xref: /freebsd/sys/kern/uipc_socket.c (revision 4f6aec90ff8521301da9c6bc04310e1ab25a410c)
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_compat.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
115 #include <sys/mac.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h>			/* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
135 #include <sys/sx.h>
136 #include <sys/sysctl.h>
137 #include <sys/uio.h>
138 #include <sys/jail.h>
139 #include <sys/syslog.h>
140 #include <netinet/in.h>
141 
142 #include <net/vnet.h>
143 
144 #include <security/mac/mac_framework.h>
145 
146 #include <vm/uma.h>
147 
148 #ifdef COMPAT_FREEBSD32
149 #include <sys/mount.h>
150 #include <sys/sysent.h>
151 #include <compat/freebsd32/freebsd32.h>
152 #endif
153 
154 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
155 		    int flags);
156 
157 static void	filt_sordetach(struct knote *kn);
158 static int	filt_soread(struct knote *kn, long hint);
159 static void	filt_sowdetach(struct knote *kn);
160 static int	filt_sowrite(struct knote *kn, long hint);
161 static int	filt_solisten(struct knote *kn, long hint);
162 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
163 fo_kqfilter_t	soo_kqfilter;
164 
165 static struct filterops solisten_filtops = {
166 	.f_isfd = 1,
167 	.f_detach = filt_sordetach,
168 	.f_event = filt_solisten,
169 };
170 static struct filterops soread_filtops = {
171 	.f_isfd = 1,
172 	.f_detach = filt_sordetach,
173 	.f_event = filt_soread,
174 };
175 static struct filterops sowrite_filtops = {
176 	.f_isfd = 1,
177 	.f_detach = filt_sowdetach,
178 	.f_event = filt_sowrite,
179 };
180 
181 so_gen_t	so_gencnt;	/* generation count for sockets */
182 
183 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
184 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
185 
186 #define	VNET_SO_ASSERT(so)						\
187 	VNET_ASSERT(curvnet != NULL,					\
188 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
189 
190 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
191 #define	V_socket_hhh		VNET(socket_hhh)
192 
193 /*
194  * Limit on the number of connections in the listen queue waiting
195  * for accept(2).
196  * NB: The orginal sysctl somaxconn is still available but hidden
197  * to prevent confusion about the actual purpose of this number.
198  */
199 static int somaxconn = SOMAXCONN;
200 
201 static int
202 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
203 {
204 	int error;
205 	int val;
206 
207 	val = somaxconn;
208 	error = sysctl_handle_int(oidp, &val, 0, req);
209 	if (error || !req->newptr )
210 		return (error);
211 
212 	if (val < 1 || val > USHRT_MAX)
213 		return (EINVAL);
214 
215 	somaxconn = val;
216 	return (0);
217 }
218 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
219     0, sizeof(int), sysctl_somaxconn, "I",
220     "Maximum listen socket pending connection accept queue size");
221 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
222     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
223     0, sizeof(int), sysctl_somaxconn, "I",
224     "Maximum listen socket pending connection accept queue size (compat)");
225 
226 static int numopensockets;
227 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
228     &numopensockets, 0, "Number of open sockets");
229 
230 /*
231  * accept_mtx locks down per-socket fields relating to accept queues.  See
232  * socketvar.h for an annotation of the protected fields of struct socket.
233  */
234 struct mtx accept_mtx;
235 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
236 
237 /*
238  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
239  * so_gencnt field.
240  */
241 static struct mtx so_global_mtx;
242 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
243 
244 /*
245  * General IPC sysctl name space, used by sockets and a variety of other IPC
246  * types.
247  */
248 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
249 
250 /*
251  * Initialize the socket subsystem and set up the socket
252  * memory allocator.
253  */
254 static uma_zone_t socket_zone;
255 int	maxsockets;
256 
257 static void
258 socket_zone_change(void *tag)
259 {
260 
261 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
262 }
263 
264 static void
265 socket_hhook_register(int subtype)
266 {
267 
268 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
269 	    &V_socket_hhh[subtype],
270 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
271 		printf("%s: WARNING: unable to register hook\n", __func__);
272 }
273 
274 static void
275 socket_hhook_deregister(int subtype)
276 {
277 
278 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
279 		printf("%s: WARNING: unable to deregister hook\n", __func__);
280 }
281 
282 static void
283 socket_init(void *tag)
284 {
285 
286 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
287 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
288 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
289 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
290 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
291 	    EVENTHANDLER_PRI_FIRST);
292 }
293 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
294 
295 static void
296 socket_vnet_init(const void *unused __unused)
297 {
298 	int i;
299 
300 	/* We expect a contiguous range */
301 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
302 		socket_hhook_register(i);
303 }
304 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
305     socket_vnet_init, NULL);
306 
307 static void
308 socket_vnet_uninit(const void *unused __unused)
309 {
310 	int i;
311 
312 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
313 		socket_hhook_deregister(i);
314 }
315 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
316     socket_vnet_uninit, NULL);
317 
318 /*
319  * Initialise maxsockets.  This SYSINIT must be run after
320  * tunable_mbinit().
321  */
322 static void
323 init_maxsockets(void *ignored)
324 {
325 
326 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
327 	maxsockets = imax(maxsockets, maxfiles);
328 }
329 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
330 
331 /*
332  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
333  * of the change so that they can update their dependent limits as required.
334  */
335 static int
336 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
337 {
338 	int error, newmaxsockets;
339 
340 	newmaxsockets = maxsockets;
341 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
342 	if (error == 0 && req->newptr) {
343 		if (newmaxsockets > maxsockets &&
344 		    newmaxsockets <= maxfiles) {
345 			maxsockets = newmaxsockets;
346 			EVENTHANDLER_INVOKE(maxsockets_change);
347 		} else
348 			error = EINVAL;
349 	}
350 	return (error);
351 }
352 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
353     &maxsockets, 0, sysctl_maxsockets, "IU",
354     "Maximum number of sockets avaliable");
355 
356 /*
357  * Socket operation routines.  These routines are called by the routines in
358  * sys_socket.c or from a system process, and implement the semantics of
359  * socket operations by switching out to the protocol specific routines.
360  */
361 
362 /*
363  * Get a socket structure from our zone, and initialize it.  Note that it
364  * would probably be better to allocate socket and PCB at the same time, but
365  * I'm not convinced that all the protocols can be easily modified to do
366  * this.
367  *
368  * soalloc() returns a socket with a ref count of 0.
369  */
370 static struct socket *
371 soalloc(struct vnet *vnet)
372 {
373 	struct socket *so;
374 
375 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
376 	if (so == NULL)
377 		return (NULL);
378 #ifdef MAC
379 	if (mac_socket_init(so, M_NOWAIT) != 0) {
380 		uma_zfree(socket_zone, so);
381 		return (NULL);
382 	}
383 #endif
384 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
385 		uma_zfree(socket_zone, so);
386 		return (NULL);
387 	}
388 
389 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
390 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
391 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
392 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
393 	TAILQ_INIT(&so->so_aiojobq);
394 	mtx_lock(&so_global_mtx);
395 	so->so_gencnt = ++so_gencnt;
396 	++numopensockets;
397 #ifdef VIMAGE
398 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
399 	    __func__, __LINE__, so));
400 	vnet->vnet_sockcnt++;
401 	so->so_vnet = vnet;
402 #endif
403 	mtx_unlock(&so_global_mtx);
404 
405 	CURVNET_SET(vnet);
406 	/* We shouldn't need the so_global_mtx */
407 	if (V_socket_hhh[HHOOK_SOCKET_CREATE]->hhh_nhooks > 0) {
408 		if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE))
409 			/* Do we need more comprehensive error returns? */
410 			so = NULL;
411 	}
412 	CURVNET_RESTORE();
413 
414 	return (so);
415 }
416 
417 /*
418  * Free the storage associated with a socket at the socket layer, tear down
419  * locks, labels, etc.  All protocol state is assumed already to have been
420  * torn down (and possibly never set up) by the caller.
421  */
422 static void
423 sodealloc(struct socket *so)
424 {
425 
426 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
427 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
428 
429 	mtx_lock(&so_global_mtx);
430 	so->so_gencnt = ++so_gencnt;
431 	--numopensockets;	/* Could be below, but faster here. */
432 #ifdef VIMAGE
433 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
434 	    __func__, __LINE__, so));
435 	so->so_vnet->vnet_sockcnt--;
436 #endif
437 	mtx_unlock(&so_global_mtx);
438 	if (so->so_rcv.sb_hiwat)
439 		(void)chgsbsize(so->so_cred->cr_uidinfo,
440 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
441 	if (so->so_snd.sb_hiwat)
442 		(void)chgsbsize(so->so_cred->cr_uidinfo,
443 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
444 	/* remove acccept filter if one is present. */
445 	if (so->so_accf != NULL)
446 		do_setopt_accept_filter(so, NULL);
447 #ifdef MAC
448 	mac_socket_destroy(so);
449 #endif
450 	CURVNET_SET(so->so_vnet);
451 	if (V_socket_hhh[HHOOK_SOCKET_CLOSE]->hhh_nhooks > 0)
452 		hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
453 	CURVNET_RESTORE();
454 
455 	crfree(so->so_cred);
456 	khelp_destroy_osd(&so->osd);
457 	sx_destroy(&so->so_snd.sb_sx);
458 	sx_destroy(&so->so_rcv.sb_sx);
459 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
460 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
461 	uma_zfree(socket_zone, so);
462 }
463 
464 /*
465  * socreate returns a socket with a ref count of 1.  The socket should be
466  * closed with soclose().
467  */
468 int
469 socreate(int dom, struct socket **aso, int type, int proto,
470     struct ucred *cred, struct thread *td)
471 {
472 	struct protosw *prp;
473 	struct socket *so;
474 	int error;
475 
476 	if (proto)
477 		prp = pffindproto(dom, proto, type);
478 	else
479 		prp = pffindtype(dom, type);
480 
481 	if (prp == NULL) {
482 		/* No support for domain. */
483 		if (pffinddomain(dom) == NULL)
484 			return (EAFNOSUPPORT);
485 		/* No support for socket type. */
486 		if (proto == 0 && type != 0)
487 			return (EPROTOTYPE);
488 		return (EPROTONOSUPPORT);
489 	}
490 	if (prp->pr_usrreqs->pru_attach == NULL ||
491 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
492 		return (EPROTONOSUPPORT);
493 
494 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
495 		return (EPROTONOSUPPORT);
496 
497 	if (prp->pr_type != type)
498 		return (EPROTOTYPE);
499 	so = soalloc(CRED_TO_VNET(cred));
500 	if (so == NULL)
501 		return (ENOBUFS);
502 
503 	TAILQ_INIT(&so->so_incomp);
504 	TAILQ_INIT(&so->so_comp);
505 	so->so_type = type;
506 	so->so_cred = crhold(cred);
507 	if ((prp->pr_domain->dom_family == PF_INET) ||
508 	    (prp->pr_domain->dom_family == PF_INET6) ||
509 	    (prp->pr_domain->dom_family == PF_ROUTE))
510 		so->so_fibnum = td->td_proc->p_fibnum;
511 	else
512 		so->so_fibnum = 0;
513 	so->so_proto = prp;
514 #ifdef MAC
515 	mac_socket_create(cred, so);
516 #endif
517 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
518 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
519 	so->so_count = 1;
520 	/*
521 	 * Auto-sizing of socket buffers is managed by the protocols and
522 	 * the appropriate flags must be set in the pru_attach function.
523 	 */
524 	CURVNET_SET(so->so_vnet);
525 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
526 	CURVNET_RESTORE();
527 	if (error) {
528 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
529 		    so->so_count));
530 		so->so_count = 0;
531 		sodealloc(so);
532 		return (error);
533 	}
534 	*aso = so;
535 	return (0);
536 }
537 
538 #ifdef REGRESSION
539 static int regression_sonewconn_earlytest = 1;
540 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
541     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
542 #endif
543 
544 /*
545  * When an attempt at a new connection is noted on a socket which accepts
546  * connections, sonewconn is called.  If the connection is possible (subject
547  * to space constraints, etc.) then we allocate a new structure, propoerly
548  * linked into the data structure of the original socket, and return this.
549  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
550  *
551  * Note: the ref count on the socket is 0 on return.
552  */
553 struct socket *
554 sonewconn(struct socket *head, int connstatus)
555 {
556 	static struct timeval lastover;
557 	static struct timeval overinterval = { 60, 0 };
558 	static int overcount;
559 
560 	struct socket *so;
561 	int over;
562 
563 	ACCEPT_LOCK();
564 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
565 	ACCEPT_UNLOCK();
566 #ifdef REGRESSION
567 	if (regression_sonewconn_earlytest && over) {
568 #else
569 	if (over) {
570 #endif
571 		overcount++;
572 
573 		if (ratecheck(&lastover, &overinterval)) {
574 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
575 			    "%i already in queue awaiting acceptance "
576 			    "(%d occurrences)\n",
577 			    __func__, head->so_pcb, head->so_qlen, overcount);
578 
579 			overcount = 0;
580 		}
581 
582 		return (NULL);
583 	}
584 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
585 	    __func__, __LINE__, head));
586 	so = soalloc(head->so_vnet);
587 	if (so == NULL) {
588 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
589 		    "limit reached or out of memory\n",
590 		    __func__, head->so_pcb);
591 		return (NULL);
592 	}
593 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
594 		connstatus = 0;
595 	so->so_head = head;
596 	so->so_type = head->so_type;
597 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
598 	so->so_linger = head->so_linger;
599 	so->so_state = head->so_state | SS_NOFDREF;
600 	so->so_fibnum = head->so_fibnum;
601 	so->so_proto = head->so_proto;
602 	so->so_cred = crhold(head->so_cred);
603 #ifdef MAC
604 	mac_socket_newconn(head, so);
605 #endif
606 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
607 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
608 	VNET_SO_ASSERT(head);
609 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
610 		sodealloc(so);
611 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
612 		    __func__, head->so_pcb);
613 		return (NULL);
614 	}
615 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
616 		sodealloc(so);
617 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
618 		    __func__, head->so_pcb);
619 		return (NULL);
620 	}
621 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
622 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
623 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
624 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
625 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
626 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
627 	so->so_state |= connstatus;
628 	ACCEPT_LOCK();
629 	/*
630 	 * The accept socket may be tearing down but we just
631 	 * won a race on the ACCEPT_LOCK.
632 	 * However, if sctp_peeloff() is called on a 1-to-many
633 	 * style socket, the SO_ACCEPTCONN doesn't need to be set.
634 	 */
635 	if (!(head->so_options & SO_ACCEPTCONN) &&
636 	    ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
637 	     (head->so_type != SOCK_SEQPACKET))) {
638 		SOCK_LOCK(so);
639 		so->so_head = NULL;
640 		sofree(so);		/* NB: returns ACCEPT_UNLOCK'ed. */
641 		return (NULL);
642 	}
643 	if (connstatus) {
644 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
645 		so->so_qstate |= SQ_COMP;
646 		head->so_qlen++;
647 	} else {
648 		/*
649 		 * Keep removing sockets from the head until there's room for
650 		 * us to insert on the tail.  In pre-locking revisions, this
651 		 * was a simple if(), but as we could be racing with other
652 		 * threads and soabort() requires dropping locks, we must
653 		 * loop waiting for the condition to be true.
654 		 */
655 		while (head->so_incqlen > head->so_qlimit) {
656 			struct socket *sp;
657 			sp = TAILQ_FIRST(&head->so_incomp);
658 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
659 			head->so_incqlen--;
660 			sp->so_qstate &= ~SQ_INCOMP;
661 			sp->so_head = NULL;
662 			ACCEPT_UNLOCK();
663 			soabort(sp);
664 			ACCEPT_LOCK();
665 		}
666 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
667 		so->so_qstate |= SQ_INCOMP;
668 		head->so_incqlen++;
669 	}
670 	ACCEPT_UNLOCK();
671 	if (connstatus) {
672 		sorwakeup(head);
673 		wakeup_one(&head->so_timeo);
674 	}
675 	return (so);
676 }
677 
678 int
679 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
680 {
681 	int error;
682 
683 	CURVNET_SET(so->so_vnet);
684 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
685 	CURVNET_RESTORE();
686 	return (error);
687 }
688 
689 int
690 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
691 {
692 	int error;
693 
694 	CURVNET_SET(so->so_vnet);
695 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
696 	CURVNET_RESTORE();
697 	return (error);
698 }
699 
700 /*
701  * solisten() transitions a socket from a non-listening state to a listening
702  * state, but can also be used to update the listen queue depth on an
703  * existing listen socket.  The protocol will call back into the sockets
704  * layer using solisten_proto_check() and solisten_proto() to check and set
705  * socket-layer listen state.  Call backs are used so that the protocol can
706  * acquire both protocol and socket layer locks in whatever order is required
707  * by the protocol.
708  *
709  * Protocol implementors are advised to hold the socket lock across the
710  * socket-layer test and set to avoid races at the socket layer.
711  */
712 int
713 solisten(struct socket *so, int backlog, struct thread *td)
714 {
715 	int error;
716 
717 	CURVNET_SET(so->so_vnet);
718 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
719 	CURVNET_RESTORE();
720 	return (error);
721 }
722 
723 int
724 solisten_proto_check(struct socket *so)
725 {
726 
727 	SOCK_LOCK_ASSERT(so);
728 
729 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
730 	    SS_ISDISCONNECTING))
731 		return (EINVAL);
732 	return (0);
733 }
734 
735 void
736 solisten_proto(struct socket *so, int backlog)
737 {
738 
739 	SOCK_LOCK_ASSERT(so);
740 
741 	if (backlog < 0 || backlog > somaxconn)
742 		backlog = somaxconn;
743 	so->so_qlimit = backlog;
744 	so->so_options |= SO_ACCEPTCONN;
745 }
746 
747 /*
748  * Evaluate the reference count and named references on a socket; if no
749  * references remain, free it.  This should be called whenever a reference is
750  * released, such as in sorele(), but also when named reference flags are
751  * cleared in socket or protocol code.
752  *
753  * sofree() will free the socket if:
754  *
755  * - There are no outstanding file descriptor references or related consumers
756  *   (so_count == 0).
757  *
758  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
759  *
760  * - The protocol does not have an outstanding strong reference on the socket
761  *   (SS_PROTOREF).
762  *
763  * - The socket is not in a completed connection queue, so a process has been
764  *   notified that it is present.  If it is removed, the user process may
765  *   block in accept() despite select() saying the socket was ready.
766  */
767 void
768 sofree(struct socket *so)
769 {
770 	struct protosw *pr = so->so_proto;
771 	struct socket *head;
772 
773 	ACCEPT_LOCK_ASSERT();
774 	SOCK_LOCK_ASSERT(so);
775 
776 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
777 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
778 		SOCK_UNLOCK(so);
779 		ACCEPT_UNLOCK();
780 		return;
781 	}
782 
783 	head = so->so_head;
784 	if (head != NULL) {
785 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
786 		    (so->so_qstate & SQ_INCOMP) != 0,
787 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
788 		    "SQ_INCOMP"));
789 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
790 		    (so->so_qstate & SQ_INCOMP) == 0,
791 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
792 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
793 		head->so_incqlen--;
794 		so->so_qstate &= ~SQ_INCOMP;
795 		so->so_head = NULL;
796 	}
797 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
798 	    (so->so_qstate & SQ_INCOMP) == 0,
799 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
800 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
801 	if (so->so_options & SO_ACCEPTCONN) {
802 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
803 		    ("sofree: so_comp populated"));
804 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
805 		    ("sofree: so_incomp populated"));
806 	}
807 	SOCK_UNLOCK(so);
808 	ACCEPT_UNLOCK();
809 
810 	VNET_SO_ASSERT(so);
811 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
812 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
813 	if (pr->pr_usrreqs->pru_detach != NULL)
814 		(*pr->pr_usrreqs->pru_detach)(so);
815 
816 	/*
817 	 * From this point on, we assume that no other references to this
818 	 * socket exist anywhere else in the stack.  Therefore, no locks need
819 	 * to be acquired or held.
820 	 *
821 	 * We used to do a lot of socket buffer and socket locking here, as
822 	 * well as invoke sorflush() and perform wakeups.  The direct call to
823 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
824 	 * necessary from sorflush().
825 	 *
826 	 * Notice that the socket buffer and kqueue state are torn down
827 	 * before calling pru_detach.  This means that protocols shold not
828 	 * assume they can perform socket wakeups, etc, in their detach code.
829 	 */
830 	sbdestroy(&so->so_snd, so);
831 	sbdestroy(&so->so_rcv, so);
832 	seldrain(&so->so_snd.sb_sel);
833 	seldrain(&so->so_rcv.sb_sel);
834 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
835 	knlist_destroy(&so->so_snd.sb_sel.si_note);
836 	sodealloc(so);
837 }
838 
839 /*
840  * Close a socket on last file table reference removal.  Initiate disconnect
841  * if connected.  Free socket when disconnect complete.
842  *
843  * This function will sorele() the socket.  Note that soclose() may be called
844  * prior to the ref count reaching zero.  The actual socket structure will
845  * not be freed until the ref count reaches zero.
846  */
847 int
848 soclose(struct socket *so)
849 {
850 	int error = 0;
851 
852 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
853 
854 	CURVNET_SET(so->so_vnet);
855 	funsetown(&so->so_sigio);
856 	if (so->so_state & SS_ISCONNECTED) {
857 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
858 			error = sodisconnect(so);
859 			if (error) {
860 				if (error == ENOTCONN)
861 					error = 0;
862 				goto drop;
863 			}
864 		}
865 		if (so->so_options & SO_LINGER) {
866 			if ((so->so_state & SS_ISDISCONNECTING) &&
867 			    (so->so_state & SS_NBIO))
868 				goto drop;
869 			while (so->so_state & SS_ISCONNECTED) {
870 				error = tsleep(&so->so_timeo,
871 				    PSOCK | PCATCH, "soclos",
872 				    so->so_linger * hz);
873 				if (error)
874 					break;
875 			}
876 		}
877 	}
878 
879 drop:
880 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
881 		(*so->so_proto->pr_usrreqs->pru_close)(so);
882 	ACCEPT_LOCK();
883 	if (so->so_options & SO_ACCEPTCONN) {
884 		struct socket *sp;
885 		/*
886 		 * Prevent new additions to the accept queues due
887 		 * to ACCEPT_LOCK races while we are draining them.
888 		 */
889 		so->so_options &= ~SO_ACCEPTCONN;
890 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
891 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
892 			so->so_incqlen--;
893 			sp->so_qstate &= ~SQ_INCOMP;
894 			sp->so_head = NULL;
895 			ACCEPT_UNLOCK();
896 			soabort(sp);
897 			ACCEPT_LOCK();
898 		}
899 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
900 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
901 			so->so_qlen--;
902 			sp->so_qstate &= ~SQ_COMP;
903 			sp->so_head = NULL;
904 			ACCEPT_UNLOCK();
905 			soabort(sp);
906 			ACCEPT_LOCK();
907 		}
908 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
909 		    ("%s: so_comp populated", __func__));
910 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
911 		    ("%s: so_incomp populated", __func__));
912 	}
913 	SOCK_LOCK(so);
914 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
915 	so->so_state |= SS_NOFDREF;
916 	sorele(so);			/* NB: Returns with ACCEPT_UNLOCK(). */
917 	CURVNET_RESTORE();
918 	return (error);
919 }
920 
921 /*
922  * soabort() is used to abruptly tear down a connection, such as when a
923  * resource limit is reached (listen queue depth exceeded), or if a listen
924  * socket is closed while there are sockets waiting to be accepted.
925  *
926  * This interface is tricky, because it is called on an unreferenced socket,
927  * and must be called only by a thread that has actually removed the socket
928  * from the listen queue it was on, or races with other threads are risked.
929  *
930  * This interface will call into the protocol code, so must not be called
931  * with any socket locks held.  Protocols do call it while holding their own
932  * recursible protocol mutexes, but this is something that should be subject
933  * to review in the future.
934  */
935 void
936 soabort(struct socket *so)
937 {
938 
939 	/*
940 	 * In as much as is possible, assert that no references to this
941 	 * socket are held.  This is not quite the same as asserting that the
942 	 * current thread is responsible for arranging for no references, but
943 	 * is as close as we can get for now.
944 	 */
945 	KASSERT(so->so_count == 0, ("soabort: so_count"));
946 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
947 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
948 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
949 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
950 	VNET_SO_ASSERT(so);
951 
952 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
953 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
954 	ACCEPT_LOCK();
955 	SOCK_LOCK(so);
956 	sofree(so);
957 }
958 
959 int
960 soaccept(struct socket *so, struct sockaddr **nam)
961 {
962 	int error;
963 
964 	SOCK_LOCK(so);
965 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
966 	so->so_state &= ~SS_NOFDREF;
967 	SOCK_UNLOCK(so);
968 
969 	CURVNET_SET(so->so_vnet);
970 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
971 	CURVNET_RESTORE();
972 	return (error);
973 }
974 
975 int
976 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
977 {
978 
979 	return (soconnectat(AT_FDCWD, so, nam, td));
980 }
981 
982 int
983 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
984 {
985 	int error;
986 
987 	if (so->so_options & SO_ACCEPTCONN)
988 		return (EOPNOTSUPP);
989 
990 	CURVNET_SET(so->so_vnet);
991 	/*
992 	 * If protocol is connection-based, can only connect once.
993 	 * Otherwise, if connected, try to disconnect first.  This allows
994 	 * user to disconnect by connecting to, e.g., a null address.
995 	 */
996 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
997 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
998 	    (error = sodisconnect(so)))) {
999 		error = EISCONN;
1000 	} else {
1001 		/*
1002 		 * Prevent accumulated error from previous connection from
1003 		 * biting us.
1004 		 */
1005 		so->so_error = 0;
1006 		if (fd == AT_FDCWD) {
1007 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1008 			    nam, td);
1009 		} else {
1010 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1011 			    so, nam, td);
1012 		}
1013 	}
1014 	CURVNET_RESTORE();
1015 
1016 	return (error);
1017 }
1018 
1019 int
1020 soconnect2(struct socket *so1, struct socket *so2)
1021 {
1022 	int error;
1023 
1024 	CURVNET_SET(so1->so_vnet);
1025 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1026 	CURVNET_RESTORE();
1027 	return (error);
1028 }
1029 
1030 int
1031 sodisconnect(struct socket *so)
1032 {
1033 	int error;
1034 
1035 	if ((so->so_state & SS_ISCONNECTED) == 0)
1036 		return (ENOTCONN);
1037 	if (so->so_state & SS_ISDISCONNECTING)
1038 		return (EALREADY);
1039 	VNET_SO_ASSERT(so);
1040 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1041 	return (error);
1042 }
1043 
1044 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1045 
1046 int
1047 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1048     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1049 {
1050 	long space;
1051 	ssize_t resid;
1052 	int clen = 0, error, dontroute;
1053 
1054 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1055 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1056 	    ("sosend_dgram: !PR_ATOMIC"));
1057 
1058 	if (uio != NULL)
1059 		resid = uio->uio_resid;
1060 	else
1061 		resid = top->m_pkthdr.len;
1062 	/*
1063 	 * In theory resid should be unsigned.  However, space must be
1064 	 * signed, as it might be less than 0 if we over-committed, and we
1065 	 * must use a signed comparison of space and resid.  On the other
1066 	 * hand, a negative resid causes us to loop sending 0-length
1067 	 * segments to the protocol.
1068 	 */
1069 	if (resid < 0) {
1070 		error = EINVAL;
1071 		goto out;
1072 	}
1073 
1074 	dontroute =
1075 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1076 	if (td != NULL)
1077 		td->td_ru.ru_msgsnd++;
1078 	if (control != NULL)
1079 		clen = control->m_len;
1080 
1081 	SOCKBUF_LOCK(&so->so_snd);
1082 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1083 		SOCKBUF_UNLOCK(&so->so_snd);
1084 		error = EPIPE;
1085 		goto out;
1086 	}
1087 	if (so->so_error) {
1088 		error = so->so_error;
1089 		so->so_error = 0;
1090 		SOCKBUF_UNLOCK(&so->so_snd);
1091 		goto out;
1092 	}
1093 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1094 		/*
1095 		 * `sendto' and `sendmsg' is allowed on a connection-based
1096 		 * socket if it supports implied connect.  Return ENOTCONN if
1097 		 * not connected and no address is supplied.
1098 		 */
1099 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1100 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1101 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1102 			    !(resid == 0 && clen != 0)) {
1103 				SOCKBUF_UNLOCK(&so->so_snd);
1104 				error = ENOTCONN;
1105 				goto out;
1106 			}
1107 		} else if (addr == NULL) {
1108 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1109 				error = ENOTCONN;
1110 			else
1111 				error = EDESTADDRREQ;
1112 			SOCKBUF_UNLOCK(&so->so_snd);
1113 			goto out;
1114 		}
1115 	}
1116 
1117 	/*
1118 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1119 	 * problem and need fixing.
1120 	 */
1121 	space = sbspace(&so->so_snd);
1122 	if (flags & MSG_OOB)
1123 		space += 1024;
1124 	space -= clen;
1125 	SOCKBUF_UNLOCK(&so->so_snd);
1126 	if (resid > space) {
1127 		error = EMSGSIZE;
1128 		goto out;
1129 	}
1130 	if (uio == NULL) {
1131 		resid = 0;
1132 		if (flags & MSG_EOR)
1133 			top->m_flags |= M_EOR;
1134 	} else {
1135 		/*
1136 		 * Copy the data from userland into a mbuf chain.
1137 		 * If no data is to be copied in, a single empty mbuf
1138 		 * is returned.
1139 		 */
1140 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1141 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1142 		if (top == NULL) {
1143 			error = EFAULT;	/* only possible error */
1144 			goto out;
1145 		}
1146 		space -= resid - uio->uio_resid;
1147 		resid = uio->uio_resid;
1148 	}
1149 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1150 	/*
1151 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1152 	 * than with.
1153 	 */
1154 	if (dontroute) {
1155 		SOCK_LOCK(so);
1156 		so->so_options |= SO_DONTROUTE;
1157 		SOCK_UNLOCK(so);
1158 	}
1159 	/*
1160 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1161 	 * of date.  We could have recieved a reset packet in an interrupt or
1162 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1163 	 * probably recheck again inside the locking protection here, but
1164 	 * there are probably other places that this also happens.  We must
1165 	 * rethink this.
1166 	 */
1167 	VNET_SO_ASSERT(so);
1168 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1169 	    (flags & MSG_OOB) ? PRUS_OOB :
1170 	/*
1171 	 * If the user set MSG_EOF, the protocol understands this flag and
1172 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1173 	 */
1174 	    ((flags & MSG_EOF) &&
1175 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1176 	     (resid <= 0)) ?
1177 		PRUS_EOF :
1178 		/* If there is more to send set PRUS_MORETOCOME */
1179 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1180 		top, addr, control, td);
1181 	if (dontroute) {
1182 		SOCK_LOCK(so);
1183 		so->so_options &= ~SO_DONTROUTE;
1184 		SOCK_UNLOCK(so);
1185 	}
1186 	clen = 0;
1187 	control = NULL;
1188 	top = NULL;
1189 out:
1190 	if (top != NULL)
1191 		m_freem(top);
1192 	if (control != NULL)
1193 		m_freem(control);
1194 	return (error);
1195 }
1196 
1197 /*
1198  * Send on a socket.  If send must go all at once and message is larger than
1199  * send buffering, then hard error.  Lock against other senders.  If must go
1200  * all at once and not enough room now, then inform user that this would
1201  * block and do nothing.  Otherwise, if nonblocking, send as much as
1202  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1203  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1204  * in mbuf chain must be small enough to send all at once.
1205  *
1206  * Returns nonzero on error, timeout or signal; callers must check for short
1207  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1208  * on return.
1209  */
1210 int
1211 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1212     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1213 {
1214 	long space;
1215 	ssize_t resid;
1216 	int clen = 0, error, dontroute;
1217 	int atomic = sosendallatonce(so) || top;
1218 
1219 	if (uio != NULL)
1220 		resid = uio->uio_resid;
1221 	else
1222 		resid = top->m_pkthdr.len;
1223 	/*
1224 	 * In theory resid should be unsigned.  However, space must be
1225 	 * signed, as it might be less than 0 if we over-committed, and we
1226 	 * must use a signed comparison of space and resid.  On the other
1227 	 * hand, a negative resid causes us to loop sending 0-length
1228 	 * segments to the protocol.
1229 	 *
1230 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1231 	 * type sockets since that's an error.
1232 	 */
1233 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1234 		error = EINVAL;
1235 		goto out;
1236 	}
1237 
1238 	dontroute =
1239 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1240 	    (so->so_proto->pr_flags & PR_ATOMIC);
1241 	if (td != NULL)
1242 		td->td_ru.ru_msgsnd++;
1243 	if (control != NULL)
1244 		clen = control->m_len;
1245 
1246 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1247 	if (error)
1248 		goto out;
1249 
1250 restart:
1251 	do {
1252 		SOCKBUF_LOCK(&so->so_snd);
1253 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1254 			SOCKBUF_UNLOCK(&so->so_snd);
1255 			error = EPIPE;
1256 			goto release;
1257 		}
1258 		if (so->so_error) {
1259 			error = so->so_error;
1260 			so->so_error = 0;
1261 			SOCKBUF_UNLOCK(&so->so_snd);
1262 			goto release;
1263 		}
1264 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1265 			/*
1266 			 * `sendto' and `sendmsg' is allowed on a connection-
1267 			 * based socket if it supports implied connect.
1268 			 * Return ENOTCONN if not connected and no address is
1269 			 * supplied.
1270 			 */
1271 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1272 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1273 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1274 				    !(resid == 0 && clen != 0)) {
1275 					SOCKBUF_UNLOCK(&so->so_snd);
1276 					error = ENOTCONN;
1277 					goto release;
1278 				}
1279 			} else if (addr == NULL) {
1280 				SOCKBUF_UNLOCK(&so->so_snd);
1281 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1282 					error = ENOTCONN;
1283 				else
1284 					error = EDESTADDRREQ;
1285 				goto release;
1286 			}
1287 		}
1288 		space = sbspace(&so->so_snd);
1289 		if (flags & MSG_OOB)
1290 			space += 1024;
1291 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1292 		    clen > so->so_snd.sb_hiwat) {
1293 			SOCKBUF_UNLOCK(&so->so_snd);
1294 			error = EMSGSIZE;
1295 			goto release;
1296 		}
1297 		if (space < resid + clen &&
1298 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1299 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1300 				SOCKBUF_UNLOCK(&so->so_snd);
1301 				error = EWOULDBLOCK;
1302 				goto release;
1303 			}
1304 			error = sbwait(&so->so_snd);
1305 			SOCKBUF_UNLOCK(&so->so_snd);
1306 			if (error)
1307 				goto release;
1308 			goto restart;
1309 		}
1310 		SOCKBUF_UNLOCK(&so->so_snd);
1311 		space -= clen;
1312 		do {
1313 			if (uio == NULL) {
1314 				resid = 0;
1315 				if (flags & MSG_EOR)
1316 					top->m_flags |= M_EOR;
1317 			} else {
1318 				/*
1319 				 * Copy the data from userland into a mbuf
1320 				 * chain.  If no data is to be copied in,
1321 				 * a single empty mbuf is returned.
1322 				 */
1323 				top = m_uiotombuf(uio, M_WAITOK, space,
1324 				    (atomic ? max_hdr : 0),
1325 				    (atomic ? M_PKTHDR : 0) |
1326 				    ((flags & MSG_EOR) ? M_EOR : 0));
1327 				if (top == NULL) {
1328 					error = EFAULT; /* only possible error */
1329 					goto release;
1330 				}
1331 				space -= resid - uio->uio_resid;
1332 				resid = uio->uio_resid;
1333 			}
1334 			if (dontroute) {
1335 				SOCK_LOCK(so);
1336 				so->so_options |= SO_DONTROUTE;
1337 				SOCK_UNLOCK(so);
1338 			}
1339 			/*
1340 			 * XXX all the SBS_CANTSENDMORE checks previously
1341 			 * done could be out of date.  We could have recieved
1342 			 * a reset packet in an interrupt or maybe we slept
1343 			 * while doing page faults in uiomove() etc.  We
1344 			 * could probably recheck again inside the locking
1345 			 * protection here, but there are probably other
1346 			 * places that this also happens.  We must rethink
1347 			 * this.
1348 			 */
1349 			VNET_SO_ASSERT(so);
1350 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1351 			    (flags & MSG_OOB) ? PRUS_OOB :
1352 			/*
1353 			 * If the user set MSG_EOF, the protocol understands
1354 			 * this flag and nothing left to send then use
1355 			 * PRU_SEND_EOF instead of PRU_SEND.
1356 			 */
1357 			    ((flags & MSG_EOF) &&
1358 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1359 			     (resid <= 0)) ?
1360 				PRUS_EOF :
1361 			/* If there is more to send set PRUS_MORETOCOME. */
1362 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1363 			    top, addr, control, td);
1364 			if (dontroute) {
1365 				SOCK_LOCK(so);
1366 				so->so_options &= ~SO_DONTROUTE;
1367 				SOCK_UNLOCK(so);
1368 			}
1369 			clen = 0;
1370 			control = NULL;
1371 			top = NULL;
1372 			if (error)
1373 				goto release;
1374 		} while (resid && space > 0);
1375 	} while (resid);
1376 
1377 release:
1378 	sbunlock(&so->so_snd);
1379 out:
1380 	if (top != NULL)
1381 		m_freem(top);
1382 	if (control != NULL)
1383 		m_freem(control);
1384 	return (error);
1385 }
1386 
1387 int
1388 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1389     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1390 {
1391 	int error;
1392 
1393 	CURVNET_SET(so->so_vnet);
1394 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1395 	    control, flags, td);
1396 	CURVNET_RESTORE();
1397 	return (error);
1398 }
1399 
1400 /*
1401  * The part of soreceive() that implements reading non-inline out-of-band
1402  * data from a socket.  For more complete comments, see soreceive(), from
1403  * which this code originated.
1404  *
1405  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1406  * unable to return an mbuf chain to the caller.
1407  */
1408 static int
1409 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1410 {
1411 	struct protosw *pr = so->so_proto;
1412 	struct mbuf *m;
1413 	int error;
1414 
1415 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1416 	VNET_SO_ASSERT(so);
1417 
1418 	m = m_get(M_WAITOK, MT_DATA);
1419 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1420 	if (error)
1421 		goto bad;
1422 	do {
1423 		error = uiomove(mtod(m, void *),
1424 		    (int) min(uio->uio_resid, m->m_len), uio);
1425 		m = m_free(m);
1426 	} while (uio->uio_resid && error == 0 && m);
1427 bad:
1428 	if (m != NULL)
1429 		m_freem(m);
1430 	return (error);
1431 }
1432 
1433 /*
1434  * Following replacement or removal of the first mbuf on the first mbuf chain
1435  * of a socket buffer, push necessary state changes back into the socket
1436  * buffer so that other consumers see the values consistently.  'nextrecord'
1437  * is the callers locally stored value of the original value of
1438  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1439  * NOTE: 'nextrecord' may be NULL.
1440  */
1441 static __inline void
1442 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1443 {
1444 
1445 	SOCKBUF_LOCK_ASSERT(sb);
1446 	/*
1447 	 * First, update for the new value of nextrecord.  If necessary, make
1448 	 * it the first record.
1449 	 */
1450 	if (sb->sb_mb != NULL)
1451 		sb->sb_mb->m_nextpkt = nextrecord;
1452 	else
1453 		sb->sb_mb = nextrecord;
1454 
1455 	/*
1456 	 * Now update any dependent socket buffer fields to reflect the new
1457 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1458 	 * addition of a second clause that takes care of the case where
1459 	 * sb_mb has been updated, but remains the last record.
1460 	 */
1461 	if (sb->sb_mb == NULL) {
1462 		sb->sb_mbtail = NULL;
1463 		sb->sb_lastrecord = NULL;
1464 	} else if (sb->sb_mb->m_nextpkt == NULL)
1465 		sb->sb_lastrecord = sb->sb_mb;
1466 }
1467 
1468 /*
1469  * Implement receive operations on a socket.  We depend on the way that
1470  * records are added to the sockbuf by sbappend.  In particular, each record
1471  * (mbufs linked through m_next) must begin with an address if the protocol
1472  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1473  * data, and then zero or more mbufs of data.  In order to allow parallelism
1474  * between network receive and copying to user space, as well as avoid
1475  * sleeping with a mutex held, we release the socket buffer mutex during the
1476  * user space copy.  Although the sockbuf is locked, new data may still be
1477  * appended, and thus we must maintain consistency of the sockbuf during that
1478  * time.
1479  *
1480  * The caller may receive the data as a single mbuf chain by supplying an
1481  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1482  * the count in uio_resid.
1483  */
1484 int
1485 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1486     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1487 {
1488 	struct mbuf *m, **mp;
1489 	int flags, error, offset;
1490 	ssize_t len;
1491 	struct protosw *pr = so->so_proto;
1492 	struct mbuf *nextrecord;
1493 	int moff, type = 0;
1494 	ssize_t orig_resid = uio->uio_resid;
1495 
1496 	mp = mp0;
1497 	if (psa != NULL)
1498 		*psa = NULL;
1499 	if (controlp != NULL)
1500 		*controlp = NULL;
1501 	if (flagsp != NULL)
1502 		flags = *flagsp &~ MSG_EOR;
1503 	else
1504 		flags = 0;
1505 	if (flags & MSG_OOB)
1506 		return (soreceive_rcvoob(so, uio, flags));
1507 	if (mp != NULL)
1508 		*mp = NULL;
1509 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1510 	    && uio->uio_resid) {
1511 		VNET_SO_ASSERT(so);
1512 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1513 	}
1514 
1515 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1516 	if (error)
1517 		return (error);
1518 
1519 restart:
1520 	SOCKBUF_LOCK(&so->so_rcv);
1521 	m = so->so_rcv.sb_mb;
1522 	/*
1523 	 * If we have less data than requested, block awaiting more (subject
1524 	 * to any timeout) if:
1525 	 *   1. the current count is less than the low water mark, or
1526 	 *   2. MSG_DONTWAIT is not set
1527 	 */
1528 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1529 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1530 	    so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1531 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1532 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1533 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1534 		    m, so->so_rcv.sb_cc));
1535 		if (so->so_error) {
1536 			if (m != NULL)
1537 				goto dontblock;
1538 			error = so->so_error;
1539 			if ((flags & MSG_PEEK) == 0)
1540 				so->so_error = 0;
1541 			SOCKBUF_UNLOCK(&so->so_rcv);
1542 			goto release;
1543 		}
1544 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1545 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1546 			if (m == NULL) {
1547 				SOCKBUF_UNLOCK(&so->so_rcv);
1548 				goto release;
1549 			} else
1550 				goto dontblock;
1551 		}
1552 		for (; m != NULL; m = m->m_next)
1553 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1554 				m = so->so_rcv.sb_mb;
1555 				goto dontblock;
1556 			}
1557 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1558 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1559 			SOCKBUF_UNLOCK(&so->so_rcv);
1560 			error = ENOTCONN;
1561 			goto release;
1562 		}
1563 		if (uio->uio_resid == 0) {
1564 			SOCKBUF_UNLOCK(&so->so_rcv);
1565 			goto release;
1566 		}
1567 		if ((so->so_state & SS_NBIO) ||
1568 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1569 			SOCKBUF_UNLOCK(&so->so_rcv);
1570 			error = EWOULDBLOCK;
1571 			goto release;
1572 		}
1573 		SBLASTRECORDCHK(&so->so_rcv);
1574 		SBLASTMBUFCHK(&so->so_rcv);
1575 		error = sbwait(&so->so_rcv);
1576 		SOCKBUF_UNLOCK(&so->so_rcv);
1577 		if (error)
1578 			goto release;
1579 		goto restart;
1580 	}
1581 dontblock:
1582 	/*
1583 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1584 	 * pointer to the next record in the socket buffer.  We must keep the
1585 	 * various socket buffer pointers and local stack versions of the
1586 	 * pointers in sync, pushing out modifications before dropping the
1587 	 * socket buffer mutex, and re-reading them when picking it up.
1588 	 *
1589 	 * Otherwise, we will race with the network stack appending new data
1590 	 * or records onto the socket buffer by using inconsistent/stale
1591 	 * versions of the field, possibly resulting in socket buffer
1592 	 * corruption.
1593 	 *
1594 	 * By holding the high-level sblock(), we prevent simultaneous
1595 	 * readers from pulling off the front of the socket buffer.
1596 	 */
1597 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1598 	if (uio->uio_td)
1599 		uio->uio_td->td_ru.ru_msgrcv++;
1600 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1601 	SBLASTRECORDCHK(&so->so_rcv);
1602 	SBLASTMBUFCHK(&so->so_rcv);
1603 	nextrecord = m->m_nextpkt;
1604 	if (pr->pr_flags & PR_ADDR) {
1605 		KASSERT(m->m_type == MT_SONAME,
1606 		    ("m->m_type == %d", m->m_type));
1607 		orig_resid = 0;
1608 		if (psa != NULL)
1609 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1610 			    M_NOWAIT);
1611 		if (flags & MSG_PEEK) {
1612 			m = m->m_next;
1613 		} else {
1614 			sbfree(&so->so_rcv, m);
1615 			so->so_rcv.sb_mb = m_free(m);
1616 			m = so->so_rcv.sb_mb;
1617 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1618 		}
1619 	}
1620 
1621 	/*
1622 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1623 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1624 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1625 	 * perform externalization (or freeing if controlp == NULL).
1626 	 */
1627 	if (m != NULL && m->m_type == MT_CONTROL) {
1628 		struct mbuf *cm = NULL, *cmn;
1629 		struct mbuf **cme = &cm;
1630 
1631 		do {
1632 			if (flags & MSG_PEEK) {
1633 				if (controlp != NULL) {
1634 					*controlp = m_copy(m, 0, m->m_len);
1635 					controlp = &(*controlp)->m_next;
1636 				}
1637 				m = m->m_next;
1638 			} else {
1639 				sbfree(&so->so_rcv, m);
1640 				so->so_rcv.sb_mb = m->m_next;
1641 				m->m_next = NULL;
1642 				*cme = m;
1643 				cme = &(*cme)->m_next;
1644 				m = so->so_rcv.sb_mb;
1645 			}
1646 		} while (m != NULL && m->m_type == MT_CONTROL);
1647 		if ((flags & MSG_PEEK) == 0)
1648 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1649 		while (cm != NULL) {
1650 			cmn = cm->m_next;
1651 			cm->m_next = NULL;
1652 			if (pr->pr_domain->dom_externalize != NULL) {
1653 				SOCKBUF_UNLOCK(&so->so_rcv);
1654 				VNET_SO_ASSERT(so);
1655 				error = (*pr->pr_domain->dom_externalize)
1656 				    (cm, controlp, flags);
1657 				SOCKBUF_LOCK(&so->so_rcv);
1658 			} else if (controlp != NULL)
1659 				*controlp = cm;
1660 			else
1661 				m_freem(cm);
1662 			if (controlp != NULL) {
1663 				orig_resid = 0;
1664 				while (*controlp != NULL)
1665 					controlp = &(*controlp)->m_next;
1666 			}
1667 			cm = cmn;
1668 		}
1669 		if (m != NULL)
1670 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1671 		else
1672 			nextrecord = so->so_rcv.sb_mb;
1673 		orig_resid = 0;
1674 	}
1675 	if (m != NULL) {
1676 		if ((flags & MSG_PEEK) == 0) {
1677 			KASSERT(m->m_nextpkt == nextrecord,
1678 			    ("soreceive: post-control, nextrecord !sync"));
1679 			if (nextrecord == NULL) {
1680 				KASSERT(so->so_rcv.sb_mb == m,
1681 				    ("soreceive: post-control, sb_mb!=m"));
1682 				KASSERT(so->so_rcv.sb_lastrecord == m,
1683 				    ("soreceive: post-control, lastrecord!=m"));
1684 			}
1685 		}
1686 		type = m->m_type;
1687 		if (type == MT_OOBDATA)
1688 			flags |= MSG_OOB;
1689 	} else {
1690 		if ((flags & MSG_PEEK) == 0) {
1691 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1692 			    ("soreceive: sb_mb != nextrecord"));
1693 			if (so->so_rcv.sb_mb == NULL) {
1694 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1695 				    ("soreceive: sb_lastercord != NULL"));
1696 			}
1697 		}
1698 	}
1699 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1700 	SBLASTRECORDCHK(&so->so_rcv);
1701 	SBLASTMBUFCHK(&so->so_rcv);
1702 
1703 	/*
1704 	 * Now continue to read any data mbufs off of the head of the socket
1705 	 * buffer until the read request is satisfied.  Note that 'type' is
1706 	 * used to store the type of any mbuf reads that have happened so far
1707 	 * such that soreceive() can stop reading if the type changes, which
1708 	 * causes soreceive() to return only one of regular data and inline
1709 	 * out-of-band data in a single socket receive operation.
1710 	 */
1711 	moff = 0;
1712 	offset = 0;
1713 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1714 		/*
1715 		 * If the type of mbuf has changed since the last mbuf
1716 		 * examined ('type'), end the receive operation.
1717 		 */
1718 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1719 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1720 			if (type != m->m_type)
1721 				break;
1722 		} else if (type == MT_OOBDATA)
1723 			break;
1724 		else
1725 		    KASSERT(m->m_type == MT_DATA,
1726 			("m->m_type == %d", m->m_type));
1727 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1728 		len = uio->uio_resid;
1729 		if (so->so_oobmark && len > so->so_oobmark - offset)
1730 			len = so->so_oobmark - offset;
1731 		if (len > m->m_len - moff)
1732 			len = m->m_len - moff;
1733 		/*
1734 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1735 		 * them out via the uio, then free.  Sockbuf must be
1736 		 * consistent here (points to current mbuf, it points to next
1737 		 * record) when we drop priority; we must note any additions
1738 		 * to the sockbuf when we block interrupts again.
1739 		 */
1740 		if (mp == NULL) {
1741 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1742 			SBLASTRECORDCHK(&so->so_rcv);
1743 			SBLASTMBUFCHK(&so->so_rcv);
1744 			SOCKBUF_UNLOCK(&so->so_rcv);
1745 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1746 			SOCKBUF_LOCK(&so->so_rcv);
1747 			if (error) {
1748 				/*
1749 				 * The MT_SONAME mbuf has already been removed
1750 				 * from the record, so it is necessary to
1751 				 * remove the data mbufs, if any, to preserve
1752 				 * the invariant in the case of PR_ADDR that
1753 				 * requires MT_SONAME mbufs at the head of
1754 				 * each record.
1755 				 */
1756 				if (m && pr->pr_flags & PR_ATOMIC &&
1757 				    ((flags & MSG_PEEK) == 0))
1758 					(void)sbdroprecord_locked(&so->so_rcv);
1759 				SOCKBUF_UNLOCK(&so->so_rcv);
1760 				goto release;
1761 			}
1762 		} else
1763 			uio->uio_resid -= len;
1764 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1765 		if (len == m->m_len - moff) {
1766 			if (m->m_flags & M_EOR)
1767 				flags |= MSG_EOR;
1768 			if (flags & MSG_PEEK) {
1769 				m = m->m_next;
1770 				moff = 0;
1771 			} else {
1772 				nextrecord = m->m_nextpkt;
1773 				sbfree(&so->so_rcv, m);
1774 				if (mp != NULL) {
1775 					m->m_nextpkt = NULL;
1776 					*mp = m;
1777 					mp = &m->m_next;
1778 					so->so_rcv.sb_mb = m = m->m_next;
1779 					*mp = NULL;
1780 				} else {
1781 					so->so_rcv.sb_mb = m_free(m);
1782 					m = so->so_rcv.sb_mb;
1783 				}
1784 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1785 				SBLASTRECORDCHK(&so->so_rcv);
1786 				SBLASTMBUFCHK(&so->so_rcv);
1787 			}
1788 		} else {
1789 			if (flags & MSG_PEEK)
1790 				moff += len;
1791 			else {
1792 				if (mp != NULL) {
1793 					if (flags & MSG_DONTWAIT) {
1794 						*mp = m_copym(m, 0, len,
1795 						    M_NOWAIT);
1796 						if (*mp == NULL) {
1797 							/*
1798 							 * m_copym() couldn't
1799 							 * allocate an mbuf.
1800 							 * Adjust uio_resid back
1801 							 * (it was adjusted
1802 							 * down by len bytes,
1803 							 * which we didn't end
1804 							 * up "copying" over).
1805 							 */
1806 							uio->uio_resid += len;
1807 							break;
1808 						}
1809 					} else {
1810 						SOCKBUF_UNLOCK(&so->so_rcv);
1811 						*mp = m_copym(m, 0, len,
1812 						    M_WAITOK);
1813 						SOCKBUF_LOCK(&so->so_rcv);
1814 					}
1815 				}
1816 				m->m_data += len;
1817 				m->m_len -= len;
1818 				so->so_rcv.sb_cc -= len;
1819 			}
1820 		}
1821 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1822 		if (so->so_oobmark) {
1823 			if ((flags & MSG_PEEK) == 0) {
1824 				so->so_oobmark -= len;
1825 				if (so->so_oobmark == 0) {
1826 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1827 					break;
1828 				}
1829 			} else {
1830 				offset += len;
1831 				if (offset == so->so_oobmark)
1832 					break;
1833 			}
1834 		}
1835 		if (flags & MSG_EOR)
1836 			break;
1837 		/*
1838 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1839 		 * must not quit until "uio->uio_resid == 0" or an error
1840 		 * termination.  If a signal/timeout occurs, return with a
1841 		 * short count but without error.  Keep sockbuf locked
1842 		 * against other readers.
1843 		 */
1844 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1845 		    !sosendallatonce(so) && nextrecord == NULL) {
1846 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1847 			if (so->so_error ||
1848 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
1849 				break;
1850 			/*
1851 			 * Notify the protocol that some data has been
1852 			 * drained before blocking.
1853 			 */
1854 			if (pr->pr_flags & PR_WANTRCVD) {
1855 				SOCKBUF_UNLOCK(&so->so_rcv);
1856 				VNET_SO_ASSERT(so);
1857 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1858 				SOCKBUF_LOCK(&so->so_rcv);
1859 			}
1860 			SBLASTRECORDCHK(&so->so_rcv);
1861 			SBLASTMBUFCHK(&so->so_rcv);
1862 			/*
1863 			 * We could receive some data while was notifying
1864 			 * the protocol. Skip blocking in this case.
1865 			 */
1866 			if (so->so_rcv.sb_mb == NULL) {
1867 				error = sbwait(&so->so_rcv);
1868 				if (error) {
1869 					SOCKBUF_UNLOCK(&so->so_rcv);
1870 					goto release;
1871 				}
1872 			}
1873 			m = so->so_rcv.sb_mb;
1874 			if (m != NULL)
1875 				nextrecord = m->m_nextpkt;
1876 		}
1877 	}
1878 
1879 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1880 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1881 		flags |= MSG_TRUNC;
1882 		if ((flags & MSG_PEEK) == 0)
1883 			(void) sbdroprecord_locked(&so->so_rcv);
1884 	}
1885 	if ((flags & MSG_PEEK) == 0) {
1886 		if (m == NULL) {
1887 			/*
1888 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1889 			 * part makes sure sb_lastrecord is up-to-date if
1890 			 * there is still data in the socket buffer.
1891 			 */
1892 			so->so_rcv.sb_mb = nextrecord;
1893 			if (so->so_rcv.sb_mb == NULL) {
1894 				so->so_rcv.sb_mbtail = NULL;
1895 				so->so_rcv.sb_lastrecord = NULL;
1896 			} else if (nextrecord->m_nextpkt == NULL)
1897 				so->so_rcv.sb_lastrecord = nextrecord;
1898 		}
1899 		SBLASTRECORDCHK(&so->so_rcv);
1900 		SBLASTMBUFCHK(&so->so_rcv);
1901 		/*
1902 		 * If soreceive() is being done from the socket callback,
1903 		 * then don't need to generate ACK to peer to update window,
1904 		 * since ACK will be generated on return to TCP.
1905 		 */
1906 		if (!(flags & MSG_SOCALLBCK) &&
1907 		    (pr->pr_flags & PR_WANTRCVD)) {
1908 			SOCKBUF_UNLOCK(&so->so_rcv);
1909 			VNET_SO_ASSERT(so);
1910 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1911 			SOCKBUF_LOCK(&so->so_rcv);
1912 		}
1913 	}
1914 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1915 	if (orig_resid == uio->uio_resid && orig_resid &&
1916 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1917 		SOCKBUF_UNLOCK(&so->so_rcv);
1918 		goto restart;
1919 	}
1920 	SOCKBUF_UNLOCK(&so->so_rcv);
1921 
1922 	if (flagsp != NULL)
1923 		*flagsp |= flags;
1924 release:
1925 	sbunlock(&so->so_rcv);
1926 	return (error);
1927 }
1928 
1929 /*
1930  * Optimized version of soreceive() for stream (TCP) sockets.
1931  * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1932  */
1933 int
1934 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1935     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1936 {
1937 	int len = 0, error = 0, flags, oresid;
1938 	struct sockbuf *sb;
1939 	struct mbuf *m, *n = NULL;
1940 
1941 	/* We only do stream sockets. */
1942 	if (so->so_type != SOCK_STREAM)
1943 		return (EINVAL);
1944 	if (psa != NULL)
1945 		*psa = NULL;
1946 	if (controlp != NULL)
1947 		return (EINVAL);
1948 	if (flagsp != NULL)
1949 		flags = *flagsp &~ MSG_EOR;
1950 	else
1951 		flags = 0;
1952 	if (flags & MSG_OOB)
1953 		return (soreceive_rcvoob(so, uio, flags));
1954 	if (mp0 != NULL)
1955 		*mp0 = NULL;
1956 
1957 	sb = &so->so_rcv;
1958 
1959 	/* Prevent other readers from entering the socket. */
1960 	error = sblock(sb, SBLOCKWAIT(flags));
1961 	if (error)
1962 		goto out;
1963 	SOCKBUF_LOCK(sb);
1964 
1965 	/* Easy one, no space to copyout anything. */
1966 	if (uio->uio_resid == 0) {
1967 		error = EINVAL;
1968 		goto out;
1969 	}
1970 	oresid = uio->uio_resid;
1971 
1972 	/* We will never ever get anything unless we are or were connected. */
1973 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1974 		error = ENOTCONN;
1975 		goto out;
1976 	}
1977 
1978 restart:
1979 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1980 
1981 	/* Abort if socket has reported problems. */
1982 	if (so->so_error) {
1983 		if (sb->sb_cc > 0)
1984 			goto deliver;
1985 		if (oresid > uio->uio_resid)
1986 			goto out;
1987 		error = so->so_error;
1988 		if (!(flags & MSG_PEEK))
1989 			so->so_error = 0;
1990 		goto out;
1991 	}
1992 
1993 	/* Door is closed.  Deliver what is left, if any. */
1994 	if (sb->sb_state & SBS_CANTRCVMORE) {
1995 		if (sb->sb_cc > 0)
1996 			goto deliver;
1997 		else
1998 			goto out;
1999 	}
2000 
2001 	/* Socket buffer is empty and we shall not block. */
2002 	if (sb->sb_cc == 0 &&
2003 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2004 		error = EAGAIN;
2005 		goto out;
2006 	}
2007 
2008 	/* Socket buffer got some data that we shall deliver now. */
2009 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
2010 	    ((sb->sb_flags & SS_NBIO) ||
2011 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2012 	     sb->sb_cc >= sb->sb_lowat ||
2013 	     sb->sb_cc >= uio->uio_resid ||
2014 	     sb->sb_cc >= sb->sb_hiwat) ) {
2015 		goto deliver;
2016 	}
2017 
2018 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2019 	if ((flags & MSG_WAITALL) &&
2020 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
2021 		goto deliver;
2022 
2023 	/*
2024 	 * Wait and block until (more) data comes in.
2025 	 * NB: Drops the sockbuf lock during wait.
2026 	 */
2027 	error = sbwait(sb);
2028 	if (error)
2029 		goto out;
2030 	goto restart;
2031 
2032 deliver:
2033 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2034 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
2035 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2036 
2037 	/* Statistics. */
2038 	if (uio->uio_td)
2039 		uio->uio_td->td_ru.ru_msgrcv++;
2040 
2041 	/* Fill uio until full or current end of socket buffer is reached. */
2042 	len = min(uio->uio_resid, sb->sb_cc);
2043 	if (mp0 != NULL) {
2044 		/* Dequeue as many mbufs as possible. */
2045 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2046 			if (*mp0 == NULL)
2047 				*mp0 = sb->sb_mb;
2048 			else
2049 				m_cat(*mp0, sb->sb_mb);
2050 			for (m = sb->sb_mb;
2051 			     m != NULL && m->m_len <= len;
2052 			     m = m->m_next) {
2053 				len -= m->m_len;
2054 				uio->uio_resid -= m->m_len;
2055 				sbfree(sb, m);
2056 				n = m;
2057 			}
2058 			n->m_next = NULL;
2059 			sb->sb_mb = m;
2060 			sb->sb_lastrecord = sb->sb_mb;
2061 			if (sb->sb_mb == NULL)
2062 				SB_EMPTY_FIXUP(sb);
2063 		}
2064 		/* Copy the remainder. */
2065 		if (len > 0) {
2066 			KASSERT(sb->sb_mb != NULL,
2067 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2068 
2069 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2070 			if (m == NULL)
2071 				len = 0;	/* Don't flush data from sockbuf. */
2072 			else
2073 				uio->uio_resid -= len;
2074 			if (*mp0 != NULL)
2075 				m_cat(*mp0, m);
2076 			else
2077 				*mp0 = m;
2078 			if (*mp0 == NULL) {
2079 				error = ENOBUFS;
2080 				goto out;
2081 			}
2082 		}
2083 	} else {
2084 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2085 		SOCKBUF_UNLOCK(sb);
2086 		error = m_mbuftouio(uio, sb->sb_mb, len);
2087 		SOCKBUF_LOCK(sb);
2088 		if (error)
2089 			goto out;
2090 	}
2091 	SBLASTRECORDCHK(sb);
2092 	SBLASTMBUFCHK(sb);
2093 
2094 	/*
2095 	 * Remove the delivered data from the socket buffer unless we
2096 	 * were only peeking.
2097 	 */
2098 	if (!(flags & MSG_PEEK)) {
2099 		if (len > 0)
2100 			sbdrop_locked(sb, len);
2101 
2102 		/* Notify protocol that we drained some data. */
2103 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2104 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2105 		     !(flags & MSG_SOCALLBCK))) {
2106 			SOCKBUF_UNLOCK(sb);
2107 			VNET_SO_ASSERT(so);
2108 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2109 			SOCKBUF_LOCK(sb);
2110 		}
2111 	}
2112 
2113 	/*
2114 	 * For MSG_WAITALL we may have to loop again and wait for
2115 	 * more data to come in.
2116 	 */
2117 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2118 		goto restart;
2119 out:
2120 	SOCKBUF_LOCK_ASSERT(sb);
2121 	SBLASTRECORDCHK(sb);
2122 	SBLASTMBUFCHK(sb);
2123 	SOCKBUF_UNLOCK(sb);
2124 	sbunlock(sb);
2125 	return (error);
2126 }
2127 
2128 /*
2129  * Optimized version of soreceive() for simple datagram cases from userspace.
2130  * Unlike in the stream case, we're able to drop a datagram if copyout()
2131  * fails, and because we handle datagrams atomically, we don't need to use a
2132  * sleep lock to prevent I/O interlacing.
2133  */
2134 int
2135 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2136     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2137 {
2138 	struct mbuf *m, *m2;
2139 	int flags, error;
2140 	ssize_t len;
2141 	struct protosw *pr = so->so_proto;
2142 	struct mbuf *nextrecord;
2143 
2144 	if (psa != NULL)
2145 		*psa = NULL;
2146 	if (controlp != NULL)
2147 		*controlp = NULL;
2148 	if (flagsp != NULL)
2149 		flags = *flagsp &~ MSG_EOR;
2150 	else
2151 		flags = 0;
2152 
2153 	/*
2154 	 * For any complicated cases, fall back to the full
2155 	 * soreceive_generic().
2156 	 */
2157 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2158 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2159 		    flagsp));
2160 
2161 	/*
2162 	 * Enforce restrictions on use.
2163 	 */
2164 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2165 	    ("soreceive_dgram: wantrcvd"));
2166 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2167 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2168 	    ("soreceive_dgram: SBS_RCVATMARK"));
2169 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2170 	    ("soreceive_dgram: P_CONNREQUIRED"));
2171 
2172 	/*
2173 	 * Loop blocking while waiting for a datagram.
2174 	 */
2175 	SOCKBUF_LOCK(&so->so_rcv);
2176 	while ((m = so->so_rcv.sb_mb) == NULL) {
2177 		KASSERT(so->so_rcv.sb_cc == 0,
2178 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2179 		    so->so_rcv.sb_cc));
2180 		if (so->so_error) {
2181 			error = so->so_error;
2182 			so->so_error = 0;
2183 			SOCKBUF_UNLOCK(&so->so_rcv);
2184 			return (error);
2185 		}
2186 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2187 		    uio->uio_resid == 0) {
2188 			SOCKBUF_UNLOCK(&so->so_rcv);
2189 			return (0);
2190 		}
2191 		if ((so->so_state & SS_NBIO) ||
2192 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2193 			SOCKBUF_UNLOCK(&so->so_rcv);
2194 			return (EWOULDBLOCK);
2195 		}
2196 		SBLASTRECORDCHK(&so->so_rcv);
2197 		SBLASTMBUFCHK(&so->so_rcv);
2198 		error = sbwait(&so->so_rcv);
2199 		if (error) {
2200 			SOCKBUF_UNLOCK(&so->so_rcv);
2201 			return (error);
2202 		}
2203 	}
2204 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2205 
2206 	if (uio->uio_td)
2207 		uio->uio_td->td_ru.ru_msgrcv++;
2208 	SBLASTRECORDCHK(&so->so_rcv);
2209 	SBLASTMBUFCHK(&so->so_rcv);
2210 	nextrecord = m->m_nextpkt;
2211 	if (nextrecord == NULL) {
2212 		KASSERT(so->so_rcv.sb_lastrecord == m,
2213 		    ("soreceive_dgram: lastrecord != m"));
2214 	}
2215 
2216 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2217 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2218 
2219 	/*
2220 	 * Pull 'm' and its chain off the front of the packet queue.
2221 	 */
2222 	so->so_rcv.sb_mb = NULL;
2223 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2224 
2225 	/*
2226 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2227 	 */
2228 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2229 		sbfree(&so->so_rcv, m2);
2230 
2231 	/*
2232 	 * Do a few last checks before we let go of the lock.
2233 	 */
2234 	SBLASTRECORDCHK(&so->so_rcv);
2235 	SBLASTMBUFCHK(&so->so_rcv);
2236 	SOCKBUF_UNLOCK(&so->so_rcv);
2237 
2238 	if (pr->pr_flags & PR_ADDR) {
2239 		KASSERT(m->m_type == MT_SONAME,
2240 		    ("m->m_type == %d", m->m_type));
2241 		if (psa != NULL)
2242 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2243 			    M_NOWAIT);
2244 		m = m_free(m);
2245 	}
2246 	if (m == NULL) {
2247 		/* XXXRW: Can this happen? */
2248 		return (0);
2249 	}
2250 
2251 	/*
2252 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2253 	 * queue.
2254 	 *
2255 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2256 	 * in the first mbuf chain on the socket buffer.  We call into the
2257 	 * protocol to perform externalization (or freeing if controlp ==
2258 	 * NULL).
2259 	 */
2260 	if (m->m_type == MT_CONTROL) {
2261 		struct mbuf *cm = NULL, *cmn;
2262 		struct mbuf **cme = &cm;
2263 
2264 		do {
2265 			m2 = m->m_next;
2266 			m->m_next = NULL;
2267 			*cme = m;
2268 			cme = &(*cme)->m_next;
2269 			m = m2;
2270 		} while (m != NULL && m->m_type == MT_CONTROL);
2271 		while (cm != NULL) {
2272 			cmn = cm->m_next;
2273 			cm->m_next = NULL;
2274 			if (pr->pr_domain->dom_externalize != NULL) {
2275 				error = (*pr->pr_domain->dom_externalize)
2276 				    (cm, controlp, flags);
2277 			} else if (controlp != NULL)
2278 				*controlp = cm;
2279 			else
2280 				m_freem(cm);
2281 			if (controlp != NULL) {
2282 				while (*controlp != NULL)
2283 					controlp = &(*controlp)->m_next;
2284 			}
2285 			cm = cmn;
2286 		}
2287 	}
2288 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2289 
2290 	while (m != NULL && uio->uio_resid > 0) {
2291 		len = uio->uio_resid;
2292 		if (len > m->m_len)
2293 			len = m->m_len;
2294 		error = uiomove(mtod(m, char *), (int)len, uio);
2295 		if (error) {
2296 			m_freem(m);
2297 			return (error);
2298 		}
2299 		if (len == m->m_len)
2300 			m = m_free(m);
2301 		else {
2302 			m->m_data += len;
2303 			m->m_len -= len;
2304 		}
2305 	}
2306 	if (m != NULL)
2307 		flags |= MSG_TRUNC;
2308 	m_freem(m);
2309 	if (flagsp != NULL)
2310 		*flagsp |= flags;
2311 	return (0);
2312 }
2313 
2314 int
2315 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2316     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2317 {
2318 	int error;
2319 
2320 	CURVNET_SET(so->so_vnet);
2321 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2322 	    controlp, flagsp));
2323 	CURVNET_RESTORE();
2324 	return (error);
2325 }
2326 
2327 int
2328 soshutdown(struct socket *so, int how)
2329 {
2330 	struct protosw *pr = so->so_proto;
2331 	int error;
2332 
2333 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2334 		return (EINVAL);
2335 
2336 	CURVNET_SET(so->so_vnet);
2337 	if (pr->pr_usrreqs->pru_flush != NULL)
2338 		(*pr->pr_usrreqs->pru_flush)(so, how);
2339 	if (how != SHUT_WR)
2340 		sorflush(so);
2341 	if (how != SHUT_RD) {
2342 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2343 		wakeup(&so->so_timeo);
2344 		CURVNET_RESTORE();
2345 		return (error);
2346 	}
2347 	wakeup(&so->so_timeo);
2348 	CURVNET_RESTORE();
2349 	return (0);
2350 }
2351 
2352 void
2353 sorflush(struct socket *so)
2354 {
2355 	struct sockbuf *sb = &so->so_rcv;
2356 	struct protosw *pr = so->so_proto;
2357 	struct sockbuf asb;
2358 
2359 	VNET_SO_ASSERT(so);
2360 
2361 	/*
2362 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2363 	 * held, and in order to generally avoid holding the lock for a long
2364 	 * time, we make a copy of the socket buffer and clear the original
2365 	 * (except locks, state).  The new socket buffer copy won't have
2366 	 * initialized locks so we can only call routines that won't use or
2367 	 * assert those locks.
2368 	 *
2369 	 * Dislodge threads currently blocked in receive and wait to acquire
2370 	 * a lock against other simultaneous readers before clearing the
2371 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2372 	 * despite any existing socket disposition on interruptable waiting.
2373 	 */
2374 	socantrcvmore(so);
2375 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2376 
2377 	/*
2378 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2379 	 * and mutex data unchanged.
2380 	 */
2381 	SOCKBUF_LOCK(sb);
2382 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2383 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2384 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2385 	bzero(&sb->sb_startzero,
2386 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2387 	SOCKBUF_UNLOCK(sb);
2388 	sbunlock(sb);
2389 
2390 	/*
2391 	 * Dispose of special rights and flush the socket buffer.  Don't call
2392 	 * any unsafe routines (that rely on locks being initialized) on asb.
2393 	 */
2394 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2395 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2396 	sbrelease_internal(&asb, so);
2397 }
2398 
2399 /*
2400  * Wrapper for Socket established helper hook.
2401  * Parameters: socket, context of the hook point, hook id.
2402  */
2403 static int inline
2404 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2405 {
2406 	struct socket_hhook_data hhook_data = {
2407 		.so = so,
2408 		.hctx = hctx,
2409 		.m = NULL
2410 	};
2411 
2412 	hhook_run_hooks(V_socket_hhh[h_id], &hhook_data, &so->osd);
2413 
2414 	/* Ugly but needed, since hhooks return void for now */
2415 	return (hhook_data.status);
2416 }
2417 
2418 /*
2419  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2420  * additional variant to handle the case where the option value needs to be
2421  * some kind of integer, but not a specific size.  In addition to their use
2422  * here, these functions are also called by the protocol-level pr_ctloutput()
2423  * routines.
2424  */
2425 int
2426 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2427 {
2428 	size_t	valsize;
2429 
2430 	/*
2431 	 * If the user gives us more than we wanted, we ignore it, but if we
2432 	 * don't get the minimum length the caller wants, we return EINVAL.
2433 	 * On success, sopt->sopt_valsize is set to however much we actually
2434 	 * retrieved.
2435 	 */
2436 	if ((valsize = sopt->sopt_valsize) < minlen)
2437 		return EINVAL;
2438 	if (valsize > len)
2439 		sopt->sopt_valsize = valsize = len;
2440 
2441 	if (sopt->sopt_td != NULL)
2442 		return (copyin(sopt->sopt_val, buf, valsize));
2443 
2444 	bcopy(sopt->sopt_val, buf, valsize);
2445 	return (0);
2446 }
2447 
2448 /*
2449  * Kernel version of setsockopt(2).
2450  *
2451  * XXX: optlen is size_t, not socklen_t
2452  */
2453 int
2454 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2455     size_t optlen)
2456 {
2457 	struct sockopt sopt;
2458 
2459 	sopt.sopt_level = level;
2460 	sopt.sopt_name = optname;
2461 	sopt.sopt_dir = SOPT_SET;
2462 	sopt.sopt_val = optval;
2463 	sopt.sopt_valsize = optlen;
2464 	sopt.sopt_td = NULL;
2465 	return (sosetopt(so, &sopt));
2466 }
2467 
2468 int
2469 sosetopt(struct socket *so, struct sockopt *sopt)
2470 {
2471 	int	error, optval;
2472 	struct	linger l;
2473 	struct	timeval tv;
2474 	sbintime_t val;
2475 	uint32_t val32;
2476 #ifdef MAC
2477 	struct mac extmac;
2478 #endif
2479 
2480 	CURVNET_SET(so->so_vnet);
2481 	error = 0;
2482 	if (sopt->sopt_level != SOL_SOCKET) {
2483 		if (so->so_proto->pr_ctloutput != NULL) {
2484 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2485 			CURVNET_RESTORE();
2486 			return (error);
2487 		}
2488 		error = ENOPROTOOPT;
2489 	} else {
2490 		switch (sopt->sopt_name) {
2491 		case SO_ACCEPTFILTER:
2492 			error = do_setopt_accept_filter(so, sopt);
2493 			if (error)
2494 				goto bad;
2495 			break;
2496 
2497 		case SO_LINGER:
2498 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2499 			if (error)
2500 				goto bad;
2501 
2502 			SOCK_LOCK(so);
2503 			so->so_linger = l.l_linger;
2504 			if (l.l_onoff)
2505 				so->so_options |= SO_LINGER;
2506 			else
2507 				so->so_options &= ~SO_LINGER;
2508 			SOCK_UNLOCK(so);
2509 			break;
2510 
2511 		case SO_DEBUG:
2512 		case SO_KEEPALIVE:
2513 		case SO_DONTROUTE:
2514 		case SO_USELOOPBACK:
2515 		case SO_BROADCAST:
2516 		case SO_REUSEADDR:
2517 		case SO_REUSEPORT:
2518 		case SO_OOBINLINE:
2519 		case SO_TIMESTAMP:
2520 		case SO_BINTIME:
2521 		case SO_NOSIGPIPE:
2522 		case SO_NO_DDP:
2523 		case SO_NO_OFFLOAD:
2524 			error = sooptcopyin(sopt, &optval, sizeof optval,
2525 			    sizeof optval);
2526 			if (error)
2527 				goto bad;
2528 			SOCK_LOCK(so);
2529 			if (optval)
2530 				so->so_options |= sopt->sopt_name;
2531 			else
2532 				so->so_options &= ~sopt->sopt_name;
2533 			SOCK_UNLOCK(so);
2534 			break;
2535 
2536 		case SO_SETFIB:
2537 			error = sooptcopyin(sopt, &optval, sizeof optval,
2538 			    sizeof optval);
2539 			if (error)
2540 				goto bad;
2541 
2542 			if (optval < 0 || optval >= rt_numfibs) {
2543 				error = EINVAL;
2544 				goto bad;
2545 			}
2546 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2547 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2548 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2549 				so->so_fibnum = optval;
2550 			else
2551 				so->so_fibnum = 0;
2552 			break;
2553 
2554 		case SO_USER_COOKIE:
2555 			error = sooptcopyin(sopt, &val32, sizeof val32,
2556 			    sizeof val32);
2557 			if (error)
2558 				goto bad;
2559 			so->so_user_cookie = val32;
2560 			break;
2561 
2562 		case SO_SNDBUF:
2563 		case SO_RCVBUF:
2564 		case SO_SNDLOWAT:
2565 		case SO_RCVLOWAT:
2566 			error = sooptcopyin(sopt, &optval, sizeof optval,
2567 			    sizeof optval);
2568 			if (error)
2569 				goto bad;
2570 
2571 			/*
2572 			 * Values < 1 make no sense for any of these options,
2573 			 * so disallow them.
2574 			 */
2575 			if (optval < 1) {
2576 				error = EINVAL;
2577 				goto bad;
2578 			}
2579 
2580 			switch (sopt->sopt_name) {
2581 			case SO_SNDBUF:
2582 			case SO_RCVBUF:
2583 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2584 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2585 				    so, curthread) == 0) {
2586 					error = ENOBUFS;
2587 					goto bad;
2588 				}
2589 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2590 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2591 				break;
2592 
2593 			/*
2594 			 * Make sure the low-water is never greater than the
2595 			 * high-water.
2596 			 */
2597 			case SO_SNDLOWAT:
2598 				SOCKBUF_LOCK(&so->so_snd);
2599 				so->so_snd.sb_lowat =
2600 				    (optval > so->so_snd.sb_hiwat) ?
2601 				    so->so_snd.sb_hiwat : optval;
2602 				SOCKBUF_UNLOCK(&so->so_snd);
2603 				break;
2604 			case SO_RCVLOWAT:
2605 				SOCKBUF_LOCK(&so->so_rcv);
2606 				so->so_rcv.sb_lowat =
2607 				    (optval > so->so_rcv.sb_hiwat) ?
2608 				    so->so_rcv.sb_hiwat : optval;
2609 				SOCKBUF_UNLOCK(&so->so_rcv);
2610 				break;
2611 			}
2612 			break;
2613 
2614 		case SO_SNDTIMEO:
2615 		case SO_RCVTIMEO:
2616 #ifdef COMPAT_FREEBSD32
2617 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2618 				struct timeval32 tv32;
2619 
2620 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2621 				    sizeof tv32);
2622 				CP(tv32, tv, tv_sec);
2623 				CP(tv32, tv, tv_usec);
2624 			} else
2625 #endif
2626 				error = sooptcopyin(sopt, &tv, sizeof tv,
2627 				    sizeof tv);
2628 			if (error)
2629 				goto bad;
2630 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2631 			    tv.tv_usec >= 1000000) {
2632 				error = EDOM;
2633 				goto bad;
2634 			}
2635 			if (tv.tv_sec > INT32_MAX)
2636 				val = SBT_MAX;
2637 			else
2638 				val = tvtosbt(tv);
2639 			switch (sopt->sopt_name) {
2640 			case SO_SNDTIMEO:
2641 				so->so_snd.sb_timeo = val;
2642 				break;
2643 			case SO_RCVTIMEO:
2644 				so->so_rcv.sb_timeo = val;
2645 				break;
2646 			}
2647 			break;
2648 
2649 		case SO_LABEL:
2650 #ifdef MAC
2651 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2652 			    sizeof extmac);
2653 			if (error)
2654 				goto bad;
2655 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2656 			    so, &extmac);
2657 #else
2658 			error = EOPNOTSUPP;
2659 #endif
2660 			break;
2661 
2662 		default:
2663 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2664 				error = hhook_run_socket(so, sopt,
2665 				    HHOOK_SOCKET_OPT);
2666 			else
2667 				error = ENOPROTOOPT;
2668 			break;
2669 		}
2670 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2671 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2672 	}
2673 bad:
2674 	CURVNET_RESTORE();
2675 	return (error);
2676 }
2677 
2678 /*
2679  * Helper routine for getsockopt.
2680  */
2681 int
2682 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2683 {
2684 	int	error;
2685 	size_t	valsize;
2686 
2687 	error = 0;
2688 
2689 	/*
2690 	 * Documented get behavior is that we always return a value, possibly
2691 	 * truncated to fit in the user's buffer.  Traditional behavior is
2692 	 * that we always tell the user precisely how much we copied, rather
2693 	 * than something useful like the total amount we had available for
2694 	 * her.  Note that this interface is not idempotent; the entire
2695 	 * answer must generated ahead of time.
2696 	 */
2697 	valsize = min(len, sopt->sopt_valsize);
2698 	sopt->sopt_valsize = valsize;
2699 	if (sopt->sopt_val != NULL) {
2700 		if (sopt->sopt_td != NULL)
2701 			error = copyout(buf, sopt->sopt_val, valsize);
2702 		else
2703 			bcopy(buf, sopt->sopt_val, valsize);
2704 	}
2705 	return (error);
2706 }
2707 
2708 int
2709 sogetopt(struct socket *so, struct sockopt *sopt)
2710 {
2711 	int	error, optval;
2712 	struct	linger l;
2713 	struct	timeval tv;
2714 #ifdef MAC
2715 	struct mac extmac;
2716 #endif
2717 
2718 	CURVNET_SET(so->so_vnet);
2719 	error = 0;
2720 	if (sopt->sopt_level != SOL_SOCKET) {
2721 		if (so->so_proto->pr_ctloutput != NULL)
2722 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2723 		else
2724 			error = ENOPROTOOPT;
2725 		CURVNET_RESTORE();
2726 		return (error);
2727 	} else {
2728 		switch (sopt->sopt_name) {
2729 		case SO_ACCEPTFILTER:
2730 			error = do_getopt_accept_filter(so, sopt);
2731 			break;
2732 
2733 		case SO_LINGER:
2734 			SOCK_LOCK(so);
2735 			l.l_onoff = so->so_options & SO_LINGER;
2736 			l.l_linger = so->so_linger;
2737 			SOCK_UNLOCK(so);
2738 			error = sooptcopyout(sopt, &l, sizeof l);
2739 			break;
2740 
2741 		case SO_USELOOPBACK:
2742 		case SO_DONTROUTE:
2743 		case SO_DEBUG:
2744 		case SO_KEEPALIVE:
2745 		case SO_REUSEADDR:
2746 		case SO_REUSEPORT:
2747 		case SO_BROADCAST:
2748 		case SO_OOBINLINE:
2749 		case SO_ACCEPTCONN:
2750 		case SO_TIMESTAMP:
2751 		case SO_BINTIME:
2752 		case SO_NOSIGPIPE:
2753 			optval = so->so_options & sopt->sopt_name;
2754 integer:
2755 			error = sooptcopyout(sopt, &optval, sizeof optval);
2756 			break;
2757 
2758 		case SO_TYPE:
2759 			optval = so->so_type;
2760 			goto integer;
2761 
2762 		case SO_PROTOCOL:
2763 			optval = so->so_proto->pr_protocol;
2764 			goto integer;
2765 
2766 		case SO_ERROR:
2767 			SOCK_LOCK(so);
2768 			optval = so->so_error;
2769 			so->so_error = 0;
2770 			SOCK_UNLOCK(so);
2771 			goto integer;
2772 
2773 		case SO_SNDBUF:
2774 			optval = so->so_snd.sb_hiwat;
2775 			goto integer;
2776 
2777 		case SO_RCVBUF:
2778 			optval = so->so_rcv.sb_hiwat;
2779 			goto integer;
2780 
2781 		case SO_SNDLOWAT:
2782 			optval = so->so_snd.sb_lowat;
2783 			goto integer;
2784 
2785 		case SO_RCVLOWAT:
2786 			optval = so->so_rcv.sb_lowat;
2787 			goto integer;
2788 
2789 		case SO_SNDTIMEO:
2790 		case SO_RCVTIMEO:
2791 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2792 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2793 #ifdef COMPAT_FREEBSD32
2794 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2795 				struct timeval32 tv32;
2796 
2797 				CP(tv, tv32, tv_sec);
2798 				CP(tv, tv32, tv_usec);
2799 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2800 			} else
2801 #endif
2802 				error = sooptcopyout(sopt, &tv, sizeof tv);
2803 			break;
2804 
2805 		case SO_LABEL:
2806 #ifdef MAC
2807 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2808 			    sizeof(extmac));
2809 			if (error)
2810 				goto bad;
2811 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2812 			    so, &extmac);
2813 			if (error)
2814 				goto bad;
2815 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2816 #else
2817 			error = EOPNOTSUPP;
2818 #endif
2819 			break;
2820 
2821 		case SO_PEERLABEL:
2822 #ifdef MAC
2823 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2824 			    sizeof(extmac));
2825 			if (error)
2826 				goto bad;
2827 			error = mac_getsockopt_peerlabel(
2828 			    sopt->sopt_td->td_ucred, so, &extmac);
2829 			if (error)
2830 				goto bad;
2831 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2832 #else
2833 			error = EOPNOTSUPP;
2834 #endif
2835 			break;
2836 
2837 		case SO_LISTENQLIMIT:
2838 			optval = so->so_qlimit;
2839 			goto integer;
2840 
2841 		case SO_LISTENQLEN:
2842 			optval = so->so_qlen;
2843 			goto integer;
2844 
2845 		case SO_LISTENINCQLEN:
2846 			optval = so->so_incqlen;
2847 			goto integer;
2848 
2849 		default:
2850 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2851 				error = hhook_run_socket(so, sopt,
2852 				    HHOOK_SOCKET_OPT);
2853 			else
2854 				error = ENOPROTOOPT;
2855 			break;
2856 		}
2857 	}
2858 #ifdef MAC
2859 bad:
2860 #endif
2861 	CURVNET_RESTORE();
2862 	return (error);
2863 }
2864 
2865 int
2866 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2867 {
2868 	struct mbuf *m, *m_prev;
2869 	int sopt_size = sopt->sopt_valsize;
2870 
2871 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2872 	if (m == NULL)
2873 		return ENOBUFS;
2874 	if (sopt_size > MLEN) {
2875 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2876 		if ((m->m_flags & M_EXT) == 0) {
2877 			m_free(m);
2878 			return ENOBUFS;
2879 		}
2880 		m->m_len = min(MCLBYTES, sopt_size);
2881 	} else {
2882 		m->m_len = min(MLEN, sopt_size);
2883 	}
2884 	sopt_size -= m->m_len;
2885 	*mp = m;
2886 	m_prev = m;
2887 
2888 	while (sopt_size) {
2889 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2890 		if (m == NULL) {
2891 			m_freem(*mp);
2892 			return ENOBUFS;
2893 		}
2894 		if (sopt_size > MLEN) {
2895 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2896 			    M_NOWAIT);
2897 			if ((m->m_flags & M_EXT) == 0) {
2898 				m_freem(m);
2899 				m_freem(*mp);
2900 				return ENOBUFS;
2901 			}
2902 			m->m_len = min(MCLBYTES, sopt_size);
2903 		} else {
2904 			m->m_len = min(MLEN, sopt_size);
2905 		}
2906 		sopt_size -= m->m_len;
2907 		m_prev->m_next = m;
2908 		m_prev = m;
2909 	}
2910 	return (0);
2911 }
2912 
2913 int
2914 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2915 {
2916 	struct mbuf *m0 = m;
2917 
2918 	if (sopt->sopt_val == NULL)
2919 		return (0);
2920 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2921 		if (sopt->sopt_td != NULL) {
2922 			int error;
2923 
2924 			error = copyin(sopt->sopt_val, mtod(m, char *),
2925 			    m->m_len);
2926 			if (error != 0) {
2927 				m_freem(m0);
2928 				return(error);
2929 			}
2930 		} else
2931 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2932 		sopt->sopt_valsize -= m->m_len;
2933 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2934 		m = m->m_next;
2935 	}
2936 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2937 		panic("ip6_sooptmcopyin");
2938 	return (0);
2939 }
2940 
2941 int
2942 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2943 {
2944 	struct mbuf *m0 = m;
2945 	size_t valsize = 0;
2946 
2947 	if (sopt->sopt_val == NULL)
2948 		return (0);
2949 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2950 		if (sopt->sopt_td != NULL) {
2951 			int error;
2952 
2953 			error = copyout(mtod(m, char *), sopt->sopt_val,
2954 			    m->m_len);
2955 			if (error != 0) {
2956 				m_freem(m0);
2957 				return(error);
2958 			}
2959 		} else
2960 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2961 		sopt->sopt_valsize -= m->m_len;
2962 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2963 		valsize += m->m_len;
2964 		m = m->m_next;
2965 	}
2966 	if (m != NULL) {
2967 		/* enough soopt buffer should be given from user-land */
2968 		m_freem(m0);
2969 		return(EINVAL);
2970 	}
2971 	sopt->sopt_valsize = valsize;
2972 	return (0);
2973 }
2974 
2975 /*
2976  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2977  * out-of-band data, which will then notify socket consumers.
2978  */
2979 void
2980 sohasoutofband(struct socket *so)
2981 {
2982 
2983 	if (so->so_sigio != NULL)
2984 		pgsigio(&so->so_sigio, SIGURG, 0);
2985 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2986 }
2987 
2988 int
2989 sopoll(struct socket *so, int events, struct ucred *active_cred,
2990     struct thread *td)
2991 {
2992 
2993 	/*
2994 	 * We do not need to set or assert curvnet as long as everyone uses
2995 	 * sopoll_generic().
2996 	 */
2997 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2998 	    td));
2999 }
3000 
3001 int
3002 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3003     struct thread *td)
3004 {
3005 	int revents = 0;
3006 
3007 	SOCKBUF_LOCK(&so->so_snd);
3008 	SOCKBUF_LOCK(&so->so_rcv);
3009 	if (events & (POLLIN | POLLRDNORM))
3010 		if (soreadabledata(so))
3011 			revents |= events & (POLLIN | POLLRDNORM);
3012 
3013 	if (events & (POLLOUT | POLLWRNORM))
3014 		if (sowriteable(so))
3015 			revents |= events & (POLLOUT | POLLWRNORM);
3016 
3017 	if (events & (POLLPRI | POLLRDBAND))
3018 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3019 			revents |= events & (POLLPRI | POLLRDBAND);
3020 
3021 	if ((events & POLLINIGNEOF) == 0) {
3022 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3023 			revents |= events & (POLLIN | POLLRDNORM);
3024 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3025 				revents |= POLLHUP;
3026 		}
3027 	}
3028 
3029 	if (revents == 0) {
3030 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3031 			selrecord(td, &so->so_rcv.sb_sel);
3032 			so->so_rcv.sb_flags |= SB_SEL;
3033 		}
3034 
3035 		if (events & (POLLOUT | POLLWRNORM)) {
3036 			selrecord(td, &so->so_snd.sb_sel);
3037 			so->so_snd.sb_flags |= SB_SEL;
3038 		}
3039 	}
3040 
3041 	SOCKBUF_UNLOCK(&so->so_rcv);
3042 	SOCKBUF_UNLOCK(&so->so_snd);
3043 	return (revents);
3044 }
3045 
3046 int
3047 soo_kqfilter(struct file *fp, struct knote *kn)
3048 {
3049 	struct socket *so = kn->kn_fp->f_data;
3050 	struct sockbuf *sb;
3051 
3052 	switch (kn->kn_filter) {
3053 	case EVFILT_READ:
3054 		if (so->so_options & SO_ACCEPTCONN)
3055 			kn->kn_fop = &solisten_filtops;
3056 		else
3057 			kn->kn_fop = &soread_filtops;
3058 		sb = &so->so_rcv;
3059 		break;
3060 	case EVFILT_WRITE:
3061 		kn->kn_fop = &sowrite_filtops;
3062 		sb = &so->so_snd;
3063 		break;
3064 	default:
3065 		return (EINVAL);
3066 	}
3067 
3068 	SOCKBUF_LOCK(sb);
3069 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3070 	sb->sb_flags |= SB_KNOTE;
3071 	SOCKBUF_UNLOCK(sb);
3072 	return (0);
3073 }
3074 
3075 /*
3076  * Some routines that return EOPNOTSUPP for entry points that are not
3077  * supported by a protocol.  Fill in as needed.
3078  */
3079 int
3080 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3081 {
3082 
3083 	return EOPNOTSUPP;
3084 }
3085 
3086 int
3087 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3088 {
3089 
3090 	return EOPNOTSUPP;
3091 }
3092 
3093 int
3094 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3095 {
3096 
3097 	return EOPNOTSUPP;
3098 }
3099 
3100 int
3101 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3102     struct thread *td)
3103 {
3104 
3105 	return EOPNOTSUPP;
3106 }
3107 
3108 int
3109 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3110 {
3111 
3112 	return EOPNOTSUPP;
3113 }
3114 
3115 int
3116 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3117     struct thread *td)
3118 {
3119 
3120 	return EOPNOTSUPP;
3121 }
3122 
3123 int
3124 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3125 {
3126 
3127 	return EOPNOTSUPP;
3128 }
3129 
3130 int
3131 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3132     struct ifnet *ifp, struct thread *td)
3133 {
3134 
3135 	return EOPNOTSUPP;
3136 }
3137 
3138 int
3139 pru_disconnect_notsupp(struct socket *so)
3140 {
3141 
3142 	return EOPNOTSUPP;
3143 }
3144 
3145 int
3146 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3147 {
3148 
3149 	return EOPNOTSUPP;
3150 }
3151 
3152 int
3153 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3154 {
3155 
3156 	return EOPNOTSUPP;
3157 }
3158 
3159 int
3160 pru_rcvd_notsupp(struct socket *so, int flags)
3161 {
3162 
3163 	return EOPNOTSUPP;
3164 }
3165 
3166 int
3167 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3168 {
3169 
3170 	return EOPNOTSUPP;
3171 }
3172 
3173 int
3174 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3175     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3176 {
3177 
3178 	return EOPNOTSUPP;
3179 }
3180 
3181 /*
3182  * This isn't really a ``null'' operation, but it's the default one and
3183  * doesn't do anything destructive.
3184  */
3185 int
3186 pru_sense_null(struct socket *so, struct stat *sb)
3187 {
3188 
3189 	sb->st_blksize = so->so_snd.sb_hiwat;
3190 	return 0;
3191 }
3192 
3193 int
3194 pru_shutdown_notsupp(struct socket *so)
3195 {
3196 
3197 	return EOPNOTSUPP;
3198 }
3199 
3200 int
3201 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3202 {
3203 
3204 	return EOPNOTSUPP;
3205 }
3206 
3207 int
3208 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3209     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3210 {
3211 
3212 	return EOPNOTSUPP;
3213 }
3214 
3215 int
3216 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3217     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3218 {
3219 
3220 	return EOPNOTSUPP;
3221 }
3222 
3223 int
3224 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3225     struct thread *td)
3226 {
3227 
3228 	return EOPNOTSUPP;
3229 }
3230 
3231 static void
3232 filt_sordetach(struct knote *kn)
3233 {
3234 	struct socket *so = kn->kn_fp->f_data;
3235 
3236 	SOCKBUF_LOCK(&so->so_rcv);
3237 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3238 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3239 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3240 	SOCKBUF_UNLOCK(&so->so_rcv);
3241 }
3242 
3243 /*ARGSUSED*/
3244 static int
3245 filt_soread(struct knote *kn, long hint)
3246 {
3247 	struct socket *so;
3248 
3249 	so = kn->kn_fp->f_data;
3250 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3251 
3252 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3253 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3254 		kn->kn_flags |= EV_EOF;
3255 		kn->kn_fflags = so->so_error;
3256 		return (1);
3257 	} else if (so->so_error)	/* temporary udp error */
3258 		return (1);
3259 
3260 	if (kn->kn_sfflags & NOTE_LOWAT) {
3261 		if (kn->kn_data >= kn->kn_sdata)
3262 			return 1;
3263 	} else {
3264 		if (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat)
3265 			return 1;
3266 	}
3267 
3268 	if (V_socket_hhh[HHOOK_FILT_SOREAD]->hhh_nhooks > 0)
3269 		/* This hook returning non-zero indicates an event, not error */
3270 		return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3271 
3272 	return (0);
3273 }
3274 
3275 static void
3276 filt_sowdetach(struct knote *kn)
3277 {
3278 	struct socket *so = kn->kn_fp->f_data;
3279 
3280 	SOCKBUF_LOCK(&so->so_snd);
3281 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3282 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3283 		so->so_snd.sb_flags &= ~SB_KNOTE;
3284 	SOCKBUF_UNLOCK(&so->so_snd);
3285 }
3286 
3287 /*ARGSUSED*/
3288 static int
3289 filt_sowrite(struct knote *kn, long hint)
3290 {
3291 	struct socket *so;
3292 
3293 	so = kn->kn_fp->f_data;
3294 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3295 	kn->kn_data = sbspace(&so->so_snd);
3296 
3297 	if (V_socket_hhh[HHOOK_FILT_SOWRITE]->hhh_nhooks > 0)
3298 		hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3299 
3300 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3301 		kn->kn_flags |= EV_EOF;
3302 		kn->kn_fflags = so->so_error;
3303 		return (1);
3304 	} else if (so->so_error)	/* temporary udp error */
3305 		return (1);
3306 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3307 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3308 		return (0);
3309 	else if (kn->kn_sfflags & NOTE_LOWAT)
3310 		return (kn->kn_data >= kn->kn_sdata);
3311 	else
3312 		return (kn->kn_data >= so->so_snd.sb_lowat);
3313 }
3314 
3315 /*ARGSUSED*/
3316 static int
3317 filt_solisten(struct knote *kn, long hint)
3318 {
3319 	struct socket *so = kn->kn_fp->f_data;
3320 
3321 	kn->kn_data = so->so_qlen;
3322 	return (!TAILQ_EMPTY(&so->so_comp));
3323 }
3324 
3325 int
3326 socheckuid(struct socket *so, uid_t uid)
3327 {
3328 
3329 	if (so == NULL)
3330 		return (EPERM);
3331 	if (so->so_cred->cr_uid != uid)
3332 		return (EPERM);
3333 	return (0);
3334 }
3335 
3336 /*
3337  * These functions are used by protocols to notify the socket layer (and its
3338  * consumers) of state changes in the sockets driven by protocol-side events.
3339  */
3340 
3341 /*
3342  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3343  *
3344  * Normal sequence from the active (originating) side is that
3345  * soisconnecting() is called during processing of connect() call, resulting
3346  * in an eventual call to soisconnected() if/when the connection is
3347  * established.  When the connection is torn down soisdisconnecting() is
3348  * called during processing of disconnect() call, and soisdisconnected() is
3349  * called when the connection to the peer is totally severed.  The semantics
3350  * of these routines are such that connectionless protocols can call
3351  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3352  * calls when setting up a ``connection'' takes no time.
3353  *
3354  * From the passive side, a socket is created with two queues of sockets:
3355  * so_incomp for connections in progress and so_comp for connections already
3356  * made and awaiting user acceptance.  As a protocol is preparing incoming
3357  * connections, it creates a socket structure queued on so_incomp by calling
3358  * sonewconn().  When the connection is established, soisconnected() is
3359  * called, and transfers the socket structure to so_comp, making it available
3360  * to accept().
3361  *
3362  * If a socket is closed with sockets on either so_incomp or so_comp, these
3363  * sockets are dropped.
3364  *
3365  * If higher-level protocols are implemented in the kernel, the wakeups done
3366  * here will sometimes cause software-interrupt process scheduling.
3367  */
3368 void
3369 soisconnecting(struct socket *so)
3370 {
3371 
3372 	SOCK_LOCK(so);
3373 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3374 	so->so_state |= SS_ISCONNECTING;
3375 	SOCK_UNLOCK(so);
3376 }
3377 
3378 void
3379 soisconnected(struct socket *so)
3380 {
3381 	struct socket *head;
3382 	int ret;
3383 
3384 restart:
3385 	ACCEPT_LOCK();
3386 	SOCK_LOCK(so);
3387 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3388 	so->so_state |= SS_ISCONNECTED;
3389 	head = so->so_head;
3390 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3391 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3392 			SOCK_UNLOCK(so);
3393 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3394 			head->so_incqlen--;
3395 			so->so_qstate &= ~SQ_INCOMP;
3396 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3397 			head->so_qlen++;
3398 			so->so_qstate |= SQ_COMP;
3399 			ACCEPT_UNLOCK();
3400 			sorwakeup(head);
3401 			wakeup_one(&head->so_timeo);
3402 		} else {
3403 			ACCEPT_UNLOCK();
3404 			soupcall_set(so, SO_RCV,
3405 			    head->so_accf->so_accept_filter->accf_callback,
3406 			    head->so_accf->so_accept_filter_arg);
3407 			so->so_options &= ~SO_ACCEPTFILTER;
3408 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3409 			    head->so_accf->so_accept_filter_arg, M_NOWAIT);
3410 			if (ret == SU_ISCONNECTED)
3411 				soupcall_clear(so, SO_RCV);
3412 			SOCK_UNLOCK(so);
3413 			if (ret == SU_ISCONNECTED)
3414 				goto restart;
3415 		}
3416 		return;
3417 	}
3418 	SOCK_UNLOCK(so);
3419 	ACCEPT_UNLOCK();
3420 	wakeup(&so->so_timeo);
3421 	sorwakeup(so);
3422 	sowwakeup(so);
3423 }
3424 
3425 void
3426 soisdisconnecting(struct socket *so)
3427 {
3428 
3429 	/*
3430 	 * Note: This code assumes that SOCK_LOCK(so) and
3431 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3432 	 */
3433 	SOCKBUF_LOCK(&so->so_rcv);
3434 	so->so_state &= ~SS_ISCONNECTING;
3435 	so->so_state |= SS_ISDISCONNECTING;
3436 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3437 	sorwakeup_locked(so);
3438 	SOCKBUF_LOCK(&so->so_snd);
3439 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3440 	sowwakeup_locked(so);
3441 	wakeup(&so->so_timeo);
3442 }
3443 
3444 void
3445 soisdisconnected(struct socket *so)
3446 {
3447 
3448 	/*
3449 	 * Note: This code assumes that SOCK_LOCK(so) and
3450 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3451 	 */
3452 	SOCKBUF_LOCK(&so->so_rcv);
3453 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3454 	so->so_state |= SS_ISDISCONNECTED;
3455 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3456 	sorwakeup_locked(so);
3457 	SOCKBUF_LOCK(&so->so_snd);
3458 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3459 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3460 	sowwakeup_locked(so);
3461 	wakeup(&so->so_timeo);
3462 }
3463 
3464 /*
3465  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3466  */
3467 struct sockaddr *
3468 sodupsockaddr(const struct sockaddr *sa, int mflags)
3469 {
3470 	struct sockaddr *sa2;
3471 
3472 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3473 	if (sa2)
3474 		bcopy(sa, sa2, sa->sa_len);
3475 	return sa2;
3476 }
3477 
3478 /*
3479  * Register per-socket buffer upcalls.
3480  */
3481 void
3482 soupcall_set(struct socket *so, int which,
3483     int (*func)(struct socket *, void *, int), void *arg)
3484 {
3485 	struct sockbuf *sb;
3486 
3487 	switch (which) {
3488 	case SO_RCV:
3489 		sb = &so->so_rcv;
3490 		break;
3491 	case SO_SND:
3492 		sb = &so->so_snd;
3493 		break;
3494 	default:
3495 		panic("soupcall_set: bad which");
3496 	}
3497 	SOCKBUF_LOCK_ASSERT(sb);
3498 #if 0
3499 	/* XXX: accf_http actually wants to do this on purpose. */
3500 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3501 #endif
3502 	sb->sb_upcall = func;
3503 	sb->sb_upcallarg = arg;
3504 	sb->sb_flags |= SB_UPCALL;
3505 }
3506 
3507 void
3508 soupcall_clear(struct socket *so, int which)
3509 {
3510 	struct sockbuf *sb;
3511 
3512 	switch (which) {
3513 	case SO_RCV:
3514 		sb = &so->so_rcv;
3515 		break;
3516 	case SO_SND:
3517 		sb = &so->so_snd;
3518 		break;
3519 	default:
3520 		panic("soupcall_clear: bad which");
3521 	}
3522 	SOCKBUF_LOCK_ASSERT(sb);
3523 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3524 	sb->sb_upcall = NULL;
3525 	sb->sb_upcallarg = NULL;
3526 	sb->sb_flags &= ~SB_UPCALL;
3527 }
3528 
3529 /*
3530  * Create an external-format (``xsocket'') structure using the information in
3531  * the kernel-format socket structure pointed to by so.  This is done to
3532  * reduce the spew of irrelevant information over this interface, to isolate
3533  * user code from changes in the kernel structure, and potentially to provide
3534  * information-hiding if we decide that some of this information should be
3535  * hidden from users.
3536  */
3537 void
3538 sotoxsocket(struct socket *so, struct xsocket *xso)
3539 {
3540 
3541 	xso->xso_len = sizeof *xso;
3542 	xso->xso_so = so;
3543 	xso->so_type = so->so_type;
3544 	xso->so_options = so->so_options;
3545 	xso->so_linger = so->so_linger;
3546 	xso->so_state = so->so_state;
3547 	xso->so_pcb = so->so_pcb;
3548 	xso->xso_protocol = so->so_proto->pr_protocol;
3549 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3550 	xso->so_qlen = so->so_qlen;
3551 	xso->so_incqlen = so->so_incqlen;
3552 	xso->so_qlimit = so->so_qlimit;
3553 	xso->so_timeo = so->so_timeo;
3554 	xso->so_error = so->so_error;
3555 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3556 	xso->so_oobmark = so->so_oobmark;
3557 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3558 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3559 	xso->so_uid = so->so_cred->cr_uid;
3560 }
3561 
3562 
3563 /*
3564  * Socket accessor functions to provide external consumers with
3565  * a safe interface to socket state
3566  *
3567  */
3568 
3569 void
3570 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3571     void *arg)
3572 {
3573 
3574 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3575 		func(so, arg);
3576 }
3577 
3578 struct sockbuf *
3579 so_sockbuf_rcv(struct socket *so)
3580 {
3581 
3582 	return (&so->so_rcv);
3583 }
3584 
3585 struct sockbuf *
3586 so_sockbuf_snd(struct socket *so)
3587 {
3588 
3589 	return (&so->so_snd);
3590 }
3591 
3592 int
3593 so_state_get(const struct socket *so)
3594 {
3595 
3596 	return (so->so_state);
3597 }
3598 
3599 void
3600 so_state_set(struct socket *so, int val)
3601 {
3602 
3603 	so->so_state = val;
3604 }
3605 
3606 int
3607 so_options_get(const struct socket *so)
3608 {
3609 
3610 	return (so->so_options);
3611 }
3612 
3613 void
3614 so_options_set(struct socket *so, int val)
3615 {
3616 
3617 	so->so_options = val;
3618 }
3619 
3620 int
3621 so_error_get(const struct socket *so)
3622 {
3623 
3624 	return (so->so_error);
3625 }
3626 
3627 void
3628 so_error_set(struct socket *so, int val)
3629 {
3630 
3631 	so->so_error = val;
3632 }
3633 
3634 int
3635 so_linger_get(const struct socket *so)
3636 {
3637 
3638 	return (so->so_linger);
3639 }
3640 
3641 void
3642 so_linger_set(struct socket *so, int val)
3643 {
3644 
3645 	so->so_linger = val;
3646 }
3647 
3648 struct protosw *
3649 so_protosw_get(const struct socket *so)
3650 {
3651 
3652 	return (so->so_proto);
3653 }
3654 
3655 void
3656 so_protosw_set(struct socket *so, struct protosw *val)
3657 {
3658 
3659 	so->so_proto = val;
3660 }
3661 
3662 void
3663 so_sorwakeup(struct socket *so)
3664 {
3665 
3666 	sorwakeup(so);
3667 }
3668 
3669 void
3670 so_sowwakeup(struct socket *so)
3671 {
3672 
3673 	sowwakeup(so);
3674 }
3675 
3676 void
3677 so_sorwakeup_locked(struct socket *so)
3678 {
3679 
3680 	sorwakeup_locked(so);
3681 }
3682 
3683 void
3684 so_sowwakeup_locked(struct socket *so)
3685 {
3686 
3687 	sowwakeup_locked(so);
3688 }
3689 
3690 void
3691 so_lock(struct socket *so)
3692 {
3693 
3694 	SOCK_LOCK(so);
3695 }
3696 
3697 void
3698 so_unlock(struct socket *so)
3699 {
3700 
3701 	SOCK_UNLOCK(so);
3702 }
3703