xref: /freebsd/sys/kern/uipc_socket.c (revision 6829dae12bb055451fa467da4589c43bd03b1e64)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2004 The FreeBSD Foundation
7  * Copyright (c) 2004-2008 Robert N. M. Watson
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
35  */
36 
37 /*
38  * Comments on the socket life cycle:
39  *
40  * soalloc() sets of socket layer state for a socket, called only by
41  * socreate() and sonewconn().  Socket layer private.
42  *
43  * sodealloc() tears down socket layer state for a socket, called only by
44  * sofree() and sonewconn().  Socket layer private.
45  *
46  * pru_attach() associates protocol layer state with an allocated socket;
47  * called only once, may fail, aborting socket allocation.  This is called
48  * from socreate() and sonewconn().  Socket layer private.
49  *
50  * pru_detach() disassociates protocol layer state from an attached socket,
51  * and will be called exactly once for sockets in which pru_attach() has
52  * been successfully called.  If pru_attach() returned an error,
53  * pru_detach() will not be called.  Socket layer private.
54  *
55  * pru_abort() and pru_close() notify the protocol layer that the last
56  * consumer of a socket is starting to tear down the socket, and that the
57  * protocol should terminate the connection.  Historically, pru_abort() also
58  * detached protocol state from the socket state, but this is no longer the
59  * case.
60  *
61  * socreate() creates a socket and attaches protocol state.  This is a public
62  * interface that may be used by socket layer consumers to create new
63  * sockets.
64  *
65  * sonewconn() creates a socket and attaches protocol state.  This is a
66  * public interface  that may be used by protocols to create new sockets when
67  * a new connection is received and will be available for accept() on a
68  * listen socket.
69  *
70  * soclose() destroys a socket after possibly waiting for it to disconnect.
71  * This is a public interface that socket consumers should use to close and
72  * release a socket when done with it.
73  *
74  * soabort() destroys a socket without waiting for it to disconnect (used
75  * only for incoming connections that are already partially or fully
76  * connected).  This is used internally by the socket layer when clearing
77  * listen socket queues (due to overflow or close on the listen socket), but
78  * is also a public interface protocols may use to abort connections in
79  * their incomplete listen queues should they no longer be required.  Sockets
80  * placed in completed connection listen queues should not be aborted for
81  * reasons described in the comment above the soclose() implementation.  This
82  * is not a general purpose close routine, and except in the specific
83  * circumstances described here, should not be used.
84  *
85  * sofree() will free a socket and its protocol state if all references on
86  * the socket have been released, and is the public interface to attempt to
87  * free a socket when a reference is removed.  This is a socket layer private
88  * interface.
89  *
90  * NOTE: In addition to socreate() and soclose(), which provide a single
91  * socket reference to the consumer to be managed as required, there are two
92  * calls to explicitly manage socket references, soref(), and sorele().
93  * Currently, these are generally required only when transitioning a socket
94  * from a listen queue to a file descriptor, in order to prevent garbage
95  * collection of the socket at an untimely moment.  For a number of reasons,
96  * these interfaces are not preferred, and should be avoided.
97  *
98  * NOTE: With regard to VNETs the general rule is that callers do not set
99  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101  * and sorflush(), which are usually called from a pre-set VNET context.
102  * sopoll() currently does not need a VNET context to be set.
103  */
104 
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
107 
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_sctp.h"
111 
112 #include <sys/param.h>
113 #include <sys/systm.h>
114 #include <sys/fcntl.h>
115 #include <sys/limits.h>
116 #include <sys/lock.h>
117 #include <sys/mac.h>
118 #include <sys/malloc.h>
119 #include <sys/mbuf.h>
120 #include <sys/mutex.h>
121 #include <sys/domain.h>
122 #include <sys/file.h>			/* for struct knote */
123 #include <sys/hhook.h>
124 #include <sys/kernel.h>
125 #include <sys/khelp.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/proc.h>
130 #include <sys/protosw.h>
131 #include <sys/socket.h>
132 #include <sys/socketvar.h>
133 #include <sys/resourcevar.h>
134 #include <net/route.h>
135 #include <sys/signalvar.h>
136 #include <sys/stat.h>
137 #include <sys/sx.h>
138 #include <sys/sysctl.h>
139 #include <sys/taskqueue.h>
140 #include <sys/uio.h>
141 #include <sys/jail.h>
142 #include <sys/syslog.h>
143 #include <netinet/in.h>
144 
145 #include <net/vnet.h>
146 
147 #include <security/mac/mac_framework.h>
148 
149 #include <vm/uma.h>
150 
151 #ifdef COMPAT_FREEBSD32
152 #include <sys/mount.h>
153 #include <sys/sysent.h>
154 #include <compat/freebsd32/freebsd32.h>
155 #endif
156 
157 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
158 		    int flags);
159 static void	so_rdknl_lock(void *);
160 static void	so_rdknl_unlock(void *);
161 static void	so_rdknl_assert_locked(void *);
162 static void	so_rdknl_assert_unlocked(void *);
163 static void	so_wrknl_lock(void *);
164 static void	so_wrknl_unlock(void *);
165 static void	so_wrknl_assert_locked(void *);
166 static void	so_wrknl_assert_unlocked(void *);
167 
168 static void	filt_sordetach(struct knote *kn);
169 static int	filt_soread(struct knote *kn, long hint);
170 static void	filt_sowdetach(struct knote *kn);
171 static int	filt_sowrite(struct knote *kn, long hint);
172 static int	filt_soempty(struct knote *kn, long hint);
173 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
174 fo_kqfilter_t	soo_kqfilter;
175 
176 static struct filterops soread_filtops = {
177 	.f_isfd = 1,
178 	.f_detach = filt_sordetach,
179 	.f_event = filt_soread,
180 };
181 static struct filterops sowrite_filtops = {
182 	.f_isfd = 1,
183 	.f_detach = filt_sowdetach,
184 	.f_event = filt_sowrite,
185 };
186 static struct filterops soempty_filtops = {
187 	.f_isfd = 1,
188 	.f_detach = filt_sowdetach,
189 	.f_event = filt_soempty,
190 };
191 
192 so_gen_t	so_gencnt;	/* generation count for sockets */
193 
194 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
195 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
196 
197 #define	VNET_SO_ASSERT(so)						\
198 	VNET_ASSERT(curvnet != NULL,					\
199 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
200 
201 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
202 #define	V_socket_hhh		VNET(socket_hhh)
203 
204 /*
205  * Limit on the number of connections in the listen queue waiting
206  * for accept(2).
207  * NB: The original sysctl somaxconn is still available but hidden
208  * to prevent confusion about the actual purpose of this number.
209  */
210 static u_int somaxconn = SOMAXCONN;
211 
212 static int
213 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
214 {
215 	int error;
216 	int val;
217 
218 	val = somaxconn;
219 	error = sysctl_handle_int(oidp, &val, 0, req);
220 	if (error || !req->newptr )
221 		return (error);
222 
223 	/*
224 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
225 	 *   3 * so_qlimit / 2
226 	 * below, will not overflow.
227          */
228 
229 	if (val < 1 || val > UINT_MAX / 3)
230 		return (EINVAL);
231 
232 	somaxconn = val;
233 	return (0);
234 }
235 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
236     0, sizeof(int), sysctl_somaxconn, "I",
237     "Maximum listen socket pending connection accept queue size");
238 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
239     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
240     0, sizeof(int), sysctl_somaxconn, "I",
241     "Maximum listen socket pending connection accept queue size (compat)");
242 
243 static int numopensockets;
244 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
245     &numopensockets, 0, "Number of open sockets");
246 
247 /*
248  * accept_mtx locks down per-socket fields relating to accept queues.  See
249  * socketvar.h for an annotation of the protected fields of struct socket.
250  */
251 struct mtx accept_mtx;
252 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
253 
254 /*
255  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
256  * so_gencnt field.
257  */
258 static struct mtx so_global_mtx;
259 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
260 
261 /*
262  * General IPC sysctl name space, used by sockets and a variety of other IPC
263  * types.
264  */
265 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
266 
267 /*
268  * Initialize the socket subsystem and set up the socket
269  * memory allocator.
270  */
271 static uma_zone_t socket_zone;
272 int	maxsockets;
273 
274 static void
275 socket_zone_change(void *tag)
276 {
277 
278 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
279 }
280 
281 static void
282 socket_hhook_register(int subtype)
283 {
284 
285 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
286 	    &V_socket_hhh[subtype],
287 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
288 		printf("%s: WARNING: unable to register hook\n", __func__);
289 }
290 
291 static void
292 socket_hhook_deregister(int subtype)
293 {
294 
295 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
296 		printf("%s: WARNING: unable to deregister hook\n", __func__);
297 }
298 
299 static void
300 socket_init(void *tag)
301 {
302 
303 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
304 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
305 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
306 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
307 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
308 	    EVENTHANDLER_PRI_FIRST);
309 }
310 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
311 
312 static void
313 socket_vnet_init(const void *unused __unused)
314 {
315 	int i;
316 
317 	/* We expect a contiguous range */
318 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
319 		socket_hhook_register(i);
320 }
321 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
322     socket_vnet_init, NULL);
323 
324 static void
325 socket_vnet_uninit(const void *unused __unused)
326 {
327 	int i;
328 
329 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
330 		socket_hhook_deregister(i);
331 }
332 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
333     socket_vnet_uninit, NULL);
334 
335 /*
336  * Initialise maxsockets.  This SYSINIT must be run after
337  * tunable_mbinit().
338  */
339 static void
340 init_maxsockets(void *ignored)
341 {
342 
343 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
344 	maxsockets = imax(maxsockets, maxfiles);
345 }
346 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
347 
348 /*
349  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
350  * of the change so that they can update their dependent limits as required.
351  */
352 static int
353 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
354 {
355 	int error, newmaxsockets;
356 
357 	newmaxsockets = maxsockets;
358 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
359 	if (error == 0 && req->newptr) {
360 		if (newmaxsockets > maxsockets &&
361 		    newmaxsockets <= maxfiles) {
362 			maxsockets = newmaxsockets;
363 			EVENTHANDLER_INVOKE(maxsockets_change);
364 		} else
365 			error = EINVAL;
366 	}
367 	return (error);
368 }
369 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
370     &maxsockets, 0, sysctl_maxsockets, "IU",
371     "Maximum number of sockets available");
372 
373 /*
374  * Socket operation routines.  These routines are called by the routines in
375  * sys_socket.c or from a system process, and implement the semantics of
376  * socket operations by switching out to the protocol specific routines.
377  */
378 
379 /*
380  * Get a socket structure from our zone, and initialize it.  Note that it
381  * would probably be better to allocate socket and PCB at the same time, but
382  * I'm not convinced that all the protocols can be easily modified to do
383  * this.
384  *
385  * soalloc() returns a socket with a ref count of 0.
386  */
387 static struct socket *
388 soalloc(struct vnet *vnet)
389 {
390 	struct socket *so;
391 
392 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
393 	if (so == NULL)
394 		return (NULL);
395 #ifdef MAC
396 	if (mac_socket_init(so, M_NOWAIT) != 0) {
397 		uma_zfree(socket_zone, so);
398 		return (NULL);
399 	}
400 #endif
401 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
402 		uma_zfree(socket_zone, so);
403 		return (NULL);
404 	}
405 
406 	/*
407 	 * The socket locking protocol allows to lock 2 sockets at a time,
408 	 * however, the first one must be a listening socket.  WITNESS lacks
409 	 * a feature to change class of an existing lock, so we use DUPOK.
410 	 */
411 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
412 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
413 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
414 	so->so_rcv.sb_sel = &so->so_rdsel;
415 	so->so_snd.sb_sel = &so->so_wrsel;
416 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
417 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
418 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
419 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
420 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
421 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
422 #ifdef VIMAGE
423 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
424 	    __func__, __LINE__, so));
425 	so->so_vnet = vnet;
426 #endif
427 	/* We shouldn't need the so_global_mtx */
428 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
429 		/* Do we need more comprehensive error returns? */
430 		uma_zfree(socket_zone, so);
431 		return (NULL);
432 	}
433 	mtx_lock(&so_global_mtx);
434 	so->so_gencnt = ++so_gencnt;
435 	++numopensockets;
436 #ifdef VIMAGE
437 	vnet->vnet_sockcnt++;
438 #endif
439 	mtx_unlock(&so_global_mtx);
440 
441 	return (so);
442 }
443 
444 /*
445  * Free the storage associated with a socket at the socket layer, tear down
446  * locks, labels, etc.  All protocol state is assumed already to have been
447  * torn down (and possibly never set up) by the caller.
448  */
449 static void
450 sodealloc(struct socket *so)
451 {
452 
453 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
454 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
455 
456 	mtx_lock(&so_global_mtx);
457 	so->so_gencnt = ++so_gencnt;
458 	--numopensockets;	/* Could be below, but faster here. */
459 #ifdef VIMAGE
460 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
461 	    __func__, __LINE__, so));
462 	so->so_vnet->vnet_sockcnt--;
463 #endif
464 	mtx_unlock(&so_global_mtx);
465 #ifdef MAC
466 	mac_socket_destroy(so);
467 #endif
468 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
469 
470 	crfree(so->so_cred);
471 	khelp_destroy_osd(&so->osd);
472 	if (SOLISTENING(so)) {
473 		if (so->sol_accept_filter != NULL)
474 			accept_filt_setopt(so, NULL);
475 	} else {
476 		if (so->so_rcv.sb_hiwat)
477 			(void)chgsbsize(so->so_cred->cr_uidinfo,
478 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
479 		if (so->so_snd.sb_hiwat)
480 			(void)chgsbsize(so->so_cred->cr_uidinfo,
481 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
482 		sx_destroy(&so->so_snd.sb_sx);
483 		sx_destroy(&so->so_rcv.sb_sx);
484 		SOCKBUF_LOCK_DESTROY(&so->so_snd);
485 		SOCKBUF_LOCK_DESTROY(&so->so_rcv);
486 	}
487 	mtx_destroy(&so->so_lock);
488 	uma_zfree(socket_zone, so);
489 }
490 
491 /*
492  * socreate returns a socket with a ref count of 1.  The socket should be
493  * closed with soclose().
494  */
495 int
496 socreate(int dom, struct socket **aso, int type, int proto,
497     struct ucred *cred, struct thread *td)
498 {
499 	struct protosw *prp;
500 	struct socket *so;
501 	int error;
502 
503 	if (proto)
504 		prp = pffindproto(dom, proto, type);
505 	else
506 		prp = pffindtype(dom, type);
507 
508 	if (prp == NULL) {
509 		/* No support for domain. */
510 		if (pffinddomain(dom) == NULL)
511 			return (EAFNOSUPPORT);
512 		/* No support for socket type. */
513 		if (proto == 0 && type != 0)
514 			return (EPROTOTYPE);
515 		return (EPROTONOSUPPORT);
516 	}
517 	if (prp->pr_usrreqs->pru_attach == NULL ||
518 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
519 		return (EPROTONOSUPPORT);
520 
521 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
522 		return (EPROTONOSUPPORT);
523 
524 	if (prp->pr_type != type)
525 		return (EPROTOTYPE);
526 	so = soalloc(CRED_TO_VNET(cred));
527 	if (so == NULL)
528 		return (ENOBUFS);
529 
530 	so->so_type = type;
531 	so->so_cred = crhold(cred);
532 	if ((prp->pr_domain->dom_family == PF_INET) ||
533 	    (prp->pr_domain->dom_family == PF_INET6) ||
534 	    (prp->pr_domain->dom_family == PF_ROUTE))
535 		so->so_fibnum = td->td_proc->p_fibnum;
536 	else
537 		so->so_fibnum = 0;
538 	so->so_proto = prp;
539 #ifdef MAC
540 	mac_socket_create(cred, so);
541 #endif
542 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
543 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
544 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
545 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
546 	/*
547 	 * Auto-sizing of socket buffers is managed by the protocols and
548 	 * the appropriate flags must be set in the pru_attach function.
549 	 */
550 	CURVNET_SET(so->so_vnet);
551 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
552 	CURVNET_RESTORE();
553 	if (error) {
554 		sodealloc(so);
555 		return (error);
556 	}
557 	soref(so);
558 	*aso = so;
559 	return (0);
560 }
561 
562 #ifdef REGRESSION
563 static int regression_sonewconn_earlytest = 1;
564 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
565     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
566 #endif
567 
568 /*
569  * When an attempt at a new connection is noted on a socket which accepts
570  * connections, sonewconn is called.  If the connection is possible (subject
571  * to space constraints, etc.) then we allocate a new structure, properly
572  * linked into the data structure of the original socket, and return this.
573  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
574  *
575  * Note: the ref count on the socket is 0 on return.
576  */
577 struct socket *
578 sonewconn(struct socket *head, int connstatus)
579 {
580 	static struct timeval lastover;
581 	static struct timeval overinterval = { 60, 0 };
582 	static int overcount;
583 
584 	struct socket *so;
585 	u_int over;
586 
587 	SOLISTEN_LOCK(head);
588 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
589 	SOLISTEN_UNLOCK(head);
590 #ifdef REGRESSION
591 	if (regression_sonewconn_earlytest && over) {
592 #else
593 	if (over) {
594 #endif
595 		overcount++;
596 
597 		if (ratecheck(&lastover, &overinterval)) {
598 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
599 			    "%i already in queue awaiting acceptance "
600 			    "(%d occurrences)\n",
601 			    __func__, head->so_pcb, head->sol_qlen, overcount);
602 
603 			overcount = 0;
604 		}
605 
606 		return (NULL);
607 	}
608 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
609 	    __func__, head));
610 	so = soalloc(head->so_vnet);
611 	if (so == NULL) {
612 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
613 		    "limit reached or out of memory\n",
614 		    __func__, head->so_pcb);
615 		return (NULL);
616 	}
617 	so->so_listen = head;
618 	so->so_type = head->so_type;
619 	so->so_linger = head->so_linger;
620 	so->so_state = head->so_state | SS_NOFDREF;
621 	so->so_fibnum = head->so_fibnum;
622 	so->so_proto = head->so_proto;
623 	so->so_cred = crhold(head->so_cred);
624 #ifdef MAC
625 	mac_socket_newconn(head, so);
626 #endif
627 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
628 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
629 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
630 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
631 	VNET_SO_ASSERT(head);
632 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
633 		sodealloc(so);
634 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
635 		    __func__, head->so_pcb);
636 		return (NULL);
637 	}
638 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
639 		sodealloc(so);
640 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
641 		    __func__, head->so_pcb);
642 		return (NULL);
643 	}
644 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
645 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
646 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
647 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
648 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
649 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
650 
651 	SOLISTEN_LOCK(head);
652 	if (head->sol_accept_filter != NULL)
653 		connstatus = 0;
654 	so->so_state |= connstatus;
655 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
656 	soref(head); /* A socket on (in)complete queue refs head. */
657 	if (connstatus) {
658 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
659 		so->so_qstate = SQ_COMP;
660 		head->sol_qlen++;
661 		solisten_wakeup(head);	/* unlocks */
662 	} else {
663 		/*
664 		 * Keep removing sockets from the head until there's room for
665 		 * us to insert on the tail.  In pre-locking revisions, this
666 		 * was a simple if(), but as we could be racing with other
667 		 * threads and soabort() requires dropping locks, we must
668 		 * loop waiting for the condition to be true.
669 		 */
670 		while (head->sol_incqlen > head->sol_qlimit) {
671 			struct socket *sp;
672 
673 			sp = TAILQ_FIRST(&head->sol_incomp);
674 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
675 			head->sol_incqlen--;
676 			SOCK_LOCK(sp);
677 			sp->so_qstate = SQ_NONE;
678 			sp->so_listen = NULL;
679 			SOCK_UNLOCK(sp);
680 			sorele(head);	/* does SOLISTEN_UNLOCK, head stays */
681 			soabort(sp);
682 			SOLISTEN_LOCK(head);
683 		}
684 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
685 		so->so_qstate = SQ_INCOMP;
686 		head->sol_incqlen++;
687 		SOLISTEN_UNLOCK(head);
688 	}
689 	return (so);
690 }
691 
692 #ifdef SCTP
693 /*
694  * Socket part of sctp_peeloff().  Detach a new socket from an
695  * association.  The new socket is returned with a reference.
696  */
697 struct socket *
698 sopeeloff(struct socket *head)
699 {
700 	struct socket *so;
701 
702 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
703 	    __func__, __LINE__, head));
704 	so = soalloc(head->so_vnet);
705 	if (so == NULL) {
706 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
707 		    "limit reached or out of memory\n",
708 		    __func__, head->so_pcb);
709 		return (NULL);
710 	}
711 	so->so_type = head->so_type;
712 	so->so_options = head->so_options;
713 	so->so_linger = head->so_linger;
714 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
715 	so->so_fibnum = head->so_fibnum;
716 	so->so_proto = head->so_proto;
717 	so->so_cred = crhold(head->so_cred);
718 #ifdef MAC
719 	mac_socket_newconn(head, so);
720 #endif
721 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
722 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
723 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
724 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
725 	VNET_SO_ASSERT(head);
726 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
727 		sodealloc(so);
728 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
729 		    __func__, head->so_pcb);
730 		return (NULL);
731 	}
732 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
733 		sodealloc(so);
734 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
735 		    __func__, head->so_pcb);
736 		return (NULL);
737 	}
738 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
739 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
740 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
741 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
742 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
743 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
744 
745 	soref(so);
746 
747 	return (so);
748 }
749 #endif	/* SCTP */
750 
751 int
752 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
753 {
754 	int error;
755 
756 	CURVNET_SET(so->so_vnet);
757 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
758 	CURVNET_RESTORE();
759 	return (error);
760 }
761 
762 int
763 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
764 {
765 	int error;
766 
767 	CURVNET_SET(so->so_vnet);
768 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
769 	CURVNET_RESTORE();
770 	return (error);
771 }
772 
773 /*
774  * solisten() transitions a socket from a non-listening state to a listening
775  * state, but can also be used to update the listen queue depth on an
776  * existing listen socket.  The protocol will call back into the sockets
777  * layer using solisten_proto_check() and solisten_proto() to check and set
778  * socket-layer listen state.  Call backs are used so that the protocol can
779  * acquire both protocol and socket layer locks in whatever order is required
780  * by the protocol.
781  *
782  * Protocol implementors are advised to hold the socket lock across the
783  * socket-layer test and set to avoid races at the socket layer.
784  */
785 int
786 solisten(struct socket *so, int backlog, struct thread *td)
787 {
788 	int error;
789 
790 	CURVNET_SET(so->so_vnet);
791 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
792 	CURVNET_RESTORE();
793 	return (error);
794 }
795 
796 int
797 solisten_proto_check(struct socket *so)
798 {
799 
800 	SOCK_LOCK_ASSERT(so);
801 
802 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
803 	    SS_ISDISCONNECTING))
804 		return (EINVAL);
805 	return (0);
806 }
807 
808 void
809 solisten_proto(struct socket *so, int backlog)
810 {
811 	int sbrcv_lowat, sbsnd_lowat;
812 	u_int sbrcv_hiwat, sbsnd_hiwat;
813 	short sbrcv_flags, sbsnd_flags;
814 	sbintime_t sbrcv_timeo, sbsnd_timeo;
815 
816 	SOCK_LOCK_ASSERT(so);
817 
818 	if (SOLISTENING(so))
819 		goto listening;
820 
821 	/*
822 	 * Change this socket to listening state.
823 	 */
824 	sbrcv_lowat = so->so_rcv.sb_lowat;
825 	sbsnd_lowat = so->so_snd.sb_lowat;
826 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
827 	sbsnd_hiwat = so->so_snd.sb_hiwat;
828 	sbrcv_flags = so->so_rcv.sb_flags;
829 	sbsnd_flags = so->so_snd.sb_flags;
830 	sbrcv_timeo = so->so_rcv.sb_timeo;
831 	sbsnd_timeo = so->so_snd.sb_timeo;
832 
833 	sbdestroy(&so->so_snd, so);
834 	sbdestroy(&so->so_rcv, so);
835 	sx_destroy(&so->so_snd.sb_sx);
836 	sx_destroy(&so->so_rcv.sb_sx);
837 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
838 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
839 
840 #ifdef INVARIANTS
841 	bzero(&so->so_rcv,
842 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
843 #endif
844 
845 	so->sol_sbrcv_lowat = sbrcv_lowat;
846 	so->sol_sbsnd_lowat = sbsnd_lowat;
847 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
848 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
849 	so->sol_sbrcv_flags = sbrcv_flags;
850 	so->sol_sbsnd_flags = sbsnd_flags;
851 	so->sol_sbrcv_timeo = sbrcv_timeo;
852 	so->sol_sbsnd_timeo = sbsnd_timeo;
853 
854 	so->sol_qlen = so->sol_incqlen = 0;
855 	TAILQ_INIT(&so->sol_incomp);
856 	TAILQ_INIT(&so->sol_comp);
857 
858 	so->sol_accept_filter = NULL;
859 	so->sol_accept_filter_arg = NULL;
860 	so->sol_accept_filter_str = NULL;
861 
862 	so->sol_upcall = NULL;
863 	so->sol_upcallarg = NULL;
864 
865 	so->so_options |= SO_ACCEPTCONN;
866 
867 listening:
868 	if (backlog < 0 || backlog > somaxconn)
869 		backlog = somaxconn;
870 	so->sol_qlimit = backlog;
871 }
872 
873 /*
874  * Wakeup listeners/subsystems once we have a complete connection.
875  * Enters with lock, returns unlocked.
876  */
877 void
878 solisten_wakeup(struct socket *sol)
879 {
880 
881 	if (sol->sol_upcall != NULL)
882 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
883 	else {
884 		selwakeuppri(&sol->so_rdsel, PSOCK);
885 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
886 	}
887 	SOLISTEN_UNLOCK(sol);
888 	wakeup_one(&sol->sol_comp);
889 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
890 		pgsigio(&sol->so_sigio, SIGIO, 0);
891 }
892 
893 /*
894  * Return single connection off a listening socket queue.  Main consumer of
895  * the function is kern_accept4().  Some modules, that do their own accept
896  * management also use the function.
897  *
898  * Listening socket must be locked on entry and is returned unlocked on
899  * return.
900  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
901  */
902 int
903 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
904 {
905 	struct socket *so;
906 	int error;
907 
908 	SOLISTEN_LOCK_ASSERT(head);
909 
910 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
911 	    head->so_error == 0) {
912 		error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
913 		    "accept", 0);
914 		if (error != 0) {
915 			SOLISTEN_UNLOCK(head);
916 			return (error);
917 		}
918 	}
919 	if (head->so_error) {
920 		error = head->so_error;
921 		head->so_error = 0;
922 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
923 		error = EWOULDBLOCK;
924 	else
925 		error = 0;
926 	if (error) {
927 		SOLISTEN_UNLOCK(head);
928 		return (error);
929 	}
930 	so = TAILQ_FIRST(&head->sol_comp);
931 	SOCK_LOCK(so);
932 	KASSERT(so->so_qstate == SQ_COMP,
933 	    ("%s: so %p not SQ_COMP", __func__, so));
934 	soref(so);
935 	head->sol_qlen--;
936 	so->so_qstate = SQ_NONE;
937 	so->so_listen = NULL;
938 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
939 	if (flags & ACCEPT4_INHERIT)
940 		so->so_state |= (head->so_state & SS_NBIO);
941 	else
942 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
943 	SOCK_UNLOCK(so);
944 	sorele(head);
945 
946 	*ret = so;
947 	return (0);
948 }
949 
950 /*
951  * Evaluate the reference count and named references on a socket; if no
952  * references remain, free it.  This should be called whenever a reference is
953  * released, such as in sorele(), but also when named reference flags are
954  * cleared in socket or protocol code.
955  *
956  * sofree() will free the socket if:
957  *
958  * - There are no outstanding file descriptor references or related consumers
959  *   (so_count == 0).
960  *
961  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
962  *
963  * - The protocol does not have an outstanding strong reference on the socket
964  *   (SS_PROTOREF).
965  *
966  * - The socket is not in a completed connection queue, so a process has been
967  *   notified that it is present.  If it is removed, the user process may
968  *   block in accept() despite select() saying the socket was ready.
969  */
970 void
971 sofree(struct socket *so)
972 {
973 	struct protosw *pr = so->so_proto;
974 
975 	SOCK_LOCK_ASSERT(so);
976 
977 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
978 	    (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
979 		SOCK_UNLOCK(so);
980 		return;
981 	}
982 
983 	if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
984 		struct socket *sol;
985 
986 		sol = so->so_listen;
987 		KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
988 
989 		/*
990 		 * To solve race between close of a listening socket and
991 		 * a socket on its incomplete queue, we need to lock both.
992 		 * The order is first listening socket, then regular.
993 		 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
994 		 * function and the listening socket are the only pointers
995 		 * to so.  To preserve so and sol, we reference both and then
996 		 * relock.
997 		 * After relock the socket may not move to so_comp since it
998 		 * doesn't have PCB already, but it may be removed from
999 		 * so_incomp. If that happens, we share responsiblity on
1000 		 * freeing the socket, but soclose() has already removed
1001 		 * it from queue.
1002 		 */
1003 		soref(sol);
1004 		soref(so);
1005 		SOCK_UNLOCK(so);
1006 		SOLISTEN_LOCK(sol);
1007 		SOCK_LOCK(so);
1008 		if (so->so_qstate == SQ_INCOMP) {
1009 			KASSERT(so->so_listen == sol,
1010 			    ("%s: so %p migrated out of sol %p",
1011 			    __func__, so, sol));
1012 			TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1013 			sol->sol_incqlen--;
1014 			/* This is guarenteed not to be the last. */
1015 			refcount_release(&sol->so_count);
1016 			so->so_qstate = SQ_NONE;
1017 			so->so_listen = NULL;
1018 		} else
1019 			KASSERT(so->so_listen == NULL,
1020 			    ("%s: so %p not on (in)comp with so_listen",
1021 			    __func__, so));
1022 		sorele(sol);
1023 		KASSERT(so->so_count == 1,
1024 		    ("%s: so %p count %u", __func__, so, so->so_count));
1025 		so->so_count = 0;
1026 	}
1027 	if (SOLISTENING(so))
1028 		so->so_error = ECONNABORTED;
1029 	SOCK_UNLOCK(so);
1030 
1031 	if (so->so_dtor != NULL)
1032 		so->so_dtor(so);
1033 
1034 	VNET_SO_ASSERT(so);
1035 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1036 		(*pr->pr_domain->dom_dispose)(so);
1037 	if (pr->pr_usrreqs->pru_detach != NULL)
1038 		(*pr->pr_usrreqs->pru_detach)(so);
1039 
1040 	/*
1041 	 * From this point on, we assume that no other references to this
1042 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1043 	 * to be acquired or held.
1044 	 *
1045 	 * We used to do a lot of socket buffer and socket locking here, as
1046 	 * well as invoke sorflush() and perform wakeups.  The direct call to
1047 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
1048 	 * necessary from sorflush().
1049 	 *
1050 	 * Notice that the socket buffer and kqueue state are torn down
1051 	 * before calling pru_detach.  This means that protocols shold not
1052 	 * assume they can perform socket wakeups, etc, in their detach code.
1053 	 */
1054 	if (!SOLISTENING(so)) {
1055 		sbdestroy(&so->so_snd, so);
1056 		sbdestroy(&so->so_rcv, so);
1057 	}
1058 	seldrain(&so->so_rdsel);
1059 	seldrain(&so->so_wrsel);
1060 	knlist_destroy(&so->so_rdsel.si_note);
1061 	knlist_destroy(&so->so_wrsel.si_note);
1062 	sodealloc(so);
1063 }
1064 
1065 /*
1066  * Close a socket on last file table reference removal.  Initiate disconnect
1067  * if connected.  Free socket when disconnect complete.
1068  *
1069  * This function will sorele() the socket.  Note that soclose() may be called
1070  * prior to the ref count reaching zero.  The actual socket structure will
1071  * not be freed until the ref count reaches zero.
1072  */
1073 int
1074 soclose(struct socket *so)
1075 {
1076 	struct accept_queue lqueue;
1077 	bool listening;
1078 	int error = 0;
1079 
1080 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1081 
1082 	CURVNET_SET(so->so_vnet);
1083 	funsetown(&so->so_sigio);
1084 	if (so->so_state & SS_ISCONNECTED) {
1085 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1086 			error = sodisconnect(so);
1087 			if (error) {
1088 				if (error == ENOTCONN)
1089 					error = 0;
1090 				goto drop;
1091 			}
1092 		}
1093 		if (so->so_options & SO_LINGER) {
1094 			if ((so->so_state & SS_ISDISCONNECTING) &&
1095 			    (so->so_state & SS_NBIO))
1096 				goto drop;
1097 			while (so->so_state & SS_ISCONNECTED) {
1098 				error = tsleep(&so->so_timeo,
1099 				    PSOCK | PCATCH, "soclos",
1100 				    so->so_linger * hz);
1101 				if (error)
1102 					break;
1103 			}
1104 		}
1105 	}
1106 
1107 drop:
1108 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1109 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1110 
1111 	SOCK_LOCK(so);
1112 	if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1113 		struct socket *sp;
1114 
1115 		TAILQ_INIT(&lqueue);
1116 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1117 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1118 
1119 		so->sol_qlen = so->sol_incqlen = 0;
1120 
1121 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1122 			SOCK_LOCK(sp);
1123 			sp->so_qstate = SQ_NONE;
1124 			sp->so_listen = NULL;
1125 			SOCK_UNLOCK(sp);
1126 			/* Guaranteed not to be the last. */
1127 			refcount_release(&so->so_count);
1128 		}
1129 	}
1130 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1131 	so->so_state |= SS_NOFDREF;
1132 	sorele(so);
1133 	if (listening) {
1134 		struct socket *sp;
1135 
1136 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1137 			SOCK_LOCK(sp);
1138 			if (sp->so_count == 0) {
1139 				SOCK_UNLOCK(sp);
1140 				soabort(sp);
1141 			} else
1142 				/* sp is now in sofree() */
1143 				SOCK_UNLOCK(sp);
1144 		}
1145 	}
1146 	CURVNET_RESTORE();
1147 	return (error);
1148 }
1149 
1150 /*
1151  * soabort() is used to abruptly tear down a connection, such as when a
1152  * resource limit is reached (listen queue depth exceeded), or if a listen
1153  * socket is closed while there are sockets waiting to be accepted.
1154  *
1155  * This interface is tricky, because it is called on an unreferenced socket,
1156  * and must be called only by a thread that has actually removed the socket
1157  * from the listen queue it was on, or races with other threads are risked.
1158  *
1159  * This interface will call into the protocol code, so must not be called
1160  * with any socket locks held.  Protocols do call it while holding their own
1161  * recursible protocol mutexes, but this is something that should be subject
1162  * to review in the future.
1163  */
1164 void
1165 soabort(struct socket *so)
1166 {
1167 
1168 	/*
1169 	 * In as much as is possible, assert that no references to this
1170 	 * socket are held.  This is not quite the same as asserting that the
1171 	 * current thread is responsible for arranging for no references, but
1172 	 * is as close as we can get for now.
1173 	 */
1174 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1175 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1176 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1177 	KASSERT(so->so_qstate == SQ_NONE, ("soabort: !SQ_NONE"));
1178 	VNET_SO_ASSERT(so);
1179 
1180 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1181 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1182 	SOCK_LOCK(so);
1183 	sofree(so);
1184 }
1185 
1186 int
1187 soaccept(struct socket *so, struct sockaddr **nam)
1188 {
1189 	int error;
1190 
1191 	SOCK_LOCK(so);
1192 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1193 	so->so_state &= ~SS_NOFDREF;
1194 	SOCK_UNLOCK(so);
1195 
1196 	CURVNET_SET(so->so_vnet);
1197 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1198 	CURVNET_RESTORE();
1199 	return (error);
1200 }
1201 
1202 int
1203 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1204 {
1205 
1206 	return (soconnectat(AT_FDCWD, so, nam, td));
1207 }
1208 
1209 int
1210 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1211 {
1212 	int error;
1213 
1214 	if (so->so_options & SO_ACCEPTCONN)
1215 		return (EOPNOTSUPP);
1216 
1217 	CURVNET_SET(so->so_vnet);
1218 	/*
1219 	 * If protocol is connection-based, can only connect once.
1220 	 * Otherwise, if connected, try to disconnect first.  This allows
1221 	 * user to disconnect by connecting to, e.g., a null address.
1222 	 */
1223 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1224 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1225 	    (error = sodisconnect(so)))) {
1226 		error = EISCONN;
1227 	} else {
1228 		/*
1229 		 * Prevent accumulated error from previous connection from
1230 		 * biting us.
1231 		 */
1232 		so->so_error = 0;
1233 		if (fd == AT_FDCWD) {
1234 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1235 			    nam, td);
1236 		} else {
1237 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1238 			    so, nam, td);
1239 		}
1240 	}
1241 	CURVNET_RESTORE();
1242 
1243 	return (error);
1244 }
1245 
1246 int
1247 soconnect2(struct socket *so1, struct socket *so2)
1248 {
1249 	int error;
1250 
1251 	CURVNET_SET(so1->so_vnet);
1252 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1253 	CURVNET_RESTORE();
1254 	return (error);
1255 }
1256 
1257 int
1258 sodisconnect(struct socket *so)
1259 {
1260 	int error;
1261 
1262 	if ((so->so_state & SS_ISCONNECTED) == 0)
1263 		return (ENOTCONN);
1264 	if (so->so_state & SS_ISDISCONNECTING)
1265 		return (EALREADY);
1266 	VNET_SO_ASSERT(so);
1267 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1268 	return (error);
1269 }
1270 
1271 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1272 
1273 int
1274 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1275     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1276 {
1277 	long space;
1278 	ssize_t resid;
1279 	int clen = 0, error, dontroute;
1280 
1281 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1282 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1283 	    ("sosend_dgram: !PR_ATOMIC"));
1284 
1285 	if (uio != NULL)
1286 		resid = uio->uio_resid;
1287 	else
1288 		resid = top->m_pkthdr.len;
1289 	/*
1290 	 * In theory resid should be unsigned.  However, space must be
1291 	 * signed, as it might be less than 0 if we over-committed, and we
1292 	 * must use a signed comparison of space and resid.  On the other
1293 	 * hand, a negative resid causes us to loop sending 0-length
1294 	 * segments to the protocol.
1295 	 */
1296 	if (resid < 0) {
1297 		error = EINVAL;
1298 		goto out;
1299 	}
1300 
1301 	dontroute =
1302 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1303 	if (td != NULL)
1304 		td->td_ru.ru_msgsnd++;
1305 	if (control != NULL)
1306 		clen = control->m_len;
1307 
1308 	SOCKBUF_LOCK(&so->so_snd);
1309 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1310 		SOCKBUF_UNLOCK(&so->so_snd);
1311 		error = EPIPE;
1312 		goto out;
1313 	}
1314 	if (so->so_error) {
1315 		error = so->so_error;
1316 		so->so_error = 0;
1317 		SOCKBUF_UNLOCK(&so->so_snd);
1318 		goto out;
1319 	}
1320 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1321 		/*
1322 		 * `sendto' and `sendmsg' is allowed on a connection-based
1323 		 * socket if it supports implied connect.  Return ENOTCONN if
1324 		 * not connected and no address is supplied.
1325 		 */
1326 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1327 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1328 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1329 			    !(resid == 0 && clen != 0)) {
1330 				SOCKBUF_UNLOCK(&so->so_snd);
1331 				error = ENOTCONN;
1332 				goto out;
1333 			}
1334 		} else if (addr == NULL) {
1335 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1336 				error = ENOTCONN;
1337 			else
1338 				error = EDESTADDRREQ;
1339 			SOCKBUF_UNLOCK(&so->so_snd);
1340 			goto out;
1341 		}
1342 	}
1343 
1344 	/*
1345 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1346 	 * problem and need fixing.
1347 	 */
1348 	space = sbspace(&so->so_snd);
1349 	if (flags & MSG_OOB)
1350 		space += 1024;
1351 	space -= clen;
1352 	SOCKBUF_UNLOCK(&so->so_snd);
1353 	if (resid > space) {
1354 		error = EMSGSIZE;
1355 		goto out;
1356 	}
1357 	if (uio == NULL) {
1358 		resid = 0;
1359 		if (flags & MSG_EOR)
1360 			top->m_flags |= M_EOR;
1361 	} else {
1362 		/*
1363 		 * Copy the data from userland into a mbuf chain.
1364 		 * If no data is to be copied in, a single empty mbuf
1365 		 * is returned.
1366 		 */
1367 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1368 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1369 		if (top == NULL) {
1370 			error = EFAULT;	/* only possible error */
1371 			goto out;
1372 		}
1373 		space -= resid - uio->uio_resid;
1374 		resid = uio->uio_resid;
1375 	}
1376 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1377 	/*
1378 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1379 	 * than with.
1380 	 */
1381 	if (dontroute) {
1382 		SOCK_LOCK(so);
1383 		so->so_options |= SO_DONTROUTE;
1384 		SOCK_UNLOCK(so);
1385 	}
1386 	/*
1387 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1388 	 * of date.  We could have received a reset packet in an interrupt or
1389 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1390 	 * probably recheck again inside the locking protection here, but
1391 	 * there are probably other places that this also happens.  We must
1392 	 * rethink this.
1393 	 */
1394 	VNET_SO_ASSERT(so);
1395 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1396 	    (flags & MSG_OOB) ? PRUS_OOB :
1397 	/*
1398 	 * If the user set MSG_EOF, the protocol understands this flag and
1399 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1400 	 */
1401 	    ((flags & MSG_EOF) &&
1402 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1403 	     (resid <= 0)) ?
1404 		PRUS_EOF :
1405 		/* If there is more to send set PRUS_MORETOCOME */
1406 		(flags & MSG_MORETOCOME) ||
1407 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1408 		top, addr, control, td);
1409 	if (dontroute) {
1410 		SOCK_LOCK(so);
1411 		so->so_options &= ~SO_DONTROUTE;
1412 		SOCK_UNLOCK(so);
1413 	}
1414 	clen = 0;
1415 	control = NULL;
1416 	top = NULL;
1417 out:
1418 	if (top != NULL)
1419 		m_freem(top);
1420 	if (control != NULL)
1421 		m_freem(control);
1422 	return (error);
1423 }
1424 
1425 /*
1426  * Send on a socket.  If send must go all at once and message is larger than
1427  * send buffering, then hard error.  Lock against other senders.  If must go
1428  * all at once and not enough room now, then inform user that this would
1429  * block and do nothing.  Otherwise, if nonblocking, send as much as
1430  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1431  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1432  * in mbuf chain must be small enough to send all at once.
1433  *
1434  * Returns nonzero on error, timeout or signal; callers must check for short
1435  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1436  * on return.
1437  */
1438 int
1439 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1440     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1441 {
1442 	long space;
1443 	ssize_t resid;
1444 	int clen = 0, error, dontroute;
1445 	int atomic = sosendallatonce(so) || top;
1446 
1447 	if (uio != NULL)
1448 		resid = uio->uio_resid;
1449 	else
1450 		resid = top->m_pkthdr.len;
1451 	/*
1452 	 * In theory resid should be unsigned.  However, space must be
1453 	 * signed, as it might be less than 0 if we over-committed, and we
1454 	 * must use a signed comparison of space and resid.  On the other
1455 	 * hand, a negative resid causes us to loop sending 0-length
1456 	 * segments to the protocol.
1457 	 *
1458 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1459 	 * type sockets since that's an error.
1460 	 */
1461 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1462 		error = EINVAL;
1463 		goto out;
1464 	}
1465 
1466 	dontroute =
1467 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1468 	    (so->so_proto->pr_flags & PR_ATOMIC);
1469 	if (td != NULL)
1470 		td->td_ru.ru_msgsnd++;
1471 	if (control != NULL)
1472 		clen = control->m_len;
1473 
1474 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1475 	if (error)
1476 		goto out;
1477 
1478 restart:
1479 	do {
1480 		SOCKBUF_LOCK(&so->so_snd);
1481 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1482 			SOCKBUF_UNLOCK(&so->so_snd);
1483 			error = EPIPE;
1484 			goto release;
1485 		}
1486 		if (so->so_error) {
1487 			error = so->so_error;
1488 			so->so_error = 0;
1489 			SOCKBUF_UNLOCK(&so->so_snd);
1490 			goto release;
1491 		}
1492 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1493 			/*
1494 			 * `sendto' and `sendmsg' is allowed on a connection-
1495 			 * based socket if it supports implied connect.
1496 			 * Return ENOTCONN if not connected and no address is
1497 			 * supplied.
1498 			 */
1499 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1500 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1501 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1502 				    !(resid == 0 && clen != 0)) {
1503 					SOCKBUF_UNLOCK(&so->so_snd);
1504 					error = ENOTCONN;
1505 					goto release;
1506 				}
1507 			} else if (addr == NULL) {
1508 				SOCKBUF_UNLOCK(&so->so_snd);
1509 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1510 					error = ENOTCONN;
1511 				else
1512 					error = EDESTADDRREQ;
1513 				goto release;
1514 			}
1515 		}
1516 		space = sbspace(&so->so_snd);
1517 		if (flags & MSG_OOB)
1518 			space += 1024;
1519 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1520 		    clen > so->so_snd.sb_hiwat) {
1521 			SOCKBUF_UNLOCK(&so->so_snd);
1522 			error = EMSGSIZE;
1523 			goto release;
1524 		}
1525 		if (space < resid + clen &&
1526 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1527 			if ((so->so_state & SS_NBIO) ||
1528 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1529 				SOCKBUF_UNLOCK(&so->so_snd);
1530 				error = EWOULDBLOCK;
1531 				goto release;
1532 			}
1533 			error = sbwait(&so->so_snd);
1534 			SOCKBUF_UNLOCK(&so->so_snd);
1535 			if (error)
1536 				goto release;
1537 			goto restart;
1538 		}
1539 		SOCKBUF_UNLOCK(&so->so_snd);
1540 		space -= clen;
1541 		do {
1542 			if (uio == NULL) {
1543 				resid = 0;
1544 				if (flags & MSG_EOR)
1545 					top->m_flags |= M_EOR;
1546 			} else {
1547 				/*
1548 				 * Copy the data from userland into a mbuf
1549 				 * chain.  If resid is 0, which can happen
1550 				 * only if we have control to send, then
1551 				 * a single empty mbuf is returned.  This
1552 				 * is a workaround to prevent protocol send
1553 				 * methods to panic.
1554 				 */
1555 				top = m_uiotombuf(uio, M_WAITOK, space,
1556 				    (atomic ? max_hdr : 0),
1557 				    (atomic ? M_PKTHDR : 0) |
1558 				    ((flags & MSG_EOR) ? M_EOR : 0));
1559 				if (top == NULL) {
1560 					error = EFAULT; /* only possible error */
1561 					goto release;
1562 				}
1563 				space -= resid - uio->uio_resid;
1564 				resid = uio->uio_resid;
1565 			}
1566 			if (dontroute) {
1567 				SOCK_LOCK(so);
1568 				so->so_options |= SO_DONTROUTE;
1569 				SOCK_UNLOCK(so);
1570 			}
1571 			/*
1572 			 * XXX all the SBS_CANTSENDMORE checks previously
1573 			 * done could be out of date.  We could have received
1574 			 * a reset packet in an interrupt or maybe we slept
1575 			 * while doing page faults in uiomove() etc.  We
1576 			 * could probably recheck again inside the locking
1577 			 * protection here, but there are probably other
1578 			 * places that this also happens.  We must rethink
1579 			 * this.
1580 			 */
1581 			VNET_SO_ASSERT(so);
1582 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1583 			    (flags & MSG_OOB) ? PRUS_OOB :
1584 			/*
1585 			 * If the user set MSG_EOF, the protocol understands
1586 			 * this flag and nothing left to send then use
1587 			 * PRU_SEND_EOF instead of PRU_SEND.
1588 			 */
1589 			    ((flags & MSG_EOF) &&
1590 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1591 			     (resid <= 0)) ?
1592 				PRUS_EOF :
1593 			/* If there is more to send set PRUS_MORETOCOME. */
1594 			    (flags & MSG_MORETOCOME) ||
1595 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1596 			    top, addr, control, td);
1597 			if (dontroute) {
1598 				SOCK_LOCK(so);
1599 				so->so_options &= ~SO_DONTROUTE;
1600 				SOCK_UNLOCK(so);
1601 			}
1602 			clen = 0;
1603 			control = NULL;
1604 			top = NULL;
1605 			if (error)
1606 				goto release;
1607 		} while (resid && space > 0);
1608 	} while (resid);
1609 
1610 release:
1611 	sbunlock(&so->so_snd);
1612 out:
1613 	if (top != NULL)
1614 		m_freem(top);
1615 	if (control != NULL)
1616 		m_freem(control);
1617 	return (error);
1618 }
1619 
1620 int
1621 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1622     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1623 {
1624 	int error;
1625 
1626 	CURVNET_SET(so->so_vnet);
1627 	if (!SOLISTENING(so))
1628 		error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1629 		    top, control, flags, td);
1630 	else {
1631 		m_freem(top);
1632 		m_freem(control);
1633 		error = ENOTCONN;
1634 	}
1635 	CURVNET_RESTORE();
1636 	return (error);
1637 }
1638 
1639 /*
1640  * The part of soreceive() that implements reading non-inline out-of-band
1641  * data from a socket.  For more complete comments, see soreceive(), from
1642  * which this code originated.
1643  *
1644  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1645  * unable to return an mbuf chain to the caller.
1646  */
1647 static int
1648 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1649 {
1650 	struct protosw *pr = so->so_proto;
1651 	struct mbuf *m;
1652 	int error;
1653 
1654 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1655 	VNET_SO_ASSERT(so);
1656 
1657 	m = m_get(M_WAITOK, MT_DATA);
1658 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1659 	if (error)
1660 		goto bad;
1661 	do {
1662 		error = uiomove(mtod(m, void *),
1663 		    (int) min(uio->uio_resid, m->m_len), uio);
1664 		m = m_free(m);
1665 	} while (uio->uio_resid && error == 0 && m);
1666 bad:
1667 	if (m != NULL)
1668 		m_freem(m);
1669 	return (error);
1670 }
1671 
1672 /*
1673  * Following replacement or removal of the first mbuf on the first mbuf chain
1674  * of a socket buffer, push necessary state changes back into the socket
1675  * buffer so that other consumers see the values consistently.  'nextrecord'
1676  * is the callers locally stored value of the original value of
1677  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1678  * NOTE: 'nextrecord' may be NULL.
1679  */
1680 static __inline void
1681 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1682 {
1683 
1684 	SOCKBUF_LOCK_ASSERT(sb);
1685 	/*
1686 	 * First, update for the new value of nextrecord.  If necessary, make
1687 	 * it the first record.
1688 	 */
1689 	if (sb->sb_mb != NULL)
1690 		sb->sb_mb->m_nextpkt = nextrecord;
1691 	else
1692 		sb->sb_mb = nextrecord;
1693 
1694 	/*
1695 	 * Now update any dependent socket buffer fields to reflect the new
1696 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1697 	 * addition of a second clause that takes care of the case where
1698 	 * sb_mb has been updated, but remains the last record.
1699 	 */
1700 	if (sb->sb_mb == NULL) {
1701 		sb->sb_mbtail = NULL;
1702 		sb->sb_lastrecord = NULL;
1703 	} else if (sb->sb_mb->m_nextpkt == NULL)
1704 		sb->sb_lastrecord = sb->sb_mb;
1705 }
1706 
1707 /*
1708  * Implement receive operations on a socket.  We depend on the way that
1709  * records are added to the sockbuf by sbappend.  In particular, each record
1710  * (mbufs linked through m_next) must begin with an address if the protocol
1711  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1712  * data, and then zero or more mbufs of data.  In order to allow parallelism
1713  * between network receive and copying to user space, as well as avoid
1714  * sleeping with a mutex held, we release the socket buffer mutex during the
1715  * user space copy.  Although the sockbuf is locked, new data may still be
1716  * appended, and thus we must maintain consistency of the sockbuf during that
1717  * time.
1718  *
1719  * The caller may receive the data as a single mbuf chain by supplying an
1720  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1721  * the count in uio_resid.
1722  */
1723 int
1724 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1725     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1726 {
1727 	struct mbuf *m, **mp;
1728 	int flags, error, offset;
1729 	ssize_t len;
1730 	struct protosw *pr = so->so_proto;
1731 	struct mbuf *nextrecord;
1732 	int moff, type = 0;
1733 	ssize_t orig_resid = uio->uio_resid;
1734 
1735 	mp = mp0;
1736 	if (psa != NULL)
1737 		*psa = NULL;
1738 	if (controlp != NULL)
1739 		*controlp = NULL;
1740 	if (flagsp != NULL)
1741 		flags = *flagsp &~ MSG_EOR;
1742 	else
1743 		flags = 0;
1744 	if (flags & MSG_OOB)
1745 		return (soreceive_rcvoob(so, uio, flags));
1746 	if (mp != NULL)
1747 		*mp = NULL;
1748 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1749 	    && uio->uio_resid) {
1750 		VNET_SO_ASSERT(so);
1751 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1752 	}
1753 
1754 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1755 	if (error)
1756 		return (error);
1757 
1758 restart:
1759 	SOCKBUF_LOCK(&so->so_rcv);
1760 	m = so->so_rcv.sb_mb;
1761 	/*
1762 	 * If we have less data than requested, block awaiting more (subject
1763 	 * to any timeout) if:
1764 	 *   1. the current count is less than the low water mark, or
1765 	 *   2. MSG_DONTWAIT is not set
1766 	 */
1767 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1768 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1769 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1770 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1771 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1772 		    ("receive: m == %p sbavail == %u",
1773 		    m, sbavail(&so->so_rcv)));
1774 		if (so->so_error) {
1775 			if (m != NULL)
1776 				goto dontblock;
1777 			error = so->so_error;
1778 			if ((flags & MSG_PEEK) == 0)
1779 				so->so_error = 0;
1780 			SOCKBUF_UNLOCK(&so->so_rcv);
1781 			goto release;
1782 		}
1783 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1784 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1785 			if (m == NULL) {
1786 				SOCKBUF_UNLOCK(&so->so_rcv);
1787 				goto release;
1788 			} else
1789 				goto dontblock;
1790 		}
1791 		for (; m != NULL; m = m->m_next)
1792 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1793 				m = so->so_rcv.sb_mb;
1794 				goto dontblock;
1795 			}
1796 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1797 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1798 			SOCKBUF_UNLOCK(&so->so_rcv);
1799 			error = ENOTCONN;
1800 			goto release;
1801 		}
1802 		if (uio->uio_resid == 0) {
1803 			SOCKBUF_UNLOCK(&so->so_rcv);
1804 			goto release;
1805 		}
1806 		if ((so->so_state & SS_NBIO) ||
1807 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1808 			SOCKBUF_UNLOCK(&so->so_rcv);
1809 			error = EWOULDBLOCK;
1810 			goto release;
1811 		}
1812 		SBLASTRECORDCHK(&so->so_rcv);
1813 		SBLASTMBUFCHK(&so->so_rcv);
1814 		error = sbwait(&so->so_rcv);
1815 		SOCKBUF_UNLOCK(&so->so_rcv);
1816 		if (error)
1817 			goto release;
1818 		goto restart;
1819 	}
1820 dontblock:
1821 	/*
1822 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1823 	 * pointer to the next record in the socket buffer.  We must keep the
1824 	 * various socket buffer pointers and local stack versions of the
1825 	 * pointers in sync, pushing out modifications before dropping the
1826 	 * socket buffer mutex, and re-reading them when picking it up.
1827 	 *
1828 	 * Otherwise, we will race with the network stack appending new data
1829 	 * or records onto the socket buffer by using inconsistent/stale
1830 	 * versions of the field, possibly resulting in socket buffer
1831 	 * corruption.
1832 	 *
1833 	 * By holding the high-level sblock(), we prevent simultaneous
1834 	 * readers from pulling off the front of the socket buffer.
1835 	 */
1836 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1837 	if (uio->uio_td)
1838 		uio->uio_td->td_ru.ru_msgrcv++;
1839 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1840 	SBLASTRECORDCHK(&so->so_rcv);
1841 	SBLASTMBUFCHK(&so->so_rcv);
1842 	nextrecord = m->m_nextpkt;
1843 	if (pr->pr_flags & PR_ADDR) {
1844 		KASSERT(m->m_type == MT_SONAME,
1845 		    ("m->m_type == %d", m->m_type));
1846 		orig_resid = 0;
1847 		if (psa != NULL)
1848 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1849 			    M_NOWAIT);
1850 		if (flags & MSG_PEEK) {
1851 			m = m->m_next;
1852 		} else {
1853 			sbfree(&so->so_rcv, m);
1854 			so->so_rcv.sb_mb = m_free(m);
1855 			m = so->so_rcv.sb_mb;
1856 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1857 		}
1858 	}
1859 
1860 	/*
1861 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1862 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1863 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1864 	 * perform externalization (or freeing if controlp == NULL).
1865 	 */
1866 	if (m != NULL && m->m_type == MT_CONTROL) {
1867 		struct mbuf *cm = NULL, *cmn;
1868 		struct mbuf **cme = &cm;
1869 
1870 		do {
1871 			if (flags & MSG_PEEK) {
1872 				if (controlp != NULL) {
1873 					*controlp = m_copym(m, 0, m->m_len,
1874 					    M_NOWAIT);
1875 					controlp = &(*controlp)->m_next;
1876 				}
1877 				m = m->m_next;
1878 			} else {
1879 				sbfree(&so->so_rcv, m);
1880 				so->so_rcv.sb_mb = m->m_next;
1881 				m->m_next = NULL;
1882 				*cme = m;
1883 				cme = &(*cme)->m_next;
1884 				m = so->so_rcv.sb_mb;
1885 			}
1886 		} while (m != NULL && m->m_type == MT_CONTROL);
1887 		if ((flags & MSG_PEEK) == 0)
1888 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1889 		while (cm != NULL) {
1890 			cmn = cm->m_next;
1891 			cm->m_next = NULL;
1892 			if (pr->pr_domain->dom_externalize != NULL) {
1893 				SOCKBUF_UNLOCK(&so->so_rcv);
1894 				VNET_SO_ASSERT(so);
1895 				error = (*pr->pr_domain->dom_externalize)
1896 				    (cm, controlp, flags);
1897 				SOCKBUF_LOCK(&so->so_rcv);
1898 			} else if (controlp != NULL)
1899 				*controlp = cm;
1900 			else
1901 				m_freem(cm);
1902 			if (controlp != NULL) {
1903 				orig_resid = 0;
1904 				while (*controlp != NULL)
1905 					controlp = &(*controlp)->m_next;
1906 			}
1907 			cm = cmn;
1908 		}
1909 		if (m != NULL)
1910 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1911 		else
1912 			nextrecord = so->so_rcv.sb_mb;
1913 		orig_resid = 0;
1914 	}
1915 	if (m != NULL) {
1916 		if ((flags & MSG_PEEK) == 0) {
1917 			KASSERT(m->m_nextpkt == nextrecord,
1918 			    ("soreceive: post-control, nextrecord !sync"));
1919 			if (nextrecord == NULL) {
1920 				KASSERT(so->so_rcv.sb_mb == m,
1921 				    ("soreceive: post-control, sb_mb!=m"));
1922 				KASSERT(so->so_rcv.sb_lastrecord == m,
1923 				    ("soreceive: post-control, lastrecord!=m"));
1924 			}
1925 		}
1926 		type = m->m_type;
1927 		if (type == MT_OOBDATA)
1928 			flags |= MSG_OOB;
1929 	} else {
1930 		if ((flags & MSG_PEEK) == 0) {
1931 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1932 			    ("soreceive: sb_mb != nextrecord"));
1933 			if (so->so_rcv.sb_mb == NULL) {
1934 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1935 				    ("soreceive: sb_lastercord != NULL"));
1936 			}
1937 		}
1938 	}
1939 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1940 	SBLASTRECORDCHK(&so->so_rcv);
1941 	SBLASTMBUFCHK(&so->so_rcv);
1942 
1943 	/*
1944 	 * Now continue to read any data mbufs off of the head of the socket
1945 	 * buffer until the read request is satisfied.  Note that 'type' is
1946 	 * used to store the type of any mbuf reads that have happened so far
1947 	 * such that soreceive() can stop reading if the type changes, which
1948 	 * causes soreceive() to return only one of regular data and inline
1949 	 * out-of-band data in a single socket receive operation.
1950 	 */
1951 	moff = 0;
1952 	offset = 0;
1953 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1954 	    && error == 0) {
1955 		/*
1956 		 * If the type of mbuf has changed since the last mbuf
1957 		 * examined ('type'), end the receive operation.
1958 		 */
1959 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1960 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1961 			if (type != m->m_type)
1962 				break;
1963 		} else if (type == MT_OOBDATA)
1964 			break;
1965 		else
1966 		    KASSERT(m->m_type == MT_DATA,
1967 			("m->m_type == %d", m->m_type));
1968 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1969 		len = uio->uio_resid;
1970 		if (so->so_oobmark && len > so->so_oobmark - offset)
1971 			len = so->so_oobmark - offset;
1972 		if (len > m->m_len - moff)
1973 			len = m->m_len - moff;
1974 		/*
1975 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1976 		 * them out via the uio, then free.  Sockbuf must be
1977 		 * consistent here (points to current mbuf, it points to next
1978 		 * record) when we drop priority; we must note any additions
1979 		 * to the sockbuf when we block interrupts again.
1980 		 */
1981 		if (mp == NULL) {
1982 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1983 			SBLASTRECORDCHK(&so->so_rcv);
1984 			SBLASTMBUFCHK(&so->so_rcv);
1985 			SOCKBUF_UNLOCK(&so->so_rcv);
1986 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1987 			SOCKBUF_LOCK(&so->so_rcv);
1988 			if (error) {
1989 				/*
1990 				 * The MT_SONAME mbuf has already been removed
1991 				 * from the record, so it is necessary to
1992 				 * remove the data mbufs, if any, to preserve
1993 				 * the invariant in the case of PR_ADDR that
1994 				 * requires MT_SONAME mbufs at the head of
1995 				 * each record.
1996 				 */
1997 				if (pr->pr_flags & PR_ATOMIC &&
1998 				    ((flags & MSG_PEEK) == 0))
1999 					(void)sbdroprecord_locked(&so->so_rcv);
2000 				SOCKBUF_UNLOCK(&so->so_rcv);
2001 				goto release;
2002 			}
2003 		} else
2004 			uio->uio_resid -= len;
2005 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2006 		if (len == m->m_len - moff) {
2007 			if (m->m_flags & M_EOR)
2008 				flags |= MSG_EOR;
2009 			if (flags & MSG_PEEK) {
2010 				m = m->m_next;
2011 				moff = 0;
2012 			} else {
2013 				nextrecord = m->m_nextpkt;
2014 				sbfree(&so->so_rcv, m);
2015 				if (mp != NULL) {
2016 					m->m_nextpkt = NULL;
2017 					*mp = m;
2018 					mp = &m->m_next;
2019 					so->so_rcv.sb_mb = m = m->m_next;
2020 					*mp = NULL;
2021 				} else {
2022 					so->so_rcv.sb_mb = m_free(m);
2023 					m = so->so_rcv.sb_mb;
2024 				}
2025 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2026 				SBLASTRECORDCHK(&so->so_rcv);
2027 				SBLASTMBUFCHK(&so->so_rcv);
2028 			}
2029 		} else {
2030 			if (flags & MSG_PEEK)
2031 				moff += len;
2032 			else {
2033 				if (mp != NULL) {
2034 					if (flags & MSG_DONTWAIT) {
2035 						*mp = m_copym(m, 0, len,
2036 						    M_NOWAIT);
2037 						if (*mp == NULL) {
2038 							/*
2039 							 * m_copym() couldn't
2040 							 * allocate an mbuf.
2041 							 * Adjust uio_resid back
2042 							 * (it was adjusted
2043 							 * down by len bytes,
2044 							 * which we didn't end
2045 							 * up "copying" over).
2046 							 */
2047 							uio->uio_resid += len;
2048 							break;
2049 						}
2050 					} else {
2051 						SOCKBUF_UNLOCK(&so->so_rcv);
2052 						*mp = m_copym(m, 0, len,
2053 						    M_WAITOK);
2054 						SOCKBUF_LOCK(&so->so_rcv);
2055 					}
2056 				}
2057 				sbcut_locked(&so->so_rcv, len);
2058 			}
2059 		}
2060 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2061 		if (so->so_oobmark) {
2062 			if ((flags & MSG_PEEK) == 0) {
2063 				so->so_oobmark -= len;
2064 				if (so->so_oobmark == 0) {
2065 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2066 					break;
2067 				}
2068 			} else {
2069 				offset += len;
2070 				if (offset == so->so_oobmark)
2071 					break;
2072 			}
2073 		}
2074 		if (flags & MSG_EOR)
2075 			break;
2076 		/*
2077 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2078 		 * must not quit until "uio->uio_resid == 0" or an error
2079 		 * termination.  If a signal/timeout occurs, return with a
2080 		 * short count but without error.  Keep sockbuf locked
2081 		 * against other readers.
2082 		 */
2083 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2084 		    !sosendallatonce(so) && nextrecord == NULL) {
2085 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2086 			if (so->so_error ||
2087 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2088 				break;
2089 			/*
2090 			 * Notify the protocol that some data has been
2091 			 * drained before blocking.
2092 			 */
2093 			if (pr->pr_flags & PR_WANTRCVD) {
2094 				SOCKBUF_UNLOCK(&so->so_rcv);
2095 				VNET_SO_ASSERT(so);
2096 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2097 				SOCKBUF_LOCK(&so->so_rcv);
2098 			}
2099 			SBLASTRECORDCHK(&so->so_rcv);
2100 			SBLASTMBUFCHK(&so->so_rcv);
2101 			/*
2102 			 * We could receive some data while was notifying
2103 			 * the protocol. Skip blocking in this case.
2104 			 */
2105 			if (so->so_rcv.sb_mb == NULL) {
2106 				error = sbwait(&so->so_rcv);
2107 				if (error) {
2108 					SOCKBUF_UNLOCK(&so->so_rcv);
2109 					goto release;
2110 				}
2111 			}
2112 			m = so->so_rcv.sb_mb;
2113 			if (m != NULL)
2114 				nextrecord = m->m_nextpkt;
2115 		}
2116 	}
2117 
2118 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2119 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2120 		flags |= MSG_TRUNC;
2121 		if ((flags & MSG_PEEK) == 0)
2122 			(void) sbdroprecord_locked(&so->so_rcv);
2123 	}
2124 	if ((flags & MSG_PEEK) == 0) {
2125 		if (m == NULL) {
2126 			/*
2127 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2128 			 * part makes sure sb_lastrecord is up-to-date if
2129 			 * there is still data in the socket buffer.
2130 			 */
2131 			so->so_rcv.sb_mb = nextrecord;
2132 			if (so->so_rcv.sb_mb == NULL) {
2133 				so->so_rcv.sb_mbtail = NULL;
2134 				so->so_rcv.sb_lastrecord = NULL;
2135 			} else if (nextrecord->m_nextpkt == NULL)
2136 				so->so_rcv.sb_lastrecord = nextrecord;
2137 		}
2138 		SBLASTRECORDCHK(&so->so_rcv);
2139 		SBLASTMBUFCHK(&so->so_rcv);
2140 		/*
2141 		 * If soreceive() is being done from the socket callback,
2142 		 * then don't need to generate ACK to peer to update window,
2143 		 * since ACK will be generated on return to TCP.
2144 		 */
2145 		if (!(flags & MSG_SOCALLBCK) &&
2146 		    (pr->pr_flags & PR_WANTRCVD)) {
2147 			SOCKBUF_UNLOCK(&so->so_rcv);
2148 			VNET_SO_ASSERT(so);
2149 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2150 			SOCKBUF_LOCK(&so->so_rcv);
2151 		}
2152 	}
2153 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2154 	if (orig_resid == uio->uio_resid && orig_resid &&
2155 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2156 		SOCKBUF_UNLOCK(&so->so_rcv);
2157 		goto restart;
2158 	}
2159 	SOCKBUF_UNLOCK(&so->so_rcv);
2160 
2161 	if (flagsp != NULL)
2162 		*flagsp |= flags;
2163 release:
2164 	sbunlock(&so->so_rcv);
2165 	return (error);
2166 }
2167 
2168 /*
2169  * Optimized version of soreceive() for stream (TCP) sockets.
2170  */
2171 int
2172 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2173     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2174 {
2175 	int len = 0, error = 0, flags, oresid;
2176 	struct sockbuf *sb;
2177 	struct mbuf *m, *n = NULL;
2178 
2179 	/* We only do stream sockets. */
2180 	if (so->so_type != SOCK_STREAM)
2181 		return (EINVAL);
2182 	if (psa != NULL)
2183 		*psa = NULL;
2184 	if (flagsp != NULL)
2185 		flags = *flagsp &~ MSG_EOR;
2186 	else
2187 		flags = 0;
2188 	if (controlp != NULL)
2189 		*controlp = NULL;
2190 	if (flags & MSG_OOB)
2191 		return (soreceive_rcvoob(so, uio, flags));
2192 	if (mp0 != NULL)
2193 		*mp0 = NULL;
2194 
2195 	sb = &so->so_rcv;
2196 
2197 	/* Prevent other readers from entering the socket. */
2198 	error = sblock(sb, SBLOCKWAIT(flags));
2199 	if (error)
2200 		goto out;
2201 	SOCKBUF_LOCK(sb);
2202 
2203 	/* Easy one, no space to copyout anything. */
2204 	if (uio->uio_resid == 0) {
2205 		error = EINVAL;
2206 		goto out;
2207 	}
2208 	oresid = uio->uio_resid;
2209 
2210 	/* We will never ever get anything unless we are or were connected. */
2211 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2212 		error = ENOTCONN;
2213 		goto out;
2214 	}
2215 
2216 restart:
2217 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2218 
2219 	/* Abort if socket has reported problems. */
2220 	if (so->so_error) {
2221 		if (sbavail(sb) > 0)
2222 			goto deliver;
2223 		if (oresid > uio->uio_resid)
2224 			goto out;
2225 		error = so->so_error;
2226 		if (!(flags & MSG_PEEK))
2227 			so->so_error = 0;
2228 		goto out;
2229 	}
2230 
2231 	/* Door is closed.  Deliver what is left, if any. */
2232 	if (sb->sb_state & SBS_CANTRCVMORE) {
2233 		if (sbavail(sb) > 0)
2234 			goto deliver;
2235 		else
2236 			goto out;
2237 	}
2238 
2239 	/* Socket buffer is empty and we shall not block. */
2240 	if (sbavail(sb) == 0 &&
2241 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2242 		error = EAGAIN;
2243 		goto out;
2244 	}
2245 
2246 	/* Socket buffer got some data that we shall deliver now. */
2247 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2248 	    ((so->so_state & SS_NBIO) ||
2249 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2250 	     sbavail(sb) >= sb->sb_lowat ||
2251 	     sbavail(sb) >= uio->uio_resid ||
2252 	     sbavail(sb) >= sb->sb_hiwat) ) {
2253 		goto deliver;
2254 	}
2255 
2256 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2257 	if ((flags & MSG_WAITALL) &&
2258 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2259 		goto deliver;
2260 
2261 	/*
2262 	 * Wait and block until (more) data comes in.
2263 	 * NB: Drops the sockbuf lock during wait.
2264 	 */
2265 	error = sbwait(sb);
2266 	if (error)
2267 		goto out;
2268 	goto restart;
2269 
2270 deliver:
2271 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2272 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2273 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2274 
2275 	/* Statistics. */
2276 	if (uio->uio_td)
2277 		uio->uio_td->td_ru.ru_msgrcv++;
2278 
2279 	/* Fill uio until full or current end of socket buffer is reached. */
2280 	len = min(uio->uio_resid, sbavail(sb));
2281 	if (mp0 != NULL) {
2282 		/* Dequeue as many mbufs as possible. */
2283 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2284 			if (*mp0 == NULL)
2285 				*mp0 = sb->sb_mb;
2286 			else
2287 				m_cat(*mp0, sb->sb_mb);
2288 			for (m = sb->sb_mb;
2289 			     m != NULL && m->m_len <= len;
2290 			     m = m->m_next) {
2291 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2292 				    ("%s: m %p not available", __func__, m));
2293 				len -= m->m_len;
2294 				uio->uio_resid -= m->m_len;
2295 				sbfree(sb, m);
2296 				n = m;
2297 			}
2298 			n->m_next = NULL;
2299 			sb->sb_mb = m;
2300 			sb->sb_lastrecord = sb->sb_mb;
2301 			if (sb->sb_mb == NULL)
2302 				SB_EMPTY_FIXUP(sb);
2303 		}
2304 		/* Copy the remainder. */
2305 		if (len > 0) {
2306 			KASSERT(sb->sb_mb != NULL,
2307 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2308 
2309 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2310 			if (m == NULL)
2311 				len = 0;	/* Don't flush data from sockbuf. */
2312 			else
2313 				uio->uio_resid -= len;
2314 			if (*mp0 != NULL)
2315 				m_cat(*mp0, m);
2316 			else
2317 				*mp0 = m;
2318 			if (*mp0 == NULL) {
2319 				error = ENOBUFS;
2320 				goto out;
2321 			}
2322 		}
2323 	} else {
2324 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2325 		SOCKBUF_UNLOCK(sb);
2326 		error = m_mbuftouio(uio, sb->sb_mb, len);
2327 		SOCKBUF_LOCK(sb);
2328 		if (error)
2329 			goto out;
2330 	}
2331 	SBLASTRECORDCHK(sb);
2332 	SBLASTMBUFCHK(sb);
2333 
2334 	/*
2335 	 * Remove the delivered data from the socket buffer unless we
2336 	 * were only peeking.
2337 	 */
2338 	if (!(flags & MSG_PEEK)) {
2339 		if (len > 0)
2340 			sbdrop_locked(sb, len);
2341 
2342 		/* Notify protocol that we drained some data. */
2343 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2344 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2345 		     !(flags & MSG_SOCALLBCK))) {
2346 			SOCKBUF_UNLOCK(sb);
2347 			VNET_SO_ASSERT(so);
2348 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2349 			SOCKBUF_LOCK(sb);
2350 		}
2351 	}
2352 
2353 	/*
2354 	 * For MSG_WAITALL we may have to loop again and wait for
2355 	 * more data to come in.
2356 	 */
2357 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2358 		goto restart;
2359 out:
2360 	SOCKBUF_LOCK_ASSERT(sb);
2361 	SBLASTRECORDCHK(sb);
2362 	SBLASTMBUFCHK(sb);
2363 	SOCKBUF_UNLOCK(sb);
2364 	sbunlock(sb);
2365 	return (error);
2366 }
2367 
2368 /*
2369  * Optimized version of soreceive() for simple datagram cases from userspace.
2370  * Unlike in the stream case, we're able to drop a datagram if copyout()
2371  * fails, and because we handle datagrams atomically, we don't need to use a
2372  * sleep lock to prevent I/O interlacing.
2373  */
2374 int
2375 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2376     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2377 {
2378 	struct mbuf *m, *m2;
2379 	int flags, error;
2380 	ssize_t len;
2381 	struct protosw *pr = so->so_proto;
2382 	struct mbuf *nextrecord;
2383 
2384 	if (psa != NULL)
2385 		*psa = NULL;
2386 	if (controlp != NULL)
2387 		*controlp = NULL;
2388 	if (flagsp != NULL)
2389 		flags = *flagsp &~ MSG_EOR;
2390 	else
2391 		flags = 0;
2392 
2393 	/*
2394 	 * For any complicated cases, fall back to the full
2395 	 * soreceive_generic().
2396 	 */
2397 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2398 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2399 		    flagsp));
2400 
2401 	/*
2402 	 * Enforce restrictions on use.
2403 	 */
2404 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2405 	    ("soreceive_dgram: wantrcvd"));
2406 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2407 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2408 	    ("soreceive_dgram: SBS_RCVATMARK"));
2409 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2410 	    ("soreceive_dgram: P_CONNREQUIRED"));
2411 
2412 	/*
2413 	 * Loop blocking while waiting for a datagram.
2414 	 */
2415 	SOCKBUF_LOCK(&so->so_rcv);
2416 	while ((m = so->so_rcv.sb_mb) == NULL) {
2417 		KASSERT(sbavail(&so->so_rcv) == 0,
2418 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2419 		    sbavail(&so->so_rcv)));
2420 		if (so->so_error) {
2421 			error = so->so_error;
2422 			so->so_error = 0;
2423 			SOCKBUF_UNLOCK(&so->so_rcv);
2424 			return (error);
2425 		}
2426 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2427 		    uio->uio_resid == 0) {
2428 			SOCKBUF_UNLOCK(&so->so_rcv);
2429 			return (0);
2430 		}
2431 		if ((so->so_state & SS_NBIO) ||
2432 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2433 			SOCKBUF_UNLOCK(&so->so_rcv);
2434 			return (EWOULDBLOCK);
2435 		}
2436 		SBLASTRECORDCHK(&so->so_rcv);
2437 		SBLASTMBUFCHK(&so->so_rcv);
2438 		error = sbwait(&so->so_rcv);
2439 		if (error) {
2440 			SOCKBUF_UNLOCK(&so->so_rcv);
2441 			return (error);
2442 		}
2443 	}
2444 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2445 
2446 	if (uio->uio_td)
2447 		uio->uio_td->td_ru.ru_msgrcv++;
2448 	SBLASTRECORDCHK(&so->so_rcv);
2449 	SBLASTMBUFCHK(&so->so_rcv);
2450 	nextrecord = m->m_nextpkt;
2451 	if (nextrecord == NULL) {
2452 		KASSERT(so->so_rcv.sb_lastrecord == m,
2453 		    ("soreceive_dgram: lastrecord != m"));
2454 	}
2455 
2456 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2457 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2458 
2459 	/*
2460 	 * Pull 'm' and its chain off the front of the packet queue.
2461 	 */
2462 	so->so_rcv.sb_mb = NULL;
2463 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2464 
2465 	/*
2466 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2467 	 */
2468 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2469 		sbfree(&so->so_rcv, m2);
2470 
2471 	/*
2472 	 * Do a few last checks before we let go of the lock.
2473 	 */
2474 	SBLASTRECORDCHK(&so->so_rcv);
2475 	SBLASTMBUFCHK(&so->so_rcv);
2476 	SOCKBUF_UNLOCK(&so->so_rcv);
2477 
2478 	if (pr->pr_flags & PR_ADDR) {
2479 		KASSERT(m->m_type == MT_SONAME,
2480 		    ("m->m_type == %d", m->m_type));
2481 		if (psa != NULL)
2482 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2483 			    M_NOWAIT);
2484 		m = m_free(m);
2485 	}
2486 	if (m == NULL) {
2487 		/* XXXRW: Can this happen? */
2488 		return (0);
2489 	}
2490 
2491 	/*
2492 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2493 	 * queue.
2494 	 *
2495 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2496 	 * in the first mbuf chain on the socket buffer.  We call into the
2497 	 * protocol to perform externalization (or freeing if controlp ==
2498 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2499 	 * MT_DATA mbufs.
2500 	 */
2501 	if (m->m_type == MT_CONTROL) {
2502 		struct mbuf *cm = NULL, *cmn;
2503 		struct mbuf **cme = &cm;
2504 
2505 		do {
2506 			m2 = m->m_next;
2507 			m->m_next = NULL;
2508 			*cme = m;
2509 			cme = &(*cme)->m_next;
2510 			m = m2;
2511 		} while (m != NULL && m->m_type == MT_CONTROL);
2512 		while (cm != NULL) {
2513 			cmn = cm->m_next;
2514 			cm->m_next = NULL;
2515 			if (pr->pr_domain->dom_externalize != NULL) {
2516 				error = (*pr->pr_domain->dom_externalize)
2517 				    (cm, controlp, flags);
2518 			} else if (controlp != NULL)
2519 				*controlp = cm;
2520 			else
2521 				m_freem(cm);
2522 			if (controlp != NULL) {
2523 				while (*controlp != NULL)
2524 					controlp = &(*controlp)->m_next;
2525 			}
2526 			cm = cmn;
2527 		}
2528 	}
2529 	KASSERT(m == NULL || m->m_type == MT_DATA,
2530 	    ("soreceive_dgram: !data"));
2531 	while (m != NULL && uio->uio_resid > 0) {
2532 		len = uio->uio_resid;
2533 		if (len > m->m_len)
2534 			len = m->m_len;
2535 		error = uiomove(mtod(m, char *), (int)len, uio);
2536 		if (error) {
2537 			m_freem(m);
2538 			return (error);
2539 		}
2540 		if (len == m->m_len)
2541 			m = m_free(m);
2542 		else {
2543 			m->m_data += len;
2544 			m->m_len -= len;
2545 		}
2546 	}
2547 	if (m != NULL) {
2548 		flags |= MSG_TRUNC;
2549 		m_freem(m);
2550 	}
2551 	if (flagsp != NULL)
2552 		*flagsp |= flags;
2553 	return (0);
2554 }
2555 
2556 int
2557 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2558     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2559 {
2560 	int error;
2561 
2562 	CURVNET_SET(so->so_vnet);
2563 	if (!SOLISTENING(so))
2564 		error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2565 		    mp0, controlp, flagsp));
2566 	else
2567 		error = ENOTCONN;
2568 	CURVNET_RESTORE();
2569 	return (error);
2570 }
2571 
2572 int
2573 soshutdown(struct socket *so, int how)
2574 {
2575 	struct protosw *pr = so->so_proto;
2576 	int error, soerror_enotconn;
2577 
2578 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2579 		return (EINVAL);
2580 
2581 	soerror_enotconn = 0;
2582 	if ((so->so_state &
2583 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2584 		/*
2585 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2586 		 * invoked on a datagram sockets, however historically we would
2587 		 * actually tear socket down. This is known to be leveraged by
2588 		 * some applications to unblock process waiting in recvXXX(2)
2589 		 * by other process that it shares that socket with. Try to meet
2590 		 * both backward-compatibility and POSIX requirements by forcing
2591 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2592 		 */
2593 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2594 			return (ENOTCONN);
2595 		soerror_enotconn = 1;
2596 	}
2597 
2598 	if (SOLISTENING(so)) {
2599 		if (how != SHUT_WR) {
2600 			SOLISTEN_LOCK(so);
2601 			so->so_error = ECONNABORTED;
2602 			solisten_wakeup(so);	/* unlocks so */
2603 		}
2604 		goto done;
2605 	}
2606 
2607 	CURVNET_SET(so->so_vnet);
2608 	if (pr->pr_usrreqs->pru_flush != NULL)
2609 		(*pr->pr_usrreqs->pru_flush)(so, how);
2610 	if (how != SHUT_WR)
2611 		sorflush(so);
2612 	if (how != SHUT_RD) {
2613 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2614 		wakeup(&so->so_timeo);
2615 		CURVNET_RESTORE();
2616 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2617 	}
2618 	wakeup(&so->so_timeo);
2619 	CURVNET_RESTORE();
2620 
2621 done:
2622 	return (soerror_enotconn ? ENOTCONN : 0);
2623 }
2624 
2625 void
2626 sorflush(struct socket *so)
2627 {
2628 	struct sockbuf *sb = &so->so_rcv;
2629 	struct protosw *pr = so->so_proto;
2630 	struct socket aso;
2631 
2632 	VNET_SO_ASSERT(so);
2633 
2634 	/*
2635 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2636 	 * held, and in order to generally avoid holding the lock for a long
2637 	 * time, we make a copy of the socket buffer and clear the original
2638 	 * (except locks, state).  The new socket buffer copy won't have
2639 	 * initialized locks so we can only call routines that won't use or
2640 	 * assert those locks.
2641 	 *
2642 	 * Dislodge threads currently blocked in receive and wait to acquire
2643 	 * a lock against other simultaneous readers before clearing the
2644 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2645 	 * despite any existing socket disposition on interruptable waiting.
2646 	 */
2647 	socantrcvmore(so);
2648 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2649 
2650 	/*
2651 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2652 	 * and mutex data unchanged.
2653 	 */
2654 	SOCKBUF_LOCK(sb);
2655 	bzero(&aso, sizeof(aso));
2656 	aso.so_pcb = so->so_pcb;
2657 	bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2658 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2659 	bzero(&sb->sb_startzero,
2660 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2661 	SOCKBUF_UNLOCK(sb);
2662 	sbunlock(sb);
2663 
2664 	/*
2665 	 * Dispose of special rights and flush the copied socket.  Don't call
2666 	 * any unsafe routines (that rely on locks being initialized) on aso.
2667 	 */
2668 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2669 		(*pr->pr_domain->dom_dispose)(&aso);
2670 	sbrelease_internal(&aso.so_rcv, so);
2671 }
2672 
2673 /*
2674  * Wrapper for Socket established helper hook.
2675  * Parameters: socket, context of the hook point, hook id.
2676  */
2677 static int inline
2678 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2679 {
2680 	struct socket_hhook_data hhook_data = {
2681 		.so = so,
2682 		.hctx = hctx,
2683 		.m = NULL,
2684 		.status = 0
2685 	};
2686 
2687 	CURVNET_SET(so->so_vnet);
2688 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2689 	CURVNET_RESTORE();
2690 
2691 	/* Ugly but needed, since hhooks return void for now */
2692 	return (hhook_data.status);
2693 }
2694 
2695 /*
2696  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2697  * additional variant to handle the case where the option value needs to be
2698  * some kind of integer, but not a specific size.  In addition to their use
2699  * here, these functions are also called by the protocol-level pr_ctloutput()
2700  * routines.
2701  */
2702 int
2703 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2704 {
2705 	size_t	valsize;
2706 
2707 	/*
2708 	 * If the user gives us more than we wanted, we ignore it, but if we
2709 	 * don't get the minimum length the caller wants, we return EINVAL.
2710 	 * On success, sopt->sopt_valsize is set to however much we actually
2711 	 * retrieved.
2712 	 */
2713 	if ((valsize = sopt->sopt_valsize) < minlen)
2714 		return EINVAL;
2715 	if (valsize > len)
2716 		sopt->sopt_valsize = valsize = len;
2717 
2718 	if (sopt->sopt_td != NULL)
2719 		return (copyin(sopt->sopt_val, buf, valsize));
2720 
2721 	bcopy(sopt->sopt_val, buf, valsize);
2722 	return (0);
2723 }
2724 
2725 /*
2726  * Kernel version of setsockopt(2).
2727  *
2728  * XXX: optlen is size_t, not socklen_t
2729  */
2730 int
2731 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2732     size_t optlen)
2733 {
2734 	struct sockopt sopt;
2735 
2736 	sopt.sopt_level = level;
2737 	sopt.sopt_name = optname;
2738 	sopt.sopt_dir = SOPT_SET;
2739 	sopt.sopt_val = optval;
2740 	sopt.sopt_valsize = optlen;
2741 	sopt.sopt_td = NULL;
2742 	return (sosetopt(so, &sopt));
2743 }
2744 
2745 int
2746 sosetopt(struct socket *so, struct sockopt *sopt)
2747 {
2748 	int	error, optval;
2749 	struct	linger l;
2750 	struct	timeval tv;
2751 	sbintime_t val;
2752 	uint32_t val32;
2753 #ifdef MAC
2754 	struct mac extmac;
2755 #endif
2756 
2757 	CURVNET_SET(so->so_vnet);
2758 	error = 0;
2759 	if (sopt->sopt_level != SOL_SOCKET) {
2760 		if (so->so_proto->pr_ctloutput != NULL)
2761 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2762 		else
2763 			error = ENOPROTOOPT;
2764 	} else {
2765 		switch (sopt->sopt_name) {
2766 		case SO_ACCEPTFILTER:
2767 			error = accept_filt_setopt(so, sopt);
2768 			if (error)
2769 				goto bad;
2770 			break;
2771 
2772 		case SO_LINGER:
2773 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2774 			if (error)
2775 				goto bad;
2776 
2777 			SOCK_LOCK(so);
2778 			so->so_linger = l.l_linger;
2779 			if (l.l_onoff)
2780 				so->so_options |= SO_LINGER;
2781 			else
2782 				so->so_options &= ~SO_LINGER;
2783 			SOCK_UNLOCK(so);
2784 			break;
2785 
2786 		case SO_DEBUG:
2787 		case SO_KEEPALIVE:
2788 		case SO_DONTROUTE:
2789 		case SO_USELOOPBACK:
2790 		case SO_BROADCAST:
2791 		case SO_REUSEADDR:
2792 		case SO_REUSEPORT:
2793 		case SO_REUSEPORT_LB:
2794 		case SO_OOBINLINE:
2795 		case SO_TIMESTAMP:
2796 		case SO_BINTIME:
2797 		case SO_NOSIGPIPE:
2798 		case SO_NO_DDP:
2799 		case SO_NO_OFFLOAD:
2800 			error = sooptcopyin(sopt, &optval, sizeof optval,
2801 			    sizeof optval);
2802 			if (error)
2803 				goto bad;
2804 			SOCK_LOCK(so);
2805 			if (optval)
2806 				so->so_options |= sopt->sopt_name;
2807 			else
2808 				so->so_options &= ~sopt->sopt_name;
2809 			SOCK_UNLOCK(so);
2810 			break;
2811 
2812 		case SO_SETFIB:
2813 			error = sooptcopyin(sopt, &optval, sizeof optval,
2814 			    sizeof optval);
2815 			if (error)
2816 				goto bad;
2817 
2818 			if (optval < 0 || optval >= rt_numfibs) {
2819 				error = EINVAL;
2820 				goto bad;
2821 			}
2822 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2823 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2824 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2825 				so->so_fibnum = optval;
2826 			else
2827 				so->so_fibnum = 0;
2828 			break;
2829 
2830 		case SO_USER_COOKIE:
2831 			error = sooptcopyin(sopt, &val32, sizeof val32,
2832 			    sizeof val32);
2833 			if (error)
2834 				goto bad;
2835 			so->so_user_cookie = val32;
2836 			break;
2837 
2838 		case SO_SNDBUF:
2839 		case SO_RCVBUF:
2840 		case SO_SNDLOWAT:
2841 		case SO_RCVLOWAT:
2842 			error = sooptcopyin(sopt, &optval, sizeof optval,
2843 			    sizeof optval);
2844 			if (error)
2845 				goto bad;
2846 
2847 			/*
2848 			 * Values < 1 make no sense for any of these options,
2849 			 * so disallow them.
2850 			 */
2851 			if (optval < 1) {
2852 				error = EINVAL;
2853 				goto bad;
2854 			}
2855 
2856 			error = sbsetopt(so, sopt->sopt_name, optval);
2857 			break;
2858 
2859 		case SO_SNDTIMEO:
2860 		case SO_RCVTIMEO:
2861 #ifdef COMPAT_FREEBSD32
2862 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2863 				struct timeval32 tv32;
2864 
2865 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2866 				    sizeof tv32);
2867 				CP(tv32, tv, tv_sec);
2868 				CP(tv32, tv, tv_usec);
2869 			} else
2870 #endif
2871 				error = sooptcopyin(sopt, &tv, sizeof tv,
2872 				    sizeof tv);
2873 			if (error)
2874 				goto bad;
2875 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2876 			    tv.tv_usec >= 1000000) {
2877 				error = EDOM;
2878 				goto bad;
2879 			}
2880 			if (tv.tv_sec > INT32_MAX)
2881 				val = SBT_MAX;
2882 			else
2883 				val = tvtosbt(tv);
2884 			switch (sopt->sopt_name) {
2885 			case SO_SNDTIMEO:
2886 				so->so_snd.sb_timeo = val;
2887 				break;
2888 			case SO_RCVTIMEO:
2889 				so->so_rcv.sb_timeo = val;
2890 				break;
2891 			}
2892 			break;
2893 
2894 		case SO_LABEL:
2895 #ifdef MAC
2896 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2897 			    sizeof extmac);
2898 			if (error)
2899 				goto bad;
2900 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2901 			    so, &extmac);
2902 #else
2903 			error = EOPNOTSUPP;
2904 #endif
2905 			break;
2906 
2907 		case SO_TS_CLOCK:
2908 			error = sooptcopyin(sopt, &optval, sizeof optval,
2909 			    sizeof optval);
2910 			if (error)
2911 				goto bad;
2912 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
2913 				error = EINVAL;
2914 				goto bad;
2915 			}
2916 			so->so_ts_clock = optval;
2917 			break;
2918 
2919 		case SO_MAX_PACING_RATE:
2920 			error = sooptcopyin(sopt, &val32, sizeof(val32),
2921 			    sizeof(val32));
2922 			if (error)
2923 				goto bad;
2924 			so->so_max_pacing_rate = val32;
2925 			break;
2926 
2927 		default:
2928 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2929 				error = hhook_run_socket(so, sopt,
2930 				    HHOOK_SOCKET_OPT);
2931 			else
2932 				error = ENOPROTOOPT;
2933 			break;
2934 		}
2935 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2936 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2937 	}
2938 bad:
2939 	CURVNET_RESTORE();
2940 	return (error);
2941 }
2942 
2943 /*
2944  * Helper routine for getsockopt.
2945  */
2946 int
2947 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2948 {
2949 	int	error;
2950 	size_t	valsize;
2951 
2952 	error = 0;
2953 
2954 	/*
2955 	 * Documented get behavior is that we always return a value, possibly
2956 	 * truncated to fit in the user's buffer.  Traditional behavior is
2957 	 * that we always tell the user precisely how much we copied, rather
2958 	 * than something useful like the total amount we had available for
2959 	 * her.  Note that this interface is not idempotent; the entire
2960 	 * answer must be generated ahead of time.
2961 	 */
2962 	valsize = min(len, sopt->sopt_valsize);
2963 	sopt->sopt_valsize = valsize;
2964 	if (sopt->sopt_val != NULL) {
2965 		if (sopt->sopt_td != NULL)
2966 			error = copyout(buf, sopt->sopt_val, valsize);
2967 		else
2968 			bcopy(buf, sopt->sopt_val, valsize);
2969 	}
2970 	return (error);
2971 }
2972 
2973 int
2974 sogetopt(struct socket *so, struct sockopt *sopt)
2975 {
2976 	int	error, optval;
2977 	struct	linger l;
2978 	struct	timeval tv;
2979 #ifdef MAC
2980 	struct mac extmac;
2981 #endif
2982 
2983 	CURVNET_SET(so->so_vnet);
2984 	error = 0;
2985 	if (sopt->sopt_level != SOL_SOCKET) {
2986 		if (so->so_proto->pr_ctloutput != NULL)
2987 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2988 		else
2989 			error = ENOPROTOOPT;
2990 		CURVNET_RESTORE();
2991 		return (error);
2992 	} else {
2993 		switch (sopt->sopt_name) {
2994 		case SO_ACCEPTFILTER:
2995 			error = accept_filt_getopt(so, sopt);
2996 			break;
2997 
2998 		case SO_LINGER:
2999 			SOCK_LOCK(so);
3000 			l.l_onoff = so->so_options & SO_LINGER;
3001 			l.l_linger = so->so_linger;
3002 			SOCK_UNLOCK(so);
3003 			error = sooptcopyout(sopt, &l, sizeof l);
3004 			break;
3005 
3006 		case SO_USELOOPBACK:
3007 		case SO_DONTROUTE:
3008 		case SO_DEBUG:
3009 		case SO_KEEPALIVE:
3010 		case SO_REUSEADDR:
3011 		case SO_REUSEPORT:
3012 		case SO_REUSEPORT_LB:
3013 		case SO_BROADCAST:
3014 		case SO_OOBINLINE:
3015 		case SO_ACCEPTCONN:
3016 		case SO_TIMESTAMP:
3017 		case SO_BINTIME:
3018 		case SO_NOSIGPIPE:
3019 			optval = so->so_options & sopt->sopt_name;
3020 integer:
3021 			error = sooptcopyout(sopt, &optval, sizeof optval);
3022 			break;
3023 
3024 		case SO_DOMAIN:
3025 			optval = so->so_proto->pr_domain->dom_family;
3026 			goto integer;
3027 
3028 		case SO_TYPE:
3029 			optval = so->so_type;
3030 			goto integer;
3031 
3032 		case SO_PROTOCOL:
3033 			optval = so->so_proto->pr_protocol;
3034 			goto integer;
3035 
3036 		case SO_ERROR:
3037 			SOCK_LOCK(so);
3038 			optval = so->so_error;
3039 			so->so_error = 0;
3040 			SOCK_UNLOCK(so);
3041 			goto integer;
3042 
3043 		case SO_SNDBUF:
3044 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3045 			    so->so_snd.sb_hiwat;
3046 			goto integer;
3047 
3048 		case SO_RCVBUF:
3049 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3050 			    so->so_rcv.sb_hiwat;
3051 			goto integer;
3052 
3053 		case SO_SNDLOWAT:
3054 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3055 			    so->so_snd.sb_lowat;
3056 			goto integer;
3057 
3058 		case SO_RCVLOWAT:
3059 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3060 			    so->so_rcv.sb_lowat;
3061 			goto integer;
3062 
3063 		case SO_SNDTIMEO:
3064 		case SO_RCVTIMEO:
3065 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3066 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3067 #ifdef COMPAT_FREEBSD32
3068 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3069 				struct timeval32 tv32;
3070 
3071 				CP(tv, tv32, tv_sec);
3072 				CP(tv, tv32, tv_usec);
3073 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3074 			} else
3075 #endif
3076 				error = sooptcopyout(sopt, &tv, sizeof tv);
3077 			break;
3078 
3079 		case SO_LABEL:
3080 #ifdef MAC
3081 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3082 			    sizeof(extmac));
3083 			if (error)
3084 				goto bad;
3085 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3086 			    so, &extmac);
3087 			if (error)
3088 				goto bad;
3089 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3090 #else
3091 			error = EOPNOTSUPP;
3092 #endif
3093 			break;
3094 
3095 		case SO_PEERLABEL:
3096 #ifdef MAC
3097 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3098 			    sizeof(extmac));
3099 			if (error)
3100 				goto bad;
3101 			error = mac_getsockopt_peerlabel(
3102 			    sopt->sopt_td->td_ucred, so, &extmac);
3103 			if (error)
3104 				goto bad;
3105 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3106 #else
3107 			error = EOPNOTSUPP;
3108 #endif
3109 			break;
3110 
3111 		case SO_LISTENQLIMIT:
3112 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3113 			goto integer;
3114 
3115 		case SO_LISTENQLEN:
3116 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3117 			goto integer;
3118 
3119 		case SO_LISTENINCQLEN:
3120 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3121 			goto integer;
3122 
3123 		case SO_TS_CLOCK:
3124 			optval = so->so_ts_clock;
3125 			goto integer;
3126 
3127 		case SO_MAX_PACING_RATE:
3128 			optval = so->so_max_pacing_rate;
3129 			goto integer;
3130 
3131 		default:
3132 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3133 				error = hhook_run_socket(so, sopt,
3134 				    HHOOK_SOCKET_OPT);
3135 			else
3136 				error = ENOPROTOOPT;
3137 			break;
3138 		}
3139 	}
3140 #ifdef MAC
3141 bad:
3142 #endif
3143 	CURVNET_RESTORE();
3144 	return (error);
3145 }
3146 
3147 int
3148 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3149 {
3150 	struct mbuf *m, *m_prev;
3151 	int sopt_size = sopt->sopt_valsize;
3152 
3153 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3154 	if (m == NULL)
3155 		return ENOBUFS;
3156 	if (sopt_size > MLEN) {
3157 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3158 		if ((m->m_flags & M_EXT) == 0) {
3159 			m_free(m);
3160 			return ENOBUFS;
3161 		}
3162 		m->m_len = min(MCLBYTES, sopt_size);
3163 	} else {
3164 		m->m_len = min(MLEN, sopt_size);
3165 	}
3166 	sopt_size -= m->m_len;
3167 	*mp = m;
3168 	m_prev = m;
3169 
3170 	while (sopt_size) {
3171 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3172 		if (m == NULL) {
3173 			m_freem(*mp);
3174 			return ENOBUFS;
3175 		}
3176 		if (sopt_size > MLEN) {
3177 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3178 			    M_NOWAIT);
3179 			if ((m->m_flags & M_EXT) == 0) {
3180 				m_freem(m);
3181 				m_freem(*mp);
3182 				return ENOBUFS;
3183 			}
3184 			m->m_len = min(MCLBYTES, sopt_size);
3185 		} else {
3186 			m->m_len = min(MLEN, sopt_size);
3187 		}
3188 		sopt_size -= m->m_len;
3189 		m_prev->m_next = m;
3190 		m_prev = m;
3191 	}
3192 	return (0);
3193 }
3194 
3195 int
3196 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3197 {
3198 	struct mbuf *m0 = m;
3199 
3200 	if (sopt->sopt_val == NULL)
3201 		return (0);
3202 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3203 		if (sopt->sopt_td != NULL) {
3204 			int error;
3205 
3206 			error = copyin(sopt->sopt_val, mtod(m, char *),
3207 			    m->m_len);
3208 			if (error != 0) {
3209 				m_freem(m0);
3210 				return(error);
3211 			}
3212 		} else
3213 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3214 		sopt->sopt_valsize -= m->m_len;
3215 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3216 		m = m->m_next;
3217 	}
3218 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3219 		panic("ip6_sooptmcopyin");
3220 	return (0);
3221 }
3222 
3223 int
3224 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3225 {
3226 	struct mbuf *m0 = m;
3227 	size_t valsize = 0;
3228 
3229 	if (sopt->sopt_val == NULL)
3230 		return (0);
3231 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3232 		if (sopt->sopt_td != NULL) {
3233 			int error;
3234 
3235 			error = copyout(mtod(m, char *), sopt->sopt_val,
3236 			    m->m_len);
3237 			if (error != 0) {
3238 				m_freem(m0);
3239 				return(error);
3240 			}
3241 		} else
3242 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3243 		sopt->sopt_valsize -= m->m_len;
3244 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3245 		valsize += m->m_len;
3246 		m = m->m_next;
3247 	}
3248 	if (m != NULL) {
3249 		/* enough soopt buffer should be given from user-land */
3250 		m_freem(m0);
3251 		return(EINVAL);
3252 	}
3253 	sopt->sopt_valsize = valsize;
3254 	return (0);
3255 }
3256 
3257 /*
3258  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3259  * out-of-band data, which will then notify socket consumers.
3260  */
3261 void
3262 sohasoutofband(struct socket *so)
3263 {
3264 
3265 	if (so->so_sigio != NULL)
3266 		pgsigio(&so->so_sigio, SIGURG, 0);
3267 	selwakeuppri(&so->so_rdsel, PSOCK);
3268 }
3269 
3270 int
3271 sopoll(struct socket *so, int events, struct ucred *active_cred,
3272     struct thread *td)
3273 {
3274 
3275 	/*
3276 	 * We do not need to set or assert curvnet as long as everyone uses
3277 	 * sopoll_generic().
3278 	 */
3279 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3280 	    td));
3281 }
3282 
3283 int
3284 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3285     struct thread *td)
3286 {
3287 	int revents;
3288 
3289 	SOCK_LOCK(so);
3290 	if (SOLISTENING(so)) {
3291 		if (!(events & (POLLIN | POLLRDNORM)))
3292 			revents = 0;
3293 		else if (!TAILQ_EMPTY(&so->sol_comp))
3294 			revents = events & (POLLIN | POLLRDNORM);
3295 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3296 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3297 		else {
3298 			selrecord(td, &so->so_rdsel);
3299 			revents = 0;
3300 		}
3301 	} else {
3302 		revents = 0;
3303 		SOCKBUF_LOCK(&so->so_snd);
3304 		SOCKBUF_LOCK(&so->so_rcv);
3305 		if (events & (POLLIN | POLLRDNORM))
3306 			if (soreadabledata(so))
3307 				revents |= events & (POLLIN | POLLRDNORM);
3308 		if (events & (POLLOUT | POLLWRNORM))
3309 			if (sowriteable(so))
3310 				revents |= events & (POLLOUT | POLLWRNORM);
3311 		if (events & (POLLPRI | POLLRDBAND))
3312 			if (so->so_oobmark ||
3313 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3314 				revents |= events & (POLLPRI | POLLRDBAND);
3315 		if ((events & POLLINIGNEOF) == 0) {
3316 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3317 				revents |= events & (POLLIN | POLLRDNORM);
3318 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3319 					revents |= POLLHUP;
3320 			}
3321 		}
3322 		if (revents == 0) {
3323 			if (events &
3324 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3325 				selrecord(td, &so->so_rdsel);
3326 				so->so_rcv.sb_flags |= SB_SEL;
3327 			}
3328 			if (events & (POLLOUT | POLLWRNORM)) {
3329 				selrecord(td, &so->so_wrsel);
3330 				so->so_snd.sb_flags |= SB_SEL;
3331 			}
3332 		}
3333 		SOCKBUF_UNLOCK(&so->so_rcv);
3334 		SOCKBUF_UNLOCK(&so->so_snd);
3335 	}
3336 	SOCK_UNLOCK(so);
3337 	return (revents);
3338 }
3339 
3340 int
3341 soo_kqfilter(struct file *fp, struct knote *kn)
3342 {
3343 	struct socket *so = kn->kn_fp->f_data;
3344 	struct sockbuf *sb;
3345 	struct knlist *knl;
3346 
3347 	switch (kn->kn_filter) {
3348 	case EVFILT_READ:
3349 		kn->kn_fop = &soread_filtops;
3350 		knl = &so->so_rdsel.si_note;
3351 		sb = &so->so_rcv;
3352 		break;
3353 	case EVFILT_WRITE:
3354 		kn->kn_fop = &sowrite_filtops;
3355 		knl = &so->so_wrsel.si_note;
3356 		sb = &so->so_snd;
3357 		break;
3358 	case EVFILT_EMPTY:
3359 		kn->kn_fop = &soempty_filtops;
3360 		knl = &so->so_wrsel.si_note;
3361 		sb = &so->so_snd;
3362 		break;
3363 	default:
3364 		return (EINVAL);
3365 	}
3366 
3367 	SOCK_LOCK(so);
3368 	if (SOLISTENING(so)) {
3369 		knlist_add(knl, kn, 1);
3370 	} else {
3371 		SOCKBUF_LOCK(sb);
3372 		knlist_add(knl, kn, 1);
3373 		sb->sb_flags |= SB_KNOTE;
3374 		SOCKBUF_UNLOCK(sb);
3375 	}
3376 	SOCK_UNLOCK(so);
3377 	return (0);
3378 }
3379 
3380 /*
3381  * Some routines that return EOPNOTSUPP for entry points that are not
3382  * supported by a protocol.  Fill in as needed.
3383  */
3384 int
3385 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3386 {
3387 
3388 	return EOPNOTSUPP;
3389 }
3390 
3391 int
3392 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3393 {
3394 
3395 	return EOPNOTSUPP;
3396 }
3397 
3398 int
3399 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3400 {
3401 
3402 	return EOPNOTSUPP;
3403 }
3404 
3405 int
3406 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3407 {
3408 
3409 	return EOPNOTSUPP;
3410 }
3411 
3412 int
3413 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3414     struct thread *td)
3415 {
3416 
3417 	return EOPNOTSUPP;
3418 }
3419 
3420 int
3421 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3422 {
3423 
3424 	return EOPNOTSUPP;
3425 }
3426 
3427 int
3428 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3429     struct thread *td)
3430 {
3431 
3432 	return EOPNOTSUPP;
3433 }
3434 
3435 int
3436 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3437 {
3438 
3439 	return EOPNOTSUPP;
3440 }
3441 
3442 int
3443 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3444     struct ifnet *ifp, struct thread *td)
3445 {
3446 
3447 	return EOPNOTSUPP;
3448 }
3449 
3450 int
3451 pru_disconnect_notsupp(struct socket *so)
3452 {
3453 
3454 	return EOPNOTSUPP;
3455 }
3456 
3457 int
3458 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3459 {
3460 
3461 	return EOPNOTSUPP;
3462 }
3463 
3464 int
3465 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3466 {
3467 
3468 	return EOPNOTSUPP;
3469 }
3470 
3471 int
3472 pru_rcvd_notsupp(struct socket *so, int flags)
3473 {
3474 
3475 	return EOPNOTSUPP;
3476 }
3477 
3478 int
3479 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3480 {
3481 
3482 	return EOPNOTSUPP;
3483 }
3484 
3485 int
3486 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3487     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3488 {
3489 
3490 	return EOPNOTSUPP;
3491 }
3492 
3493 int
3494 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3495 {
3496 
3497 	return (EOPNOTSUPP);
3498 }
3499 
3500 /*
3501  * This isn't really a ``null'' operation, but it's the default one and
3502  * doesn't do anything destructive.
3503  */
3504 int
3505 pru_sense_null(struct socket *so, struct stat *sb)
3506 {
3507 
3508 	sb->st_blksize = so->so_snd.sb_hiwat;
3509 	return 0;
3510 }
3511 
3512 int
3513 pru_shutdown_notsupp(struct socket *so)
3514 {
3515 
3516 	return EOPNOTSUPP;
3517 }
3518 
3519 int
3520 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3521 {
3522 
3523 	return EOPNOTSUPP;
3524 }
3525 
3526 int
3527 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3528     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3529 {
3530 
3531 	return EOPNOTSUPP;
3532 }
3533 
3534 int
3535 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3536     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3537 {
3538 
3539 	return EOPNOTSUPP;
3540 }
3541 
3542 int
3543 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3544     struct thread *td)
3545 {
3546 
3547 	return EOPNOTSUPP;
3548 }
3549 
3550 static void
3551 filt_sordetach(struct knote *kn)
3552 {
3553 	struct socket *so = kn->kn_fp->f_data;
3554 
3555 	so_rdknl_lock(so);
3556 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3557 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3558 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3559 	so_rdknl_unlock(so);
3560 }
3561 
3562 /*ARGSUSED*/
3563 static int
3564 filt_soread(struct knote *kn, long hint)
3565 {
3566 	struct socket *so;
3567 
3568 	so = kn->kn_fp->f_data;
3569 
3570 	if (SOLISTENING(so)) {
3571 		SOCK_LOCK_ASSERT(so);
3572 		kn->kn_data = so->sol_qlen;
3573 		if (so->so_error) {
3574 			kn->kn_flags |= EV_EOF;
3575 			kn->kn_fflags = so->so_error;
3576 			return (1);
3577 		}
3578 		return (!TAILQ_EMPTY(&so->sol_comp));
3579 	}
3580 
3581 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3582 
3583 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3584 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3585 		kn->kn_flags |= EV_EOF;
3586 		kn->kn_fflags = so->so_error;
3587 		return (1);
3588 	} else if (so->so_error)	/* temporary udp error */
3589 		return (1);
3590 
3591 	if (kn->kn_sfflags & NOTE_LOWAT) {
3592 		if (kn->kn_data >= kn->kn_sdata)
3593 			return (1);
3594 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3595 		return (1);
3596 
3597 	/* This hook returning non-zero indicates an event, not error */
3598 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3599 }
3600 
3601 static void
3602 filt_sowdetach(struct knote *kn)
3603 {
3604 	struct socket *so = kn->kn_fp->f_data;
3605 
3606 	so_wrknl_lock(so);
3607 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3608 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3609 		so->so_snd.sb_flags &= ~SB_KNOTE;
3610 	so_wrknl_unlock(so);
3611 }
3612 
3613 /*ARGSUSED*/
3614 static int
3615 filt_sowrite(struct knote *kn, long hint)
3616 {
3617 	struct socket *so;
3618 
3619 	so = kn->kn_fp->f_data;
3620 
3621 	if (SOLISTENING(so))
3622 		return (0);
3623 
3624 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3625 	kn->kn_data = sbspace(&so->so_snd);
3626 
3627 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3628 
3629 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3630 		kn->kn_flags |= EV_EOF;
3631 		kn->kn_fflags = so->so_error;
3632 		return (1);
3633 	} else if (so->so_error)	/* temporary udp error */
3634 		return (1);
3635 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3636 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3637 		return (0);
3638 	else if (kn->kn_sfflags & NOTE_LOWAT)
3639 		return (kn->kn_data >= kn->kn_sdata);
3640 	else
3641 		return (kn->kn_data >= so->so_snd.sb_lowat);
3642 }
3643 
3644 static int
3645 filt_soempty(struct knote *kn, long hint)
3646 {
3647 	struct socket *so;
3648 
3649 	so = kn->kn_fp->f_data;
3650 
3651 	if (SOLISTENING(so))
3652 		return (1);
3653 
3654 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3655 	kn->kn_data = sbused(&so->so_snd);
3656 
3657 	if (kn->kn_data == 0)
3658 		return (1);
3659 	else
3660 		return (0);
3661 }
3662 
3663 int
3664 socheckuid(struct socket *so, uid_t uid)
3665 {
3666 
3667 	if (so == NULL)
3668 		return (EPERM);
3669 	if (so->so_cred->cr_uid != uid)
3670 		return (EPERM);
3671 	return (0);
3672 }
3673 
3674 /*
3675  * These functions are used by protocols to notify the socket layer (and its
3676  * consumers) of state changes in the sockets driven by protocol-side events.
3677  */
3678 
3679 /*
3680  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3681  *
3682  * Normal sequence from the active (originating) side is that
3683  * soisconnecting() is called during processing of connect() call, resulting
3684  * in an eventual call to soisconnected() if/when the connection is
3685  * established.  When the connection is torn down soisdisconnecting() is
3686  * called during processing of disconnect() call, and soisdisconnected() is
3687  * called when the connection to the peer is totally severed.  The semantics
3688  * of these routines are such that connectionless protocols can call
3689  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3690  * calls when setting up a ``connection'' takes no time.
3691  *
3692  * From the passive side, a socket is created with two queues of sockets:
3693  * so_incomp for connections in progress and so_comp for connections already
3694  * made and awaiting user acceptance.  As a protocol is preparing incoming
3695  * connections, it creates a socket structure queued on so_incomp by calling
3696  * sonewconn().  When the connection is established, soisconnected() is
3697  * called, and transfers the socket structure to so_comp, making it available
3698  * to accept().
3699  *
3700  * If a socket is closed with sockets on either so_incomp or so_comp, these
3701  * sockets are dropped.
3702  *
3703  * If higher-level protocols are implemented in the kernel, the wakeups done
3704  * here will sometimes cause software-interrupt process scheduling.
3705  */
3706 void
3707 soisconnecting(struct socket *so)
3708 {
3709 
3710 	SOCK_LOCK(so);
3711 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3712 	so->so_state |= SS_ISCONNECTING;
3713 	SOCK_UNLOCK(so);
3714 }
3715 
3716 void
3717 soisconnected(struct socket *so)
3718 {
3719 
3720 	SOCK_LOCK(so);
3721 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3722 	so->so_state |= SS_ISCONNECTED;
3723 
3724 	if (so->so_qstate == SQ_INCOMP) {
3725 		struct socket *head = so->so_listen;
3726 		int ret;
3727 
3728 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3729 		/*
3730 		 * Promoting a socket from incomplete queue to complete, we
3731 		 * need to go through reverse order of locking.  We first do
3732 		 * trylock, and if that doesn't succeed, we go the hard way
3733 		 * leaving a reference and rechecking consistency after proper
3734 		 * locking.
3735 		 */
3736 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3737 			soref(head);
3738 			SOCK_UNLOCK(so);
3739 			SOLISTEN_LOCK(head);
3740 			SOCK_LOCK(so);
3741 			if (__predict_false(head != so->so_listen)) {
3742 				/*
3743 				 * The socket went off the listen queue,
3744 				 * should be lost race to close(2) of sol.
3745 				 * The socket is about to soabort().
3746 				 */
3747 				SOCK_UNLOCK(so);
3748 				sorele(head);
3749 				return;
3750 			}
3751 			/* Not the last one, as so holds a ref. */
3752 			refcount_release(&head->so_count);
3753 		}
3754 again:
3755 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3756 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3757 			head->sol_incqlen--;
3758 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3759 			head->sol_qlen++;
3760 			so->so_qstate = SQ_COMP;
3761 			SOCK_UNLOCK(so);
3762 			solisten_wakeup(head);	/* unlocks */
3763 		} else {
3764 			SOCKBUF_LOCK(&so->so_rcv);
3765 			soupcall_set(so, SO_RCV,
3766 			    head->sol_accept_filter->accf_callback,
3767 			    head->sol_accept_filter_arg);
3768 			so->so_options &= ~SO_ACCEPTFILTER;
3769 			ret = head->sol_accept_filter->accf_callback(so,
3770 			    head->sol_accept_filter_arg, M_NOWAIT);
3771 			if (ret == SU_ISCONNECTED) {
3772 				soupcall_clear(so, SO_RCV);
3773 				SOCKBUF_UNLOCK(&so->so_rcv);
3774 				goto again;
3775 			}
3776 			SOCKBUF_UNLOCK(&so->so_rcv);
3777 			SOCK_UNLOCK(so);
3778 			SOLISTEN_UNLOCK(head);
3779 		}
3780 		return;
3781 	}
3782 	SOCK_UNLOCK(so);
3783 	wakeup(&so->so_timeo);
3784 	sorwakeup(so);
3785 	sowwakeup(so);
3786 }
3787 
3788 void
3789 soisdisconnecting(struct socket *so)
3790 {
3791 
3792 	SOCK_LOCK(so);
3793 	so->so_state &= ~SS_ISCONNECTING;
3794 	so->so_state |= SS_ISDISCONNECTING;
3795 
3796 	if (!SOLISTENING(so)) {
3797 		SOCKBUF_LOCK(&so->so_rcv);
3798 		socantrcvmore_locked(so);
3799 		SOCKBUF_LOCK(&so->so_snd);
3800 		socantsendmore_locked(so);
3801 	}
3802 	SOCK_UNLOCK(so);
3803 	wakeup(&so->so_timeo);
3804 }
3805 
3806 void
3807 soisdisconnected(struct socket *so)
3808 {
3809 
3810 	SOCK_LOCK(so);
3811 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3812 	so->so_state |= SS_ISDISCONNECTED;
3813 
3814 	if (!SOLISTENING(so)) {
3815 		SOCK_UNLOCK(so);
3816 		SOCKBUF_LOCK(&so->so_rcv);
3817 		socantrcvmore_locked(so);
3818 		SOCKBUF_LOCK(&so->so_snd);
3819 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3820 		socantsendmore_locked(so);
3821 	} else
3822 		SOCK_UNLOCK(so);
3823 	wakeup(&so->so_timeo);
3824 }
3825 
3826 /*
3827  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3828  */
3829 struct sockaddr *
3830 sodupsockaddr(const struct sockaddr *sa, int mflags)
3831 {
3832 	struct sockaddr *sa2;
3833 
3834 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3835 	if (sa2)
3836 		bcopy(sa, sa2, sa->sa_len);
3837 	return sa2;
3838 }
3839 
3840 /*
3841  * Register per-socket destructor.
3842  */
3843 void
3844 sodtor_set(struct socket *so, so_dtor_t *func)
3845 {
3846 
3847 	SOCK_LOCK_ASSERT(so);
3848 	so->so_dtor = func;
3849 }
3850 
3851 /*
3852  * Register per-socket buffer upcalls.
3853  */
3854 void
3855 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
3856 {
3857 	struct sockbuf *sb;
3858 
3859 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3860 
3861 	switch (which) {
3862 	case SO_RCV:
3863 		sb = &so->so_rcv;
3864 		break;
3865 	case SO_SND:
3866 		sb = &so->so_snd;
3867 		break;
3868 	default:
3869 		panic("soupcall_set: bad which");
3870 	}
3871 	SOCKBUF_LOCK_ASSERT(sb);
3872 	sb->sb_upcall = func;
3873 	sb->sb_upcallarg = arg;
3874 	sb->sb_flags |= SB_UPCALL;
3875 }
3876 
3877 void
3878 soupcall_clear(struct socket *so, int which)
3879 {
3880 	struct sockbuf *sb;
3881 
3882 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
3883 
3884 	switch (which) {
3885 	case SO_RCV:
3886 		sb = &so->so_rcv;
3887 		break;
3888 	case SO_SND:
3889 		sb = &so->so_snd;
3890 		break;
3891 	default:
3892 		panic("soupcall_clear: bad which");
3893 	}
3894 	SOCKBUF_LOCK_ASSERT(sb);
3895 	KASSERT(sb->sb_upcall != NULL,
3896 	    ("%s: so %p no upcall to clear", __func__, so));
3897 	sb->sb_upcall = NULL;
3898 	sb->sb_upcallarg = NULL;
3899 	sb->sb_flags &= ~SB_UPCALL;
3900 }
3901 
3902 void
3903 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
3904 {
3905 
3906 	SOLISTEN_LOCK_ASSERT(so);
3907 	so->sol_upcall = func;
3908 	so->sol_upcallarg = arg;
3909 }
3910 
3911 static void
3912 so_rdknl_lock(void *arg)
3913 {
3914 	struct socket *so = arg;
3915 
3916 	if (SOLISTENING(so))
3917 		SOCK_LOCK(so);
3918 	else
3919 		SOCKBUF_LOCK(&so->so_rcv);
3920 }
3921 
3922 static void
3923 so_rdknl_unlock(void *arg)
3924 {
3925 	struct socket *so = arg;
3926 
3927 	if (SOLISTENING(so))
3928 		SOCK_UNLOCK(so);
3929 	else
3930 		SOCKBUF_UNLOCK(&so->so_rcv);
3931 }
3932 
3933 static void
3934 so_rdknl_assert_locked(void *arg)
3935 {
3936 	struct socket *so = arg;
3937 
3938 	if (SOLISTENING(so))
3939 		SOCK_LOCK_ASSERT(so);
3940 	else
3941 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3942 }
3943 
3944 static void
3945 so_rdknl_assert_unlocked(void *arg)
3946 {
3947 	struct socket *so = arg;
3948 
3949 	if (SOLISTENING(so))
3950 		SOCK_UNLOCK_ASSERT(so);
3951 	else
3952 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
3953 }
3954 
3955 static void
3956 so_wrknl_lock(void *arg)
3957 {
3958 	struct socket *so = arg;
3959 
3960 	if (SOLISTENING(so))
3961 		SOCK_LOCK(so);
3962 	else
3963 		SOCKBUF_LOCK(&so->so_snd);
3964 }
3965 
3966 static void
3967 so_wrknl_unlock(void *arg)
3968 {
3969 	struct socket *so = arg;
3970 
3971 	if (SOLISTENING(so))
3972 		SOCK_UNLOCK(so);
3973 	else
3974 		SOCKBUF_UNLOCK(&so->so_snd);
3975 }
3976 
3977 static void
3978 so_wrknl_assert_locked(void *arg)
3979 {
3980 	struct socket *so = arg;
3981 
3982 	if (SOLISTENING(so))
3983 		SOCK_LOCK_ASSERT(so);
3984 	else
3985 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
3986 }
3987 
3988 static void
3989 so_wrknl_assert_unlocked(void *arg)
3990 {
3991 	struct socket *so = arg;
3992 
3993 	if (SOLISTENING(so))
3994 		SOCK_UNLOCK_ASSERT(so);
3995 	else
3996 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
3997 }
3998 
3999 /*
4000  * Create an external-format (``xsocket'') structure using the information in
4001  * the kernel-format socket structure pointed to by so.  This is done to
4002  * reduce the spew of irrelevant information over this interface, to isolate
4003  * user code from changes in the kernel structure, and potentially to provide
4004  * information-hiding if we decide that some of this information should be
4005  * hidden from users.
4006  */
4007 void
4008 sotoxsocket(struct socket *so, struct xsocket *xso)
4009 {
4010 
4011 	bzero(xso, sizeof(*xso));
4012 	xso->xso_len = sizeof *xso;
4013 	xso->xso_so = (uintptr_t)so;
4014 	xso->so_type = so->so_type;
4015 	xso->so_options = so->so_options;
4016 	xso->so_linger = so->so_linger;
4017 	xso->so_state = so->so_state;
4018 	xso->so_pcb = (uintptr_t)so->so_pcb;
4019 	xso->xso_protocol = so->so_proto->pr_protocol;
4020 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4021 	xso->so_timeo = so->so_timeo;
4022 	xso->so_error = so->so_error;
4023 	xso->so_uid = so->so_cred->cr_uid;
4024 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4025 	if (SOLISTENING(so)) {
4026 		xso->so_qlen = so->sol_qlen;
4027 		xso->so_incqlen = so->sol_incqlen;
4028 		xso->so_qlimit = so->sol_qlimit;
4029 		xso->so_oobmark = 0;
4030 	} else {
4031 		xso->so_state |= so->so_qstate;
4032 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4033 		xso->so_oobmark = so->so_oobmark;
4034 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4035 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4036 	}
4037 }
4038 
4039 struct sockbuf *
4040 so_sockbuf_rcv(struct socket *so)
4041 {
4042 
4043 	return (&so->so_rcv);
4044 }
4045 
4046 struct sockbuf *
4047 so_sockbuf_snd(struct socket *so)
4048 {
4049 
4050 	return (&so->so_snd);
4051 }
4052 
4053 int
4054 so_state_get(const struct socket *so)
4055 {
4056 
4057 	return (so->so_state);
4058 }
4059 
4060 void
4061 so_state_set(struct socket *so, int val)
4062 {
4063 
4064 	so->so_state = val;
4065 }
4066 
4067 int
4068 so_options_get(const struct socket *so)
4069 {
4070 
4071 	return (so->so_options);
4072 }
4073 
4074 void
4075 so_options_set(struct socket *so, int val)
4076 {
4077 
4078 	so->so_options = val;
4079 }
4080 
4081 int
4082 so_error_get(const struct socket *so)
4083 {
4084 
4085 	return (so->so_error);
4086 }
4087 
4088 void
4089 so_error_set(struct socket *so, int val)
4090 {
4091 
4092 	so->so_error = val;
4093 }
4094 
4095 int
4096 so_linger_get(const struct socket *so)
4097 {
4098 
4099 	return (so->so_linger);
4100 }
4101 
4102 void
4103 so_linger_set(struct socket *so, int val)
4104 {
4105 
4106 	so->so_linger = val;
4107 }
4108 
4109 struct protosw *
4110 so_protosw_get(const struct socket *so)
4111 {
4112 
4113 	return (so->so_proto);
4114 }
4115 
4116 void
4117 so_protosw_set(struct socket *so, struct protosw *val)
4118 {
4119 
4120 	so->so_proto = val;
4121 }
4122 
4123 void
4124 so_sorwakeup(struct socket *so)
4125 {
4126 
4127 	sorwakeup(so);
4128 }
4129 
4130 void
4131 so_sowwakeup(struct socket *so)
4132 {
4133 
4134 	sowwakeup(so);
4135 }
4136 
4137 void
4138 so_sorwakeup_locked(struct socket *so)
4139 {
4140 
4141 	sorwakeup_locked(so);
4142 }
4143 
4144 void
4145 so_sowwakeup_locked(struct socket *so)
4146 {
4147 
4148 	sowwakeup_locked(so);
4149 }
4150 
4151 void
4152 so_lock(struct socket *so)
4153 {
4154 
4155 	SOCK_LOCK(so);
4156 }
4157 
4158 void
4159 so_unlock(struct socket *so)
4160 {
4161 
4162 	SOCK_UNLOCK(so);
4163 }
4164