xref: /freebsd/sys/kern/uipc_socket.c (revision 732a02b4e77866604a120a275c082bb6221bd2ff)
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_kern_tls.h"
111 #include "opt_sctp.h"
112 
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/fcntl.h>
116 #include <sys/limits.h>
117 #include <sys/lock.h>
118 #include <sys/mac.h>
119 #include <sys/malloc.h>
120 #include <sys/mbuf.h>
121 #include <sys/mutex.h>
122 #include <sys/domain.h>
123 #include <sys/file.h>			/* for struct knote */
124 #include <sys/hhook.h>
125 #include <sys/kernel.h>
126 #include <sys/khelp.h>
127 #include <sys/ktls.h>
128 #include <sys/event.h>
129 #include <sys/eventhandler.h>
130 #include <sys/poll.h>
131 #include <sys/proc.h>
132 #include <sys/protosw.h>
133 #include <sys/sbuf.h>
134 #include <sys/socket.h>
135 #include <sys/socketvar.h>
136 #include <sys/resourcevar.h>
137 #include <net/route.h>
138 #include <sys/signalvar.h>
139 #include <sys/stat.h>
140 #include <sys/sx.h>
141 #include <sys/sysctl.h>
142 #include <sys/taskqueue.h>
143 #include <sys/uio.h>
144 #include <sys/un.h>
145 #include <sys/unpcb.h>
146 #include <sys/jail.h>
147 #include <sys/syslog.h>
148 #include <netinet/in.h>
149 #include <netinet/in_pcb.h>
150 #include <netinet/tcp.h>
151 
152 #include <net/vnet.h>
153 
154 #include <security/mac/mac_framework.h>
155 
156 #include <vm/uma.h>
157 
158 #ifdef COMPAT_FREEBSD32
159 #include <sys/mount.h>
160 #include <sys/sysent.h>
161 #include <compat/freebsd32/freebsd32.h>
162 #endif
163 
164 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
165 		    int flags);
166 static void	so_rdknl_lock(void *);
167 static void	so_rdknl_unlock(void *);
168 static void	so_rdknl_assert_locked(void *);
169 static void	so_rdknl_assert_unlocked(void *);
170 static void	so_wrknl_lock(void *);
171 static void	so_wrknl_unlock(void *);
172 static void	so_wrknl_assert_locked(void *);
173 static void	so_wrknl_assert_unlocked(void *);
174 
175 static void	filt_sordetach(struct knote *kn);
176 static int	filt_soread(struct knote *kn, long hint);
177 static void	filt_sowdetach(struct knote *kn);
178 static int	filt_sowrite(struct knote *kn, long hint);
179 static int	filt_soempty(struct knote *kn, long hint);
180 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
181 fo_kqfilter_t	soo_kqfilter;
182 
183 static struct filterops soread_filtops = {
184 	.f_isfd = 1,
185 	.f_detach = filt_sordetach,
186 	.f_event = filt_soread,
187 };
188 static struct filterops sowrite_filtops = {
189 	.f_isfd = 1,
190 	.f_detach = filt_sowdetach,
191 	.f_event = filt_sowrite,
192 };
193 static struct filterops soempty_filtops = {
194 	.f_isfd = 1,
195 	.f_detach = filt_sowdetach,
196 	.f_event = filt_soempty,
197 };
198 
199 so_gen_t	so_gencnt;	/* generation count for sockets */
200 
201 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
202 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
203 
204 #define	VNET_SO_ASSERT(so)						\
205 	VNET_ASSERT(curvnet != NULL,					\
206 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
207 
208 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
209 #define	V_socket_hhh		VNET(socket_hhh)
210 
211 /*
212  * Limit on the number of connections in the listen queue waiting
213  * for accept(2).
214  * NB: The original sysctl somaxconn is still available but hidden
215  * to prevent confusion about the actual purpose of this number.
216  */
217 static u_int somaxconn = SOMAXCONN;
218 
219 static int
220 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
221 {
222 	int error;
223 	int val;
224 
225 	val = somaxconn;
226 	error = sysctl_handle_int(oidp, &val, 0, req);
227 	if (error || !req->newptr )
228 		return (error);
229 
230 	/*
231 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
232 	 *   3 * so_qlimit / 2
233 	 * below, will not overflow.
234          */
235 
236 	if (val < 1 || val > UINT_MAX / 3)
237 		return (EINVAL);
238 
239 	somaxconn = val;
240 	return (0);
241 }
242 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
243     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
244     sysctl_somaxconn, "I",
245     "Maximum listen socket pending connection accept queue size");
246 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
247     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
248     sizeof(int), sysctl_somaxconn, "I",
249     "Maximum listen socket pending connection accept queue size (compat)");
250 
251 static int numopensockets;
252 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
253     &numopensockets, 0, "Number of open sockets");
254 
255 /*
256  * accept_mtx locks down per-socket fields relating to accept queues.  See
257  * socketvar.h for an annotation of the protected fields of struct socket.
258  */
259 struct mtx accept_mtx;
260 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
261 
262 /*
263  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
264  * so_gencnt field.
265  */
266 static struct mtx so_global_mtx;
267 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
268 
269 /*
270  * General IPC sysctl name space, used by sockets and a variety of other IPC
271  * types.
272  */
273 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
274     "IPC");
275 
276 /*
277  * Initialize the socket subsystem and set up the socket
278  * memory allocator.
279  */
280 static uma_zone_t socket_zone;
281 int	maxsockets;
282 
283 static void
284 socket_zone_change(void *tag)
285 {
286 
287 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
288 }
289 
290 static void
291 socket_hhook_register(int subtype)
292 {
293 
294 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
295 	    &V_socket_hhh[subtype],
296 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
297 		printf("%s: WARNING: unable to register hook\n", __func__);
298 }
299 
300 static void
301 socket_hhook_deregister(int subtype)
302 {
303 
304 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
305 		printf("%s: WARNING: unable to deregister hook\n", __func__);
306 }
307 
308 static void
309 socket_init(void *tag)
310 {
311 
312 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
313 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
314 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
315 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
316 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
317 	    EVENTHANDLER_PRI_FIRST);
318 }
319 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
320 
321 static void
322 socket_vnet_init(const void *unused __unused)
323 {
324 	int i;
325 
326 	/* We expect a contiguous range */
327 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
328 		socket_hhook_register(i);
329 }
330 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
331     socket_vnet_init, NULL);
332 
333 static void
334 socket_vnet_uninit(const void *unused __unused)
335 {
336 	int i;
337 
338 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
339 		socket_hhook_deregister(i);
340 }
341 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
342     socket_vnet_uninit, NULL);
343 
344 /*
345  * Initialise maxsockets.  This SYSINIT must be run after
346  * tunable_mbinit().
347  */
348 static void
349 init_maxsockets(void *ignored)
350 {
351 
352 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
353 	maxsockets = imax(maxsockets, maxfiles);
354 }
355 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
356 
357 /*
358  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
359  * of the change so that they can update their dependent limits as required.
360  */
361 static int
362 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
363 {
364 	int error, newmaxsockets;
365 
366 	newmaxsockets = maxsockets;
367 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
368 	if (error == 0 && req->newptr) {
369 		if (newmaxsockets > maxsockets &&
370 		    newmaxsockets <= maxfiles) {
371 			maxsockets = newmaxsockets;
372 			EVENTHANDLER_INVOKE(maxsockets_change);
373 		} else
374 			error = EINVAL;
375 	}
376 	return (error);
377 }
378 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
379     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
380     sysctl_maxsockets, "IU",
381     "Maximum number of sockets available");
382 
383 /*
384  * Socket operation routines.  These routines are called by the routines in
385  * sys_socket.c or from a system process, and implement the semantics of
386  * socket operations by switching out to the protocol specific routines.
387  */
388 
389 /*
390  * Get a socket structure from our zone, and initialize it.  Note that it
391  * would probably be better to allocate socket and PCB at the same time, but
392  * I'm not convinced that all the protocols can be easily modified to do
393  * this.
394  *
395  * soalloc() returns a socket with a ref count of 0.
396  */
397 static struct socket *
398 soalloc(struct vnet *vnet)
399 {
400 	struct socket *so;
401 
402 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
403 	if (so == NULL)
404 		return (NULL);
405 #ifdef MAC
406 	if (mac_socket_init(so, M_NOWAIT) != 0) {
407 		uma_zfree(socket_zone, so);
408 		return (NULL);
409 	}
410 #endif
411 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
412 		uma_zfree(socket_zone, so);
413 		return (NULL);
414 	}
415 
416 	/*
417 	 * The socket locking protocol allows to lock 2 sockets at a time,
418 	 * however, the first one must be a listening socket.  WITNESS lacks
419 	 * a feature to change class of an existing lock, so we use DUPOK.
420 	 */
421 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
422 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
423 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
424 	so->so_rcv.sb_sel = &so->so_rdsel;
425 	so->so_snd.sb_sel = &so->so_wrsel;
426 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
427 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
428 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
429 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
430 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
431 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
432 #ifdef VIMAGE
433 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
434 	    __func__, __LINE__, so));
435 	so->so_vnet = vnet;
436 #endif
437 	/* We shouldn't need the so_global_mtx */
438 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
439 		/* Do we need more comprehensive error returns? */
440 		uma_zfree(socket_zone, so);
441 		return (NULL);
442 	}
443 	mtx_lock(&so_global_mtx);
444 	so->so_gencnt = ++so_gencnt;
445 	++numopensockets;
446 #ifdef VIMAGE
447 	vnet->vnet_sockcnt++;
448 #endif
449 	mtx_unlock(&so_global_mtx);
450 
451 	return (so);
452 }
453 
454 /*
455  * Free the storage associated with a socket at the socket layer, tear down
456  * locks, labels, etc.  All protocol state is assumed already to have been
457  * torn down (and possibly never set up) by the caller.
458  */
459 static void
460 sodealloc(struct socket *so)
461 {
462 
463 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
464 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
465 
466 	mtx_lock(&so_global_mtx);
467 	so->so_gencnt = ++so_gencnt;
468 	--numopensockets;	/* Could be below, but faster here. */
469 #ifdef VIMAGE
470 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
471 	    __func__, __LINE__, so));
472 	so->so_vnet->vnet_sockcnt--;
473 #endif
474 	mtx_unlock(&so_global_mtx);
475 #ifdef MAC
476 	mac_socket_destroy(so);
477 #endif
478 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
479 
480 	crfree(so->so_cred);
481 	khelp_destroy_osd(&so->osd);
482 	if (SOLISTENING(so)) {
483 		if (so->sol_accept_filter != NULL)
484 			accept_filt_setopt(so, NULL);
485 	} else {
486 		if (so->so_rcv.sb_hiwat)
487 			(void)chgsbsize(so->so_cred->cr_uidinfo,
488 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
489 		if (so->so_snd.sb_hiwat)
490 			(void)chgsbsize(so->so_cred->cr_uidinfo,
491 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
492 		sx_destroy(&so->so_snd.sb_sx);
493 		sx_destroy(&so->so_rcv.sb_sx);
494 		SOCKBUF_LOCK_DESTROY(&so->so_snd);
495 		SOCKBUF_LOCK_DESTROY(&so->so_rcv);
496 	}
497 	mtx_destroy(&so->so_lock);
498 	uma_zfree(socket_zone, so);
499 }
500 
501 /*
502  * socreate returns a socket with a ref count of 1.  The socket should be
503  * closed with soclose().
504  */
505 int
506 socreate(int dom, struct socket **aso, int type, int proto,
507     struct ucred *cred, struct thread *td)
508 {
509 	struct protosw *prp;
510 	struct socket *so;
511 	int error;
512 
513 	if (proto)
514 		prp = pffindproto(dom, proto, type);
515 	else
516 		prp = pffindtype(dom, type);
517 
518 	if (prp == NULL) {
519 		/* No support for domain. */
520 		if (pffinddomain(dom) == NULL)
521 			return (EAFNOSUPPORT);
522 		/* No support for socket type. */
523 		if (proto == 0 && type != 0)
524 			return (EPROTOTYPE);
525 		return (EPROTONOSUPPORT);
526 	}
527 	if (prp->pr_usrreqs->pru_attach == NULL ||
528 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
529 		return (EPROTONOSUPPORT);
530 
531 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
532 		return (EPROTONOSUPPORT);
533 
534 	if (prp->pr_type != type)
535 		return (EPROTOTYPE);
536 	so = soalloc(CRED_TO_VNET(cred));
537 	if (so == NULL)
538 		return (ENOBUFS);
539 
540 	so->so_type = type;
541 	so->so_cred = crhold(cred);
542 	if ((prp->pr_domain->dom_family == PF_INET) ||
543 	    (prp->pr_domain->dom_family == PF_INET6) ||
544 	    (prp->pr_domain->dom_family == PF_ROUTE))
545 		so->so_fibnum = td->td_proc->p_fibnum;
546 	else
547 		so->so_fibnum = 0;
548 	so->so_proto = prp;
549 #ifdef MAC
550 	mac_socket_create(cred, so);
551 #endif
552 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
553 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
554 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
555 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
556 	/*
557 	 * Auto-sizing of socket buffers is managed by the protocols and
558 	 * the appropriate flags must be set in the pru_attach function.
559 	 */
560 	CURVNET_SET(so->so_vnet);
561 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
562 	CURVNET_RESTORE();
563 	if (error) {
564 		sodealloc(so);
565 		return (error);
566 	}
567 	soref(so);
568 	*aso = so;
569 	return (0);
570 }
571 
572 #ifdef REGRESSION
573 static int regression_sonewconn_earlytest = 1;
574 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
575     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
576 #endif
577 
578 static struct timeval overinterval = { 60, 0 };
579 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
580     &overinterval,
581     "Delay in seconds between warnings for listen socket overflows");
582 
583 /*
584  * When an attempt at a new connection is noted on a socket which accepts
585  * connections, sonewconn is called.  If the connection is possible (subject
586  * to space constraints, etc.) then we allocate a new structure, properly
587  * linked into the data structure of the original socket, and return this.
588  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
589  *
590  * Note: the ref count on the socket is 0 on return.
591  */
592 struct socket *
593 sonewconn(struct socket *head, int connstatus)
594 {
595 	struct sbuf descrsb;
596 	struct socket *so;
597 	int len, overcount;
598 	u_int qlen;
599 	const char localprefix[] = "local:";
600 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
601 #if defined(INET6)
602 	char addrbuf[INET6_ADDRSTRLEN];
603 #elif defined(INET)
604 	char addrbuf[INET_ADDRSTRLEN];
605 #endif
606 	bool dolog, over;
607 
608 	SOLISTEN_LOCK(head);
609 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
610 #ifdef REGRESSION
611 	if (regression_sonewconn_earlytest && over) {
612 #else
613 	if (over) {
614 #endif
615 		head->sol_overcount++;
616 		dolog = !!ratecheck(&head->sol_lastover, &overinterval);
617 
618 		/*
619 		 * If we're going to log, copy the overflow count and queue
620 		 * length from the listen socket before dropping the lock.
621 		 * Also, reset the overflow count.
622 		 */
623 		if (dolog) {
624 			overcount = head->sol_overcount;
625 			head->sol_overcount = 0;
626 			qlen = head->sol_qlen;
627 		}
628 		SOLISTEN_UNLOCK(head);
629 
630 		if (dolog) {
631 			/*
632 			 * Try to print something descriptive about the
633 			 * socket for the error message.
634 			 */
635 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
636 			    SBUF_FIXEDLEN);
637 			switch (head->so_proto->pr_domain->dom_family) {
638 #if defined(INET) || defined(INET6)
639 #ifdef INET
640 			case AF_INET:
641 #endif
642 #ifdef INET6
643 			case AF_INET6:
644 				if (head->so_proto->pr_domain->dom_family ==
645 				    AF_INET6 ||
646 				    (sotoinpcb(head)->inp_inc.inc_flags &
647 				    INC_ISIPV6)) {
648 					ip6_sprintf(addrbuf,
649 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
650 					sbuf_printf(&descrsb, "[%s]", addrbuf);
651 				} else
652 #endif
653 				{
654 #ifdef INET
655 					inet_ntoa_r(
656 					    sotoinpcb(head)->inp_inc.inc_laddr,
657 					    addrbuf);
658 					sbuf_cat(&descrsb, addrbuf);
659 #endif
660 				}
661 				sbuf_printf(&descrsb, ":%hu (proto %u)",
662 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
663 				    head->so_proto->pr_protocol);
664 				break;
665 #endif /* INET || INET6 */
666 			case AF_UNIX:
667 				sbuf_cat(&descrsb, localprefix);
668 				if (sotounpcb(head)->unp_addr != NULL)
669 					len =
670 					    sotounpcb(head)->unp_addr->sun_len -
671 					    offsetof(struct sockaddr_un,
672 					    sun_path);
673 				else
674 					len = 0;
675 				if (len > 0)
676 					sbuf_bcat(&descrsb,
677 					    sotounpcb(head)->unp_addr->sun_path,
678 					    len);
679 				else
680 					sbuf_cat(&descrsb, "(unknown)");
681 				break;
682 			}
683 
684 			/*
685 			 * If we can't print something more specific, at least
686 			 * print the domain name.
687 			 */
688 			if (sbuf_finish(&descrsb) != 0 ||
689 			    sbuf_len(&descrsb) <= 0) {
690 				sbuf_clear(&descrsb);
691 				sbuf_cat(&descrsb,
692 				    head->so_proto->pr_domain->dom_name ?:
693 				    "unknown");
694 				sbuf_finish(&descrsb);
695 			}
696 			KASSERT(sbuf_len(&descrsb) > 0,
697 			    ("%s: sbuf creation failed", __func__));
698 			log(LOG_DEBUG,
699 			    "%s: pcb %p (%s): Listen queue overflow: "
700 			    "%i already in queue awaiting acceptance "
701 			    "(%d occurrences)\n",
702 			    __func__, head->so_pcb, sbuf_data(&descrsb),
703 			    qlen, overcount);
704 			sbuf_delete(&descrsb);
705 
706 			overcount = 0;
707 		}
708 
709 		return (NULL);
710 	}
711 	SOLISTEN_UNLOCK(head);
712 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
713 	    __func__, head));
714 	so = soalloc(head->so_vnet);
715 	if (so == NULL) {
716 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
717 		    "limit reached or out of memory\n",
718 		    __func__, head->so_pcb);
719 		return (NULL);
720 	}
721 	so->so_listen = head;
722 	so->so_type = head->so_type;
723 	so->so_linger = head->so_linger;
724 	so->so_state = head->so_state | SS_NOFDREF;
725 	so->so_fibnum = head->so_fibnum;
726 	so->so_proto = head->so_proto;
727 	so->so_cred = crhold(head->so_cred);
728 #ifdef MAC
729 	mac_socket_newconn(head, so);
730 #endif
731 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
732 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
733 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
734 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
735 	VNET_SO_ASSERT(head);
736 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
737 		sodealloc(so);
738 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
739 		    __func__, head->so_pcb);
740 		return (NULL);
741 	}
742 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
743 		sodealloc(so);
744 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
745 		    __func__, head->so_pcb);
746 		return (NULL);
747 	}
748 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
749 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
750 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
751 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
752 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
753 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
754 
755 	SOLISTEN_LOCK(head);
756 	if (head->sol_accept_filter != NULL)
757 		connstatus = 0;
758 	so->so_state |= connstatus;
759 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
760 	soref(head); /* A socket on (in)complete queue refs head. */
761 	if (connstatus) {
762 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
763 		so->so_qstate = SQ_COMP;
764 		head->sol_qlen++;
765 		solisten_wakeup(head);	/* unlocks */
766 	} else {
767 		/*
768 		 * Keep removing sockets from the head until there's room for
769 		 * us to insert on the tail.  In pre-locking revisions, this
770 		 * was a simple if(), but as we could be racing with other
771 		 * threads and soabort() requires dropping locks, we must
772 		 * loop waiting for the condition to be true.
773 		 */
774 		while (head->sol_incqlen > head->sol_qlimit) {
775 			struct socket *sp;
776 
777 			sp = TAILQ_FIRST(&head->sol_incomp);
778 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
779 			head->sol_incqlen--;
780 			SOCK_LOCK(sp);
781 			sp->so_qstate = SQ_NONE;
782 			sp->so_listen = NULL;
783 			SOCK_UNLOCK(sp);
784 			sorele(head);	/* does SOLISTEN_UNLOCK, head stays */
785 			soabort(sp);
786 			SOLISTEN_LOCK(head);
787 		}
788 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
789 		so->so_qstate = SQ_INCOMP;
790 		head->sol_incqlen++;
791 		SOLISTEN_UNLOCK(head);
792 	}
793 	return (so);
794 }
795 
796 #ifdef SCTP
797 /*
798  * Socket part of sctp_peeloff().  Detach a new socket from an
799  * association.  The new socket is returned with a reference.
800  */
801 struct socket *
802 sopeeloff(struct socket *head)
803 {
804 	struct socket *so;
805 
806 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
807 	    __func__, __LINE__, head));
808 	so = soalloc(head->so_vnet);
809 	if (so == NULL) {
810 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
811 		    "limit reached or out of memory\n",
812 		    __func__, head->so_pcb);
813 		return (NULL);
814 	}
815 	so->so_type = head->so_type;
816 	so->so_options = head->so_options;
817 	so->so_linger = head->so_linger;
818 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
819 	so->so_fibnum = head->so_fibnum;
820 	so->so_proto = head->so_proto;
821 	so->so_cred = crhold(head->so_cred);
822 #ifdef MAC
823 	mac_socket_newconn(head, so);
824 #endif
825 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
826 	    so_rdknl_assert_locked, so_rdknl_assert_unlocked);
827 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
828 	    so_wrknl_assert_locked, so_wrknl_assert_unlocked);
829 	VNET_SO_ASSERT(head);
830 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
831 		sodealloc(so);
832 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
833 		    __func__, head->so_pcb);
834 		return (NULL);
835 	}
836 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
837 		sodealloc(so);
838 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
839 		    __func__, head->so_pcb);
840 		return (NULL);
841 	}
842 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
843 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
844 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
845 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
846 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
847 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
848 
849 	soref(so);
850 
851 	return (so);
852 }
853 #endif	/* SCTP */
854 
855 int
856 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
857 {
858 	int error;
859 
860 	CURVNET_SET(so->so_vnet);
861 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
862 	CURVNET_RESTORE();
863 	return (error);
864 }
865 
866 int
867 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
868 {
869 	int error;
870 
871 	CURVNET_SET(so->so_vnet);
872 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
873 	CURVNET_RESTORE();
874 	return (error);
875 }
876 
877 /*
878  * solisten() transitions a socket from a non-listening state to a listening
879  * state, but can also be used to update the listen queue depth on an
880  * existing listen socket.  The protocol will call back into the sockets
881  * layer using solisten_proto_check() and solisten_proto() to check and set
882  * socket-layer listen state.  Call backs are used so that the protocol can
883  * acquire both protocol and socket layer locks in whatever order is required
884  * by the protocol.
885  *
886  * Protocol implementors are advised to hold the socket lock across the
887  * socket-layer test and set to avoid races at the socket layer.
888  */
889 int
890 solisten(struct socket *so, int backlog, struct thread *td)
891 {
892 	int error;
893 
894 	CURVNET_SET(so->so_vnet);
895 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
896 	CURVNET_RESTORE();
897 	return (error);
898 }
899 
900 int
901 solisten_proto_check(struct socket *so)
902 {
903 
904 	SOCK_LOCK_ASSERT(so);
905 
906 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
907 	    SS_ISDISCONNECTING))
908 		return (EINVAL);
909 	return (0);
910 }
911 
912 void
913 solisten_proto(struct socket *so, int backlog)
914 {
915 	int sbrcv_lowat, sbsnd_lowat;
916 	u_int sbrcv_hiwat, sbsnd_hiwat;
917 	short sbrcv_flags, sbsnd_flags;
918 	sbintime_t sbrcv_timeo, sbsnd_timeo;
919 
920 	SOCK_LOCK_ASSERT(so);
921 
922 	if (SOLISTENING(so))
923 		goto listening;
924 
925 	/*
926 	 * Change this socket to listening state.
927 	 */
928 	sbrcv_lowat = so->so_rcv.sb_lowat;
929 	sbsnd_lowat = so->so_snd.sb_lowat;
930 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
931 	sbsnd_hiwat = so->so_snd.sb_hiwat;
932 	sbrcv_flags = so->so_rcv.sb_flags;
933 	sbsnd_flags = so->so_snd.sb_flags;
934 	sbrcv_timeo = so->so_rcv.sb_timeo;
935 	sbsnd_timeo = so->so_snd.sb_timeo;
936 
937 	sbdestroy(&so->so_snd, so);
938 	sbdestroy(&so->so_rcv, so);
939 	sx_destroy(&so->so_snd.sb_sx);
940 	sx_destroy(&so->so_rcv.sb_sx);
941 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
942 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
943 
944 #ifdef INVARIANTS
945 	bzero(&so->so_rcv,
946 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
947 #endif
948 
949 	so->sol_sbrcv_lowat = sbrcv_lowat;
950 	so->sol_sbsnd_lowat = sbsnd_lowat;
951 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
952 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
953 	so->sol_sbrcv_flags = sbrcv_flags;
954 	so->sol_sbsnd_flags = sbsnd_flags;
955 	so->sol_sbrcv_timeo = sbrcv_timeo;
956 	so->sol_sbsnd_timeo = sbsnd_timeo;
957 
958 	so->sol_qlen = so->sol_incqlen = 0;
959 	TAILQ_INIT(&so->sol_incomp);
960 	TAILQ_INIT(&so->sol_comp);
961 
962 	so->sol_accept_filter = NULL;
963 	so->sol_accept_filter_arg = NULL;
964 	so->sol_accept_filter_str = NULL;
965 
966 	so->sol_upcall = NULL;
967 	so->sol_upcallarg = NULL;
968 
969 	so->so_options |= SO_ACCEPTCONN;
970 
971 listening:
972 	if (backlog < 0 || backlog > somaxconn)
973 		backlog = somaxconn;
974 	so->sol_qlimit = backlog;
975 }
976 
977 /*
978  * Wakeup listeners/subsystems once we have a complete connection.
979  * Enters with lock, returns unlocked.
980  */
981 void
982 solisten_wakeup(struct socket *sol)
983 {
984 
985 	if (sol->sol_upcall != NULL)
986 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
987 	else {
988 		selwakeuppri(&sol->so_rdsel, PSOCK);
989 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
990 	}
991 	SOLISTEN_UNLOCK(sol);
992 	wakeup_one(&sol->sol_comp);
993 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
994 		pgsigio(&sol->so_sigio, SIGIO, 0);
995 }
996 
997 /*
998  * Return single connection off a listening socket queue.  Main consumer of
999  * the function is kern_accept4().  Some modules, that do their own accept
1000  * management also use the function.
1001  *
1002  * Listening socket must be locked on entry and is returned unlocked on
1003  * return.
1004  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1005  */
1006 int
1007 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1008 {
1009 	struct socket *so;
1010 	int error;
1011 
1012 	SOLISTEN_LOCK_ASSERT(head);
1013 
1014 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1015 	    head->so_error == 0) {
1016 		error = msleep(&head->sol_comp, &head->so_lock, PSOCK | PCATCH,
1017 		    "accept", 0);
1018 		if (error != 0) {
1019 			SOLISTEN_UNLOCK(head);
1020 			return (error);
1021 		}
1022 	}
1023 	if (head->so_error) {
1024 		error = head->so_error;
1025 		head->so_error = 0;
1026 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1027 		error = EWOULDBLOCK;
1028 	else
1029 		error = 0;
1030 	if (error) {
1031 		SOLISTEN_UNLOCK(head);
1032 		return (error);
1033 	}
1034 	so = TAILQ_FIRST(&head->sol_comp);
1035 	SOCK_LOCK(so);
1036 	KASSERT(so->so_qstate == SQ_COMP,
1037 	    ("%s: so %p not SQ_COMP", __func__, so));
1038 	soref(so);
1039 	head->sol_qlen--;
1040 	so->so_qstate = SQ_NONE;
1041 	so->so_listen = NULL;
1042 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1043 	if (flags & ACCEPT4_INHERIT)
1044 		so->so_state |= (head->so_state & SS_NBIO);
1045 	else
1046 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1047 	SOCK_UNLOCK(so);
1048 	sorele(head);
1049 
1050 	*ret = so;
1051 	return (0);
1052 }
1053 
1054 /*
1055  * Evaluate the reference count and named references on a socket; if no
1056  * references remain, free it.  This should be called whenever a reference is
1057  * released, such as in sorele(), but also when named reference flags are
1058  * cleared in socket or protocol code.
1059  *
1060  * sofree() will free the socket if:
1061  *
1062  * - There are no outstanding file descriptor references or related consumers
1063  *   (so_count == 0).
1064  *
1065  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1066  *
1067  * - The protocol does not have an outstanding strong reference on the socket
1068  *   (SS_PROTOREF).
1069  *
1070  * - The socket is not in a completed connection queue, so a process has been
1071  *   notified that it is present.  If it is removed, the user process may
1072  *   block in accept() despite select() saying the socket was ready.
1073  */
1074 void
1075 sofree(struct socket *so)
1076 {
1077 	struct protosw *pr = so->so_proto;
1078 
1079 	SOCK_LOCK_ASSERT(so);
1080 
1081 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
1082 	    (so->so_state & SS_PROTOREF) || (so->so_qstate == SQ_COMP)) {
1083 		SOCK_UNLOCK(so);
1084 		return;
1085 	}
1086 
1087 	if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1088 		struct socket *sol;
1089 
1090 		sol = so->so_listen;
1091 		KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1092 
1093 		/*
1094 		 * To solve race between close of a listening socket and
1095 		 * a socket on its incomplete queue, we need to lock both.
1096 		 * The order is first listening socket, then regular.
1097 		 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1098 		 * function and the listening socket are the only pointers
1099 		 * to so.  To preserve so and sol, we reference both and then
1100 		 * relock.
1101 		 * After relock the socket may not move to so_comp since it
1102 		 * doesn't have PCB already, but it may be removed from
1103 		 * so_incomp. If that happens, we share responsiblity on
1104 		 * freeing the socket, but soclose() has already removed
1105 		 * it from queue.
1106 		 */
1107 		soref(sol);
1108 		soref(so);
1109 		SOCK_UNLOCK(so);
1110 		SOLISTEN_LOCK(sol);
1111 		SOCK_LOCK(so);
1112 		if (so->so_qstate == SQ_INCOMP) {
1113 			KASSERT(so->so_listen == sol,
1114 			    ("%s: so %p migrated out of sol %p",
1115 			    __func__, so, sol));
1116 			TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1117 			sol->sol_incqlen--;
1118 			/* This is guarenteed not to be the last. */
1119 			refcount_release(&sol->so_count);
1120 			so->so_qstate = SQ_NONE;
1121 			so->so_listen = NULL;
1122 		} else
1123 			KASSERT(so->so_listen == NULL,
1124 			    ("%s: so %p not on (in)comp with so_listen",
1125 			    __func__, so));
1126 		sorele(sol);
1127 		KASSERT(so->so_count == 1,
1128 		    ("%s: so %p count %u", __func__, so, so->so_count));
1129 		so->so_count = 0;
1130 	}
1131 	if (SOLISTENING(so))
1132 		so->so_error = ECONNABORTED;
1133 	SOCK_UNLOCK(so);
1134 
1135 	if (so->so_dtor != NULL)
1136 		so->so_dtor(so);
1137 
1138 	VNET_SO_ASSERT(so);
1139 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1140 		(*pr->pr_domain->dom_dispose)(so);
1141 	if (pr->pr_usrreqs->pru_detach != NULL)
1142 		(*pr->pr_usrreqs->pru_detach)(so);
1143 
1144 	/*
1145 	 * From this point on, we assume that no other references to this
1146 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1147 	 * to be acquired or held.
1148 	 *
1149 	 * We used to do a lot of socket buffer and socket locking here, as
1150 	 * well as invoke sorflush() and perform wakeups.  The direct call to
1151 	 * dom_dispose() and sbdestroy() are an inlining of what was
1152 	 * necessary from sorflush().
1153 	 *
1154 	 * Notice that the socket buffer and kqueue state are torn down
1155 	 * before calling pru_detach.  This means that protocols shold not
1156 	 * assume they can perform socket wakeups, etc, in their detach code.
1157 	 */
1158 	if (!SOLISTENING(so)) {
1159 		sbdestroy(&so->so_snd, so);
1160 		sbdestroy(&so->so_rcv, so);
1161 	}
1162 	seldrain(&so->so_rdsel);
1163 	seldrain(&so->so_wrsel);
1164 	knlist_destroy(&so->so_rdsel.si_note);
1165 	knlist_destroy(&so->so_wrsel.si_note);
1166 	sodealloc(so);
1167 }
1168 
1169 /*
1170  * Close a socket on last file table reference removal.  Initiate disconnect
1171  * if connected.  Free socket when disconnect complete.
1172  *
1173  * This function will sorele() the socket.  Note that soclose() may be called
1174  * prior to the ref count reaching zero.  The actual socket structure will
1175  * not be freed until the ref count reaches zero.
1176  */
1177 int
1178 soclose(struct socket *so)
1179 {
1180 	struct accept_queue lqueue;
1181 	bool listening;
1182 	int error = 0;
1183 
1184 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1185 
1186 	CURVNET_SET(so->so_vnet);
1187 	funsetown(&so->so_sigio);
1188 	if (so->so_state & SS_ISCONNECTED) {
1189 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1190 			error = sodisconnect(so);
1191 			if (error) {
1192 				if (error == ENOTCONN)
1193 					error = 0;
1194 				goto drop;
1195 			}
1196 		}
1197 		if (so->so_options & SO_LINGER) {
1198 			if ((so->so_state & SS_ISDISCONNECTING) &&
1199 			    (so->so_state & SS_NBIO))
1200 				goto drop;
1201 			while (so->so_state & SS_ISCONNECTED) {
1202 				error = tsleep(&so->so_timeo,
1203 				    PSOCK | PCATCH, "soclos",
1204 				    so->so_linger * hz);
1205 				if (error)
1206 					break;
1207 			}
1208 		}
1209 	}
1210 
1211 drop:
1212 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1213 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1214 
1215 	SOCK_LOCK(so);
1216 	if ((listening = (so->so_options & SO_ACCEPTCONN))) {
1217 		struct socket *sp;
1218 
1219 		TAILQ_INIT(&lqueue);
1220 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1221 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1222 
1223 		so->sol_qlen = so->sol_incqlen = 0;
1224 
1225 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1226 			SOCK_LOCK(sp);
1227 			sp->so_qstate = SQ_NONE;
1228 			sp->so_listen = NULL;
1229 			SOCK_UNLOCK(sp);
1230 			/* Guaranteed not to be the last. */
1231 			refcount_release(&so->so_count);
1232 		}
1233 	}
1234 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1235 	so->so_state |= SS_NOFDREF;
1236 	sorele(so);
1237 	if (listening) {
1238 		struct socket *sp, *tsp;
1239 
1240 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1241 			SOCK_LOCK(sp);
1242 			if (sp->so_count == 0) {
1243 				SOCK_UNLOCK(sp);
1244 				soabort(sp);
1245 			} else
1246 				/* sp is now in sofree() */
1247 				SOCK_UNLOCK(sp);
1248 		}
1249 	}
1250 	CURVNET_RESTORE();
1251 	return (error);
1252 }
1253 
1254 /*
1255  * soabort() is used to abruptly tear down a connection, such as when a
1256  * resource limit is reached (listen queue depth exceeded), or if a listen
1257  * socket is closed while there are sockets waiting to be accepted.
1258  *
1259  * This interface is tricky, because it is called on an unreferenced socket,
1260  * and must be called only by a thread that has actually removed the socket
1261  * from the listen queue it was on, or races with other threads are risked.
1262  *
1263  * This interface will call into the protocol code, so must not be called
1264  * with any socket locks held.  Protocols do call it while holding their own
1265  * recursible protocol mutexes, but this is something that should be subject
1266  * to review in the future.
1267  */
1268 void
1269 soabort(struct socket *so)
1270 {
1271 
1272 	/*
1273 	 * In as much as is possible, assert that no references to this
1274 	 * socket are held.  This is not quite the same as asserting that the
1275 	 * current thread is responsible for arranging for no references, but
1276 	 * is as close as we can get for now.
1277 	 */
1278 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1279 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1280 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1281 	VNET_SO_ASSERT(so);
1282 
1283 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1284 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1285 	SOCK_LOCK(so);
1286 	sofree(so);
1287 }
1288 
1289 int
1290 soaccept(struct socket *so, struct sockaddr **nam)
1291 {
1292 	int error;
1293 
1294 	SOCK_LOCK(so);
1295 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1296 	so->so_state &= ~SS_NOFDREF;
1297 	SOCK_UNLOCK(so);
1298 
1299 	CURVNET_SET(so->so_vnet);
1300 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1301 	CURVNET_RESTORE();
1302 	return (error);
1303 }
1304 
1305 int
1306 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1307 {
1308 
1309 	return (soconnectat(AT_FDCWD, so, nam, td));
1310 }
1311 
1312 int
1313 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1314 {
1315 	int error;
1316 
1317 	if (so->so_options & SO_ACCEPTCONN)
1318 		return (EOPNOTSUPP);
1319 
1320 	CURVNET_SET(so->so_vnet);
1321 	/*
1322 	 * If protocol is connection-based, can only connect once.
1323 	 * Otherwise, if connected, try to disconnect first.  This allows
1324 	 * user to disconnect by connecting to, e.g., a null address.
1325 	 */
1326 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1327 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1328 	    (error = sodisconnect(so)))) {
1329 		error = EISCONN;
1330 	} else {
1331 		/*
1332 		 * Prevent accumulated error from previous connection from
1333 		 * biting us.
1334 		 */
1335 		so->so_error = 0;
1336 		if (fd == AT_FDCWD) {
1337 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1338 			    nam, td);
1339 		} else {
1340 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1341 			    so, nam, td);
1342 		}
1343 	}
1344 	CURVNET_RESTORE();
1345 
1346 	return (error);
1347 }
1348 
1349 int
1350 soconnect2(struct socket *so1, struct socket *so2)
1351 {
1352 	int error;
1353 
1354 	CURVNET_SET(so1->so_vnet);
1355 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1356 	CURVNET_RESTORE();
1357 	return (error);
1358 }
1359 
1360 int
1361 sodisconnect(struct socket *so)
1362 {
1363 	int error;
1364 
1365 	if ((so->so_state & SS_ISCONNECTED) == 0)
1366 		return (ENOTCONN);
1367 	if (so->so_state & SS_ISDISCONNECTING)
1368 		return (EALREADY);
1369 	VNET_SO_ASSERT(so);
1370 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1371 	return (error);
1372 }
1373 
1374 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1375 
1376 int
1377 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1378     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1379 {
1380 	long space;
1381 	ssize_t resid;
1382 	int clen = 0, error, dontroute;
1383 
1384 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1385 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1386 	    ("sosend_dgram: !PR_ATOMIC"));
1387 
1388 	if (uio != NULL)
1389 		resid = uio->uio_resid;
1390 	else
1391 		resid = top->m_pkthdr.len;
1392 	/*
1393 	 * In theory resid should be unsigned.  However, space must be
1394 	 * signed, as it might be less than 0 if we over-committed, and we
1395 	 * must use a signed comparison of space and resid.  On the other
1396 	 * hand, a negative resid causes us to loop sending 0-length
1397 	 * segments to the protocol.
1398 	 */
1399 	if (resid < 0) {
1400 		error = EINVAL;
1401 		goto out;
1402 	}
1403 
1404 	dontroute =
1405 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1406 	if (td != NULL)
1407 		td->td_ru.ru_msgsnd++;
1408 	if (control != NULL)
1409 		clen = control->m_len;
1410 
1411 	SOCKBUF_LOCK(&so->so_snd);
1412 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1413 		SOCKBUF_UNLOCK(&so->so_snd);
1414 		error = EPIPE;
1415 		goto out;
1416 	}
1417 	if (so->so_error) {
1418 		error = so->so_error;
1419 		so->so_error = 0;
1420 		SOCKBUF_UNLOCK(&so->so_snd);
1421 		goto out;
1422 	}
1423 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1424 		/*
1425 		 * `sendto' and `sendmsg' is allowed on a connection-based
1426 		 * socket if it supports implied connect.  Return ENOTCONN if
1427 		 * not connected and no address is supplied.
1428 		 */
1429 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1430 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1431 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1432 			    !(resid == 0 && clen != 0)) {
1433 				SOCKBUF_UNLOCK(&so->so_snd);
1434 				error = ENOTCONN;
1435 				goto out;
1436 			}
1437 		} else if (addr == NULL) {
1438 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1439 				error = ENOTCONN;
1440 			else
1441 				error = EDESTADDRREQ;
1442 			SOCKBUF_UNLOCK(&so->so_snd);
1443 			goto out;
1444 		}
1445 	}
1446 
1447 	/*
1448 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1449 	 * problem and need fixing.
1450 	 */
1451 	space = sbspace(&so->so_snd);
1452 	if (flags & MSG_OOB)
1453 		space += 1024;
1454 	space -= clen;
1455 	SOCKBUF_UNLOCK(&so->so_snd);
1456 	if (resid > space) {
1457 		error = EMSGSIZE;
1458 		goto out;
1459 	}
1460 	if (uio == NULL) {
1461 		resid = 0;
1462 		if (flags & MSG_EOR)
1463 			top->m_flags |= M_EOR;
1464 	} else {
1465 		/*
1466 		 * Copy the data from userland into a mbuf chain.
1467 		 * If no data is to be copied in, a single empty mbuf
1468 		 * is returned.
1469 		 */
1470 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1471 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1472 		if (top == NULL) {
1473 			error = EFAULT;	/* only possible error */
1474 			goto out;
1475 		}
1476 		space -= resid - uio->uio_resid;
1477 		resid = uio->uio_resid;
1478 	}
1479 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1480 	/*
1481 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1482 	 * than with.
1483 	 */
1484 	if (dontroute) {
1485 		SOCK_LOCK(so);
1486 		so->so_options |= SO_DONTROUTE;
1487 		SOCK_UNLOCK(so);
1488 	}
1489 	/*
1490 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1491 	 * of date.  We could have received a reset packet in an interrupt or
1492 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1493 	 * probably recheck again inside the locking protection here, but
1494 	 * there are probably other places that this also happens.  We must
1495 	 * rethink this.
1496 	 */
1497 	VNET_SO_ASSERT(so);
1498 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1499 	    (flags & MSG_OOB) ? PRUS_OOB :
1500 	/*
1501 	 * If the user set MSG_EOF, the protocol understands this flag and
1502 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1503 	 */
1504 	    ((flags & MSG_EOF) &&
1505 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1506 	     (resid <= 0)) ?
1507 		PRUS_EOF :
1508 		/* If there is more to send set PRUS_MORETOCOME */
1509 		(flags & MSG_MORETOCOME) ||
1510 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1511 		top, addr, control, td);
1512 	if (dontroute) {
1513 		SOCK_LOCK(so);
1514 		so->so_options &= ~SO_DONTROUTE;
1515 		SOCK_UNLOCK(so);
1516 	}
1517 	clen = 0;
1518 	control = NULL;
1519 	top = NULL;
1520 out:
1521 	if (top != NULL)
1522 		m_freem(top);
1523 	if (control != NULL)
1524 		m_freem(control);
1525 	return (error);
1526 }
1527 
1528 /*
1529  * Send on a socket.  If send must go all at once and message is larger than
1530  * send buffering, then hard error.  Lock against other senders.  If must go
1531  * all at once and not enough room now, then inform user that this would
1532  * block and do nothing.  Otherwise, if nonblocking, send as much as
1533  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1534  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1535  * in mbuf chain must be small enough to send all at once.
1536  *
1537  * Returns nonzero on error, timeout or signal; callers must check for short
1538  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1539  * on return.
1540  */
1541 int
1542 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1543     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1544 {
1545 	long space;
1546 	ssize_t resid;
1547 	int clen = 0, error, dontroute;
1548 	int atomic = sosendallatonce(so) || top;
1549 	int pru_flag;
1550 #ifdef KERN_TLS
1551 	struct ktls_session *tls;
1552 	int tls_enq_cnt, tls_pruflag;
1553 	uint8_t tls_rtype;
1554 
1555 	tls = NULL;
1556 	tls_rtype = TLS_RLTYPE_APP;
1557 #endif
1558 	if (uio != NULL)
1559 		resid = uio->uio_resid;
1560 	else
1561 		resid = top->m_pkthdr.len;
1562 	/*
1563 	 * In theory resid should be unsigned.  However, space must be
1564 	 * signed, as it might be less than 0 if we over-committed, and we
1565 	 * must use a signed comparison of space and resid.  On the other
1566 	 * hand, a negative resid causes us to loop sending 0-length
1567 	 * segments to the protocol.
1568 	 *
1569 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1570 	 * type sockets since that's an error.
1571 	 */
1572 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1573 		error = EINVAL;
1574 		goto out;
1575 	}
1576 
1577 	dontroute =
1578 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1579 	    (so->so_proto->pr_flags & PR_ATOMIC);
1580 	if (td != NULL)
1581 		td->td_ru.ru_msgsnd++;
1582 	if (control != NULL)
1583 		clen = control->m_len;
1584 
1585 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1586 	if (error)
1587 		goto out;
1588 
1589 #ifdef KERN_TLS
1590 	tls_pruflag = 0;
1591 	tls = ktls_hold(so->so_snd.sb_tls_info);
1592 	if (tls != NULL) {
1593 		if (tls->mode == TCP_TLS_MODE_SW)
1594 			tls_pruflag = PRUS_NOTREADY;
1595 
1596 		if (control != NULL) {
1597 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1598 
1599 			if (clen >= sizeof(*cm) &&
1600 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1601 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1602 				clen = 0;
1603 				m_freem(control);
1604 				control = NULL;
1605 				atomic = 1;
1606 			}
1607 		}
1608 	}
1609 #endif
1610 
1611 restart:
1612 	do {
1613 		SOCKBUF_LOCK(&so->so_snd);
1614 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1615 			SOCKBUF_UNLOCK(&so->so_snd);
1616 			error = EPIPE;
1617 			goto release;
1618 		}
1619 		if (so->so_error) {
1620 			error = so->so_error;
1621 			so->so_error = 0;
1622 			SOCKBUF_UNLOCK(&so->so_snd);
1623 			goto release;
1624 		}
1625 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1626 			/*
1627 			 * `sendto' and `sendmsg' is allowed on a connection-
1628 			 * based socket if it supports implied connect.
1629 			 * Return ENOTCONN if not connected and no address is
1630 			 * supplied.
1631 			 */
1632 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1633 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1634 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1635 				    !(resid == 0 && clen != 0)) {
1636 					SOCKBUF_UNLOCK(&so->so_snd);
1637 					error = ENOTCONN;
1638 					goto release;
1639 				}
1640 			} else if (addr == NULL) {
1641 				SOCKBUF_UNLOCK(&so->so_snd);
1642 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1643 					error = ENOTCONN;
1644 				else
1645 					error = EDESTADDRREQ;
1646 				goto release;
1647 			}
1648 		}
1649 		space = sbspace(&so->so_snd);
1650 		if (flags & MSG_OOB)
1651 			space += 1024;
1652 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1653 		    clen > so->so_snd.sb_hiwat) {
1654 			SOCKBUF_UNLOCK(&so->so_snd);
1655 			error = EMSGSIZE;
1656 			goto release;
1657 		}
1658 		if (space < resid + clen &&
1659 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1660 			if ((so->so_state & SS_NBIO) ||
1661 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1662 				SOCKBUF_UNLOCK(&so->so_snd);
1663 				error = EWOULDBLOCK;
1664 				goto release;
1665 			}
1666 			error = sbwait(&so->so_snd);
1667 			SOCKBUF_UNLOCK(&so->so_snd);
1668 			if (error)
1669 				goto release;
1670 			goto restart;
1671 		}
1672 		SOCKBUF_UNLOCK(&so->so_snd);
1673 		space -= clen;
1674 		do {
1675 			if (uio == NULL) {
1676 				resid = 0;
1677 				if (flags & MSG_EOR)
1678 					top->m_flags |= M_EOR;
1679 			} else {
1680 				/*
1681 				 * Copy the data from userland into a mbuf
1682 				 * chain.  If resid is 0, which can happen
1683 				 * only if we have control to send, then
1684 				 * a single empty mbuf is returned.  This
1685 				 * is a workaround to prevent protocol send
1686 				 * methods to panic.
1687 				 */
1688 #ifdef KERN_TLS
1689 				if (tls != NULL) {
1690 					top = m_uiotombuf(uio, M_WAITOK, space,
1691 					    tls->params.max_frame_len,
1692 					    M_NOMAP |
1693 					    ((flags & MSG_EOR) ? M_EOR : 0));
1694 					if (top != NULL) {
1695 						ktls_frame(top, tls,
1696 						    &tls_enq_cnt, tls_rtype);
1697 					}
1698 					tls_rtype = TLS_RLTYPE_APP;
1699 				} else
1700 #endif
1701 					top = m_uiotombuf(uio, M_WAITOK, space,
1702 					    (atomic ? max_hdr : 0),
1703 					    (atomic ? M_PKTHDR : 0) |
1704 					    ((flags & MSG_EOR) ? M_EOR : 0));
1705 				if (top == NULL) {
1706 					error = EFAULT; /* only possible error */
1707 					goto release;
1708 				}
1709 				space -= resid - uio->uio_resid;
1710 				resid = uio->uio_resid;
1711 			}
1712 			if (dontroute) {
1713 				SOCK_LOCK(so);
1714 				so->so_options |= SO_DONTROUTE;
1715 				SOCK_UNLOCK(so);
1716 			}
1717 			/*
1718 			 * XXX all the SBS_CANTSENDMORE checks previously
1719 			 * done could be out of date.  We could have received
1720 			 * a reset packet in an interrupt or maybe we slept
1721 			 * while doing page faults in uiomove() etc.  We
1722 			 * could probably recheck again inside the locking
1723 			 * protection here, but there are probably other
1724 			 * places that this also happens.  We must rethink
1725 			 * this.
1726 			 */
1727 			VNET_SO_ASSERT(so);
1728 
1729 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1730 			/*
1731 			 * If the user set MSG_EOF, the protocol understands
1732 			 * this flag and nothing left to send then use
1733 			 * PRU_SEND_EOF instead of PRU_SEND.
1734 			 */
1735 			    ((flags & MSG_EOF) &&
1736 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1737 			     (resid <= 0)) ?
1738 				PRUS_EOF :
1739 			/* If there is more to send set PRUS_MORETOCOME. */
1740 			    (flags & MSG_MORETOCOME) ||
1741 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1742 
1743 #ifdef KERN_TLS
1744 			pru_flag |= tls_pruflag;
1745 #endif
1746 
1747 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1748 			    pru_flag, top, addr, control, td);
1749 
1750 			if (dontroute) {
1751 				SOCK_LOCK(so);
1752 				so->so_options &= ~SO_DONTROUTE;
1753 				SOCK_UNLOCK(so);
1754 			}
1755 
1756 #ifdef KERN_TLS
1757 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1758 				/*
1759 				 * Note that error is intentionally
1760 				 * ignored.
1761 				 *
1762 				 * Like sendfile(), we rely on the
1763 				 * completion routine (pru_ready())
1764 				 * to free the mbufs in the event that
1765 				 * pru_send() encountered an error and
1766 				 * did not append them to the sockbuf.
1767 				 */
1768 				soref(so);
1769 				ktls_enqueue(top, so, tls_enq_cnt);
1770 			}
1771 #endif
1772 			clen = 0;
1773 			control = NULL;
1774 			top = NULL;
1775 			if (error)
1776 				goto release;
1777 		} while (resid && space > 0);
1778 	} while (resid);
1779 
1780 release:
1781 	sbunlock(&so->so_snd);
1782 out:
1783 #ifdef KERN_TLS
1784 	if (tls != NULL)
1785 		ktls_free(tls);
1786 #endif
1787 	if (top != NULL)
1788 		m_freem(top);
1789 	if (control != NULL)
1790 		m_freem(control);
1791 	return (error);
1792 }
1793 
1794 int
1795 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1796     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1797 {
1798 	int error;
1799 
1800 	CURVNET_SET(so->so_vnet);
1801 	if (!SOLISTENING(so))
1802 		error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1803 		    top, control, flags, td);
1804 	else {
1805 		m_freem(top);
1806 		m_freem(control);
1807 		error = ENOTCONN;
1808 	}
1809 	CURVNET_RESTORE();
1810 	return (error);
1811 }
1812 
1813 /*
1814  * The part of soreceive() that implements reading non-inline out-of-band
1815  * data from a socket.  For more complete comments, see soreceive(), from
1816  * which this code originated.
1817  *
1818  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1819  * unable to return an mbuf chain to the caller.
1820  */
1821 static int
1822 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1823 {
1824 	struct protosw *pr = so->so_proto;
1825 	struct mbuf *m;
1826 	int error;
1827 
1828 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1829 	VNET_SO_ASSERT(so);
1830 
1831 	m = m_get(M_WAITOK, MT_DATA);
1832 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1833 	if (error)
1834 		goto bad;
1835 	do {
1836 		error = uiomove(mtod(m, void *),
1837 		    (int) min(uio->uio_resid, m->m_len), uio);
1838 		m = m_free(m);
1839 	} while (uio->uio_resid && error == 0 && m);
1840 bad:
1841 	if (m != NULL)
1842 		m_freem(m);
1843 	return (error);
1844 }
1845 
1846 /*
1847  * Following replacement or removal of the first mbuf on the first mbuf chain
1848  * of a socket buffer, push necessary state changes back into the socket
1849  * buffer so that other consumers see the values consistently.  'nextrecord'
1850  * is the callers locally stored value of the original value of
1851  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1852  * NOTE: 'nextrecord' may be NULL.
1853  */
1854 static __inline void
1855 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1856 {
1857 
1858 	SOCKBUF_LOCK_ASSERT(sb);
1859 	/*
1860 	 * First, update for the new value of nextrecord.  If necessary, make
1861 	 * it the first record.
1862 	 */
1863 	if (sb->sb_mb != NULL)
1864 		sb->sb_mb->m_nextpkt = nextrecord;
1865 	else
1866 		sb->sb_mb = nextrecord;
1867 
1868 	/*
1869 	 * Now update any dependent socket buffer fields to reflect the new
1870 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1871 	 * addition of a second clause that takes care of the case where
1872 	 * sb_mb has been updated, but remains the last record.
1873 	 */
1874 	if (sb->sb_mb == NULL) {
1875 		sb->sb_mbtail = NULL;
1876 		sb->sb_lastrecord = NULL;
1877 	} else if (sb->sb_mb->m_nextpkt == NULL)
1878 		sb->sb_lastrecord = sb->sb_mb;
1879 }
1880 
1881 /*
1882  * Implement receive operations on a socket.  We depend on the way that
1883  * records are added to the sockbuf by sbappend.  In particular, each record
1884  * (mbufs linked through m_next) must begin with an address if the protocol
1885  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1886  * data, and then zero or more mbufs of data.  In order to allow parallelism
1887  * between network receive and copying to user space, as well as avoid
1888  * sleeping with a mutex held, we release the socket buffer mutex during the
1889  * user space copy.  Although the sockbuf is locked, new data may still be
1890  * appended, and thus we must maintain consistency of the sockbuf during that
1891  * time.
1892  *
1893  * The caller may receive the data as a single mbuf chain by supplying an
1894  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1895  * the count in uio_resid.
1896  */
1897 int
1898 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1899     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1900 {
1901 	struct mbuf *m, **mp;
1902 	int flags, error, offset;
1903 	ssize_t len;
1904 	struct protosw *pr = so->so_proto;
1905 	struct mbuf *nextrecord;
1906 	int moff, type = 0;
1907 	ssize_t orig_resid = uio->uio_resid;
1908 
1909 	mp = mp0;
1910 	if (psa != NULL)
1911 		*psa = NULL;
1912 	if (controlp != NULL)
1913 		*controlp = NULL;
1914 	if (flagsp != NULL)
1915 		flags = *flagsp &~ MSG_EOR;
1916 	else
1917 		flags = 0;
1918 	if (flags & MSG_OOB)
1919 		return (soreceive_rcvoob(so, uio, flags));
1920 	if (mp != NULL)
1921 		*mp = NULL;
1922 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1923 	    && uio->uio_resid) {
1924 		VNET_SO_ASSERT(so);
1925 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1926 	}
1927 
1928 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1929 	if (error)
1930 		return (error);
1931 
1932 restart:
1933 	SOCKBUF_LOCK(&so->so_rcv);
1934 	m = so->so_rcv.sb_mb;
1935 	/*
1936 	 * If we have less data than requested, block awaiting more (subject
1937 	 * to any timeout) if:
1938 	 *   1. the current count is less than the low water mark, or
1939 	 *   2. MSG_DONTWAIT is not set
1940 	 */
1941 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1942 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1943 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1944 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1945 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1946 		    ("receive: m == %p sbavail == %u",
1947 		    m, sbavail(&so->so_rcv)));
1948 		if (so->so_error) {
1949 			if (m != NULL)
1950 				goto dontblock;
1951 			error = so->so_error;
1952 			if ((flags & MSG_PEEK) == 0)
1953 				so->so_error = 0;
1954 			SOCKBUF_UNLOCK(&so->so_rcv);
1955 			goto release;
1956 		}
1957 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1958 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1959 			if (m == NULL) {
1960 				SOCKBUF_UNLOCK(&so->so_rcv);
1961 				goto release;
1962 			} else
1963 				goto dontblock;
1964 		}
1965 		for (; m != NULL; m = m->m_next)
1966 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1967 				m = so->so_rcv.sb_mb;
1968 				goto dontblock;
1969 			}
1970 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1971 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1972 			SOCKBUF_UNLOCK(&so->so_rcv);
1973 			error = ENOTCONN;
1974 			goto release;
1975 		}
1976 		if (uio->uio_resid == 0) {
1977 			SOCKBUF_UNLOCK(&so->so_rcv);
1978 			goto release;
1979 		}
1980 		if ((so->so_state & SS_NBIO) ||
1981 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1982 			SOCKBUF_UNLOCK(&so->so_rcv);
1983 			error = EWOULDBLOCK;
1984 			goto release;
1985 		}
1986 		SBLASTRECORDCHK(&so->so_rcv);
1987 		SBLASTMBUFCHK(&so->so_rcv);
1988 		error = sbwait(&so->so_rcv);
1989 		SOCKBUF_UNLOCK(&so->so_rcv);
1990 		if (error)
1991 			goto release;
1992 		goto restart;
1993 	}
1994 dontblock:
1995 	/*
1996 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1997 	 * pointer to the next record in the socket buffer.  We must keep the
1998 	 * various socket buffer pointers and local stack versions of the
1999 	 * pointers in sync, pushing out modifications before dropping the
2000 	 * socket buffer mutex, and re-reading them when picking it up.
2001 	 *
2002 	 * Otherwise, we will race with the network stack appending new data
2003 	 * or records onto the socket buffer by using inconsistent/stale
2004 	 * versions of the field, possibly resulting in socket buffer
2005 	 * corruption.
2006 	 *
2007 	 * By holding the high-level sblock(), we prevent simultaneous
2008 	 * readers from pulling off the front of the socket buffer.
2009 	 */
2010 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2011 	if (uio->uio_td)
2012 		uio->uio_td->td_ru.ru_msgrcv++;
2013 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2014 	SBLASTRECORDCHK(&so->so_rcv);
2015 	SBLASTMBUFCHK(&so->so_rcv);
2016 	nextrecord = m->m_nextpkt;
2017 	if (pr->pr_flags & PR_ADDR) {
2018 		KASSERT(m->m_type == MT_SONAME,
2019 		    ("m->m_type == %d", m->m_type));
2020 		orig_resid = 0;
2021 		if (psa != NULL)
2022 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2023 			    M_NOWAIT);
2024 		if (flags & MSG_PEEK) {
2025 			m = m->m_next;
2026 		} else {
2027 			sbfree(&so->so_rcv, m);
2028 			so->so_rcv.sb_mb = m_free(m);
2029 			m = so->so_rcv.sb_mb;
2030 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2031 		}
2032 	}
2033 
2034 	/*
2035 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2036 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2037 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2038 	 * perform externalization (or freeing if controlp == NULL).
2039 	 */
2040 	if (m != NULL && m->m_type == MT_CONTROL) {
2041 		struct mbuf *cm = NULL, *cmn;
2042 		struct mbuf **cme = &cm;
2043 
2044 		do {
2045 			if (flags & MSG_PEEK) {
2046 				if (controlp != NULL) {
2047 					*controlp = m_copym(m, 0, m->m_len,
2048 					    M_NOWAIT);
2049 					controlp = &(*controlp)->m_next;
2050 				}
2051 				m = m->m_next;
2052 			} else {
2053 				sbfree(&so->so_rcv, m);
2054 				so->so_rcv.sb_mb = m->m_next;
2055 				m->m_next = NULL;
2056 				*cme = m;
2057 				cme = &(*cme)->m_next;
2058 				m = so->so_rcv.sb_mb;
2059 			}
2060 		} while (m != NULL && m->m_type == MT_CONTROL);
2061 		if ((flags & MSG_PEEK) == 0)
2062 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2063 		while (cm != NULL) {
2064 			cmn = cm->m_next;
2065 			cm->m_next = NULL;
2066 			if (pr->pr_domain->dom_externalize != NULL) {
2067 				SOCKBUF_UNLOCK(&so->so_rcv);
2068 				VNET_SO_ASSERT(so);
2069 				error = (*pr->pr_domain->dom_externalize)
2070 				    (cm, controlp, flags);
2071 				SOCKBUF_LOCK(&so->so_rcv);
2072 			} else if (controlp != NULL)
2073 				*controlp = cm;
2074 			else
2075 				m_freem(cm);
2076 			if (controlp != NULL) {
2077 				orig_resid = 0;
2078 				while (*controlp != NULL)
2079 					controlp = &(*controlp)->m_next;
2080 			}
2081 			cm = cmn;
2082 		}
2083 		if (m != NULL)
2084 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2085 		else
2086 			nextrecord = so->so_rcv.sb_mb;
2087 		orig_resid = 0;
2088 	}
2089 	if (m != NULL) {
2090 		if ((flags & MSG_PEEK) == 0) {
2091 			KASSERT(m->m_nextpkt == nextrecord,
2092 			    ("soreceive: post-control, nextrecord !sync"));
2093 			if (nextrecord == NULL) {
2094 				KASSERT(so->so_rcv.sb_mb == m,
2095 				    ("soreceive: post-control, sb_mb!=m"));
2096 				KASSERT(so->so_rcv.sb_lastrecord == m,
2097 				    ("soreceive: post-control, lastrecord!=m"));
2098 			}
2099 		}
2100 		type = m->m_type;
2101 		if (type == MT_OOBDATA)
2102 			flags |= MSG_OOB;
2103 	} else {
2104 		if ((flags & MSG_PEEK) == 0) {
2105 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2106 			    ("soreceive: sb_mb != nextrecord"));
2107 			if (so->so_rcv.sb_mb == NULL) {
2108 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2109 				    ("soreceive: sb_lastercord != NULL"));
2110 			}
2111 		}
2112 	}
2113 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2114 	SBLASTRECORDCHK(&so->so_rcv);
2115 	SBLASTMBUFCHK(&so->so_rcv);
2116 
2117 	/*
2118 	 * Now continue to read any data mbufs off of the head of the socket
2119 	 * buffer until the read request is satisfied.  Note that 'type' is
2120 	 * used to store the type of any mbuf reads that have happened so far
2121 	 * such that soreceive() can stop reading if the type changes, which
2122 	 * causes soreceive() to return only one of regular data and inline
2123 	 * out-of-band data in a single socket receive operation.
2124 	 */
2125 	moff = 0;
2126 	offset = 0;
2127 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2128 	    && error == 0) {
2129 		/*
2130 		 * If the type of mbuf has changed since the last mbuf
2131 		 * examined ('type'), end the receive operation.
2132 		 */
2133 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2134 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2135 			if (type != m->m_type)
2136 				break;
2137 		} else if (type == MT_OOBDATA)
2138 			break;
2139 		else
2140 		    KASSERT(m->m_type == MT_DATA,
2141 			("m->m_type == %d", m->m_type));
2142 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2143 		len = uio->uio_resid;
2144 		if (so->so_oobmark && len > so->so_oobmark - offset)
2145 			len = so->so_oobmark - offset;
2146 		if (len > m->m_len - moff)
2147 			len = m->m_len - moff;
2148 		/*
2149 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2150 		 * them out via the uio, then free.  Sockbuf must be
2151 		 * consistent here (points to current mbuf, it points to next
2152 		 * record) when we drop priority; we must note any additions
2153 		 * to the sockbuf when we block interrupts again.
2154 		 */
2155 		if (mp == NULL) {
2156 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2157 			SBLASTRECORDCHK(&so->so_rcv);
2158 			SBLASTMBUFCHK(&so->so_rcv);
2159 			SOCKBUF_UNLOCK(&so->so_rcv);
2160 			if ((m->m_flags & M_NOMAP) != 0)
2161 				error = m_unmappedtouio(m, moff, uio, (int)len);
2162 			else
2163 				error = uiomove(mtod(m, char *) + moff,
2164 				    (int)len, uio);
2165 			SOCKBUF_LOCK(&so->so_rcv);
2166 			if (error) {
2167 				/*
2168 				 * The MT_SONAME mbuf has already been removed
2169 				 * from the record, so it is necessary to
2170 				 * remove the data mbufs, if any, to preserve
2171 				 * the invariant in the case of PR_ADDR that
2172 				 * requires MT_SONAME mbufs at the head of
2173 				 * each record.
2174 				 */
2175 				if (pr->pr_flags & PR_ATOMIC &&
2176 				    ((flags & MSG_PEEK) == 0))
2177 					(void)sbdroprecord_locked(&so->so_rcv);
2178 				SOCKBUF_UNLOCK(&so->so_rcv);
2179 				goto release;
2180 			}
2181 		} else
2182 			uio->uio_resid -= len;
2183 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2184 		if (len == m->m_len - moff) {
2185 			if (m->m_flags & M_EOR)
2186 				flags |= MSG_EOR;
2187 			if (flags & MSG_PEEK) {
2188 				m = m->m_next;
2189 				moff = 0;
2190 			} else {
2191 				nextrecord = m->m_nextpkt;
2192 				sbfree(&so->so_rcv, m);
2193 				if (mp != NULL) {
2194 					m->m_nextpkt = NULL;
2195 					*mp = m;
2196 					mp = &m->m_next;
2197 					so->so_rcv.sb_mb = m = m->m_next;
2198 					*mp = NULL;
2199 				} else {
2200 					so->so_rcv.sb_mb = m_free(m);
2201 					m = so->so_rcv.sb_mb;
2202 				}
2203 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2204 				SBLASTRECORDCHK(&so->so_rcv);
2205 				SBLASTMBUFCHK(&so->so_rcv);
2206 			}
2207 		} else {
2208 			if (flags & MSG_PEEK)
2209 				moff += len;
2210 			else {
2211 				if (mp != NULL) {
2212 					if (flags & MSG_DONTWAIT) {
2213 						*mp = m_copym(m, 0, len,
2214 						    M_NOWAIT);
2215 						if (*mp == NULL) {
2216 							/*
2217 							 * m_copym() couldn't
2218 							 * allocate an mbuf.
2219 							 * Adjust uio_resid back
2220 							 * (it was adjusted
2221 							 * down by len bytes,
2222 							 * which we didn't end
2223 							 * up "copying" over).
2224 							 */
2225 							uio->uio_resid += len;
2226 							break;
2227 						}
2228 					} else {
2229 						SOCKBUF_UNLOCK(&so->so_rcv);
2230 						*mp = m_copym(m, 0, len,
2231 						    M_WAITOK);
2232 						SOCKBUF_LOCK(&so->so_rcv);
2233 					}
2234 				}
2235 				sbcut_locked(&so->so_rcv, len);
2236 			}
2237 		}
2238 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2239 		if (so->so_oobmark) {
2240 			if ((flags & MSG_PEEK) == 0) {
2241 				so->so_oobmark -= len;
2242 				if (so->so_oobmark == 0) {
2243 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2244 					break;
2245 				}
2246 			} else {
2247 				offset += len;
2248 				if (offset == so->so_oobmark)
2249 					break;
2250 			}
2251 		}
2252 		if (flags & MSG_EOR)
2253 			break;
2254 		/*
2255 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2256 		 * must not quit until "uio->uio_resid == 0" or an error
2257 		 * termination.  If a signal/timeout occurs, return with a
2258 		 * short count but without error.  Keep sockbuf locked
2259 		 * against other readers.
2260 		 */
2261 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2262 		    !sosendallatonce(so) && nextrecord == NULL) {
2263 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2264 			if (so->so_error ||
2265 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2266 				break;
2267 			/*
2268 			 * Notify the protocol that some data has been
2269 			 * drained before blocking.
2270 			 */
2271 			if (pr->pr_flags & PR_WANTRCVD) {
2272 				SOCKBUF_UNLOCK(&so->so_rcv);
2273 				VNET_SO_ASSERT(so);
2274 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2275 				SOCKBUF_LOCK(&so->so_rcv);
2276 			}
2277 			SBLASTRECORDCHK(&so->so_rcv);
2278 			SBLASTMBUFCHK(&so->so_rcv);
2279 			/*
2280 			 * We could receive some data while was notifying
2281 			 * the protocol. Skip blocking in this case.
2282 			 */
2283 			if (so->so_rcv.sb_mb == NULL) {
2284 				error = sbwait(&so->so_rcv);
2285 				if (error) {
2286 					SOCKBUF_UNLOCK(&so->so_rcv);
2287 					goto release;
2288 				}
2289 			}
2290 			m = so->so_rcv.sb_mb;
2291 			if (m != NULL)
2292 				nextrecord = m->m_nextpkt;
2293 		}
2294 	}
2295 
2296 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2297 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2298 		flags |= MSG_TRUNC;
2299 		if ((flags & MSG_PEEK) == 0)
2300 			(void) sbdroprecord_locked(&so->so_rcv);
2301 	}
2302 	if ((flags & MSG_PEEK) == 0) {
2303 		if (m == NULL) {
2304 			/*
2305 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2306 			 * part makes sure sb_lastrecord is up-to-date if
2307 			 * there is still data in the socket buffer.
2308 			 */
2309 			so->so_rcv.sb_mb = nextrecord;
2310 			if (so->so_rcv.sb_mb == NULL) {
2311 				so->so_rcv.sb_mbtail = NULL;
2312 				so->so_rcv.sb_lastrecord = NULL;
2313 			} else if (nextrecord->m_nextpkt == NULL)
2314 				so->so_rcv.sb_lastrecord = nextrecord;
2315 		}
2316 		SBLASTRECORDCHK(&so->so_rcv);
2317 		SBLASTMBUFCHK(&so->so_rcv);
2318 		/*
2319 		 * If soreceive() is being done from the socket callback,
2320 		 * then don't need to generate ACK to peer to update window,
2321 		 * since ACK will be generated on return to TCP.
2322 		 */
2323 		if (!(flags & MSG_SOCALLBCK) &&
2324 		    (pr->pr_flags & PR_WANTRCVD)) {
2325 			SOCKBUF_UNLOCK(&so->so_rcv);
2326 			VNET_SO_ASSERT(so);
2327 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2328 			SOCKBUF_LOCK(&so->so_rcv);
2329 		}
2330 	}
2331 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2332 	if (orig_resid == uio->uio_resid && orig_resid &&
2333 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2334 		SOCKBUF_UNLOCK(&so->so_rcv);
2335 		goto restart;
2336 	}
2337 	SOCKBUF_UNLOCK(&so->so_rcv);
2338 
2339 	if (flagsp != NULL)
2340 		*flagsp |= flags;
2341 release:
2342 	sbunlock(&so->so_rcv);
2343 	return (error);
2344 }
2345 
2346 /*
2347  * Optimized version of soreceive() for stream (TCP) sockets.
2348  */
2349 int
2350 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2351     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2352 {
2353 	int len = 0, error = 0, flags, oresid;
2354 	struct sockbuf *sb;
2355 	struct mbuf *m, *n = NULL;
2356 
2357 	/* We only do stream sockets. */
2358 	if (so->so_type != SOCK_STREAM)
2359 		return (EINVAL);
2360 	if (psa != NULL)
2361 		*psa = NULL;
2362 	if (flagsp != NULL)
2363 		flags = *flagsp &~ MSG_EOR;
2364 	else
2365 		flags = 0;
2366 	if (controlp != NULL)
2367 		*controlp = NULL;
2368 	if (flags & MSG_OOB)
2369 		return (soreceive_rcvoob(so, uio, flags));
2370 	if (mp0 != NULL)
2371 		*mp0 = NULL;
2372 
2373 	sb = &so->so_rcv;
2374 
2375 	/* Prevent other readers from entering the socket. */
2376 	error = sblock(sb, SBLOCKWAIT(flags));
2377 	if (error)
2378 		return (error);
2379 	SOCKBUF_LOCK(sb);
2380 
2381 	/* Easy one, no space to copyout anything. */
2382 	if (uio->uio_resid == 0) {
2383 		error = EINVAL;
2384 		goto out;
2385 	}
2386 	oresid = uio->uio_resid;
2387 
2388 	/* We will never ever get anything unless we are or were connected. */
2389 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2390 		error = ENOTCONN;
2391 		goto out;
2392 	}
2393 
2394 restart:
2395 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2396 
2397 	/* Abort if socket has reported problems. */
2398 	if (so->so_error) {
2399 		if (sbavail(sb) > 0)
2400 			goto deliver;
2401 		if (oresid > uio->uio_resid)
2402 			goto out;
2403 		error = so->so_error;
2404 		if (!(flags & MSG_PEEK))
2405 			so->so_error = 0;
2406 		goto out;
2407 	}
2408 
2409 	/* Door is closed.  Deliver what is left, if any. */
2410 	if (sb->sb_state & SBS_CANTRCVMORE) {
2411 		if (sbavail(sb) > 0)
2412 			goto deliver;
2413 		else
2414 			goto out;
2415 	}
2416 
2417 	/* Socket buffer is empty and we shall not block. */
2418 	if (sbavail(sb) == 0 &&
2419 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2420 		error = EAGAIN;
2421 		goto out;
2422 	}
2423 
2424 	/* Socket buffer got some data that we shall deliver now. */
2425 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2426 	    ((so->so_state & SS_NBIO) ||
2427 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2428 	     sbavail(sb) >= sb->sb_lowat ||
2429 	     sbavail(sb) >= uio->uio_resid ||
2430 	     sbavail(sb) >= sb->sb_hiwat) ) {
2431 		goto deliver;
2432 	}
2433 
2434 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2435 	if ((flags & MSG_WAITALL) &&
2436 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2437 		goto deliver;
2438 
2439 	/*
2440 	 * Wait and block until (more) data comes in.
2441 	 * NB: Drops the sockbuf lock during wait.
2442 	 */
2443 	error = sbwait(sb);
2444 	if (error)
2445 		goto out;
2446 	goto restart;
2447 
2448 deliver:
2449 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2450 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2451 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2452 
2453 	/* Statistics. */
2454 	if (uio->uio_td)
2455 		uio->uio_td->td_ru.ru_msgrcv++;
2456 
2457 	/* Fill uio until full or current end of socket buffer is reached. */
2458 	len = min(uio->uio_resid, sbavail(sb));
2459 	if (mp0 != NULL) {
2460 		/* Dequeue as many mbufs as possible. */
2461 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2462 			if (*mp0 == NULL)
2463 				*mp0 = sb->sb_mb;
2464 			else
2465 				m_cat(*mp0, sb->sb_mb);
2466 			for (m = sb->sb_mb;
2467 			     m != NULL && m->m_len <= len;
2468 			     m = m->m_next) {
2469 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2470 				    ("%s: m %p not available", __func__, m));
2471 				len -= m->m_len;
2472 				uio->uio_resid -= m->m_len;
2473 				sbfree(sb, m);
2474 				n = m;
2475 			}
2476 			n->m_next = NULL;
2477 			sb->sb_mb = m;
2478 			sb->sb_lastrecord = sb->sb_mb;
2479 			if (sb->sb_mb == NULL)
2480 				SB_EMPTY_FIXUP(sb);
2481 		}
2482 		/* Copy the remainder. */
2483 		if (len > 0) {
2484 			KASSERT(sb->sb_mb != NULL,
2485 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2486 
2487 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2488 			if (m == NULL)
2489 				len = 0;	/* Don't flush data from sockbuf. */
2490 			else
2491 				uio->uio_resid -= len;
2492 			if (*mp0 != NULL)
2493 				m_cat(*mp0, m);
2494 			else
2495 				*mp0 = m;
2496 			if (*mp0 == NULL) {
2497 				error = ENOBUFS;
2498 				goto out;
2499 			}
2500 		}
2501 	} else {
2502 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2503 		SOCKBUF_UNLOCK(sb);
2504 		error = m_mbuftouio(uio, sb->sb_mb, len);
2505 		SOCKBUF_LOCK(sb);
2506 		if (error)
2507 			goto out;
2508 	}
2509 	SBLASTRECORDCHK(sb);
2510 	SBLASTMBUFCHK(sb);
2511 
2512 	/*
2513 	 * Remove the delivered data from the socket buffer unless we
2514 	 * were only peeking.
2515 	 */
2516 	if (!(flags & MSG_PEEK)) {
2517 		if (len > 0)
2518 			sbdrop_locked(sb, len);
2519 
2520 		/* Notify protocol that we drained some data. */
2521 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2522 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2523 		     !(flags & MSG_SOCALLBCK))) {
2524 			SOCKBUF_UNLOCK(sb);
2525 			VNET_SO_ASSERT(so);
2526 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2527 			SOCKBUF_LOCK(sb);
2528 		}
2529 	}
2530 
2531 	/*
2532 	 * For MSG_WAITALL we may have to loop again and wait for
2533 	 * more data to come in.
2534 	 */
2535 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2536 		goto restart;
2537 out:
2538 	SOCKBUF_LOCK_ASSERT(sb);
2539 	SBLASTRECORDCHK(sb);
2540 	SBLASTMBUFCHK(sb);
2541 	SOCKBUF_UNLOCK(sb);
2542 	sbunlock(sb);
2543 	return (error);
2544 }
2545 
2546 /*
2547  * Optimized version of soreceive() for simple datagram cases from userspace.
2548  * Unlike in the stream case, we're able to drop a datagram if copyout()
2549  * fails, and because we handle datagrams atomically, we don't need to use a
2550  * sleep lock to prevent I/O interlacing.
2551  */
2552 int
2553 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2554     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2555 {
2556 	struct mbuf *m, *m2;
2557 	int flags, error;
2558 	ssize_t len;
2559 	struct protosw *pr = so->so_proto;
2560 	struct mbuf *nextrecord;
2561 
2562 	if (psa != NULL)
2563 		*psa = NULL;
2564 	if (controlp != NULL)
2565 		*controlp = NULL;
2566 	if (flagsp != NULL)
2567 		flags = *flagsp &~ MSG_EOR;
2568 	else
2569 		flags = 0;
2570 
2571 	/*
2572 	 * For any complicated cases, fall back to the full
2573 	 * soreceive_generic().
2574 	 */
2575 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2576 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2577 		    flagsp));
2578 
2579 	/*
2580 	 * Enforce restrictions on use.
2581 	 */
2582 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2583 	    ("soreceive_dgram: wantrcvd"));
2584 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2585 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2586 	    ("soreceive_dgram: SBS_RCVATMARK"));
2587 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2588 	    ("soreceive_dgram: P_CONNREQUIRED"));
2589 
2590 	/*
2591 	 * Loop blocking while waiting for a datagram.
2592 	 */
2593 	SOCKBUF_LOCK(&so->so_rcv);
2594 	while ((m = so->so_rcv.sb_mb) == NULL) {
2595 		KASSERT(sbavail(&so->so_rcv) == 0,
2596 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2597 		    sbavail(&so->so_rcv)));
2598 		if (so->so_error) {
2599 			error = so->so_error;
2600 			so->so_error = 0;
2601 			SOCKBUF_UNLOCK(&so->so_rcv);
2602 			return (error);
2603 		}
2604 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2605 		    uio->uio_resid == 0) {
2606 			SOCKBUF_UNLOCK(&so->so_rcv);
2607 			return (0);
2608 		}
2609 		if ((so->so_state & SS_NBIO) ||
2610 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2611 			SOCKBUF_UNLOCK(&so->so_rcv);
2612 			return (EWOULDBLOCK);
2613 		}
2614 		SBLASTRECORDCHK(&so->so_rcv);
2615 		SBLASTMBUFCHK(&so->so_rcv);
2616 		error = sbwait(&so->so_rcv);
2617 		if (error) {
2618 			SOCKBUF_UNLOCK(&so->so_rcv);
2619 			return (error);
2620 		}
2621 	}
2622 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2623 
2624 	if (uio->uio_td)
2625 		uio->uio_td->td_ru.ru_msgrcv++;
2626 	SBLASTRECORDCHK(&so->so_rcv);
2627 	SBLASTMBUFCHK(&so->so_rcv);
2628 	nextrecord = m->m_nextpkt;
2629 	if (nextrecord == NULL) {
2630 		KASSERT(so->so_rcv.sb_lastrecord == m,
2631 		    ("soreceive_dgram: lastrecord != m"));
2632 	}
2633 
2634 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2635 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2636 
2637 	/*
2638 	 * Pull 'm' and its chain off the front of the packet queue.
2639 	 */
2640 	so->so_rcv.sb_mb = NULL;
2641 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2642 
2643 	/*
2644 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2645 	 */
2646 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2647 		sbfree(&so->so_rcv, m2);
2648 
2649 	/*
2650 	 * Do a few last checks before we let go of the lock.
2651 	 */
2652 	SBLASTRECORDCHK(&so->so_rcv);
2653 	SBLASTMBUFCHK(&so->so_rcv);
2654 	SOCKBUF_UNLOCK(&so->so_rcv);
2655 
2656 	if (pr->pr_flags & PR_ADDR) {
2657 		KASSERT(m->m_type == MT_SONAME,
2658 		    ("m->m_type == %d", m->m_type));
2659 		if (psa != NULL)
2660 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2661 			    M_NOWAIT);
2662 		m = m_free(m);
2663 	}
2664 	if (m == NULL) {
2665 		/* XXXRW: Can this happen? */
2666 		return (0);
2667 	}
2668 
2669 	/*
2670 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2671 	 * queue.
2672 	 *
2673 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2674 	 * in the first mbuf chain on the socket buffer.  We call into the
2675 	 * protocol to perform externalization (or freeing if controlp ==
2676 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2677 	 * MT_DATA mbufs.
2678 	 */
2679 	if (m->m_type == MT_CONTROL) {
2680 		struct mbuf *cm = NULL, *cmn;
2681 		struct mbuf **cme = &cm;
2682 
2683 		do {
2684 			m2 = m->m_next;
2685 			m->m_next = NULL;
2686 			*cme = m;
2687 			cme = &(*cme)->m_next;
2688 			m = m2;
2689 		} while (m != NULL && m->m_type == MT_CONTROL);
2690 		while (cm != NULL) {
2691 			cmn = cm->m_next;
2692 			cm->m_next = NULL;
2693 			if (pr->pr_domain->dom_externalize != NULL) {
2694 				error = (*pr->pr_domain->dom_externalize)
2695 				    (cm, controlp, flags);
2696 			} else if (controlp != NULL)
2697 				*controlp = cm;
2698 			else
2699 				m_freem(cm);
2700 			if (controlp != NULL) {
2701 				while (*controlp != NULL)
2702 					controlp = &(*controlp)->m_next;
2703 			}
2704 			cm = cmn;
2705 		}
2706 	}
2707 	KASSERT(m == NULL || m->m_type == MT_DATA,
2708 	    ("soreceive_dgram: !data"));
2709 	while (m != NULL && uio->uio_resid > 0) {
2710 		len = uio->uio_resid;
2711 		if (len > m->m_len)
2712 			len = m->m_len;
2713 		error = uiomove(mtod(m, char *), (int)len, uio);
2714 		if (error) {
2715 			m_freem(m);
2716 			return (error);
2717 		}
2718 		if (len == m->m_len)
2719 			m = m_free(m);
2720 		else {
2721 			m->m_data += len;
2722 			m->m_len -= len;
2723 		}
2724 	}
2725 	if (m != NULL) {
2726 		flags |= MSG_TRUNC;
2727 		m_freem(m);
2728 	}
2729 	if (flagsp != NULL)
2730 		*flagsp |= flags;
2731 	return (0);
2732 }
2733 
2734 int
2735 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2736     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2737 {
2738 	int error;
2739 
2740 	CURVNET_SET(so->so_vnet);
2741 	if (!SOLISTENING(so))
2742 		error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2743 		    mp0, controlp, flagsp));
2744 	else
2745 		error = ENOTCONN;
2746 	CURVNET_RESTORE();
2747 	return (error);
2748 }
2749 
2750 int
2751 soshutdown(struct socket *so, int how)
2752 {
2753 	struct protosw *pr = so->so_proto;
2754 	int error, soerror_enotconn;
2755 
2756 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2757 		return (EINVAL);
2758 
2759 	soerror_enotconn = 0;
2760 	if ((so->so_state &
2761 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2762 		/*
2763 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2764 		 * invoked on a datagram sockets, however historically we would
2765 		 * actually tear socket down. This is known to be leveraged by
2766 		 * some applications to unblock process waiting in recvXXX(2)
2767 		 * by other process that it shares that socket with. Try to meet
2768 		 * both backward-compatibility and POSIX requirements by forcing
2769 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2770 		 */
2771 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so))
2772 			return (ENOTCONN);
2773 		soerror_enotconn = 1;
2774 	}
2775 
2776 	if (SOLISTENING(so)) {
2777 		if (how != SHUT_WR) {
2778 			SOLISTEN_LOCK(so);
2779 			so->so_error = ECONNABORTED;
2780 			solisten_wakeup(so);	/* unlocks so */
2781 		}
2782 		goto done;
2783 	}
2784 
2785 	CURVNET_SET(so->so_vnet);
2786 	if (pr->pr_usrreqs->pru_flush != NULL)
2787 		(*pr->pr_usrreqs->pru_flush)(so, how);
2788 	if (how != SHUT_WR)
2789 		sorflush(so);
2790 	if (how != SHUT_RD) {
2791 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2792 		wakeup(&so->so_timeo);
2793 		CURVNET_RESTORE();
2794 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2795 	}
2796 	wakeup(&so->so_timeo);
2797 	CURVNET_RESTORE();
2798 
2799 done:
2800 	return (soerror_enotconn ? ENOTCONN : 0);
2801 }
2802 
2803 void
2804 sorflush(struct socket *so)
2805 {
2806 	struct sockbuf *sb = &so->so_rcv;
2807 	struct protosw *pr = so->so_proto;
2808 	struct socket aso;
2809 
2810 	VNET_SO_ASSERT(so);
2811 
2812 	/*
2813 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2814 	 * held, and in order to generally avoid holding the lock for a long
2815 	 * time, we make a copy of the socket buffer and clear the original
2816 	 * (except locks, state).  The new socket buffer copy won't have
2817 	 * initialized locks so we can only call routines that won't use or
2818 	 * assert those locks.
2819 	 *
2820 	 * Dislodge threads currently blocked in receive and wait to acquire
2821 	 * a lock against other simultaneous readers before clearing the
2822 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2823 	 * despite any existing socket disposition on interruptable waiting.
2824 	 */
2825 	socantrcvmore(so);
2826 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2827 
2828 	/*
2829 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2830 	 * and mutex data unchanged.
2831 	 */
2832 	SOCKBUF_LOCK(sb);
2833 	bzero(&aso, sizeof(aso));
2834 	aso.so_pcb = so->so_pcb;
2835 	bcopy(&sb->sb_startzero, &aso.so_rcv.sb_startzero,
2836 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2837 	bzero(&sb->sb_startzero,
2838 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2839 	SOCKBUF_UNLOCK(sb);
2840 	sbunlock(sb);
2841 
2842 	/*
2843 	 * Dispose of special rights and flush the copied socket.  Don't call
2844 	 * any unsafe routines (that rely on locks being initialized) on aso.
2845 	 */
2846 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2847 		(*pr->pr_domain->dom_dispose)(&aso);
2848 	sbrelease_internal(&aso.so_rcv, so);
2849 }
2850 
2851 /*
2852  * Wrapper for Socket established helper hook.
2853  * Parameters: socket, context of the hook point, hook id.
2854  */
2855 static int inline
2856 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2857 {
2858 	struct socket_hhook_data hhook_data = {
2859 		.so = so,
2860 		.hctx = hctx,
2861 		.m = NULL,
2862 		.status = 0
2863 	};
2864 
2865 	CURVNET_SET(so->so_vnet);
2866 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2867 	CURVNET_RESTORE();
2868 
2869 	/* Ugly but needed, since hhooks return void for now */
2870 	return (hhook_data.status);
2871 }
2872 
2873 /*
2874  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2875  * additional variant to handle the case where the option value needs to be
2876  * some kind of integer, but not a specific size.  In addition to their use
2877  * here, these functions are also called by the protocol-level pr_ctloutput()
2878  * routines.
2879  */
2880 int
2881 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2882 {
2883 	size_t	valsize;
2884 
2885 	/*
2886 	 * If the user gives us more than we wanted, we ignore it, but if we
2887 	 * don't get the minimum length the caller wants, we return EINVAL.
2888 	 * On success, sopt->sopt_valsize is set to however much we actually
2889 	 * retrieved.
2890 	 */
2891 	if ((valsize = sopt->sopt_valsize) < minlen)
2892 		return EINVAL;
2893 	if (valsize > len)
2894 		sopt->sopt_valsize = valsize = len;
2895 
2896 	if (sopt->sopt_td != NULL)
2897 		return (copyin(sopt->sopt_val, buf, valsize));
2898 
2899 	bcopy(sopt->sopt_val, buf, valsize);
2900 	return (0);
2901 }
2902 
2903 /*
2904  * Kernel version of setsockopt(2).
2905  *
2906  * XXX: optlen is size_t, not socklen_t
2907  */
2908 int
2909 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2910     size_t optlen)
2911 {
2912 	struct sockopt sopt;
2913 
2914 	sopt.sopt_level = level;
2915 	sopt.sopt_name = optname;
2916 	sopt.sopt_dir = SOPT_SET;
2917 	sopt.sopt_val = optval;
2918 	sopt.sopt_valsize = optlen;
2919 	sopt.sopt_td = NULL;
2920 	return (sosetopt(so, &sopt));
2921 }
2922 
2923 int
2924 sosetopt(struct socket *so, struct sockopt *sopt)
2925 {
2926 	int	error, optval;
2927 	struct	linger l;
2928 	struct	timeval tv;
2929 	sbintime_t val;
2930 	uint32_t val32;
2931 #ifdef MAC
2932 	struct mac extmac;
2933 #endif
2934 
2935 	CURVNET_SET(so->so_vnet);
2936 	error = 0;
2937 	if (sopt->sopt_level != SOL_SOCKET) {
2938 		if (so->so_proto->pr_ctloutput != NULL)
2939 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2940 		else
2941 			error = ENOPROTOOPT;
2942 	} else {
2943 		switch (sopt->sopt_name) {
2944 		case SO_ACCEPTFILTER:
2945 			error = accept_filt_setopt(so, sopt);
2946 			if (error)
2947 				goto bad;
2948 			break;
2949 
2950 		case SO_LINGER:
2951 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2952 			if (error)
2953 				goto bad;
2954 			if (l.l_linger < 0 ||
2955 			    l.l_linger > USHRT_MAX ||
2956 			    l.l_linger > (INT_MAX / hz)) {
2957 				error = EDOM;
2958 				goto bad;
2959 			}
2960 			SOCK_LOCK(so);
2961 			so->so_linger = l.l_linger;
2962 			if (l.l_onoff)
2963 				so->so_options |= SO_LINGER;
2964 			else
2965 				so->so_options &= ~SO_LINGER;
2966 			SOCK_UNLOCK(so);
2967 			break;
2968 
2969 		case SO_DEBUG:
2970 		case SO_KEEPALIVE:
2971 		case SO_DONTROUTE:
2972 		case SO_USELOOPBACK:
2973 		case SO_BROADCAST:
2974 		case SO_REUSEADDR:
2975 		case SO_REUSEPORT:
2976 		case SO_REUSEPORT_LB:
2977 		case SO_OOBINLINE:
2978 		case SO_TIMESTAMP:
2979 		case SO_BINTIME:
2980 		case SO_NOSIGPIPE:
2981 		case SO_NO_DDP:
2982 		case SO_NO_OFFLOAD:
2983 			error = sooptcopyin(sopt, &optval, sizeof optval,
2984 			    sizeof optval);
2985 			if (error)
2986 				goto bad;
2987 			SOCK_LOCK(so);
2988 			if (optval)
2989 				so->so_options |= sopt->sopt_name;
2990 			else
2991 				so->so_options &= ~sopt->sopt_name;
2992 			SOCK_UNLOCK(so);
2993 			break;
2994 
2995 		case SO_SETFIB:
2996 			error = sooptcopyin(sopt, &optval, sizeof optval,
2997 			    sizeof optval);
2998 			if (error)
2999 				goto bad;
3000 
3001 			if (optval < 0 || optval >= rt_numfibs) {
3002 				error = EINVAL;
3003 				goto bad;
3004 			}
3005 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3006 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3007 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3008 				so->so_fibnum = optval;
3009 			else
3010 				so->so_fibnum = 0;
3011 			break;
3012 
3013 		case SO_USER_COOKIE:
3014 			error = sooptcopyin(sopt, &val32, sizeof val32,
3015 			    sizeof val32);
3016 			if (error)
3017 				goto bad;
3018 			so->so_user_cookie = val32;
3019 			break;
3020 
3021 		case SO_SNDBUF:
3022 		case SO_RCVBUF:
3023 		case SO_SNDLOWAT:
3024 		case SO_RCVLOWAT:
3025 			error = sooptcopyin(sopt, &optval, sizeof optval,
3026 			    sizeof optval);
3027 			if (error)
3028 				goto bad;
3029 
3030 			/*
3031 			 * Values < 1 make no sense for any of these options,
3032 			 * so disallow them.
3033 			 */
3034 			if (optval < 1) {
3035 				error = EINVAL;
3036 				goto bad;
3037 			}
3038 
3039 			error = sbsetopt(so, sopt->sopt_name, optval);
3040 			break;
3041 
3042 		case SO_SNDTIMEO:
3043 		case SO_RCVTIMEO:
3044 #ifdef COMPAT_FREEBSD32
3045 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3046 				struct timeval32 tv32;
3047 
3048 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3049 				    sizeof tv32);
3050 				CP(tv32, tv, tv_sec);
3051 				CP(tv32, tv, tv_usec);
3052 			} else
3053 #endif
3054 				error = sooptcopyin(sopt, &tv, sizeof tv,
3055 				    sizeof tv);
3056 			if (error)
3057 				goto bad;
3058 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3059 			    tv.tv_usec >= 1000000) {
3060 				error = EDOM;
3061 				goto bad;
3062 			}
3063 			if (tv.tv_sec > INT32_MAX)
3064 				val = SBT_MAX;
3065 			else
3066 				val = tvtosbt(tv);
3067 			switch (sopt->sopt_name) {
3068 			case SO_SNDTIMEO:
3069 				so->so_snd.sb_timeo = val;
3070 				break;
3071 			case SO_RCVTIMEO:
3072 				so->so_rcv.sb_timeo = val;
3073 				break;
3074 			}
3075 			break;
3076 
3077 		case SO_LABEL:
3078 #ifdef MAC
3079 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3080 			    sizeof extmac);
3081 			if (error)
3082 				goto bad;
3083 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3084 			    so, &extmac);
3085 #else
3086 			error = EOPNOTSUPP;
3087 #endif
3088 			break;
3089 
3090 		case SO_TS_CLOCK:
3091 			error = sooptcopyin(sopt, &optval, sizeof optval,
3092 			    sizeof optval);
3093 			if (error)
3094 				goto bad;
3095 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3096 				error = EINVAL;
3097 				goto bad;
3098 			}
3099 			so->so_ts_clock = optval;
3100 			break;
3101 
3102 		case SO_MAX_PACING_RATE:
3103 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3104 			    sizeof(val32));
3105 			if (error)
3106 				goto bad;
3107 			so->so_max_pacing_rate = val32;
3108 			break;
3109 
3110 		default:
3111 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3112 				error = hhook_run_socket(so, sopt,
3113 				    HHOOK_SOCKET_OPT);
3114 			else
3115 				error = ENOPROTOOPT;
3116 			break;
3117 		}
3118 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3119 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3120 	}
3121 bad:
3122 	CURVNET_RESTORE();
3123 	return (error);
3124 }
3125 
3126 /*
3127  * Helper routine for getsockopt.
3128  */
3129 int
3130 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3131 {
3132 	int	error;
3133 	size_t	valsize;
3134 
3135 	error = 0;
3136 
3137 	/*
3138 	 * Documented get behavior is that we always return a value, possibly
3139 	 * truncated to fit in the user's buffer.  Traditional behavior is
3140 	 * that we always tell the user precisely how much we copied, rather
3141 	 * than something useful like the total amount we had available for
3142 	 * her.  Note that this interface is not idempotent; the entire
3143 	 * answer must be generated ahead of time.
3144 	 */
3145 	valsize = min(len, sopt->sopt_valsize);
3146 	sopt->sopt_valsize = valsize;
3147 	if (sopt->sopt_val != NULL) {
3148 		if (sopt->sopt_td != NULL)
3149 			error = copyout(buf, sopt->sopt_val, valsize);
3150 		else
3151 			bcopy(buf, sopt->sopt_val, valsize);
3152 	}
3153 	return (error);
3154 }
3155 
3156 int
3157 sogetopt(struct socket *so, struct sockopt *sopt)
3158 {
3159 	int	error, optval;
3160 	struct	linger l;
3161 	struct	timeval tv;
3162 #ifdef MAC
3163 	struct mac extmac;
3164 #endif
3165 
3166 	CURVNET_SET(so->so_vnet);
3167 	error = 0;
3168 	if (sopt->sopt_level != SOL_SOCKET) {
3169 		if (so->so_proto->pr_ctloutput != NULL)
3170 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3171 		else
3172 			error = ENOPROTOOPT;
3173 		CURVNET_RESTORE();
3174 		return (error);
3175 	} else {
3176 		switch (sopt->sopt_name) {
3177 		case SO_ACCEPTFILTER:
3178 			error = accept_filt_getopt(so, sopt);
3179 			break;
3180 
3181 		case SO_LINGER:
3182 			SOCK_LOCK(so);
3183 			l.l_onoff = so->so_options & SO_LINGER;
3184 			l.l_linger = so->so_linger;
3185 			SOCK_UNLOCK(so);
3186 			error = sooptcopyout(sopt, &l, sizeof l);
3187 			break;
3188 
3189 		case SO_USELOOPBACK:
3190 		case SO_DONTROUTE:
3191 		case SO_DEBUG:
3192 		case SO_KEEPALIVE:
3193 		case SO_REUSEADDR:
3194 		case SO_REUSEPORT:
3195 		case SO_REUSEPORT_LB:
3196 		case SO_BROADCAST:
3197 		case SO_OOBINLINE:
3198 		case SO_ACCEPTCONN:
3199 		case SO_TIMESTAMP:
3200 		case SO_BINTIME:
3201 		case SO_NOSIGPIPE:
3202 			optval = so->so_options & sopt->sopt_name;
3203 integer:
3204 			error = sooptcopyout(sopt, &optval, sizeof optval);
3205 			break;
3206 
3207 		case SO_DOMAIN:
3208 			optval = so->so_proto->pr_domain->dom_family;
3209 			goto integer;
3210 
3211 		case SO_TYPE:
3212 			optval = so->so_type;
3213 			goto integer;
3214 
3215 		case SO_PROTOCOL:
3216 			optval = so->so_proto->pr_protocol;
3217 			goto integer;
3218 
3219 		case SO_ERROR:
3220 			SOCK_LOCK(so);
3221 			optval = so->so_error;
3222 			so->so_error = 0;
3223 			SOCK_UNLOCK(so);
3224 			goto integer;
3225 
3226 		case SO_SNDBUF:
3227 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3228 			    so->so_snd.sb_hiwat;
3229 			goto integer;
3230 
3231 		case SO_RCVBUF:
3232 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3233 			    so->so_rcv.sb_hiwat;
3234 			goto integer;
3235 
3236 		case SO_SNDLOWAT:
3237 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3238 			    so->so_snd.sb_lowat;
3239 			goto integer;
3240 
3241 		case SO_RCVLOWAT:
3242 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3243 			    so->so_rcv.sb_lowat;
3244 			goto integer;
3245 
3246 		case SO_SNDTIMEO:
3247 		case SO_RCVTIMEO:
3248 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3249 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3250 #ifdef COMPAT_FREEBSD32
3251 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3252 				struct timeval32 tv32;
3253 
3254 				CP(tv, tv32, tv_sec);
3255 				CP(tv, tv32, tv_usec);
3256 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3257 			} else
3258 #endif
3259 				error = sooptcopyout(sopt, &tv, sizeof tv);
3260 			break;
3261 
3262 		case SO_LABEL:
3263 #ifdef MAC
3264 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3265 			    sizeof(extmac));
3266 			if (error)
3267 				goto bad;
3268 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3269 			    so, &extmac);
3270 			if (error)
3271 				goto bad;
3272 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3273 #else
3274 			error = EOPNOTSUPP;
3275 #endif
3276 			break;
3277 
3278 		case SO_PEERLABEL:
3279 #ifdef MAC
3280 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3281 			    sizeof(extmac));
3282 			if (error)
3283 				goto bad;
3284 			error = mac_getsockopt_peerlabel(
3285 			    sopt->sopt_td->td_ucred, so, &extmac);
3286 			if (error)
3287 				goto bad;
3288 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3289 #else
3290 			error = EOPNOTSUPP;
3291 #endif
3292 			break;
3293 
3294 		case SO_LISTENQLIMIT:
3295 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3296 			goto integer;
3297 
3298 		case SO_LISTENQLEN:
3299 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3300 			goto integer;
3301 
3302 		case SO_LISTENINCQLEN:
3303 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3304 			goto integer;
3305 
3306 		case SO_TS_CLOCK:
3307 			optval = so->so_ts_clock;
3308 			goto integer;
3309 
3310 		case SO_MAX_PACING_RATE:
3311 			optval = so->so_max_pacing_rate;
3312 			goto integer;
3313 
3314 		default:
3315 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3316 				error = hhook_run_socket(so, sopt,
3317 				    HHOOK_SOCKET_OPT);
3318 			else
3319 				error = ENOPROTOOPT;
3320 			break;
3321 		}
3322 	}
3323 #ifdef MAC
3324 bad:
3325 #endif
3326 	CURVNET_RESTORE();
3327 	return (error);
3328 }
3329 
3330 int
3331 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3332 {
3333 	struct mbuf *m, *m_prev;
3334 	int sopt_size = sopt->sopt_valsize;
3335 
3336 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3337 	if (m == NULL)
3338 		return ENOBUFS;
3339 	if (sopt_size > MLEN) {
3340 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3341 		if ((m->m_flags & M_EXT) == 0) {
3342 			m_free(m);
3343 			return ENOBUFS;
3344 		}
3345 		m->m_len = min(MCLBYTES, sopt_size);
3346 	} else {
3347 		m->m_len = min(MLEN, sopt_size);
3348 	}
3349 	sopt_size -= m->m_len;
3350 	*mp = m;
3351 	m_prev = m;
3352 
3353 	while (sopt_size) {
3354 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3355 		if (m == NULL) {
3356 			m_freem(*mp);
3357 			return ENOBUFS;
3358 		}
3359 		if (sopt_size > MLEN) {
3360 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3361 			    M_NOWAIT);
3362 			if ((m->m_flags & M_EXT) == 0) {
3363 				m_freem(m);
3364 				m_freem(*mp);
3365 				return ENOBUFS;
3366 			}
3367 			m->m_len = min(MCLBYTES, sopt_size);
3368 		} else {
3369 			m->m_len = min(MLEN, sopt_size);
3370 		}
3371 		sopt_size -= m->m_len;
3372 		m_prev->m_next = m;
3373 		m_prev = m;
3374 	}
3375 	return (0);
3376 }
3377 
3378 int
3379 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3380 {
3381 	struct mbuf *m0 = m;
3382 
3383 	if (sopt->sopt_val == NULL)
3384 		return (0);
3385 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3386 		if (sopt->sopt_td != NULL) {
3387 			int error;
3388 
3389 			error = copyin(sopt->sopt_val, mtod(m, char *),
3390 			    m->m_len);
3391 			if (error != 0) {
3392 				m_freem(m0);
3393 				return(error);
3394 			}
3395 		} else
3396 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3397 		sopt->sopt_valsize -= m->m_len;
3398 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3399 		m = m->m_next;
3400 	}
3401 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3402 		panic("ip6_sooptmcopyin");
3403 	return (0);
3404 }
3405 
3406 int
3407 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3408 {
3409 	struct mbuf *m0 = m;
3410 	size_t valsize = 0;
3411 
3412 	if (sopt->sopt_val == NULL)
3413 		return (0);
3414 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3415 		if (sopt->sopt_td != NULL) {
3416 			int error;
3417 
3418 			error = copyout(mtod(m, char *), sopt->sopt_val,
3419 			    m->m_len);
3420 			if (error != 0) {
3421 				m_freem(m0);
3422 				return(error);
3423 			}
3424 		} else
3425 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3426 		sopt->sopt_valsize -= m->m_len;
3427 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3428 		valsize += m->m_len;
3429 		m = m->m_next;
3430 	}
3431 	if (m != NULL) {
3432 		/* enough soopt buffer should be given from user-land */
3433 		m_freem(m0);
3434 		return(EINVAL);
3435 	}
3436 	sopt->sopt_valsize = valsize;
3437 	return (0);
3438 }
3439 
3440 /*
3441  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3442  * out-of-band data, which will then notify socket consumers.
3443  */
3444 void
3445 sohasoutofband(struct socket *so)
3446 {
3447 
3448 	if (so->so_sigio != NULL)
3449 		pgsigio(&so->so_sigio, SIGURG, 0);
3450 	selwakeuppri(&so->so_rdsel, PSOCK);
3451 }
3452 
3453 int
3454 sopoll(struct socket *so, int events, struct ucred *active_cred,
3455     struct thread *td)
3456 {
3457 
3458 	/*
3459 	 * We do not need to set or assert curvnet as long as everyone uses
3460 	 * sopoll_generic().
3461 	 */
3462 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3463 	    td));
3464 }
3465 
3466 int
3467 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3468     struct thread *td)
3469 {
3470 	int revents;
3471 
3472 	SOCK_LOCK(so);
3473 	if (SOLISTENING(so)) {
3474 		if (!(events & (POLLIN | POLLRDNORM)))
3475 			revents = 0;
3476 		else if (!TAILQ_EMPTY(&so->sol_comp))
3477 			revents = events & (POLLIN | POLLRDNORM);
3478 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3479 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3480 		else {
3481 			selrecord(td, &so->so_rdsel);
3482 			revents = 0;
3483 		}
3484 	} else {
3485 		revents = 0;
3486 		SOCKBUF_LOCK(&so->so_snd);
3487 		SOCKBUF_LOCK(&so->so_rcv);
3488 		if (events & (POLLIN | POLLRDNORM))
3489 			if (soreadabledata(so))
3490 				revents |= events & (POLLIN | POLLRDNORM);
3491 		if (events & (POLLOUT | POLLWRNORM))
3492 			if (sowriteable(so))
3493 				revents |= events & (POLLOUT | POLLWRNORM);
3494 		if (events & (POLLPRI | POLLRDBAND))
3495 			if (so->so_oobmark ||
3496 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3497 				revents |= events & (POLLPRI | POLLRDBAND);
3498 		if ((events & POLLINIGNEOF) == 0) {
3499 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3500 				revents |= events & (POLLIN | POLLRDNORM);
3501 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3502 					revents |= POLLHUP;
3503 			}
3504 		}
3505 		if (revents == 0) {
3506 			if (events &
3507 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3508 				selrecord(td, &so->so_rdsel);
3509 				so->so_rcv.sb_flags |= SB_SEL;
3510 			}
3511 			if (events & (POLLOUT | POLLWRNORM)) {
3512 				selrecord(td, &so->so_wrsel);
3513 				so->so_snd.sb_flags |= SB_SEL;
3514 			}
3515 		}
3516 		SOCKBUF_UNLOCK(&so->so_rcv);
3517 		SOCKBUF_UNLOCK(&so->so_snd);
3518 	}
3519 	SOCK_UNLOCK(so);
3520 	return (revents);
3521 }
3522 
3523 int
3524 soo_kqfilter(struct file *fp, struct knote *kn)
3525 {
3526 	struct socket *so = kn->kn_fp->f_data;
3527 	struct sockbuf *sb;
3528 	struct knlist *knl;
3529 
3530 	switch (kn->kn_filter) {
3531 	case EVFILT_READ:
3532 		kn->kn_fop = &soread_filtops;
3533 		knl = &so->so_rdsel.si_note;
3534 		sb = &so->so_rcv;
3535 		break;
3536 	case EVFILT_WRITE:
3537 		kn->kn_fop = &sowrite_filtops;
3538 		knl = &so->so_wrsel.si_note;
3539 		sb = &so->so_snd;
3540 		break;
3541 	case EVFILT_EMPTY:
3542 		kn->kn_fop = &soempty_filtops;
3543 		knl = &so->so_wrsel.si_note;
3544 		sb = &so->so_snd;
3545 		break;
3546 	default:
3547 		return (EINVAL);
3548 	}
3549 
3550 	SOCK_LOCK(so);
3551 	if (SOLISTENING(so)) {
3552 		knlist_add(knl, kn, 1);
3553 	} else {
3554 		SOCKBUF_LOCK(sb);
3555 		knlist_add(knl, kn, 1);
3556 		sb->sb_flags |= SB_KNOTE;
3557 		SOCKBUF_UNLOCK(sb);
3558 	}
3559 	SOCK_UNLOCK(so);
3560 	return (0);
3561 }
3562 
3563 /*
3564  * Some routines that return EOPNOTSUPP for entry points that are not
3565  * supported by a protocol.  Fill in as needed.
3566  */
3567 int
3568 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3569 {
3570 
3571 	return EOPNOTSUPP;
3572 }
3573 
3574 int
3575 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3576 {
3577 
3578 	return EOPNOTSUPP;
3579 }
3580 
3581 int
3582 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3583 {
3584 
3585 	return EOPNOTSUPP;
3586 }
3587 
3588 int
3589 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3590 {
3591 
3592 	return EOPNOTSUPP;
3593 }
3594 
3595 int
3596 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3597     struct thread *td)
3598 {
3599 
3600 	return EOPNOTSUPP;
3601 }
3602 
3603 int
3604 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3605 {
3606 
3607 	return EOPNOTSUPP;
3608 }
3609 
3610 int
3611 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3612     struct thread *td)
3613 {
3614 
3615 	return EOPNOTSUPP;
3616 }
3617 
3618 int
3619 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3620 {
3621 
3622 	return EOPNOTSUPP;
3623 }
3624 
3625 int
3626 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3627     struct ifnet *ifp, struct thread *td)
3628 {
3629 
3630 	return EOPNOTSUPP;
3631 }
3632 
3633 int
3634 pru_disconnect_notsupp(struct socket *so)
3635 {
3636 
3637 	return EOPNOTSUPP;
3638 }
3639 
3640 int
3641 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3642 {
3643 
3644 	return EOPNOTSUPP;
3645 }
3646 
3647 int
3648 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3649 {
3650 
3651 	return EOPNOTSUPP;
3652 }
3653 
3654 int
3655 pru_rcvd_notsupp(struct socket *so, int flags)
3656 {
3657 
3658 	return EOPNOTSUPP;
3659 }
3660 
3661 int
3662 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3663 {
3664 
3665 	return EOPNOTSUPP;
3666 }
3667 
3668 int
3669 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3670     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3671 {
3672 
3673 	return EOPNOTSUPP;
3674 }
3675 
3676 int
3677 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3678 {
3679 
3680 	return (EOPNOTSUPP);
3681 }
3682 
3683 /*
3684  * This isn't really a ``null'' operation, but it's the default one and
3685  * doesn't do anything destructive.
3686  */
3687 int
3688 pru_sense_null(struct socket *so, struct stat *sb)
3689 {
3690 
3691 	sb->st_blksize = so->so_snd.sb_hiwat;
3692 	return 0;
3693 }
3694 
3695 int
3696 pru_shutdown_notsupp(struct socket *so)
3697 {
3698 
3699 	return EOPNOTSUPP;
3700 }
3701 
3702 int
3703 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3704 {
3705 
3706 	return EOPNOTSUPP;
3707 }
3708 
3709 int
3710 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3711     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3712 {
3713 
3714 	return EOPNOTSUPP;
3715 }
3716 
3717 int
3718 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3719     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3720 {
3721 
3722 	return EOPNOTSUPP;
3723 }
3724 
3725 int
3726 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3727     struct thread *td)
3728 {
3729 
3730 	return EOPNOTSUPP;
3731 }
3732 
3733 static void
3734 filt_sordetach(struct knote *kn)
3735 {
3736 	struct socket *so = kn->kn_fp->f_data;
3737 
3738 	so_rdknl_lock(so);
3739 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3740 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3741 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3742 	so_rdknl_unlock(so);
3743 }
3744 
3745 /*ARGSUSED*/
3746 static int
3747 filt_soread(struct knote *kn, long hint)
3748 {
3749 	struct socket *so;
3750 
3751 	so = kn->kn_fp->f_data;
3752 
3753 	if (SOLISTENING(so)) {
3754 		SOCK_LOCK_ASSERT(so);
3755 		kn->kn_data = so->sol_qlen;
3756 		if (so->so_error) {
3757 			kn->kn_flags |= EV_EOF;
3758 			kn->kn_fflags = so->so_error;
3759 			return (1);
3760 		}
3761 		return (!TAILQ_EMPTY(&so->sol_comp));
3762 	}
3763 
3764 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3765 
3766 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3767 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3768 		kn->kn_flags |= EV_EOF;
3769 		kn->kn_fflags = so->so_error;
3770 		return (1);
3771 	} else if (so->so_error)	/* temporary udp error */
3772 		return (1);
3773 
3774 	if (kn->kn_sfflags & NOTE_LOWAT) {
3775 		if (kn->kn_data >= kn->kn_sdata)
3776 			return (1);
3777 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3778 		return (1);
3779 
3780 	/* This hook returning non-zero indicates an event, not error */
3781 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3782 }
3783 
3784 static void
3785 filt_sowdetach(struct knote *kn)
3786 {
3787 	struct socket *so = kn->kn_fp->f_data;
3788 
3789 	so_wrknl_lock(so);
3790 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3791 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3792 		so->so_snd.sb_flags &= ~SB_KNOTE;
3793 	so_wrknl_unlock(so);
3794 }
3795 
3796 /*ARGSUSED*/
3797 static int
3798 filt_sowrite(struct knote *kn, long hint)
3799 {
3800 	struct socket *so;
3801 
3802 	so = kn->kn_fp->f_data;
3803 
3804 	if (SOLISTENING(so))
3805 		return (0);
3806 
3807 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3808 	kn->kn_data = sbspace(&so->so_snd);
3809 
3810 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3811 
3812 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3813 		kn->kn_flags |= EV_EOF;
3814 		kn->kn_fflags = so->so_error;
3815 		return (1);
3816 	} else if (so->so_error)	/* temporary udp error */
3817 		return (1);
3818 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3819 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3820 		return (0);
3821 	else if (kn->kn_sfflags & NOTE_LOWAT)
3822 		return (kn->kn_data >= kn->kn_sdata);
3823 	else
3824 		return (kn->kn_data >= so->so_snd.sb_lowat);
3825 }
3826 
3827 static int
3828 filt_soempty(struct knote *kn, long hint)
3829 {
3830 	struct socket *so;
3831 
3832 	so = kn->kn_fp->f_data;
3833 
3834 	if (SOLISTENING(so))
3835 		return (1);
3836 
3837 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3838 	kn->kn_data = sbused(&so->so_snd);
3839 
3840 	if (kn->kn_data == 0)
3841 		return (1);
3842 	else
3843 		return (0);
3844 }
3845 
3846 int
3847 socheckuid(struct socket *so, uid_t uid)
3848 {
3849 
3850 	if (so == NULL)
3851 		return (EPERM);
3852 	if (so->so_cred->cr_uid != uid)
3853 		return (EPERM);
3854 	return (0);
3855 }
3856 
3857 /*
3858  * These functions are used by protocols to notify the socket layer (and its
3859  * consumers) of state changes in the sockets driven by protocol-side events.
3860  */
3861 
3862 /*
3863  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3864  *
3865  * Normal sequence from the active (originating) side is that
3866  * soisconnecting() is called during processing of connect() call, resulting
3867  * in an eventual call to soisconnected() if/when the connection is
3868  * established.  When the connection is torn down soisdisconnecting() is
3869  * called during processing of disconnect() call, and soisdisconnected() is
3870  * called when the connection to the peer is totally severed.  The semantics
3871  * of these routines are such that connectionless protocols can call
3872  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3873  * calls when setting up a ``connection'' takes no time.
3874  *
3875  * From the passive side, a socket is created with two queues of sockets:
3876  * so_incomp for connections in progress and so_comp for connections already
3877  * made and awaiting user acceptance.  As a protocol is preparing incoming
3878  * connections, it creates a socket structure queued on so_incomp by calling
3879  * sonewconn().  When the connection is established, soisconnected() is
3880  * called, and transfers the socket structure to so_comp, making it available
3881  * to accept().
3882  *
3883  * If a socket is closed with sockets on either so_incomp or so_comp, these
3884  * sockets are dropped.
3885  *
3886  * If higher-level protocols are implemented in the kernel, the wakeups done
3887  * here will sometimes cause software-interrupt process scheduling.
3888  */
3889 void
3890 soisconnecting(struct socket *so)
3891 {
3892 
3893 	SOCK_LOCK(so);
3894 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3895 	so->so_state |= SS_ISCONNECTING;
3896 	SOCK_UNLOCK(so);
3897 }
3898 
3899 void
3900 soisconnected(struct socket *so)
3901 {
3902 
3903 	SOCK_LOCK(so);
3904 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3905 	so->so_state |= SS_ISCONNECTED;
3906 
3907 	if (so->so_qstate == SQ_INCOMP) {
3908 		struct socket *head = so->so_listen;
3909 		int ret;
3910 
3911 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3912 		/*
3913 		 * Promoting a socket from incomplete queue to complete, we
3914 		 * need to go through reverse order of locking.  We first do
3915 		 * trylock, and if that doesn't succeed, we go the hard way
3916 		 * leaving a reference and rechecking consistency after proper
3917 		 * locking.
3918 		 */
3919 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3920 			soref(head);
3921 			SOCK_UNLOCK(so);
3922 			SOLISTEN_LOCK(head);
3923 			SOCK_LOCK(so);
3924 			if (__predict_false(head != so->so_listen)) {
3925 				/*
3926 				 * The socket went off the listen queue,
3927 				 * should be lost race to close(2) of sol.
3928 				 * The socket is about to soabort().
3929 				 */
3930 				SOCK_UNLOCK(so);
3931 				sorele(head);
3932 				return;
3933 			}
3934 			/* Not the last one, as so holds a ref. */
3935 			refcount_release(&head->so_count);
3936 		}
3937 again:
3938 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3939 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3940 			head->sol_incqlen--;
3941 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3942 			head->sol_qlen++;
3943 			so->so_qstate = SQ_COMP;
3944 			SOCK_UNLOCK(so);
3945 			solisten_wakeup(head);	/* unlocks */
3946 		} else {
3947 			SOCKBUF_LOCK(&so->so_rcv);
3948 			soupcall_set(so, SO_RCV,
3949 			    head->sol_accept_filter->accf_callback,
3950 			    head->sol_accept_filter_arg);
3951 			so->so_options &= ~SO_ACCEPTFILTER;
3952 			ret = head->sol_accept_filter->accf_callback(so,
3953 			    head->sol_accept_filter_arg, M_NOWAIT);
3954 			if (ret == SU_ISCONNECTED) {
3955 				soupcall_clear(so, SO_RCV);
3956 				SOCKBUF_UNLOCK(&so->so_rcv);
3957 				goto again;
3958 			}
3959 			SOCKBUF_UNLOCK(&so->so_rcv);
3960 			SOCK_UNLOCK(so);
3961 			SOLISTEN_UNLOCK(head);
3962 		}
3963 		return;
3964 	}
3965 	SOCK_UNLOCK(so);
3966 	wakeup(&so->so_timeo);
3967 	sorwakeup(so);
3968 	sowwakeup(so);
3969 }
3970 
3971 void
3972 soisdisconnecting(struct socket *so)
3973 {
3974 
3975 	SOCK_LOCK(so);
3976 	so->so_state &= ~SS_ISCONNECTING;
3977 	so->so_state |= SS_ISDISCONNECTING;
3978 
3979 	if (!SOLISTENING(so)) {
3980 		SOCKBUF_LOCK(&so->so_rcv);
3981 		socantrcvmore_locked(so);
3982 		SOCKBUF_LOCK(&so->so_snd);
3983 		socantsendmore_locked(so);
3984 	}
3985 	SOCK_UNLOCK(so);
3986 	wakeup(&so->so_timeo);
3987 }
3988 
3989 void
3990 soisdisconnected(struct socket *so)
3991 {
3992 
3993 	SOCK_LOCK(so);
3994 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3995 	so->so_state |= SS_ISDISCONNECTED;
3996 
3997 	if (!SOLISTENING(so)) {
3998 		SOCK_UNLOCK(so);
3999 		SOCKBUF_LOCK(&so->so_rcv);
4000 		socantrcvmore_locked(so);
4001 		SOCKBUF_LOCK(&so->so_snd);
4002 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4003 		socantsendmore_locked(so);
4004 	} else
4005 		SOCK_UNLOCK(so);
4006 	wakeup(&so->so_timeo);
4007 }
4008 
4009 /*
4010  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4011  */
4012 struct sockaddr *
4013 sodupsockaddr(const struct sockaddr *sa, int mflags)
4014 {
4015 	struct sockaddr *sa2;
4016 
4017 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4018 	if (sa2)
4019 		bcopy(sa, sa2, sa->sa_len);
4020 	return sa2;
4021 }
4022 
4023 /*
4024  * Register per-socket destructor.
4025  */
4026 void
4027 sodtor_set(struct socket *so, so_dtor_t *func)
4028 {
4029 
4030 	SOCK_LOCK_ASSERT(so);
4031 	so->so_dtor = func;
4032 }
4033 
4034 /*
4035  * Register per-socket buffer upcalls.
4036  */
4037 void
4038 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4039 {
4040 	struct sockbuf *sb;
4041 
4042 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4043 
4044 	switch (which) {
4045 	case SO_RCV:
4046 		sb = &so->so_rcv;
4047 		break;
4048 	case SO_SND:
4049 		sb = &so->so_snd;
4050 		break;
4051 	default:
4052 		panic("soupcall_set: bad which");
4053 	}
4054 	SOCKBUF_LOCK_ASSERT(sb);
4055 	sb->sb_upcall = func;
4056 	sb->sb_upcallarg = arg;
4057 	sb->sb_flags |= SB_UPCALL;
4058 }
4059 
4060 void
4061 soupcall_clear(struct socket *so, int which)
4062 {
4063 	struct sockbuf *sb;
4064 
4065 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4066 
4067 	switch (which) {
4068 	case SO_RCV:
4069 		sb = &so->so_rcv;
4070 		break;
4071 	case SO_SND:
4072 		sb = &so->so_snd;
4073 		break;
4074 	default:
4075 		panic("soupcall_clear: bad which");
4076 	}
4077 	SOCKBUF_LOCK_ASSERT(sb);
4078 	KASSERT(sb->sb_upcall != NULL,
4079 	    ("%s: so %p no upcall to clear", __func__, so));
4080 	sb->sb_upcall = NULL;
4081 	sb->sb_upcallarg = NULL;
4082 	sb->sb_flags &= ~SB_UPCALL;
4083 }
4084 
4085 void
4086 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4087 {
4088 
4089 	SOLISTEN_LOCK_ASSERT(so);
4090 	so->sol_upcall = func;
4091 	so->sol_upcallarg = arg;
4092 }
4093 
4094 static void
4095 so_rdknl_lock(void *arg)
4096 {
4097 	struct socket *so = arg;
4098 
4099 	if (SOLISTENING(so))
4100 		SOCK_LOCK(so);
4101 	else
4102 		SOCKBUF_LOCK(&so->so_rcv);
4103 }
4104 
4105 static void
4106 so_rdknl_unlock(void *arg)
4107 {
4108 	struct socket *so = arg;
4109 
4110 	if (SOLISTENING(so))
4111 		SOCK_UNLOCK(so);
4112 	else
4113 		SOCKBUF_UNLOCK(&so->so_rcv);
4114 }
4115 
4116 static void
4117 so_rdknl_assert_locked(void *arg)
4118 {
4119 	struct socket *so = arg;
4120 
4121 	if (SOLISTENING(so))
4122 		SOCK_LOCK_ASSERT(so);
4123 	else
4124 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4125 }
4126 
4127 static void
4128 so_rdknl_assert_unlocked(void *arg)
4129 {
4130 	struct socket *so = arg;
4131 
4132 	if (SOLISTENING(so))
4133 		SOCK_UNLOCK_ASSERT(so);
4134 	else
4135 		SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4136 }
4137 
4138 static void
4139 so_wrknl_lock(void *arg)
4140 {
4141 	struct socket *so = arg;
4142 
4143 	if (SOLISTENING(so))
4144 		SOCK_LOCK(so);
4145 	else
4146 		SOCKBUF_LOCK(&so->so_snd);
4147 }
4148 
4149 static void
4150 so_wrknl_unlock(void *arg)
4151 {
4152 	struct socket *so = arg;
4153 
4154 	if (SOLISTENING(so))
4155 		SOCK_UNLOCK(so);
4156 	else
4157 		SOCKBUF_UNLOCK(&so->so_snd);
4158 }
4159 
4160 static void
4161 so_wrknl_assert_locked(void *arg)
4162 {
4163 	struct socket *so = arg;
4164 
4165 	if (SOLISTENING(so))
4166 		SOCK_LOCK_ASSERT(so);
4167 	else
4168 		SOCKBUF_LOCK_ASSERT(&so->so_snd);
4169 }
4170 
4171 static void
4172 so_wrknl_assert_unlocked(void *arg)
4173 {
4174 	struct socket *so = arg;
4175 
4176 	if (SOLISTENING(so))
4177 		SOCK_UNLOCK_ASSERT(so);
4178 	else
4179 		SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4180 }
4181 
4182 /*
4183  * Create an external-format (``xsocket'') structure using the information in
4184  * the kernel-format socket structure pointed to by so.  This is done to
4185  * reduce the spew of irrelevant information over this interface, to isolate
4186  * user code from changes in the kernel structure, and potentially to provide
4187  * information-hiding if we decide that some of this information should be
4188  * hidden from users.
4189  */
4190 void
4191 sotoxsocket(struct socket *so, struct xsocket *xso)
4192 {
4193 
4194 	bzero(xso, sizeof(*xso));
4195 	xso->xso_len = sizeof *xso;
4196 	xso->xso_so = (uintptr_t)so;
4197 	xso->so_type = so->so_type;
4198 	xso->so_options = so->so_options;
4199 	xso->so_linger = so->so_linger;
4200 	xso->so_state = so->so_state;
4201 	xso->so_pcb = (uintptr_t)so->so_pcb;
4202 	xso->xso_protocol = so->so_proto->pr_protocol;
4203 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4204 	xso->so_timeo = so->so_timeo;
4205 	xso->so_error = so->so_error;
4206 	xso->so_uid = so->so_cred->cr_uid;
4207 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4208 	if (SOLISTENING(so)) {
4209 		xso->so_qlen = so->sol_qlen;
4210 		xso->so_incqlen = so->sol_incqlen;
4211 		xso->so_qlimit = so->sol_qlimit;
4212 		xso->so_oobmark = 0;
4213 	} else {
4214 		xso->so_state |= so->so_qstate;
4215 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4216 		xso->so_oobmark = so->so_oobmark;
4217 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4218 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4219 	}
4220 }
4221 
4222 struct sockbuf *
4223 so_sockbuf_rcv(struct socket *so)
4224 {
4225 
4226 	return (&so->so_rcv);
4227 }
4228 
4229 struct sockbuf *
4230 so_sockbuf_snd(struct socket *so)
4231 {
4232 
4233 	return (&so->so_snd);
4234 }
4235 
4236 int
4237 so_state_get(const struct socket *so)
4238 {
4239 
4240 	return (so->so_state);
4241 }
4242 
4243 void
4244 so_state_set(struct socket *so, int val)
4245 {
4246 
4247 	so->so_state = val;
4248 }
4249 
4250 int
4251 so_options_get(const struct socket *so)
4252 {
4253 
4254 	return (so->so_options);
4255 }
4256 
4257 void
4258 so_options_set(struct socket *so, int val)
4259 {
4260 
4261 	so->so_options = val;
4262 }
4263 
4264 int
4265 so_error_get(const struct socket *so)
4266 {
4267 
4268 	return (so->so_error);
4269 }
4270 
4271 void
4272 so_error_set(struct socket *so, int val)
4273 {
4274 
4275 	so->so_error = val;
4276 }
4277 
4278 int
4279 so_linger_get(const struct socket *so)
4280 {
4281 
4282 	return (so->so_linger);
4283 }
4284 
4285 void
4286 so_linger_set(struct socket *so, int val)
4287 {
4288 
4289 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4290 	    ("%s: val %d out of range", __func__, val));
4291 
4292 	so->so_linger = val;
4293 }
4294 
4295 struct protosw *
4296 so_protosw_get(const struct socket *so)
4297 {
4298 
4299 	return (so->so_proto);
4300 }
4301 
4302 void
4303 so_protosw_set(struct socket *so, struct protosw *val)
4304 {
4305 
4306 	so->so_proto = val;
4307 }
4308 
4309 void
4310 so_sorwakeup(struct socket *so)
4311 {
4312 
4313 	sorwakeup(so);
4314 }
4315 
4316 void
4317 so_sowwakeup(struct socket *so)
4318 {
4319 
4320 	sowwakeup(so);
4321 }
4322 
4323 void
4324 so_sorwakeup_locked(struct socket *so)
4325 {
4326 
4327 	sorwakeup_locked(so);
4328 }
4329 
4330 void
4331 so_sowwakeup_locked(struct socket *so)
4332 {
4333 
4334 	sowwakeup_locked(so);
4335 }
4336 
4337 void
4338 so_lock(struct socket *so)
4339 {
4340 
4341 	SOCK_LOCK(so);
4342 }
4343 
4344 void
4345 so_unlock(struct socket *so)
4346 {
4347 
4348 	SOCK_UNLOCK(so);
4349 }
4350