
2808 #endif
2809 			break;
2810 
2811 		case SO_LISTENQLIMIT:
2812 			optval = so->so_qlimit;
2813 			goto integer;
2814 
2815 		case SO_LISTENQLEN:
2816 			optval = so->so_qlen;
2817 			goto integer;
2818 
2819 		case SO_LISTENINCQLEN:
2820 			optval = so->so_incqlen;
2821 			goto integer;
2822 
2823 		default:
2824 			error = ENOPROTOOPT;
2825 			break;
2826 		}
2827 	}
2828 #ifdef MAC
2829 bad:
2830 #endif
2831 	CURVNET_RESTORE();
2832 	return (error);
2833 }
2834 
2835 /* XXX; prepare mbuf for (__FreeBSD__ < 3) routines. */
2836 int
2837 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2838 {
2839 	struct mbuf *m, *m_prev;
2840 	int sopt_size = sopt->sopt_valsize;
2841 
2842 	MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2843 	if (m == NULL)
2844 		return ENOBUFS;
2845 	if (sopt_size > MLEN) {
2846 		MCLGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT);
2847 		if ((m->m_flags & M_EXT) == 0) {
2848 			m_free(m);
2849 			return ENOBUFS;
2850 		}
2851 		m->m_len = min(MCLBYTES, sopt_size);
2852 	} else {
2853 		m->m_len = min(MLEN, sopt_size);
2854 	}
2855 	sopt_size -= m->m_len;
2856 	*mp = m;
2857 	m_prev = m;
2858 
2859 	while (sopt_size) {
2860 		MGET(m, sopt->sopt_td ? M_WAIT : M_DONTWAIT, MT_DATA);
2861 		if (m == NULL) {
2862 			m_freem(*mp);
2863 			return ENOBUFS;
2864 		}
2865 		if (sopt_size > MLEN) {
2866 			MCLGET(m, sopt->sopt_td != NULL ? M_WAIT :
2867 			    M_DONTWAIT);
2868 			if ((m->m_flags & M_EXT) == 0) {
2869 				m_freem(m);
2870 				m_freem(*mp);
2871 				return ENOBUFS;
2872 			}
2873 			m->m_len = min(MCLBYTES, sopt_size);
2874 		} else {
2875 			m->m_len = min(MLEN, sopt_size);
2876 		}
2877 		sopt_size -= m->m_len;
2878 		m_prev->m_next = m;
2879 		m_prev = m;
2880 	}
2881 	return (0);
2882 }
2883 
2884 /* XXX; copyin sopt data into mbuf chain for (__FreeBSD__ < 3) routines. */
2885 int
2886 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2887 {
2888 	struct mbuf *m0 = m;
2889 
2890 	if (sopt->sopt_val == NULL)
2891 		return (0);
2892 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2893 		if (sopt->sopt_td != NULL) {
2894 			int error;
2895 
2896 			error = copyin(sopt->sopt_val, mtod(m, char *),
2897 				       m->m_len);
2898 			if (error != 0) {
2899 				m_freem(m0);
2900 				return(error);
2901 			}
2902 		} else
2903 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2904 		sopt->sopt_valsize -= m->m_len;
2905 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2906 		m = m->m_next;
2907 	}
2908 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2909 		panic("ip6_sooptmcopyin");
2910 	return (0);
2911 }
2912 
2913 /* XXX; copyout mbuf chain data into soopt for (__FreeBSD__ < 3) routines. */
2914 int
2915 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2916 {
2917 	struct mbuf *m0 = m;
2918 	size_t valsize = 0;
2919 
2920 	if (sopt->sopt_val == NULL)
2921 		return (0);
2922 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2923 		if (sopt->sopt_td != NULL) {
2924 			int error;
2925 
2926 			error = copyout(mtod(m, char *), sopt->sopt_val,
2927 				       m->m_len);
2928 			if (error != 0) {
2929 				m_freem(m0);
2930 				return(error);
2931 			}
2932 		} else
2933 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2934 	       sopt->sopt_valsize -= m->m_len;
2935 	       sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2936 	       valsize += m->m_len;
2937 	       m = m->m_next;
2938 	}
2939 	if (m != NULL) {
2940 		/* enough soopt buffer should be given from user-land */
2941 		m_freem(m0);
2942 		return(EINVAL);
2943 	}
2944 	sopt->sopt_valsize = valsize;
2945 	return (0);
2946 }
2947 
2948 /*
2949  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2950  * out-of-band data, which will then notify socket consumers.
2951  */
2952 void
2953 sohasoutofband(struct socket *so)
2954 {
2955 
2956 	if (so->so_sigio != NULL)
2957 		pgsigio(&so->so_sigio, SIGURG, 0);
2958 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2959 }
2960 
2961 int
2962 sopoll(struct socket *so, int events, struct ucred *active_cred,
2963     struct thread *td)
2964 {
2965 
2966 	/*
2967 	 * We do not need to set or assert curvnet as long as everyone uses
2968 	 * sopoll_generic().
2969 	 */
2970 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2971 	    td));
2972 }
2973 
2974 int
2975 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2976     struct thread *td)
2977 {
2978 	int revents = 0;
2979 
2980 	SOCKBUF_LOCK(&so->so_snd);
2981 	SOCKBUF_LOCK(&so->so_rcv);
2982 	if (events & (POLLIN | POLLRDNORM))
2983 		if (soreadabledata(so))
2984 			revents |= events & (POLLIN | POLLRDNORM);
2985 
2986 	if (events & (POLLOUT | POLLWRNORM))
2987 		if (sowriteable(so))
2988 			revents |= events & (POLLOUT | POLLWRNORM);
2989 
2990 	if (events & (POLLPRI | POLLRDBAND))
2991 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2992 			revents |= events & (POLLPRI | POLLRDBAND);
2993 
2994 	if ((events & POLLINIGNEOF) == 0) {
2995 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2996 			revents |= events & (POLLIN | POLLRDNORM);
2997 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2998 				revents |= POLLHUP;
2999 		}
3000 	}
3001 
3002 	if (revents == 0) {
3003 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3004 			selrecord(td, &so->so_rcv.sb_sel);
3005 			so->so_rcv.sb_flags |= SB_SEL;
3006 		}
3007 
3008 		if (events & (POLLOUT | POLLWRNORM)) {
3009 			selrecord(td, &so->so_snd.sb_sel);
3010 			so->so_snd.sb_flags |= SB_SEL;
3011 		}
3012 	}
3013 
3014 	SOCKBUF_UNLOCK(&so->so_rcv);
3015 	SOCKBUF_UNLOCK(&so->so_snd);
3016 	return (revents);
3017 }
3018 
3019 int
3020 soo_kqfilter(struct file *fp, struct knote *kn)
3021 {
3022 	struct socket *so = kn->kn_fp->f_data;
3023 	struct sockbuf *sb;
3024 
3025 	switch (kn->kn_filter) {
3026 	case EVFILT_READ:
3027 		if (so->so_options & SO_ACCEPTCONN)
3028 			kn->kn_fop = &solisten_filtops;
3029 		else
3030 			kn->kn_fop = &soread_filtops;
3031 		sb = &so->so_rcv;
3032 		break;
3033 	case EVFILT_WRITE:
3034 		kn->kn_fop = &sowrite_filtops;
3035 		sb = &so->so_snd;
3036 		break;
3037 	default:
3038 		return (EINVAL);
3039 	}
3040 
3041 	SOCKBUF_LOCK(sb);
3042 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3043 	sb->sb_flags |= SB_KNOTE;
3044 	SOCKBUF_UNLOCK(sb);
3045 	return (0);
3046 }
3047 
3048 /*
3049  * Some routines that return EOPNOTSUPP for entry points that are not
3050  * supported by a protocol.  Fill in as needed.
3051  */
3052 int
3053 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3054 {
3055 
3056 	return EOPNOTSUPP;
3057 }
3058 
3059 int
3060 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3061 {
3062 
3063 	return EOPNOTSUPP;
3064 }
3065 
3066 int
3067 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3068 {
3069 
3070 	return EOPNOTSUPP;
3071 }
3072 
3073 int
3074 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3075 {
3076 
3077 	return EOPNOTSUPP;
3078 }
3079 
3080 int
3081 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3082 {
3083 
3084 	return EOPNOTSUPP;
3085 }
3086 
3087 int
3088 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3089     struct ifnet *ifp, struct thread *td)
3090 {
3091 
3092 	return EOPNOTSUPP;
3093 }
3094 
3095 int
3096 pru_disconnect_notsupp(struct socket *so)
3097 {
3098 
3099 	return EOPNOTSUPP;
3100 }
3101 
3102 int
3103 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3104 {
3105 
3106 	return EOPNOTSUPP;
3107 }
3108 
3109 int
3110 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3111 {
3112 
3113 	return EOPNOTSUPP;
3114 }
3115 
3116 int
3117 pru_rcvd_notsupp(struct socket *so, int flags)
3118 {
3119 
3120 	return EOPNOTSUPP;
3121 }
3122 
3123 int
3124 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3125 {
3126 
3127 	return EOPNOTSUPP;
3128 }
3129 
3130 int
3131 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3132     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3133 {
3134 
3135 	return EOPNOTSUPP;
3136 }
3137 
3138 /*
3139  * This isn't really a ``null'' operation, but it's the default one and
3140  * doesn't do anything destructive.
3141  */
3142 int
3143 pru_sense_null(struct socket *so, struct stat *sb)
3144 {
3145 
3146 	sb->st_blksize = so->so_snd.sb_hiwat;
3147 	return 0;
3148 }
3149 
3150 int
3151 pru_shutdown_notsupp(struct socket *so)
3152 {
3153 
3154 	return EOPNOTSUPP;
3155 }
3156 
3157 int
3158 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3159 {
3160 
3161 	return EOPNOTSUPP;
3162 }
3163 
3164 int
3165 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3166     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3167 {
3168 
3169 	return EOPNOTSUPP;
3170 }
3171 
3172 int
3173 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3174     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3175 {
3176 
3177 	return EOPNOTSUPP;
3178 }
3179 
3180 int
3181 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3182     struct thread *td)
3183 {
3184 
3185 	return EOPNOTSUPP;
3186 }
3187 
3188 static void
3189 filt_sordetach(struct knote *kn)
3190 {
3191 	struct socket *so = kn->kn_fp->f_data;
3192 
3193 	SOCKBUF_LOCK(&so->so_rcv);
3194 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3195 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3196 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3197 	SOCKBUF_UNLOCK(&so->so_rcv);
3198 }
3199 
3200 /*ARGSUSED*/
3201 static int
3202 filt_soread(struct knote *kn, long hint)
3203 {
3204 	struct socket *so;
3205 
3206 	so = kn->kn_fp->f_data;
3207 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3208 
3209 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3210 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3211 		kn->kn_flags |= EV_EOF;
3212 		kn->kn_fflags = so->so_error;
3213 		return (1);
3214 	} else if (so->so_error)	/* temporary udp error */
3215 		return (1);
3216 	else if (kn->kn_sfflags & NOTE_LOWAT)
3217 		return (kn->kn_data >= kn->kn_sdata);
3218 	else
3219 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3220 }
3221 
3222 static void
3223 filt_sowdetach(struct knote *kn)
3224 {
3225 	struct socket *so = kn->kn_fp->f_data;
3226 
3227 	SOCKBUF_LOCK(&so->so_snd);
3228 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3229 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3230 		so->so_snd.sb_flags &= ~SB_KNOTE;
3231 	SOCKBUF_UNLOCK(&so->so_snd);
3232 }
3233 
3234 /*ARGSUSED*/
3235 static int
3236 filt_sowrite(struct knote *kn, long hint)
3237 {
3238 	struct socket *so;
3239 
3240 	so = kn->kn_fp->f_data;
3241 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3242 	kn->kn_data = sbspace(&so->so_snd);
3243 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3244 		kn->kn_flags |= EV_EOF;
3245 		kn->kn_fflags = so->so_error;
3246 		return (1);
3247 	} else if (so->so_error)	/* temporary udp error */
3248 		return (1);
3249 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3250 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3251 		return (0);
3252 	else if (kn->kn_sfflags & NOTE_LOWAT)
3253 		return (kn->kn_data >= kn->kn_sdata);
3254 	else
3255 		return (kn->kn_data >= so->so_snd.sb_lowat);
3256 }
3257 
3258 /*ARGSUSED*/
3259 static int
3260 filt_solisten(struct knote *kn, long hint)
3261 {
3262 	struct socket *so = kn->kn_fp->f_data;
3263 
3264 	kn->kn_data = so->so_qlen;
3265 	return (! TAILQ_EMPTY(&so->so_comp));
3266 }
3267 
3268 int
3269 socheckuid(struct socket *so, uid_t uid)
3270 {
3271 
3272 	if (so == NULL)
3273 		return (EPERM);
3274 	if (so->so_cred->cr_uid != uid)
3275 		return (EPERM);
3276 	return (0);
3277 }
3278 
3279 static int
3280 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
3281 {
3282 	int error;
3283 	int val;
3284 
3285 	val = somaxconn;
3286 	error = sysctl_handle_int(oidp, &val, 0, req);
3287 	if (error || !req->newptr )
3288 		return (error);
3289 
3290 	if (val < 1 || val > USHRT_MAX)
3291 		return (EINVAL);
3292 
3293 	somaxconn = val;
3294 	return (0);
3295 }
3296 
3297 /*
3298  * These functions are used by protocols to notify the socket layer (and its
3299  * consumers) of state changes in the sockets driven by protocol-side events.
3300  */
3301 
3302 /*
3303  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3304  *
3305  * Normal sequence from the active (originating) side is that
3306  * soisconnecting() is called during processing of connect() call, resulting
3307  * in an eventual call to soisconnected() if/when the connection is
3308  * established.  When the connection is torn down soisdisconnecting() is
3309  * called during processing of disconnect() call, and soisdisconnected() is
3310  * called when the connection to the peer is totally severed.  The semantics
3311  * of these routines are such that connectionless protocols can call
3312  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3313  * calls when setting up a ``connection'' takes no time.
3314  *
3315  * From the passive side, a socket is created with two queues of sockets:
3316  * so_incomp for connections in progress and so_comp for connections already
3317  * made and awaiting user acceptance.  As a protocol is preparing incoming
3318  * connections, it creates a socket structure queued on so_incomp by calling
3319  * sonewconn().  When the connection is established, soisconnected() is
3320  * called, and transfers the socket structure to so_comp, making it available
3321  * to accept().
3322  *
3323  * If a socket is closed with sockets on either so_incomp or so_comp, these
3324  * sockets are dropped.
3325  *
3326  * If higher-level protocols are implemented in the kernel, the wakeups done
3327  * here will sometimes cause software-interrupt process scheduling.
3328  */
3329 void
3330 soisconnecting(struct socket *so)
3331 {
3332 
3333 	SOCK_LOCK(so);
3334 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3335 	so->so_state |= SS_ISCONNECTING;
3336 	SOCK_UNLOCK(so);
3337 }
3338 
3339 void
3340 soisconnected(struct socket *so)
3341 {
3342 	struct socket *head;
3343 	int ret;
3344 
3345 restart:
3346 	ACCEPT_LOCK();
3347 	SOCK_LOCK(so);
3348 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3349 	so->so_state |= SS_ISCONNECTED;
3350 	head = so->so_head;
3351 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3352 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3353 			SOCK_UNLOCK(so);
3354 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3355 			head->so_incqlen--;
3356 			so->so_qstate &= ~SQ_INCOMP;
3357 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3358 			head->so_qlen++;
3359 			so->so_qstate |= SQ_COMP;
3360 			ACCEPT_UNLOCK();
3361 			sorwakeup(head);
3362 			wakeup_one(&head->so_timeo);
3363 		} else {
3364 			ACCEPT_UNLOCK();
3365 			soupcall_set(so, SO_RCV,
3366 			    head->so_accf->so_accept_filter->accf_callback,
3367 			    head->so_accf->so_accept_filter_arg);
3368 			so->so_options &= ~SO_ACCEPTFILTER;
3369 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3370 			    head->so_accf->so_accept_filter_arg, M_DONTWAIT);
3371 			if (ret == SU_ISCONNECTED)
3372 				soupcall_clear(so, SO_RCV);
3373 			SOCK_UNLOCK(so);
3374 			if (ret == SU_ISCONNECTED)
3375 				goto restart;
3376 		}
3377 		return;
3378 	}
3379 	SOCK_UNLOCK(so);
3380 	ACCEPT_UNLOCK();
3381 	wakeup(&so->so_timeo);
3382 	sorwakeup(so);
3383 	sowwakeup(so);
3384 }
3385 
3386 void
3387 soisdisconnecting(struct socket *so)
3388 {
3389 
3390 	/*
3391 	 * Note: This code assumes that SOCK_LOCK(so) and
3392 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3393 	 */
3394 	SOCKBUF_LOCK(&so->so_rcv);
3395 	so->so_state &= ~SS_ISCONNECTING;
3396 	so->so_state |= SS_ISDISCONNECTING;
3397 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3398 	sorwakeup_locked(so);
3399 	SOCKBUF_LOCK(&so->so_snd);
3400 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3401 	sowwakeup_locked(so);
3402 	wakeup(&so->so_timeo);
3403 }
3404 
3405 void
3406 soisdisconnected(struct socket *so)
3407 {
3408 
3409 	/*
3410 	 * Note: This code assumes that SOCK_LOCK(so) and
3411 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3412 	 */
3413 	SOCKBUF_LOCK(&so->so_rcv);
3414 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3415 	so->so_state |= SS_ISDISCONNECTED;
3416 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3417 	sorwakeup_locked(so);
3418 	SOCKBUF_LOCK(&so->so_snd);
3419 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3420 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3421 	sowwakeup_locked(so);
3422 	wakeup(&so->so_timeo);
3423 }
3424 
3425 /*
3426  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3427  */
3428 struct sockaddr *
3429 sodupsockaddr(const struct sockaddr *sa, int mflags)
3430 {
3431 	struct sockaddr *sa2;
3432 
3433 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3434 	if (sa2)
3435 		bcopy(sa, sa2, sa->sa_len);
3436 	return sa2;
3437 }
3438 
3439 /*
3440  * Register per-socket buffer upcalls.
3441  */
3442 void
3443 soupcall_set(struct socket *so, int which,
3444     int (*func)(struct socket *, void *, int), void *arg)
3445 {
3446 	struct sockbuf *sb;
3447 
3448 	switch (which) {
3449 	case SO_RCV:
3450 		sb = &so->so_rcv;
3451 		break;
3452 	case SO_SND:
3453 		sb = &so->so_snd;
3454 		break;
3455 	default:
3456 		panic("soupcall_set: bad which");
3457 	}
3458 	SOCKBUF_LOCK_ASSERT(sb);
3459 #if 0
3460 	/* XXX: accf_http actually wants to do this on purpose. */
3461 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3462 #endif
3463 	sb->sb_upcall = func;
3464 	sb->sb_upcallarg = arg;
3465 	sb->sb_flags |= SB_UPCALL;
3466 }
3467 
3468 void
3469 soupcall_clear(struct socket *so, int which)
3470 {
3471 	struct sockbuf *sb;
3472 
3473 	switch (which) {
3474 	case SO_RCV:
3475 		sb = &so->so_rcv;
3476 		break;
3477 	case SO_SND:
3478 		sb = &so->so_snd;
3479 		break;
3480 	default:
3481 		panic("soupcall_clear: bad which");
3482 	}
3483 	SOCKBUF_LOCK_ASSERT(sb);
3484 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3485 	sb->sb_upcall = NULL;
3486 	sb->sb_upcallarg = NULL;
3487 	sb->sb_flags &= ~SB_UPCALL;
3488 }
3489 
3490 /*
3491  * Create an external-format (``xsocket'') structure using the information in
3492  * the kernel-format socket structure pointed to by so.  This is done to
3493  * reduce the spew of irrelevant information over this interface, to isolate
3494  * user code from changes in the kernel structure, and potentially to provide
3495  * information-hiding if we decide that some of this information should be
3496  * hidden from users.
3497  */
3498 void
3499 sotoxsocket(struct socket *so, struct xsocket *xso)
3500 {
3501 
3502 	xso->xso_len = sizeof *xso;
3503 	xso->xso_so = so;
3504 	xso->so_type = so->so_type;
3505 	xso->so_options = so->so_options;
3506 	xso->so_linger = so->so_linger;
3507 	xso->so_state = so->so_state;
3508 	xso->so_pcb = so->so_pcb;
3509 	xso->xso_protocol = so->so_proto->pr_protocol;
3510 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3511 	xso->so_qlen = so->so_qlen;
3512 	xso->so_incqlen = so->so_incqlen;
3513 	xso->so_qlimit = so->so_qlimit;
3514 	xso->so_timeo = so->so_timeo;
3515 	xso->so_error = so->so_error;
3516 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3517 	xso->so_oobmark = so->so_oobmark;
3518 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3519 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3520 	xso->so_uid = so->so_cred->cr_uid;
3521 }
3522 
3523 
3524 /*
3525  * Socket accessor functions to provide external consumers with
3526  * a safe interface to socket state
3527  *
3528  */
3529 
3530 void
3531 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *), void *arg)
3532 {
3533 
3534 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3535 		func(so, arg);
3536 }
3537 
3538 struct sockbuf *
3539 so_sockbuf_rcv(struct socket *so)
3540 {
3541 
3542 	return (&so->so_rcv);
3543 }
3544 
3545 struct sockbuf *
3546 so_sockbuf_snd(struct socket *so)
3547 {
3548 
3549 	return (&so->so_snd);
3550 }
3551 
3552 int
3553 so_state_get(const struct socket *so)
3554 {
3555 
3556 	return (so->so_state);
3557 }
3558 
3559 void
3560 so_state_set(struct socket *so, int val)
3561 {
3562 
3563 	so->so_state = val;
3564 }
3565 
3566 int
3567 so_options_get(const struct socket *so)
3568 {
3569 
3570 	return (so->so_options);
3571 }
3572 
3573 void
3574 so_options_set(struct socket *so, int val)
3575 {
3576 
3577 	so->so_options = val;
3578 }
3579 
3580 int
3581 so_error_get(const struct socket *so)
3582 {
3583 
3584 	return (so->so_error);
3585 }
3586 
3587 void
3588 so_error_set(struct socket *so, int val)
3589 {
3590 
3591 	so->so_error = val;
3592 }
3593 
3594 int
3595 so_linger_get(const struct socket *so)
3596 {
3597 
3598 	return (so->so_linger);
3599 }
3600 
3601 void
3602 so_linger_set(struct socket *so, int val)
3603 {
3604 
3605 	so->so_linger = val;
3606 }
3607 
3608 struct protosw *
3609 so_protosw_get(const struct socket *so)
3610 {
3611 
3612 	return (so->so_proto);
3613 }
3614 
3615 void
3616 so_protosw_set(struct socket *so, struct protosw *val)
3617 {
3618 
3619 	so->so_proto = val;
3620 }
3621 
3622 void
3623 so_sorwakeup(struct socket *so)
3624 {
3625 
3626 	sorwakeup(so);
3627 }
3628 
3629 void
3630 so_sowwakeup(struct socket *so)
3631 {
3632 
3633 	sowwakeup(so);
3634 }
3635 
3636 void
3637 so_sorwakeup_locked(struct socket *so)
3638 {
3639 
3640 	sorwakeup_locked(so);
3641 }
3642 
3643 void
3644 so_sowwakeup_locked(struct socket *so)
3645 {
3646 
3647 	sowwakeup_locked(so);
3648 }
3649 
3650 void
3651 so_lock(struct socket *so)
3652 {
3653 	SOCK_LOCK(so);
3654 }
3655 
3656 void
3657 so_unlock(struct socket *so)
3658 {
3659 	SOCK_UNLOCK(so);
3660 }
3661