xref: /freebsd/sys/kern/uipc_socket.c (revision f5463265955b829775bbb32e1fd0bc11dafc36ce)
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 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  *
96  * NOTE: With regard to VNETs the general rule is that callers do not set
97  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn(),
99  * which are usually called from a pre-set VNET context. sopoll() currently
100  * does not need a VNET context to be set.
101  */
102 
103 #include <sys/cdefs.h>
104 #include "opt_inet.h"
105 #include "opt_inet6.h"
106 #include "opt_kern_tls.h"
107 #include "opt_sctp.h"
108 
109 #include <sys/param.h>
110 #include <sys/systm.h>
111 #include <sys/capsicum.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
115 #include <sys/mac.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h>			/* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/ktls.h>
125 #include <sys/event.h>
126 #include <sys/eventhandler.h>
127 #include <sys/poll.h>
128 #include <sys/proc.h>
129 #include <sys/protosw.h>
130 #include <sys/sbuf.h>
131 #include <sys/socket.h>
132 #include <sys/socketvar.h>
133 #include <sys/resourcevar.h>
134 #include <net/route.h>
135 #include <sys/signalvar.h>
136 #include <sys/stat.h>
137 #include <sys/sx.h>
138 #include <sys/sysctl.h>
139 #include <sys/taskqueue.h>
140 #include <sys/uio.h>
141 #include <sys/un.h>
142 #include <sys/unpcb.h>
143 #include <sys/jail.h>
144 #include <sys/syslog.h>
145 #include <netinet/in.h>
146 #include <netinet/in_pcb.h>
147 #include <netinet/tcp.h>
148 
149 #include <net/vnet.h>
150 
151 #include <security/mac/mac_framework.h>
152 
153 #include <vm/uma.h>
154 
155 #ifdef COMPAT_FREEBSD32
156 #include <sys/mount.h>
157 #include <sys/sysent.h>
158 #include <compat/freebsd32/freebsd32.h>
159 #endif
160 
161 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
162 		    int flags);
163 static void	so_rdknl_lock(void *);
164 static void	so_rdknl_unlock(void *);
165 static void	so_rdknl_assert_lock(void *, int);
166 static void	so_wrknl_lock(void *);
167 static void	so_wrknl_unlock(void *);
168 static void	so_wrknl_assert_lock(void *, int);
169 
170 static void	filt_sordetach(struct knote *kn);
171 static int	filt_soread(struct knote *kn, long hint);
172 static void	filt_sowdetach(struct knote *kn);
173 static int	filt_sowrite(struct knote *kn, long hint);
174 static int	filt_soempty(struct knote *kn, long hint);
175 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
176 fo_kqfilter_t	soo_kqfilter;
177 
178 static struct filterops soread_filtops = {
179 	.f_isfd = 1,
180 	.f_detach = filt_sordetach,
181 	.f_event = filt_soread,
182 };
183 static struct filterops sowrite_filtops = {
184 	.f_isfd = 1,
185 	.f_detach = filt_sowdetach,
186 	.f_event = filt_sowrite,
187 };
188 static struct filterops soempty_filtops = {
189 	.f_isfd = 1,
190 	.f_detach = filt_sowdetach,
191 	.f_event = filt_soempty,
192 };
193 
194 so_gen_t	so_gencnt;	/* generation count for sockets */
195 
196 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
197 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
198 
199 #define	VNET_SO_ASSERT(so)						\
200 	VNET_ASSERT(curvnet != NULL,					\
201 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
202 
203 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
204 #define	V_socket_hhh		VNET(socket_hhh)
205 
206 /*
207  * Limit on the number of connections in the listen queue waiting
208  * for accept(2).
209  * NB: The original sysctl somaxconn is still available but hidden
210  * to prevent confusion about the actual purpose of this number.
211  */
212 static u_int somaxconn = SOMAXCONN;
213 
214 static int
215 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
216 {
217 	int error;
218 	int val;
219 
220 	val = somaxconn;
221 	error = sysctl_handle_int(oidp, &val, 0, req);
222 	if (error || !req->newptr )
223 		return (error);
224 
225 	/*
226 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
227 	 *   3 * so_qlimit / 2
228 	 * below, will not overflow.
229          */
230 
231 	if (val < 1 || val > UINT_MAX / 3)
232 		return (EINVAL);
233 
234 	somaxconn = val;
235 	return (0);
236 }
237 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
238     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
239     sysctl_somaxconn, "I",
240     "Maximum listen socket pending connection accept queue size");
241 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
242     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
243     sizeof(int), sysctl_somaxconn, "I",
244     "Maximum listen socket pending connection accept queue size (compat)");
245 
246 static int numopensockets;
247 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
248     &numopensockets, 0, "Number of open sockets");
249 
250 /*
251  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
252  * so_gencnt field.
253  */
254 static struct mtx so_global_mtx;
255 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
256 
257 /*
258  * General IPC sysctl name space, used by sockets and a variety of other IPC
259  * types.
260  */
261 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
262     "IPC");
263 
264 /*
265  * Initialize the socket subsystem and set up the socket
266  * memory allocator.
267  */
268 static uma_zone_t socket_zone;
269 int	maxsockets;
270 
271 static void
272 socket_zone_change(void *tag)
273 {
274 
275 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
276 }
277 
278 static void
279 socket_hhook_register(int subtype)
280 {
281 
282 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
283 	    &V_socket_hhh[subtype],
284 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
285 		printf("%s: WARNING: unable to register hook\n", __func__);
286 }
287 
288 static void
289 socket_hhook_deregister(int subtype)
290 {
291 
292 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
293 		printf("%s: WARNING: unable to deregister hook\n", __func__);
294 }
295 
296 static void
297 socket_init(void *tag)
298 {
299 
300 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
301 	    NULL, NULL, UMA_ALIGN_PTR, 0);
302 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
303 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
304 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
305 	    EVENTHANDLER_PRI_FIRST);
306 }
307 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
308 
309 static void
310 socket_vnet_init(const void *unused __unused)
311 {
312 	int i;
313 
314 	/* We expect a contiguous range */
315 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
316 		socket_hhook_register(i);
317 }
318 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
319     socket_vnet_init, NULL);
320 
321 static void
322 socket_vnet_uninit(const void *unused __unused)
323 {
324 	int i;
325 
326 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 		socket_hhook_deregister(i);
328 }
329 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330     socket_vnet_uninit, NULL);
331 
332 /*
333  * Initialise maxsockets.  This SYSINIT must be run after
334  * tunable_mbinit().
335  */
336 static void
337 init_maxsockets(void *ignored)
338 {
339 
340 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
341 	maxsockets = imax(maxsockets, maxfiles);
342 }
343 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
344 
345 /*
346  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
347  * of the change so that they can update their dependent limits as required.
348  */
349 static int
350 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
351 {
352 	int error, newmaxsockets;
353 
354 	newmaxsockets = maxsockets;
355 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
356 	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
357 		if (newmaxsockets > maxsockets &&
358 		    newmaxsockets <= maxfiles) {
359 			maxsockets = newmaxsockets;
360 			EVENTHANDLER_INVOKE(maxsockets_change);
361 		} else
362 			error = EINVAL;
363 	}
364 	return (error);
365 }
366 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
367     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE,
368     &maxsockets, 0, sysctl_maxsockets, "IU",
369     "Maximum number of sockets available");
370 
371 /*
372  * Socket operation routines.  These routines are called by the routines in
373  * sys_socket.c or from a system process, and implement the semantics of
374  * socket operations by switching out to the protocol specific routines.
375  */
376 
377 /*
378  * Get a socket structure from our zone, and initialize it.  Note that it
379  * would probably be better to allocate socket and PCB at the same time, but
380  * I'm not convinced that all the protocols can be easily modified to do
381  * this.
382  *
383  * soalloc() returns a socket with a ref count of 0.
384  */
385 static struct socket *
386 soalloc(struct vnet *vnet)
387 {
388 	struct socket *so;
389 
390 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
391 	if (so == NULL)
392 		return (NULL);
393 #ifdef MAC
394 	if (mac_socket_init(so, M_NOWAIT) != 0) {
395 		uma_zfree(socket_zone, so);
396 		return (NULL);
397 	}
398 #endif
399 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
400 		uma_zfree(socket_zone, so);
401 		return (NULL);
402 	}
403 
404 	/*
405 	 * The socket locking protocol allows to lock 2 sockets at a time,
406 	 * however, the first one must be a listening socket.  WITNESS lacks
407 	 * a feature to change class of an existing lock, so we use DUPOK.
408 	 */
409 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
410 	mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
411 	mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
412 	so->so_rcv.sb_sel = &so->so_rdsel;
413 	so->so_snd.sb_sel = &so->so_wrsel;
414 	sx_init(&so->so_snd_sx, "so_snd_sx");
415 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
416 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
417 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
418 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
419 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
420 #ifdef VIMAGE
421 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
422 	    __func__, __LINE__, so));
423 	so->so_vnet = vnet;
424 #endif
425 	/* We shouldn't need the so_global_mtx */
426 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
427 		/* Do we need more comprehensive error returns? */
428 		uma_zfree(socket_zone, so);
429 		return (NULL);
430 	}
431 	mtx_lock(&so_global_mtx);
432 	so->so_gencnt = ++so_gencnt;
433 	++numopensockets;
434 #ifdef VIMAGE
435 	vnet->vnet_sockcnt++;
436 #endif
437 	mtx_unlock(&so_global_mtx);
438 
439 	return (so);
440 }
441 
442 /*
443  * Free the storage associated with a socket at the socket layer, tear down
444  * locks, labels, etc.  All protocol state is assumed already to have been
445  * torn down (and possibly never set up) by the caller.
446  */
447 void
448 sodealloc(struct socket *so)
449 {
450 
451 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
452 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
453 
454 	mtx_lock(&so_global_mtx);
455 	so->so_gencnt = ++so_gencnt;
456 	--numopensockets;	/* Could be below, but faster here. */
457 #ifdef VIMAGE
458 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
459 	    __func__, __LINE__, so));
460 	so->so_vnet->vnet_sockcnt--;
461 #endif
462 	mtx_unlock(&so_global_mtx);
463 #ifdef MAC
464 	mac_socket_destroy(so);
465 #endif
466 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
467 
468 	khelp_destroy_osd(&so->osd);
469 	if (SOLISTENING(so)) {
470 		if (so->sol_accept_filter != NULL)
471 			accept_filt_setopt(so, NULL);
472 	} else {
473 		if (so->so_rcv.sb_hiwat)
474 			(void)chgsbsize(so->so_cred->cr_uidinfo,
475 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
476 		if (so->so_snd.sb_hiwat)
477 			(void)chgsbsize(so->so_cred->cr_uidinfo,
478 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
479 		sx_destroy(&so->so_snd_sx);
480 		sx_destroy(&so->so_rcv_sx);
481 		mtx_destroy(&so->so_snd_mtx);
482 		mtx_destroy(&so->so_rcv_mtx);
483 	}
484 	crfree(so->so_cred);
485 	mtx_destroy(&so->so_lock);
486 	uma_zfree(socket_zone, so);
487 }
488 
489 /*
490  * socreate returns a socket with a ref count of 1 and a file descriptor
491  * reference.  The socket should be closed with soclose().
492  */
493 int
494 socreate(int dom, struct socket **aso, int type, int proto,
495     struct ucred *cred, struct thread *td)
496 {
497 	struct protosw *prp;
498 	struct socket *so;
499 	int error;
500 
501 	/*
502 	 * XXX: divert(4) historically abused PF_INET.  Keep this compatibility
503 	 * shim until all applications have been updated.
504 	 */
505 	if (__predict_false(dom == PF_INET && type == SOCK_RAW &&
506 	    proto == IPPROTO_DIVERT)) {
507 		dom = PF_DIVERT;
508 		printf("%s uses obsolete way to create divert(4) socket\n",
509 		    td->td_proc->p_comm);
510 	}
511 
512 	prp = pffindproto(dom, type, proto);
513 	if (prp == NULL) {
514 		/* No support for domain. */
515 		if (pffinddomain(dom) == NULL)
516 			return (EAFNOSUPPORT);
517 		/* No support for socket type. */
518 		if (proto == 0 && type != 0)
519 			return (EPROTOTYPE);
520 		return (EPROTONOSUPPORT);
521 	}
522 
523 	MPASS(prp->pr_attach);
524 
525 	if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
526 		return (ECAPMODE);
527 
528 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
529 		return (EPROTONOSUPPORT);
530 
531 	so = soalloc(CRED_TO_VNET(cred));
532 	if (so == NULL)
533 		return (ENOBUFS);
534 
535 	so->so_type = type;
536 	so->so_cred = crhold(cred);
537 	if ((prp->pr_domain->dom_family == PF_INET) ||
538 	    (prp->pr_domain->dom_family == PF_INET6) ||
539 	    (prp->pr_domain->dom_family == PF_ROUTE))
540 		so->so_fibnum = td->td_proc->p_fibnum;
541 	else
542 		so->so_fibnum = 0;
543 	so->so_proto = prp;
544 #ifdef MAC
545 	mac_socket_create(cred, so);
546 #endif
547 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
548 	    so_rdknl_assert_lock);
549 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
550 	    so_wrknl_assert_lock);
551 	if ((prp->pr_flags & PR_SOCKBUF) == 0) {
552 		so->so_snd.sb_mtx = &so->so_snd_mtx;
553 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
554 	}
555 	/*
556 	 * Auto-sizing of socket buffers is managed by the protocols and
557 	 * the appropriate flags must be set in the pru_attach function.
558 	 */
559 	CURVNET_SET(so->so_vnet);
560 	error = prp->pr_attach(so, proto, td);
561 	CURVNET_RESTORE();
562 	if (error) {
563 		sodealloc(so);
564 		return (error);
565 	}
566 	soref(so);
567 	*aso = so;
568 	return (0);
569 }
570 
571 #ifdef REGRESSION
572 static int regression_sonewconn_earlytest = 1;
573 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
574     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
575 #endif
576 
577 static int sooverprio = LOG_DEBUG;
578 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
579     &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
580 
581 static struct timeval overinterval = { 60, 0 };
582 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
583     &overinterval,
584     "Delay in seconds between warnings for listen socket overflows");
585 
586 /*
587  * When an attempt at a new connection is noted on a socket which supports
588  * accept(2), the protocol has two options:
589  * 1) Call legacy sonewconn() function, which would call protocol attach
590  *    method, same as used for socket(2).
591  * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
592  *    and then call solisten_enqueue().
593  *
594  * Note: the ref count on the socket is 0 on return.
595  */
596 struct socket *
597 solisten_clone(struct socket *head)
598 {
599 	struct sbuf descrsb;
600 	struct socket *so;
601 	int len, overcount;
602 	u_int qlen;
603 	const char localprefix[] = "local:";
604 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
605 #if defined(INET6)
606 	char addrbuf[INET6_ADDRSTRLEN];
607 #elif defined(INET)
608 	char addrbuf[INET_ADDRSTRLEN];
609 #endif
610 	bool dolog, over;
611 
612 	SOLISTEN_LOCK(head);
613 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
614 #ifdef REGRESSION
615 	if (regression_sonewconn_earlytest && over) {
616 #else
617 	if (over) {
618 #endif
619 		head->sol_overcount++;
620 		dolog = (sooverprio >= 0) &&
621 			!!ratecheck(&head->sol_lastover, &overinterval);
622 
623 		/*
624 		 * If we're going to log, copy the overflow count and queue
625 		 * length from the listen socket before dropping the lock.
626 		 * Also, reset the overflow count.
627 		 */
628 		if (dolog) {
629 			overcount = head->sol_overcount;
630 			head->sol_overcount = 0;
631 			qlen = head->sol_qlen;
632 		}
633 		SOLISTEN_UNLOCK(head);
634 
635 		if (dolog) {
636 			/*
637 			 * Try to print something descriptive about the
638 			 * socket for the error message.
639 			 */
640 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
641 			    SBUF_FIXEDLEN);
642 			switch (head->so_proto->pr_domain->dom_family) {
643 #if defined(INET) || defined(INET6)
644 #ifdef INET
645 			case AF_INET:
646 #endif
647 #ifdef INET6
648 			case AF_INET6:
649 				if (head->so_proto->pr_domain->dom_family ==
650 				    AF_INET6 ||
651 				    (sotoinpcb(head)->inp_inc.inc_flags &
652 				    INC_ISIPV6)) {
653 					ip6_sprintf(addrbuf,
654 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
655 					sbuf_printf(&descrsb, "[%s]", addrbuf);
656 				} else
657 #endif
658 				{
659 #ifdef INET
660 					inet_ntoa_r(
661 					    sotoinpcb(head)->inp_inc.inc_laddr,
662 					    addrbuf);
663 					sbuf_cat(&descrsb, addrbuf);
664 #endif
665 				}
666 				sbuf_printf(&descrsb, ":%hu (proto %u)",
667 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
668 				    head->so_proto->pr_protocol);
669 				break;
670 #endif /* INET || INET6 */
671 			case AF_UNIX:
672 				sbuf_cat(&descrsb, localprefix);
673 				if (sotounpcb(head)->unp_addr != NULL)
674 					len =
675 					    sotounpcb(head)->unp_addr->sun_len -
676 					    offsetof(struct sockaddr_un,
677 					    sun_path);
678 				else
679 					len = 0;
680 				if (len > 0)
681 					sbuf_bcat(&descrsb,
682 					    sotounpcb(head)->unp_addr->sun_path,
683 					    len);
684 				else
685 					sbuf_cat(&descrsb, "(unknown)");
686 				break;
687 			}
688 
689 			/*
690 			 * If we can't print something more specific, at least
691 			 * print the domain name.
692 			 */
693 			if (sbuf_finish(&descrsb) != 0 ||
694 			    sbuf_len(&descrsb) <= 0) {
695 				sbuf_clear(&descrsb);
696 				sbuf_cat(&descrsb,
697 				    head->so_proto->pr_domain->dom_name ?:
698 				    "unknown");
699 				sbuf_finish(&descrsb);
700 			}
701 			KASSERT(sbuf_len(&descrsb) > 0,
702 			    ("%s: sbuf creation failed", __func__));
703 			/*
704 			 * Preserve the historic listen queue overflow log
705 			 * message, that starts with "sonewconn:".  It has
706 			 * been known to sysadmins for years and also test
707 			 * sys/kern/sonewconn_overflow checks for it.
708 			 */
709 			if (head->so_cred == 0) {
710 				log(LOG_PRI(sooverprio),
711 				    "sonewconn: pcb %p (%s): "
712 				    "Listen queue overflow: %i already in "
713 				    "queue awaiting acceptance (%d "
714 				    "occurrences)\n", head->so_pcb,
715 				    sbuf_data(&descrsb),
716 			    	qlen, overcount);
717 			} else {
718 				log(LOG_PRI(sooverprio),
719 				    "sonewconn: pcb %p (%s): "
720 				    "Listen queue overflow: "
721 				    "%i already in queue awaiting acceptance "
722 				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
723 				    head->so_pcb, sbuf_data(&descrsb), qlen,
724 				    overcount, head->so_cred->cr_uid,
725 				    head->so_cred->cr_rgid,
726 				    head->so_cred->cr_prison ?
727 					head->so_cred->cr_prison->pr_name :
728 					"not_jailed");
729 			}
730 			sbuf_delete(&descrsb);
731 
732 			overcount = 0;
733 		}
734 
735 		return (NULL);
736 	}
737 	SOLISTEN_UNLOCK(head);
738 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
739 	    __func__, head));
740 	so = soalloc(head->so_vnet);
741 	if (so == NULL) {
742 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
743 		    "limit reached or out of memory\n",
744 		    __func__, head->so_pcb);
745 		return (NULL);
746 	}
747 	so->so_listen = head;
748 	so->so_type = head->so_type;
749 	/*
750 	 * POSIX is ambiguous on what options an accept(2)ed socket should
751 	 * inherit from the listener.  Words "create a new socket" may be
752 	 * interpreted as not inheriting anything.  Best programming practice
753 	 * for application developers is to not rely on such inheritance.
754 	 * FreeBSD had historically inherited all so_options excluding
755 	 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options,
756 	 * including those completely irrelevant to a new born socket.  For
757 	 * compatibility with older versions we will inherit a list of
758 	 * meaningful options.
759 	 */
760 	so->so_options = head->so_options & (SO_KEEPALIVE | SO_DONTROUTE |
761 	    SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE);
762 	so->so_linger = head->so_linger;
763 	so->so_state = head->so_state;
764 	so->so_fibnum = head->so_fibnum;
765 	so->so_proto = head->so_proto;
766 	so->so_cred = crhold(head->so_cred);
767 #ifdef MAC
768 	mac_socket_newconn(head, so);
769 #endif
770 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
771 	    so_rdknl_assert_lock);
772 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
773 	    so_wrknl_assert_lock);
774 	VNET_SO_ASSERT(head);
775 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
776 		sodealloc(so);
777 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
778 		    __func__, head->so_pcb);
779 		return (NULL);
780 	}
781 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
782 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
783 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
784 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
785 	so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
786 	so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE;
787 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
788 		so->so_snd.sb_mtx = &so->so_snd_mtx;
789 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
790 	}
791 
792 	return (so);
793 }
794 
795 /* Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED. */
796 struct socket *
797 sonewconn(struct socket *head, int connstatus)
798 {
799 	struct socket *so;
800 
801 	if ((so = solisten_clone(head)) == NULL)
802 		return (NULL);
803 
804 	if (so->so_proto->pr_attach(so, 0, NULL) != 0) {
805 		sodealloc(so);
806 		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
807 		    __func__, head->so_pcb);
808 		return (NULL);
809 	}
810 
811 	(void)solisten_enqueue(so, connstatus);
812 
813 	return (so);
814 }
815 
816 /*
817  * Enqueue socket cloned by solisten_clone() to the listen queue of the
818  * listener it has been cloned from.
819  *
820  * Return 'true' if socket landed on complete queue, otherwise 'false'.
821  */
822 bool
823 solisten_enqueue(struct socket *so, int connstatus)
824 {
825 	struct socket *head = so->so_listen;
826 
827 	MPASS(refcount_load(&so->so_count) == 0);
828 	refcount_init(&so->so_count, 1);
829 
830 	SOLISTEN_LOCK(head);
831 	if (head->sol_accept_filter != NULL)
832 		connstatus = 0;
833 	so->so_state |= connstatus;
834 	soref(head); /* A socket on (in)complete queue refs head. */
835 	if (connstatus) {
836 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
837 		so->so_qstate = SQ_COMP;
838 		head->sol_qlen++;
839 		solisten_wakeup(head);	/* unlocks */
840 		return (true);
841 	} else {
842 		/*
843 		 * Keep removing sockets from the head until there's room for
844 		 * us to insert on the tail.  In pre-locking revisions, this
845 		 * was a simple if(), but as we could be racing with other
846 		 * threads and soabort() requires dropping locks, we must
847 		 * loop waiting for the condition to be true.
848 		 */
849 		while (head->sol_incqlen > head->sol_qlimit) {
850 			struct socket *sp;
851 
852 			sp = TAILQ_FIRST(&head->sol_incomp);
853 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
854 			head->sol_incqlen--;
855 			SOCK_LOCK(sp);
856 			sp->so_qstate = SQ_NONE;
857 			sp->so_listen = NULL;
858 			SOCK_UNLOCK(sp);
859 			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
860 			soabort(sp);
861 			SOLISTEN_LOCK(head);
862 		}
863 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
864 		so->so_qstate = SQ_INCOMP;
865 		head->sol_incqlen++;
866 		SOLISTEN_UNLOCK(head);
867 		return (false);
868 	}
869 }
870 
871 #if defined(SCTP) || defined(SCTP_SUPPORT)
872 /*
873  * Socket part of sctp_peeloff().  Detach a new socket from an
874  * association.  The new socket is returned with a reference.
875  *
876  * XXXGL: reduce copy-paste with solisten_clone().
877  */
878 struct socket *
879 sopeeloff(struct socket *head)
880 {
881 	struct socket *so;
882 
883 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
884 	    __func__, __LINE__, head));
885 	so = soalloc(head->so_vnet);
886 	if (so == NULL) {
887 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
888 		    "limit reached or out of memory\n",
889 		    __func__, head->so_pcb);
890 		return (NULL);
891 	}
892 	so->so_type = head->so_type;
893 	so->so_options = head->so_options;
894 	so->so_linger = head->so_linger;
895 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
896 	so->so_fibnum = head->so_fibnum;
897 	so->so_proto = head->so_proto;
898 	so->so_cred = crhold(head->so_cred);
899 #ifdef MAC
900 	mac_socket_newconn(head, so);
901 #endif
902 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
903 	    so_rdknl_assert_lock);
904 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
905 	    so_wrknl_assert_lock);
906 	VNET_SO_ASSERT(head);
907 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
908 		sodealloc(so);
909 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
910 		    __func__, head->so_pcb);
911 		return (NULL);
912 	}
913 	if ((*so->so_proto->pr_attach)(so, 0, NULL)) {
914 		sodealloc(so);
915 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
916 		    __func__, head->so_pcb);
917 		return (NULL);
918 	}
919 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
920 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
921 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
922 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
923 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
924 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
925 
926 	soref(so);
927 
928 	return (so);
929 }
930 #endif	/* SCTP */
931 
932 int
933 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
934 {
935 	int error;
936 
937 	CURVNET_SET(so->so_vnet);
938 	error = so->so_proto->pr_bind(so, nam, td);
939 	CURVNET_RESTORE();
940 	return (error);
941 }
942 
943 int
944 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
945 {
946 	int error;
947 
948 	CURVNET_SET(so->so_vnet);
949 	error = so->so_proto->pr_bindat(fd, so, nam, td);
950 	CURVNET_RESTORE();
951 	return (error);
952 }
953 
954 /*
955  * solisten() transitions a socket from a non-listening state to a listening
956  * state, but can also be used to update the listen queue depth on an
957  * existing listen socket.  The protocol will call back into the sockets
958  * layer using solisten_proto_check() and solisten_proto() to check and set
959  * socket-layer listen state.  Call backs are used so that the protocol can
960  * acquire both protocol and socket layer locks in whatever order is required
961  * by the protocol.
962  *
963  * Protocol implementors are advised to hold the socket lock across the
964  * socket-layer test and set to avoid races at the socket layer.
965  */
966 int
967 solisten(struct socket *so, int backlog, struct thread *td)
968 {
969 	int error;
970 
971 	CURVNET_SET(so->so_vnet);
972 	error = so->so_proto->pr_listen(so, backlog, td);
973 	CURVNET_RESTORE();
974 	return (error);
975 }
976 
977 /*
978  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
979  * order to interlock with socket I/O.
980  */
981 int
982 solisten_proto_check(struct socket *so)
983 {
984 	SOCK_LOCK_ASSERT(so);
985 
986 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
987 	    SS_ISDISCONNECTING)) != 0)
988 		return (EINVAL);
989 
990 	/*
991 	 * Sleeping is not permitted here, so simply fail if userspace is
992 	 * attempting to transmit or receive on the socket.  This kind of
993 	 * transient failure is not ideal, but it should occur only if userspace
994 	 * is misusing the socket interfaces.
995 	 */
996 	if (!sx_try_xlock(&so->so_snd_sx))
997 		return (EAGAIN);
998 	if (!sx_try_xlock(&so->so_rcv_sx)) {
999 		sx_xunlock(&so->so_snd_sx);
1000 		return (EAGAIN);
1001 	}
1002 	mtx_lock(&so->so_snd_mtx);
1003 	mtx_lock(&so->so_rcv_mtx);
1004 
1005 	/* Interlock with soo_aio_queue() and KTLS. */
1006 	if (!SOLISTENING(so)) {
1007 		bool ktls;
1008 
1009 #ifdef KERN_TLS
1010 		ktls = so->so_snd.sb_tls_info != NULL ||
1011 		    so->so_rcv.sb_tls_info != NULL;
1012 #else
1013 		ktls = false;
1014 #endif
1015 		if (ktls ||
1016 		    (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
1017 		    (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
1018 			solisten_proto_abort(so);
1019 			return (EINVAL);
1020 		}
1021 	}
1022 
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Undo the setup done by solisten_proto_check().
1028  */
1029 void
1030 solisten_proto_abort(struct socket *so)
1031 {
1032 	mtx_unlock(&so->so_snd_mtx);
1033 	mtx_unlock(&so->so_rcv_mtx);
1034 	sx_xunlock(&so->so_snd_sx);
1035 	sx_xunlock(&so->so_rcv_sx);
1036 }
1037 
1038 void
1039 solisten_proto(struct socket *so, int backlog)
1040 {
1041 	int sbrcv_lowat, sbsnd_lowat;
1042 	u_int sbrcv_hiwat, sbsnd_hiwat;
1043 	short sbrcv_flags, sbsnd_flags;
1044 	sbintime_t sbrcv_timeo, sbsnd_timeo;
1045 
1046 	SOCK_LOCK_ASSERT(so);
1047 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1048 	    SS_ISDISCONNECTING)) == 0,
1049 	    ("%s: bad socket state %p", __func__, so));
1050 
1051 	if (SOLISTENING(so))
1052 		goto listening;
1053 
1054 	/*
1055 	 * Change this socket to listening state.
1056 	 */
1057 	sbrcv_lowat = so->so_rcv.sb_lowat;
1058 	sbsnd_lowat = so->so_snd.sb_lowat;
1059 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
1060 	sbsnd_hiwat = so->so_snd.sb_hiwat;
1061 	sbrcv_flags = so->so_rcv.sb_flags;
1062 	sbsnd_flags = so->so_snd.sb_flags;
1063 	sbrcv_timeo = so->so_rcv.sb_timeo;
1064 	sbsnd_timeo = so->so_snd.sb_timeo;
1065 
1066 	sbdestroy(so, SO_SND);
1067 	sbdestroy(so, SO_RCV);
1068 
1069 #ifdef INVARIANTS
1070 	bzero(&so->so_rcv,
1071 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
1072 #endif
1073 
1074 	so->sol_sbrcv_lowat = sbrcv_lowat;
1075 	so->sol_sbsnd_lowat = sbsnd_lowat;
1076 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
1077 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
1078 	so->sol_sbrcv_flags = sbrcv_flags;
1079 	so->sol_sbsnd_flags = sbsnd_flags;
1080 	so->sol_sbrcv_timeo = sbrcv_timeo;
1081 	so->sol_sbsnd_timeo = sbsnd_timeo;
1082 
1083 	so->sol_qlen = so->sol_incqlen = 0;
1084 	TAILQ_INIT(&so->sol_incomp);
1085 	TAILQ_INIT(&so->sol_comp);
1086 
1087 	so->sol_accept_filter = NULL;
1088 	so->sol_accept_filter_arg = NULL;
1089 	so->sol_accept_filter_str = NULL;
1090 
1091 	so->sol_upcall = NULL;
1092 	so->sol_upcallarg = NULL;
1093 
1094 	so->so_options |= SO_ACCEPTCONN;
1095 
1096 listening:
1097 	if (backlog < 0 || backlog > somaxconn)
1098 		backlog = somaxconn;
1099 	so->sol_qlimit = backlog;
1100 
1101 	mtx_unlock(&so->so_snd_mtx);
1102 	mtx_unlock(&so->so_rcv_mtx);
1103 	sx_xunlock(&so->so_snd_sx);
1104 	sx_xunlock(&so->so_rcv_sx);
1105 }
1106 
1107 /*
1108  * Wakeup listeners/subsystems once we have a complete connection.
1109  * Enters with lock, returns unlocked.
1110  */
1111 void
1112 solisten_wakeup(struct socket *sol)
1113 {
1114 
1115 	if (sol->sol_upcall != NULL)
1116 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1117 	else {
1118 		selwakeuppri(&sol->so_rdsel, PSOCK);
1119 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1120 	}
1121 	SOLISTEN_UNLOCK(sol);
1122 	wakeup_one(&sol->sol_comp);
1123 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1124 		pgsigio(&sol->so_sigio, SIGIO, 0);
1125 }
1126 
1127 /*
1128  * Return single connection off a listening socket queue.  Main consumer of
1129  * the function is kern_accept4().  Some modules, that do their own accept
1130  * management also use the function.  The socket reference held by the
1131  * listen queue is handed to the caller.
1132  *
1133  * Listening socket must be locked on entry and is returned unlocked on
1134  * return.
1135  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1136  */
1137 int
1138 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1139 {
1140 	struct socket *so;
1141 	int error;
1142 
1143 	SOLISTEN_LOCK_ASSERT(head);
1144 
1145 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1146 	    head->so_error == 0) {
1147 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1148 		    "accept", 0);
1149 		if (error != 0) {
1150 			SOLISTEN_UNLOCK(head);
1151 			return (error);
1152 		}
1153 	}
1154 	if (head->so_error) {
1155 		error = head->so_error;
1156 		head->so_error = 0;
1157 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1158 		error = EWOULDBLOCK;
1159 	else
1160 		error = 0;
1161 	if (error) {
1162 		SOLISTEN_UNLOCK(head);
1163 		return (error);
1164 	}
1165 	so = TAILQ_FIRST(&head->sol_comp);
1166 	SOCK_LOCK(so);
1167 	KASSERT(so->so_qstate == SQ_COMP,
1168 	    ("%s: so %p not SQ_COMP", __func__, so));
1169 	head->sol_qlen--;
1170 	so->so_qstate = SQ_NONE;
1171 	so->so_listen = NULL;
1172 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1173 	if (flags & ACCEPT4_INHERIT)
1174 		so->so_state |= (head->so_state & SS_NBIO);
1175 	else
1176 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1177 	SOCK_UNLOCK(so);
1178 	sorele_locked(head);
1179 
1180 	*ret = so;
1181 	return (0);
1182 }
1183 
1184 /*
1185  * Free socket upon release of the very last reference.
1186  */
1187 static void
1188 sofree(struct socket *so)
1189 {
1190 	struct protosw *pr = so->so_proto;
1191 
1192 	SOCK_LOCK_ASSERT(so);
1193 	KASSERT(refcount_load(&so->so_count) == 0,
1194 	    ("%s: so %p has references", __func__, so));
1195 	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1196 	    ("%s: so %p is on listen queue", __func__, so));
1197 
1198 	SOCK_UNLOCK(so);
1199 
1200 	if (so->so_dtor != NULL)
1201 		so->so_dtor(so);
1202 
1203 	VNET_SO_ASSERT(so);
1204 	if (pr->pr_detach != NULL)
1205 		pr->pr_detach(so);
1206 
1207 	/*
1208 	 * From this point on, we assume that no other references to this
1209 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1210 	 * to be acquired or held.
1211 	 */
1212 	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1213 		sbdestroy(so, SO_SND);
1214 		sbdestroy(so, SO_RCV);
1215 	}
1216 	seldrain(&so->so_rdsel);
1217 	seldrain(&so->so_wrsel);
1218 	knlist_destroy(&so->so_rdsel.si_note);
1219 	knlist_destroy(&so->so_wrsel.si_note);
1220 	sodealloc(so);
1221 }
1222 
1223 /*
1224  * Release a reference on a socket while holding the socket lock.
1225  * Unlocks the socket lock before returning.
1226  */
1227 void
1228 sorele_locked(struct socket *so)
1229 {
1230 	SOCK_LOCK_ASSERT(so);
1231 	if (refcount_release(&so->so_count))
1232 		sofree(so);
1233 	else
1234 		SOCK_UNLOCK(so);
1235 }
1236 
1237 /*
1238  * Close a socket on last file table reference removal.  Initiate disconnect
1239  * if connected.  Free socket when disconnect complete.
1240  *
1241  * This function will sorele() the socket.  Note that soclose() may be called
1242  * prior to the ref count reaching zero.  The actual socket structure will
1243  * not be freed until the ref count reaches zero.
1244  */
1245 int
1246 soclose(struct socket *so)
1247 {
1248 	struct accept_queue lqueue;
1249 	int error = 0;
1250 	bool listening, last __diagused;
1251 
1252 	CURVNET_SET(so->so_vnet);
1253 	funsetown(&so->so_sigio);
1254 	if (so->so_state & SS_ISCONNECTED) {
1255 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1256 			error = sodisconnect(so);
1257 			if (error) {
1258 				if (error == ENOTCONN)
1259 					error = 0;
1260 				goto drop;
1261 			}
1262 		}
1263 
1264 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1265 			if ((so->so_state & SS_ISDISCONNECTING) &&
1266 			    (so->so_state & SS_NBIO))
1267 				goto drop;
1268 			while (so->so_state & SS_ISCONNECTED) {
1269 				error = tsleep(&so->so_timeo,
1270 				    PSOCK | PCATCH, "soclos",
1271 				    so->so_linger * hz);
1272 				if (error)
1273 					break;
1274 			}
1275 		}
1276 	}
1277 
1278 drop:
1279 	if (so->so_proto->pr_close != NULL)
1280 		so->so_proto->pr_close(so);
1281 
1282 	SOCK_LOCK(so);
1283 	if ((listening = SOLISTENING(so))) {
1284 		struct socket *sp;
1285 
1286 		TAILQ_INIT(&lqueue);
1287 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1288 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1289 
1290 		so->sol_qlen = so->sol_incqlen = 0;
1291 
1292 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1293 			SOCK_LOCK(sp);
1294 			sp->so_qstate = SQ_NONE;
1295 			sp->so_listen = NULL;
1296 			SOCK_UNLOCK(sp);
1297 			last = refcount_release(&so->so_count);
1298 			KASSERT(!last, ("%s: released last reference for %p",
1299 			    __func__, so));
1300 		}
1301 	}
1302 	sorele_locked(so);
1303 	if (listening) {
1304 		struct socket *sp, *tsp;
1305 
1306 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1307 			soabort(sp);
1308 	}
1309 	CURVNET_RESTORE();
1310 	return (error);
1311 }
1312 
1313 /*
1314  * soabort() is used to abruptly tear down a connection, such as when a
1315  * resource limit is reached (listen queue depth exceeded), or if a listen
1316  * socket is closed while there are sockets waiting to be accepted.
1317  *
1318  * This interface is tricky, because it is called on an unreferenced socket,
1319  * and must be called only by a thread that has actually removed the socket
1320  * from the listen queue it was on.  Likely this thread holds the last
1321  * reference on the socket and soabort() will proceed with sofree().  But
1322  * it might be not the last, as the sockets on the listen queues are seen
1323  * from the protocol side.
1324  *
1325  * This interface will call into the protocol code, so must not be called
1326  * with any socket locks held.  Protocols do call it while holding their own
1327  * recursible protocol mutexes, but this is something that should be subject
1328  * to review in the future.
1329  *
1330  * Usually socket should have a single reference left, but this is not a
1331  * requirement.  In the past, when we have had named references for file
1332  * descriptor and protocol, we asserted that none of them are being held.
1333  */
1334 void
1335 soabort(struct socket *so)
1336 {
1337 
1338 	VNET_SO_ASSERT(so);
1339 
1340 	if (so->so_proto->pr_abort != NULL)
1341 		so->so_proto->pr_abort(so);
1342 	SOCK_LOCK(so);
1343 	sorele_locked(so);
1344 }
1345 
1346 int
1347 soaccept(struct socket *so, struct sockaddr *sa)
1348 {
1349 #ifdef INVARIANTS
1350 	u_char len = sa->sa_len;
1351 #endif
1352 	int error;
1353 
1354 	CURVNET_SET(so->so_vnet);
1355 	error = so->so_proto->pr_accept(so, sa);
1356 	KASSERT(sa->sa_len <= len,
1357 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1358 	CURVNET_RESTORE();
1359 	return (error);
1360 }
1361 
1362 int
1363 sopeeraddr(struct socket *so, struct sockaddr *sa)
1364 {
1365 #ifdef INVARIANTS
1366 	u_char len = sa->sa_len;
1367 #endif
1368 	int error;
1369 
1370 	CURVNET_SET(so->so_vnet);
1371 	error = so->so_proto->pr_peeraddr(so, sa);
1372 	KASSERT(sa->sa_len <= len,
1373 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1374 	CURVNET_RESTORE();
1375 
1376 	return (error);
1377 }
1378 
1379 int
1380 sosockaddr(struct socket *so, struct sockaddr *sa)
1381 {
1382 #ifdef INVARIANTS
1383 	u_char len = sa->sa_len;
1384 #endif
1385 	int error;
1386 
1387 	CURVNET_SET(so->so_vnet);
1388 	error = so->so_proto->pr_sockaddr(so, sa);
1389 	KASSERT(sa->sa_len <= len,
1390 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1391 	CURVNET_RESTORE();
1392 
1393 	return (error);
1394 }
1395 
1396 int
1397 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1398 {
1399 
1400 	return (soconnectat(AT_FDCWD, so, nam, td));
1401 }
1402 
1403 int
1404 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1405 {
1406 	int error;
1407 
1408 	CURVNET_SET(so->so_vnet);
1409 
1410 	/*
1411 	 * If protocol is connection-based, can only connect once.
1412 	 * Otherwise, if connected, try to disconnect first.  This allows
1413 	 * user to disconnect by connecting to, e.g., a null address.
1414 	 *
1415 	 * Note, this check is racy and may need to be re-evaluated at the
1416 	 * protocol layer.
1417 	 */
1418 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1419 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1420 	    (error = sodisconnect(so)))) {
1421 		error = EISCONN;
1422 	} else {
1423 		/*
1424 		 * Prevent accumulated error from previous connection from
1425 		 * biting us.
1426 		 */
1427 		so->so_error = 0;
1428 		if (fd == AT_FDCWD) {
1429 			error = so->so_proto->pr_connect(so, nam, td);
1430 		} else {
1431 			error = so->so_proto->pr_connectat(fd, so, nam, td);
1432 		}
1433 	}
1434 	CURVNET_RESTORE();
1435 
1436 	return (error);
1437 }
1438 
1439 int
1440 soconnect2(struct socket *so1, struct socket *so2)
1441 {
1442 	int error;
1443 
1444 	CURVNET_SET(so1->so_vnet);
1445 	error = so1->so_proto->pr_connect2(so1, so2);
1446 	CURVNET_RESTORE();
1447 	return (error);
1448 }
1449 
1450 int
1451 sodisconnect(struct socket *so)
1452 {
1453 	int error;
1454 
1455 	if ((so->so_state & SS_ISCONNECTED) == 0)
1456 		return (ENOTCONN);
1457 	if (so->so_state & SS_ISDISCONNECTING)
1458 		return (EALREADY);
1459 	VNET_SO_ASSERT(so);
1460 	error = so->so_proto->pr_disconnect(so);
1461 	return (error);
1462 }
1463 
1464 int
1465 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1466     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1467 {
1468 	long space;
1469 	ssize_t resid;
1470 	int clen = 0, error, dontroute;
1471 
1472 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1473 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1474 	    ("sosend_dgram: !PR_ATOMIC"));
1475 
1476 	if (uio != NULL)
1477 		resid = uio->uio_resid;
1478 	else
1479 		resid = top->m_pkthdr.len;
1480 	/*
1481 	 * In theory resid should be unsigned.  However, space must be
1482 	 * signed, as it might be less than 0 if we over-committed, and we
1483 	 * must use a signed comparison of space and resid.  On the other
1484 	 * hand, a negative resid causes us to loop sending 0-length
1485 	 * segments to the protocol.
1486 	 */
1487 	if (resid < 0) {
1488 		error = EINVAL;
1489 		goto out;
1490 	}
1491 
1492 	dontroute =
1493 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1494 	if (td != NULL)
1495 		td->td_ru.ru_msgsnd++;
1496 	if (control != NULL)
1497 		clen = control->m_len;
1498 
1499 	SOCKBUF_LOCK(&so->so_snd);
1500 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1501 		SOCKBUF_UNLOCK(&so->so_snd);
1502 		error = EPIPE;
1503 		goto out;
1504 	}
1505 	if (so->so_error) {
1506 		error = so->so_error;
1507 		so->so_error = 0;
1508 		SOCKBUF_UNLOCK(&so->so_snd);
1509 		goto out;
1510 	}
1511 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1512 		/*
1513 		 * `sendto' and `sendmsg' is allowed on a connection-based
1514 		 * socket if it supports implied connect.  Return ENOTCONN if
1515 		 * not connected and no address is supplied.
1516 		 */
1517 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1518 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1519 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1520 			    !(resid == 0 && clen != 0)) {
1521 				SOCKBUF_UNLOCK(&so->so_snd);
1522 				error = ENOTCONN;
1523 				goto out;
1524 			}
1525 		} else if (addr == NULL) {
1526 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1527 				error = ENOTCONN;
1528 			else
1529 				error = EDESTADDRREQ;
1530 			SOCKBUF_UNLOCK(&so->so_snd);
1531 			goto out;
1532 		}
1533 	}
1534 
1535 	/*
1536 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1537 	 * problem and need fixing.
1538 	 */
1539 	space = sbspace(&so->so_snd);
1540 	if (flags & MSG_OOB)
1541 		space += 1024;
1542 	space -= clen;
1543 	SOCKBUF_UNLOCK(&so->so_snd);
1544 	if (resid > space) {
1545 		error = EMSGSIZE;
1546 		goto out;
1547 	}
1548 	if (uio == NULL) {
1549 		resid = 0;
1550 		if (flags & MSG_EOR)
1551 			top->m_flags |= M_EOR;
1552 	} else {
1553 		/*
1554 		 * Copy the data from userland into a mbuf chain.
1555 		 * If no data is to be copied in, a single empty mbuf
1556 		 * is returned.
1557 		 */
1558 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1559 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1560 		if (top == NULL) {
1561 			error = EFAULT;	/* only possible error */
1562 			goto out;
1563 		}
1564 		space -= resid - uio->uio_resid;
1565 		resid = uio->uio_resid;
1566 	}
1567 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1568 	/*
1569 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1570 	 * than with.
1571 	 */
1572 	if (dontroute) {
1573 		SOCK_LOCK(so);
1574 		so->so_options |= SO_DONTROUTE;
1575 		SOCK_UNLOCK(so);
1576 	}
1577 	/*
1578 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1579 	 * of date.  We could have received a reset packet in an interrupt or
1580 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1581 	 * probably recheck again inside the locking protection here, but
1582 	 * there are probably other places that this also happens.  We must
1583 	 * rethink this.
1584 	 */
1585 	VNET_SO_ASSERT(so);
1586 	error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1587 	/*
1588 	 * If the user set MSG_EOF, the protocol understands this flag and
1589 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1590 	 */
1591 	    ((flags & MSG_EOF) &&
1592 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1593 	     (resid <= 0)) ?
1594 		PRUS_EOF :
1595 		/* If there is more to send set PRUS_MORETOCOME */
1596 		(flags & MSG_MORETOCOME) ||
1597 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1598 		top, addr, control, td);
1599 	if (dontroute) {
1600 		SOCK_LOCK(so);
1601 		so->so_options &= ~SO_DONTROUTE;
1602 		SOCK_UNLOCK(so);
1603 	}
1604 	clen = 0;
1605 	control = NULL;
1606 	top = NULL;
1607 out:
1608 	if (top != NULL)
1609 		m_freem(top);
1610 	if (control != NULL)
1611 		m_freem(control);
1612 	return (error);
1613 }
1614 
1615 /*
1616  * Send on a socket.  If send must go all at once and message is larger than
1617  * send buffering, then hard error.  Lock against other senders.  If must go
1618  * all at once and not enough room now, then inform user that this would
1619  * block and do nothing.  Otherwise, if nonblocking, send as much as
1620  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1621  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1622  * in mbuf chain must be small enough to send all at once.
1623  *
1624  * Returns nonzero on error, timeout or signal; callers must check for short
1625  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1626  * on return.
1627  */
1628 int
1629 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1630     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1631 {
1632 	long space;
1633 	ssize_t resid;
1634 	int clen = 0, error, dontroute;
1635 	int atomic = sosendallatonce(so) || top;
1636 	int pr_send_flag;
1637 #ifdef KERN_TLS
1638 	struct ktls_session *tls;
1639 	int tls_enq_cnt, tls_send_flag;
1640 	uint8_t tls_rtype;
1641 
1642 	tls = NULL;
1643 	tls_rtype = TLS_RLTYPE_APP;
1644 #endif
1645 	if (uio != NULL)
1646 		resid = uio->uio_resid;
1647 	else if ((top->m_flags & M_PKTHDR) != 0)
1648 		resid = top->m_pkthdr.len;
1649 	else
1650 		resid = m_length(top, NULL);
1651 	/*
1652 	 * In theory resid should be unsigned.  However, space must be
1653 	 * signed, as it might be less than 0 if we over-committed, and we
1654 	 * must use a signed comparison of space and resid.  On the other
1655 	 * hand, a negative resid causes us to loop sending 0-length
1656 	 * segments to the protocol.
1657 	 *
1658 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1659 	 * type sockets since that's an error.
1660 	 */
1661 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1662 		error = EINVAL;
1663 		goto out;
1664 	}
1665 
1666 	dontroute =
1667 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1668 	    (so->so_proto->pr_flags & PR_ATOMIC);
1669 	if (td != NULL)
1670 		td->td_ru.ru_msgsnd++;
1671 	if (control != NULL)
1672 		clen = control->m_len;
1673 
1674 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1675 	if (error)
1676 		goto out;
1677 
1678 #ifdef KERN_TLS
1679 	tls_send_flag = 0;
1680 	tls = ktls_hold(so->so_snd.sb_tls_info);
1681 	if (tls != NULL) {
1682 		if (tls->mode == TCP_TLS_MODE_SW)
1683 			tls_send_flag = PRUS_NOTREADY;
1684 
1685 		if (control != NULL) {
1686 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1687 
1688 			if (clen >= sizeof(*cm) &&
1689 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1690 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1691 				clen = 0;
1692 				m_freem(control);
1693 				control = NULL;
1694 				atomic = 1;
1695 			}
1696 		}
1697 
1698 		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1699 			error = EINVAL;
1700 			goto release;
1701 		}
1702 	}
1703 #endif
1704 
1705 restart:
1706 	do {
1707 		SOCKBUF_LOCK(&so->so_snd);
1708 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1709 			SOCKBUF_UNLOCK(&so->so_snd);
1710 			error = EPIPE;
1711 			goto release;
1712 		}
1713 		if (so->so_error) {
1714 			error = so->so_error;
1715 			so->so_error = 0;
1716 			SOCKBUF_UNLOCK(&so->so_snd);
1717 			goto release;
1718 		}
1719 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1720 			/*
1721 			 * `sendto' and `sendmsg' is allowed on a connection-
1722 			 * based socket if it supports implied connect.
1723 			 * Return ENOTCONN if not connected and no address is
1724 			 * supplied.
1725 			 */
1726 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1727 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1728 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1729 				    !(resid == 0 && clen != 0)) {
1730 					SOCKBUF_UNLOCK(&so->so_snd);
1731 					error = ENOTCONN;
1732 					goto release;
1733 				}
1734 			} else if (addr == NULL) {
1735 				SOCKBUF_UNLOCK(&so->so_snd);
1736 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1737 					error = ENOTCONN;
1738 				else
1739 					error = EDESTADDRREQ;
1740 				goto release;
1741 			}
1742 		}
1743 		space = sbspace(&so->so_snd);
1744 		if (flags & MSG_OOB)
1745 			space += 1024;
1746 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1747 		    clen > so->so_snd.sb_hiwat) {
1748 			SOCKBUF_UNLOCK(&so->so_snd);
1749 			error = EMSGSIZE;
1750 			goto release;
1751 		}
1752 		if (space < resid + clen &&
1753 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1754 			if ((so->so_state & SS_NBIO) ||
1755 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1756 				SOCKBUF_UNLOCK(&so->so_snd);
1757 				error = EWOULDBLOCK;
1758 				goto release;
1759 			}
1760 			error = sbwait(so, SO_SND);
1761 			SOCKBUF_UNLOCK(&so->so_snd);
1762 			if (error)
1763 				goto release;
1764 			goto restart;
1765 		}
1766 		SOCKBUF_UNLOCK(&so->so_snd);
1767 		space -= clen;
1768 		do {
1769 			if (uio == NULL) {
1770 				resid = 0;
1771 				if (flags & MSG_EOR)
1772 					top->m_flags |= M_EOR;
1773 #ifdef KERN_TLS
1774 				if (tls != NULL) {
1775 					ktls_frame(top, tls, &tls_enq_cnt,
1776 					    tls_rtype);
1777 					tls_rtype = TLS_RLTYPE_APP;
1778 				}
1779 #endif
1780 			} else {
1781 				/*
1782 				 * Copy the data from userland into a mbuf
1783 				 * chain.  If resid is 0, which can happen
1784 				 * only if we have control to send, then
1785 				 * a single empty mbuf is returned.  This
1786 				 * is a workaround to prevent protocol send
1787 				 * methods to panic.
1788 				 */
1789 #ifdef KERN_TLS
1790 				if (tls != NULL) {
1791 					top = m_uiotombuf(uio, M_WAITOK, space,
1792 					    tls->params.max_frame_len,
1793 					    M_EXTPG |
1794 					    ((flags & MSG_EOR) ? M_EOR : 0));
1795 					if (top != NULL) {
1796 						ktls_frame(top, tls,
1797 						    &tls_enq_cnt, tls_rtype);
1798 					}
1799 					tls_rtype = TLS_RLTYPE_APP;
1800 				} else
1801 #endif
1802 					top = m_uiotombuf(uio, M_WAITOK, space,
1803 					    (atomic ? max_hdr : 0),
1804 					    (atomic ? M_PKTHDR : 0) |
1805 					    ((flags & MSG_EOR) ? M_EOR : 0));
1806 				if (top == NULL) {
1807 					error = EFAULT; /* only possible error */
1808 					goto release;
1809 				}
1810 				space -= resid - uio->uio_resid;
1811 				resid = uio->uio_resid;
1812 			}
1813 			if (dontroute) {
1814 				SOCK_LOCK(so);
1815 				so->so_options |= SO_DONTROUTE;
1816 				SOCK_UNLOCK(so);
1817 			}
1818 			/*
1819 			 * XXX all the SBS_CANTSENDMORE checks previously
1820 			 * done could be out of date.  We could have received
1821 			 * a reset packet in an interrupt or maybe we slept
1822 			 * while doing page faults in uiomove() etc.  We
1823 			 * could probably recheck again inside the locking
1824 			 * protection here, but there are probably other
1825 			 * places that this also happens.  We must rethink
1826 			 * this.
1827 			 */
1828 			VNET_SO_ASSERT(so);
1829 
1830 			pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
1831 			/*
1832 			 * If the user set MSG_EOF, the protocol understands
1833 			 * this flag and nothing left to send then use
1834 			 * PRU_SEND_EOF instead of PRU_SEND.
1835 			 */
1836 			    ((flags & MSG_EOF) &&
1837 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1838 			     (resid <= 0)) ?
1839 				PRUS_EOF :
1840 			/* If there is more to send set PRUS_MORETOCOME. */
1841 			    (flags & MSG_MORETOCOME) ||
1842 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1843 
1844 #ifdef KERN_TLS
1845 			pr_send_flag |= tls_send_flag;
1846 #endif
1847 
1848 			error = so->so_proto->pr_send(so, pr_send_flag, top,
1849 			    addr, control, td);
1850 
1851 			if (dontroute) {
1852 				SOCK_LOCK(so);
1853 				so->so_options &= ~SO_DONTROUTE;
1854 				SOCK_UNLOCK(so);
1855 			}
1856 
1857 #ifdef KERN_TLS
1858 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1859 				if (error != 0) {
1860 					m_freem(top);
1861 					top = NULL;
1862 				} else {
1863 					soref(so);
1864 					ktls_enqueue(top, so, tls_enq_cnt);
1865 				}
1866 			}
1867 #endif
1868 			clen = 0;
1869 			control = NULL;
1870 			top = NULL;
1871 			if (error)
1872 				goto release;
1873 		} while (resid && space > 0);
1874 	} while (resid);
1875 
1876 release:
1877 	SOCK_IO_SEND_UNLOCK(so);
1878 out:
1879 #ifdef KERN_TLS
1880 	if (tls != NULL)
1881 		ktls_free(tls);
1882 #endif
1883 	if (top != NULL)
1884 		m_freem(top);
1885 	if (control != NULL)
1886 		m_freem(control);
1887 	return (error);
1888 }
1889 
1890 /*
1891  * Send to a socket from a kernel thread.
1892  *
1893  * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
1894  * Exception is nfs/bootp_subr.c.  It is arguable that the VNET context needs
1895  * to be set at all.  This function should just boil down to a static inline
1896  * calling the protocol method.
1897  */
1898 int
1899 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1900     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1901 {
1902 	int error;
1903 
1904 	CURVNET_SET(so->so_vnet);
1905 	error = so->so_proto->pr_sosend(so, addr, uio,
1906 	    top, control, flags, td);
1907 	CURVNET_RESTORE();
1908 	return (error);
1909 }
1910 
1911 /*
1912  * send(2), write(2) or aio_write(2) on a socket.
1913  */
1914 int
1915 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1916     struct mbuf *control, int flags, struct proc *userproc)
1917 {
1918 	struct thread *td;
1919 	ssize_t len;
1920 	int error;
1921 
1922 	td = uio->uio_td;
1923 	len = uio->uio_resid;
1924 	CURVNET_SET(so->so_vnet);
1925 	error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
1926 	    td);
1927 	CURVNET_RESTORE();
1928 	if (error != 0) {
1929 		/*
1930 		 * Clear transient errors for stream protocols if they made
1931 		 * some progress.  Make exclusion for aio(4) that would
1932 		 * schedule a new write in case of EWOULDBLOCK and clear
1933 		 * error itself.  See soaio_process_job().
1934 		 */
1935 		if (uio->uio_resid != len &&
1936 		    (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
1937 		    userproc == NULL &&
1938 		    (error == ERESTART || error == EINTR ||
1939 		    error == EWOULDBLOCK))
1940 			error = 0;
1941 		/* Generation of SIGPIPE can be controlled per socket. */
1942 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
1943 		    (flags & MSG_NOSIGNAL) == 0) {
1944 			if (userproc != NULL) {
1945 				/* aio(4) job */
1946 				PROC_LOCK(userproc);
1947 				kern_psignal(userproc, SIGPIPE);
1948 				PROC_UNLOCK(userproc);
1949 			} else {
1950 				PROC_LOCK(td->td_proc);
1951 				tdsignal(td, SIGPIPE);
1952 				PROC_UNLOCK(td->td_proc);
1953 			}
1954 		}
1955 	}
1956 	return (error);
1957 }
1958 
1959 /*
1960  * The part of soreceive() that implements reading non-inline out-of-band
1961  * data from a socket.  For more complete comments, see soreceive(), from
1962  * which this code originated.
1963  *
1964  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1965  * unable to return an mbuf chain to the caller.
1966  */
1967 static int
1968 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1969 {
1970 	struct protosw *pr = so->so_proto;
1971 	struct mbuf *m;
1972 	int error;
1973 
1974 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1975 	VNET_SO_ASSERT(so);
1976 
1977 	m = m_get(M_WAITOK, MT_DATA);
1978 	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
1979 	if (error)
1980 		goto bad;
1981 	do {
1982 		error = uiomove(mtod(m, void *),
1983 		    (int) min(uio->uio_resid, m->m_len), uio);
1984 		m = m_free(m);
1985 	} while (uio->uio_resid && error == 0 && m);
1986 bad:
1987 	if (m != NULL)
1988 		m_freem(m);
1989 	return (error);
1990 }
1991 
1992 /*
1993  * Following replacement or removal of the first mbuf on the first mbuf chain
1994  * of a socket buffer, push necessary state changes back into the socket
1995  * buffer so that other consumers see the values consistently.  'nextrecord'
1996  * is the callers locally stored value of the original value of
1997  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1998  * NOTE: 'nextrecord' may be NULL.
1999  */
2000 static __inline void
2001 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
2002 {
2003 
2004 	SOCKBUF_LOCK_ASSERT(sb);
2005 	/*
2006 	 * First, update for the new value of nextrecord.  If necessary, make
2007 	 * it the first record.
2008 	 */
2009 	if (sb->sb_mb != NULL)
2010 		sb->sb_mb->m_nextpkt = nextrecord;
2011 	else
2012 		sb->sb_mb = nextrecord;
2013 
2014 	/*
2015 	 * Now update any dependent socket buffer fields to reflect the new
2016 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
2017 	 * addition of a second clause that takes care of the case where
2018 	 * sb_mb has been updated, but remains the last record.
2019 	 */
2020 	if (sb->sb_mb == NULL) {
2021 		sb->sb_mbtail = NULL;
2022 		sb->sb_lastrecord = NULL;
2023 	} else if (sb->sb_mb->m_nextpkt == NULL)
2024 		sb->sb_lastrecord = sb->sb_mb;
2025 }
2026 
2027 /*
2028  * Implement receive operations on a socket.  We depend on the way that
2029  * records are added to the sockbuf by sbappend.  In particular, each record
2030  * (mbufs linked through m_next) must begin with an address if the protocol
2031  * so specifies, followed by an optional mbuf or mbufs containing ancillary
2032  * data, and then zero or more mbufs of data.  In order to allow parallelism
2033  * between network receive and copying to user space, as well as avoid
2034  * sleeping with a mutex held, we release the socket buffer mutex during the
2035  * user space copy.  Although the sockbuf is locked, new data may still be
2036  * appended, and thus we must maintain consistency of the sockbuf during that
2037  * time.
2038  *
2039  * The caller may receive the data as a single mbuf chain by supplying an
2040  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
2041  * the count in uio_resid.
2042  */
2043 int
2044 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
2045     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2046 {
2047 	struct mbuf *m, **mp;
2048 	int flags, error, offset;
2049 	ssize_t len;
2050 	struct protosw *pr = so->so_proto;
2051 	struct mbuf *nextrecord;
2052 	int moff, type = 0;
2053 	ssize_t orig_resid = uio->uio_resid;
2054 	bool report_real_len = false;
2055 
2056 	mp = mp0;
2057 	if (psa != NULL)
2058 		*psa = NULL;
2059 	if (controlp != NULL)
2060 		*controlp = NULL;
2061 	if (flagsp != NULL) {
2062 		report_real_len = *flagsp & MSG_TRUNC;
2063 		*flagsp &= ~MSG_TRUNC;
2064 		flags = *flagsp &~ MSG_EOR;
2065 	} else
2066 		flags = 0;
2067 	if (flags & MSG_OOB)
2068 		return (soreceive_rcvoob(so, uio, flags));
2069 	if (mp != NULL)
2070 		*mp = NULL;
2071 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
2072 	    && uio->uio_resid) {
2073 		VNET_SO_ASSERT(so);
2074 		pr->pr_rcvd(so, 0);
2075 	}
2076 
2077 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2078 	if (error)
2079 		return (error);
2080 
2081 restart:
2082 	SOCKBUF_LOCK(&so->so_rcv);
2083 	m = so->so_rcv.sb_mb;
2084 	/*
2085 	 * If we have less data than requested, block awaiting more (subject
2086 	 * to any timeout) if:
2087 	 *   1. the current count is less than the low water mark, or
2088 	 *   2. MSG_DONTWAIT is not set
2089 	 */
2090 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2091 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
2092 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2093 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2094 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
2095 		    ("receive: m == %p sbavail == %u",
2096 		    m, sbavail(&so->so_rcv)));
2097 		if (so->so_error || so->so_rerror) {
2098 			if (m != NULL)
2099 				goto dontblock;
2100 			if (so->so_error)
2101 				error = so->so_error;
2102 			else
2103 				error = so->so_rerror;
2104 			if ((flags & MSG_PEEK) == 0) {
2105 				if (so->so_error)
2106 					so->so_error = 0;
2107 				else
2108 					so->so_rerror = 0;
2109 			}
2110 			SOCKBUF_UNLOCK(&so->so_rcv);
2111 			goto release;
2112 		}
2113 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2114 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2115 			if (m != NULL)
2116 				goto dontblock;
2117 #ifdef KERN_TLS
2118 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2119 			    so->so_rcv.sb_tlscc == 0) {
2120 #else
2121 			else {
2122 #endif
2123 				SOCKBUF_UNLOCK(&so->so_rcv);
2124 				goto release;
2125 			}
2126 		}
2127 		for (; m != NULL; m = m->m_next)
2128 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2129 				m = so->so_rcv.sb_mb;
2130 				goto dontblock;
2131 			}
2132 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2133 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2134 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2135 			SOCKBUF_UNLOCK(&so->so_rcv);
2136 			error = ENOTCONN;
2137 			goto release;
2138 		}
2139 		if (uio->uio_resid == 0 && !report_real_len) {
2140 			SOCKBUF_UNLOCK(&so->so_rcv);
2141 			goto release;
2142 		}
2143 		if ((so->so_state & SS_NBIO) ||
2144 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2145 			SOCKBUF_UNLOCK(&so->so_rcv);
2146 			error = EWOULDBLOCK;
2147 			goto release;
2148 		}
2149 		SBLASTRECORDCHK(&so->so_rcv);
2150 		SBLASTMBUFCHK(&so->so_rcv);
2151 		error = sbwait(so, SO_RCV);
2152 		SOCKBUF_UNLOCK(&so->so_rcv);
2153 		if (error)
2154 			goto release;
2155 		goto restart;
2156 	}
2157 dontblock:
2158 	/*
2159 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2160 	 * pointer to the next record in the socket buffer.  We must keep the
2161 	 * various socket buffer pointers and local stack versions of the
2162 	 * pointers in sync, pushing out modifications before dropping the
2163 	 * socket buffer mutex, and re-reading them when picking it up.
2164 	 *
2165 	 * Otherwise, we will race with the network stack appending new data
2166 	 * or records onto the socket buffer by using inconsistent/stale
2167 	 * versions of the field, possibly resulting in socket buffer
2168 	 * corruption.
2169 	 *
2170 	 * By holding the high-level sblock(), we prevent simultaneous
2171 	 * readers from pulling off the front of the socket buffer.
2172 	 */
2173 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2174 	if (uio->uio_td)
2175 		uio->uio_td->td_ru.ru_msgrcv++;
2176 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2177 	SBLASTRECORDCHK(&so->so_rcv);
2178 	SBLASTMBUFCHK(&so->so_rcv);
2179 	nextrecord = m->m_nextpkt;
2180 	if (pr->pr_flags & PR_ADDR) {
2181 		KASSERT(m->m_type == MT_SONAME,
2182 		    ("m->m_type == %d", m->m_type));
2183 		orig_resid = 0;
2184 		if (psa != NULL)
2185 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2186 			    M_NOWAIT);
2187 		if (flags & MSG_PEEK) {
2188 			m = m->m_next;
2189 		} else {
2190 			sbfree(&so->so_rcv, m);
2191 			so->so_rcv.sb_mb = m_free(m);
2192 			m = so->so_rcv.sb_mb;
2193 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2194 		}
2195 	}
2196 
2197 	/*
2198 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2199 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2200 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2201 	 * perform externalization (or freeing if controlp == NULL).
2202 	 */
2203 	if (m != NULL && m->m_type == MT_CONTROL) {
2204 		struct mbuf *cm = NULL, *cmn;
2205 		struct mbuf **cme = &cm;
2206 #ifdef KERN_TLS
2207 		struct cmsghdr *cmsg;
2208 		struct tls_get_record tgr;
2209 
2210 		/*
2211 		 * For MSG_TLSAPPDATA, check for an alert record.
2212 		 * If found, return ENXIO without removing
2213 		 * it from the receive queue.  This allows a subsequent
2214 		 * call without MSG_TLSAPPDATA to receive it.
2215 		 * Note that, for TLS, there should only be a single
2216 		 * control mbuf with the TLS_GET_RECORD message in it.
2217 		 */
2218 		if (flags & MSG_TLSAPPDATA) {
2219 			cmsg = mtod(m, struct cmsghdr *);
2220 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2221 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2222 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2223 				if (__predict_false(tgr.tls_type ==
2224 				    TLS_RLTYPE_ALERT)) {
2225 					SOCKBUF_UNLOCK(&so->so_rcv);
2226 					error = ENXIO;
2227 					goto release;
2228 				}
2229 			}
2230 		}
2231 #endif
2232 
2233 		do {
2234 			if (flags & MSG_PEEK) {
2235 				if (controlp != NULL) {
2236 					*controlp = m_copym(m, 0, m->m_len,
2237 					    M_NOWAIT);
2238 					controlp = &(*controlp)->m_next;
2239 				}
2240 				m = m->m_next;
2241 			} else {
2242 				sbfree(&so->so_rcv, m);
2243 				so->so_rcv.sb_mb = m->m_next;
2244 				m->m_next = NULL;
2245 				*cme = m;
2246 				cme = &(*cme)->m_next;
2247 				m = so->so_rcv.sb_mb;
2248 			}
2249 		} while (m != NULL && m->m_type == MT_CONTROL);
2250 		if ((flags & MSG_PEEK) == 0)
2251 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2252 		while (cm != NULL) {
2253 			cmn = cm->m_next;
2254 			cm->m_next = NULL;
2255 			if (pr->pr_domain->dom_externalize != NULL) {
2256 				SOCKBUF_UNLOCK(&so->so_rcv);
2257 				VNET_SO_ASSERT(so);
2258 				error = (*pr->pr_domain->dom_externalize)
2259 				    (cm, controlp, flags);
2260 				SOCKBUF_LOCK(&so->so_rcv);
2261 			} else if (controlp != NULL)
2262 				*controlp = cm;
2263 			else
2264 				m_freem(cm);
2265 			if (controlp != NULL) {
2266 				while (*controlp != NULL)
2267 					controlp = &(*controlp)->m_next;
2268 			}
2269 			cm = cmn;
2270 		}
2271 		if (m != NULL)
2272 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2273 		else
2274 			nextrecord = so->so_rcv.sb_mb;
2275 		orig_resid = 0;
2276 	}
2277 	if (m != NULL) {
2278 		if ((flags & MSG_PEEK) == 0) {
2279 			KASSERT(m->m_nextpkt == nextrecord,
2280 			    ("soreceive: post-control, nextrecord !sync"));
2281 			if (nextrecord == NULL) {
2282 				KASSERT(so->so_rcv.sb_mb == m,
2283 				    ("soreceive: post-control, sb_mb!=m"));
2284 				KASSERT(so->so_rcv.sb_lastrecord == m,
2285 				    ("soreceive: post-control, lastrecord!=m"));
2286 			}
2287 		}
2288 		type = m->m_type;
2289 		if (type == MT_OOBDATA)
2290 			flags |= MSG_OOB;
2291 	} else {
2292 		if ((flags & MSG_PEEK) == 0) {
2293 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2294 			    ("soreceive: sb_mb != nextrecord"));
2295 			if (so->so_rcv.sb_mb == NULL) {
2296 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2297 				    ("soreceive: sb_lastercord != NULL"));
2298 			}
2299 		}
2300 	}
2301 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2302 	SBLASTRECORDCHK(&so->so_rcv);
2303 	SBLASTMBUFCHK(&so->so_rcv);
2304 
2305 	/*
2306 	 * Now continue to read any data mbufs off of the head of the socket
2307 	 * buffer until the read request is satisfied.  Note that 'type' is
2308 	 * used to store the type of any mbuf reads that have happened so far
2309 	 * such that soreceive() can stop reading if the type changes, which
2310 	 * causes soreceive() to return only one of regular data and inline
2311 	 * out-of-band data in a single socket receive operation.
2312 	 */
2313 	moff = 0;
2314 	offset = 0;
2315 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2316 	    && error == 0) {
2317 		/*
2318 		 * If the type of mbuf has changed since the last mbuf
2319 		 * examined ('type'), end the receive operation.
2320 		 */
2321 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2322 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2323 			if (type != m->m_type)
2324 				break;
2325 		} else if (type == MT_OOBDATA)
2326 			break;
2327 		else
2328 		    KASSERT(m->m_type == MT_DATA,
2329 			("m->m_type == %d", m->m_type));
2330 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2331 		len = uio->uio_resid;
2332 		if (so->so_oobmark && len > so->so_oobmark - offset)
2333 			len = so->so_oobmark - offset;
2334 		if (len > m->m_len - moff)
2335 			len = m->m_len - moff;
2336 		/*
2337 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2338 		 * them out via the uio, then free.  Sockbuf must be
2339 		 * consistent here (points to current mbuf, it points to next
2340 		 * record) when we drop priority; we must note any additions
2341 		 * to the sockbuf when we block interrupts again.
2342 		 */
2343 		if (mp == NULL) {
2344 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2345 			SBLASTRECORDCHK(&so->so_rcv);
2346 			SBLASTMBUFCHK(&so->so_rcv);
2347 			SOCKBUF_UNLOCK(&so->so_rcv);
2348 			if ((m->m_flags & M_EXTPG) != 0)
2349 				error = m_unmapped_uiomove(m, moff, uio,
2350 				    (int)len);
2351 			else
2352 				error = uiomove(mtod(m, char *) + moff,
2353 				    (int)len, uio);
2354 			SOCKBUF_LOCK(&so->so_rcv);
2355 			if (error) {
2356 				/*
2357 				 * The MT_SONAME mbuf has already been removed
2358 				 * from the record, so it is necessary to
2359 				 * remove the data mbufs, if any, to preserve
2360 				 * the invariant in the case of PR_ADDR that
2361 				 * requires MT_SONAME mbufs at the head of
2362 				 * each record.
2363 				 */
2364 				if (pr->pr_flags & PR_ATOMIC &&
2365 				    ((flags & MSG_PEEK) == 0))
2366 					(void)sbdroprecord_locked(&so->so_rcv);
2367 				SOCKBUF_UNLOCK(&so->so_rcv);
2368 				goto release;
2369 			}
2370 		} else
2371 			uio->uio_resid -= len;
2372 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2373 		if (len == m->m_len - moff) {
2374 			if (m->m_flags & M_EOR)
2375 				flags |= MSG_EOR;
2376 			if (flags & MSG_PEEK) {
2377 				m = m->m_next;
2378 				moff = 0;
2379 			} else {
2380 				nextrecord = m->m_nextpkt;
2381 				sbfree(&so->so_rcv, m);
2382 				if (mp != NULL) {
2383 					m->m_nextpkt = NULL;
2384 					*mp = m;
2385 					mp = &m->m_next;
2386 					so->so_rcv.sb_mb = m = m->m_next;
2387 					*mp = NULL;
2388 				} else {
2389 					so->so_rcv.sb_mb = m_free(m);
2390 					m = so->so_rcv.sb_mb;
2391 				}
2392 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2393 				SBLASTRECORDCHK(&so->so_rcv);
2394 				SBLASTMBUFCHK(&so->so_rcv);
2395 			}
2396 		} else {
2397 			if (flags & MSG_PEEK)
2398 				moff += len;
2399 			else {
2400 				if (mp != NULL) {
2401 					if (flags & MSG_DONTWAIT) {
2402 						*mp = m_copym(m, 0, len,
2403 						    M_NOWAIT);
2404 						if (*mp == NULL) {
2405 							/*
2406 							 * m_copym() couldn't
2407 							 * allocate an mbuf.
2408 							 * Adjust uio_resid back
2409 							 * (it was adjusted
2410 							 * down by len bytes,
2411 							 * which we didn't end
2412 							 * up "copying" over).
2413 							 */
2414 							uio->uio_resid += len;
2415 							break;
2416 						}
2417 					} else {
2418 						SOCKBUF_UNLOCK(&so->so_rcv);
2419 						*mp = m_copym(m, 0, len,
2420 						    M_WAITOK);
2421 						SOCKBUF_LOCK(&so->so_rcv);
2422 					}
2423 				}
2424 				sbcut_locked(&so->so_rcv, len);
2425 			}
2426 		}
2427 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2428 		if (so->so_oobmark) {
2429 			if ((flags & MSG_PEEK) == 0) {
2430 				so->so_oobmark -= len;
2431 				if (so->so_oobmark == 0) {
2432 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2433 					break;
2434 				}
2435 			} else {
2436 				offset += len;
2437 				if (offset == so->so_oobmark)
2438 					break;
2439 			}
2440 		}
2441 		if (flags & MSG_EOR)
2442 			break;
2443 		/*
2444 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2445 		 * must not quit until "uio->uio_resid == 0" or an error
2446 		 * termination.  If a signal/timeout occurs, return with a
2447 		 * short count but without error.  Keep sockbuf locked
2448 		 * against other readers.
2449 		 */
2450 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2451 		    !sosendallatonce(so) && nextrecord == NULL) {
2452 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2453 			if (so->so_error || so->so_rerror ||
2454 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2455 				break;
2456 			/*
2457 			 * Notify the protocol that some data has been
2458 			 * drained before blocking.
2459 			 */
2460 			if (pr->pr_flags & PR_WANTRCVD) {
2461 				SOCKBUF_UNLOCK(&so->so_rcv);
2462 				VNET_SO_ASSERT(so);
2463 				pr->pr_rcvd(so, flags);
2464 				SOCKBUF_LOCK(&so->so_rcv);
2465 				if (__predict_false(so->so_rcv.sb_mb == NULL &&
2466 				    (so->so_error || so->so_rerror ||
2467 				    so->so_rcv.sb_state & SBS_CANTRCVMORE)))
2468 					break;
2469 			}
2470 			SBLASTRECORDCHK(&so->so_rcv);
2471 			SBLASTMBUFCHK(&so->so_rcv);
2472 			/*
2473 			 * We could receive some data while was notifying
2474 			 * the protocol. Skip blocking in this case.
2475 			 */
2476 			if (so->so_rcv.sb_mb == NULL) {
2477 				error = sbwait(so, SO_RCV);
2478 				if (error) {
2479 					SOCKBUF_UNLOCK(&so->so_rcv);
2480 					goto release;
2481 				}
2482 			}
2483 			m = so->so_rcv.sb_mb;
2484 			if (m != NULL)
2485 				nextrecord = m->m_nextpkt;
2486 		}
2487 	}
2488 
2489 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2490 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2491 		if (report_real_len)
2492 			uio->uio_resid -= m_length(m, NULL) - moff;
2493 		flags |= MSG_TRUNC;
2494 		if ((flags & MSG_PEEK) == 0)
2495 			(void) sbdroprecord_locked(&so->so_rcv);
2496 	}
2497 	if ((flags & MSG_PEEK) == 0) {
2498 		if (m == NULL) {
2499 			/*
2500 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2501 			 * part makes sure sb_lastrecord is up-to-date if
2502 			 * there is still data in the socket buffer.
2503 			 */
2504 			so->so_rcv.sb_mb = nextrecord;
2505 			if (so->so_rcv.sb_mb == NULL) {
2506 				so->so_rcv.sb_mbtail = NULL;
2507 				so->so_rcv.sb_lastrecord = NULL;
2508 			} else if (nextrecord->m_nextpkt == NULL)
2509 				so->so_rcv.sb_lastrecord = nextrecord;
2510 		}
2511 		SBLASTRECORDCHK(&so->so_rcv);
2512 		SBLASTMBUFCHK(&so->so_rcv);
2513 		/*
2514 		 * If soreceive() is being done from the socket callback,
2515 		 * then don't need to generate ACK to peer to update window,
2516 		 * since ACK will be generated on return to TCP.
2517 		 */
2518 		if (!(flags & MSG_SOCALLBCK) &&
2519 		    (pr->pr_flags & PR_WANTRCVD)) {
2520 			SOCKBUF_UNLOCK(&so->so_rcv);
2521 			VNET_SO_ASSERT(so);
2522 			pr->pr_rcvd(so, flags);
2523 			SOCKBUF_LOCK(&so->so_rcv);
2524 		}
2525 	}
2526 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2527 	if (orig_resid == uio->uio_resid && orig_resid &&
2528 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2529 		SOCKBUF_UNLOCK(&so->so_rcv);
2530 		goto restart;
2531 	}
2532 	SOCKBUF_UNLOCK(&so->so_rcv);
2533 
2534 	if (flagsp != NULL)
2535 		*flagsp |= flags;
2536 release:
2537 	SOCK_IO_RECV_UNLOCK(so);
2538 	return (error);
2539 }
2540 
2541 /*
2542  * Optimized version of soreceive() for stream (TCP) sockets.
2543  */
2544 int
2545 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2546     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2547 {
2548 	int len = 0, error = 0, flags, oresid;
2549 	struct sockbuf *sb;
2550 	struct mbuf *m, *n = NULL;
2551 
2552 	/* We only do stream sockets. */
2553 	if (so->so_type != SOCK_STREAM)
2554 		return (EINVAL);
2555 	if (psa != NULL)
2556 		*psa = NULL;
2557 	if (flagsp != NULL)
2558 		flags = *flagsp &~ MSG_EOR;
2559 	else
2560 		flags = 0;
2561 	if (controlp != NULL)
2562 		*controlp = NULL;
2563 	if (flags & MSG_OOB)
2564 		return (soreceive_rcvoob(so, uio, flags));
2565 	if (mp0 != NULL)
2566 		*mp0 = NULL;
2567 
2568 	sb = &so->so_rcv;
2569 
2570 #ifdef KERN_TLS
2571 	/*
2572 	 * KTLS store TLS records as records with a control message to
2573 	 * describe the framing.
2574 	 *
2575 	 * We check once here before acquiring locks to optimize the
2576 	 * common case.
2577 	 */
2578 	if (sb->sb_tls_info != NULL)
2579 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2580 		    flagsp));
2581 #endif
2582 
2583 	/* Prevent other readers from entering the socket. */
2584 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2585 	if (error)
2586 		return (error);
2587 	SOCKBUF_LOCK(sb);
2588 
2589 #ifdef KERN_TLS
2590 	if (sb->sb_tls_info != NULL) {
2591 		SOCKBUF_UNLOCK(sb);
2592 		SOCK_IO_RECV_UNLOCK(so);
2593 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2594 		    flagsp));
2595 	}
2596 #endif
2597 
2598 	/* Easy one, no space to copyout anything. */
2599 	if (uio->uio_resid == 0) {
2600 		error = EINVAL;
2601 		goto out;
2602 	}
2603 	oresid = uio->uio_resid;
2604 
2605 	/* We will never ever get anything unless we are or were connected. */
2606 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2607 		error = ENOTCONN;
2608 		goto out;
2609 	}
2610 
2611 restart:
2612 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2613 
2614 	/* Abort if socket has reported problems. */
2615 	if (so->so_error) {
2616 		if (sbavail(sb) > 0)
2617 			goto deliver;
2618 		if (oresid > uio->uio_resid)
2619 			goto out;
2620 		error = so->so_error;
2621 		if (!(flags & MSG_PEEK))
2622 			so->so_error = 0;
2623 		goto out;
2624 	}
2625 
2626 	/* Door is closed.  Deliver what is left, if any. */
2627 	if (sb->sb_state & SBS_CANTRCVMORE) {
2628 		if (sbavail(sb) > 0)
2629 			goto deliver;
2630 		else
2631 			goto out;
2632 	}
2633 
2634 	/* Socket buffer is empty and we shall not block. */
2635 	if (sbavail(sb) == 0 &&
2636 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2637 		error = EAGAIN;
2638 		goto out;
2639 	}
2640 
2641 	/* Socket buffer got some data that we shall deliver now. */
2642 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2643 	    ((so->so_state & SS_NBIO) ||
2644 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2645 	     sbavail(sb) >= sb->sb_lowat ||
2646 	     sbavail(sb) >= uio->uio_resid ||
2647 	     sbavail(sb) >= sb->sb_hiwat) ) {
2648 		goto deliver;
2649 	}
2650 
2651 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2652 	if ((flags & MSG_WAITALL) &&
2653 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2654 		goto deliver;
2655 
2656 	/*
2657 	 * Wait and block until (more) data comes in.
2658 	 * NB: Drops the sockbuf lock during wait.
2659 	 */
2660 	error = sbwait(so, SO_RCV);
2661 	if (error)
2662 		goto out;
2663 	goto restart;
2664 
2665 deliver:
2666 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2667 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2668 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2669 
2670 	/* Statistics. */
2671 	if (uio->uio_td)
2672 		uio->uio_td->td_ru.ru_msgrcv++;
2673 
2674 	/* Fill uio until full or current end of socket buffer is reached. */
2675 	len = min(uio->uio_resid, sbavail(sb));
2676 	if (mp0 != NULL) {
2677 		/* Dequeue as many mbufs as possible. */
2678 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2679 			if (*mp0 == NULL)
2680 				*mp0 = sb->sb_mb;
2681 			else
2682 				m_cat(*mp0, sb->sb_mb);
2683 			for (m = sb->sb_mb;
2684 			     m != NULL && m->m_len <= len;
2685 			     m = m->m_next) {
2686 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2687 				    ("%s: m %p not available", __func__, m));
2688 				len -= m->m_len;
2689 				uio->uio_resid -= m->m_len;
2690 				sbfree(sb, m);
2691 				n = m;
2692 			}
2693 			n->m_next = NULL;
2694 			sb->sb_mb = m;
2695 			sb->sb_lastrecord = sb->sb_mb;
2696 			if (sb->sb_mb == NULL)
2697 				SB_EMPTY_FIXUP(sb);
2698 		}
2699 		/* Copy the remainder. */
2700 		if (len > 0) {
2701 			KASSERT(sb->sb_mb != NULL,
2702 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2703 
2704 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2705 			if (m == NULL)
2706 				len = 0;	/* Don't flush data from sockbuf. */
2707 			else
2708 				uio->uio_resid -= len;
2709 			if (*mp0 != NULL)
2710 				m_cat(*mp0, m);
2711 			else
2712 				*mp0 = m;
2713 			if (*mp0 == NULL) {
2714 				error = ENOBUFS;
2715 				goto out;
2716 			}
2717 		}
2718 	} else {
2719 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2720 		SOCKBUF_UNLOCK(sb);
2721 		error = m_mbuftouio(uio, sb->sb_mb, len);
2722 		SOCKBUF_LOCK(sb);
2723 		if (error)
2724 			goto out;
2725 	}
2726 	SBLASTRECORDCHK(sb);
2727 	SBLASTMBUFCHK(sb);
2728 
2729 	/*
2730 	 * Remove the delivered data from the socket buffer unless we
2731 	 * were only peeking.
2732 	 */
2733 	if (!(flags & MSG_PEEK)) {
2734 		if (len > 0)
2735 			sbdrop_locked(sb, len);
2736 
2737 		/* Notify protocol that we drained some data. */
2738 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2739 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2740 		     !(flags & MSG_SOCALLBCK))) {
2741 			SOCKBUF_UNLOCK(sb);
2742 			VNET_SO_ASSERT(so);
2743 			so->so_proto->pr_rcvd(so, flags);
2744 			SOCKBUF_LOCK(sb);
2745 		}
2746 	}
2747 
2748 	/*
2749 	 * For MSG_WAITALL we may have to loop again and wait for
2750 	 * more data to come in.
2751 	 */
2752 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2753 		goto restart;
2754 out:
2755 	SBLASTRECORDCHK(sb);
2756 	SBLASTMBUFCHK(sb);
2757 	SOCKBUF_UNLOCK(sb);
2758 	SOCK_IO_RECV_UNLOCK(so);
2759 	return (error);
2760 }
2761 
2762 /*
2763  * Optimized version of soreceive() for simple datagram cases from userspace.
2764  * Unlike in the stream case, we're able to drop a datagram if copyout()
2765  * fails, and because we handle datagrams atomically, we don't need to use a
2766  * sleep lock to prevent I/O interlacing.
2767  */
2768 int
2769 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2770     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2771 {
2772 	struct mbuf *m, *m2;
2773 	int flags, error;
2774 	ssize_t len;
2775 	struct protosw *pr = so->so_proto;
2776 	struct mbuf *nextrecord;
2777 
2778 	if (psa != NULL)
2779 		*psa = NULL;
2780 	if (controlp != NULL)
2781 		*controlp = NULL;
2782 	if (flagsp != NULL)
2783 		flags = *flagsp &~ MSG_EOR;
2784 	else
2785 		flags = 0;
2786 
2787 	/*
2788 	 * For any complicated cases, fall back to the full
2789 	 * soreceive_generic().
2790 	 */
2791 	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2792 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2793 		    flagsp));
2794 
2795 	/*
2796 	 * Enforce restrictions on use.
2797 	 */
2798 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2799 	    ("soreceive_dgram: wantrcvd"));
2800 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2801 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2802 	    ("soreceive_dgram: SBS_RCVATMARK"));
2803 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2804 	    ("soreceive_dgram: P_CONNREQUIRED"));
2805 
2806 	/*
2807 	 * Loop blocking while waiting for a datagram.
2808 	 */
2809 	SOCKBUF_LOCK(&so->so_rcv);
2810 	while ((m = so->so_rcv.sb_mb) == NULL) {
2811 		KASSERT(sbavail(&so->so_rcv) == 0,
2812 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2813 		    sbavail(&so->so_rcv)));
2814 		if (so->so_error) {
2815 			error = so->so_error;
2816 			so->so_error = 0;
2817 			SOCKBUF_UNLOCK(&so->so_rcv);
2818 			return (error);
2819 		}
2820 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2821 		    uio->uio_resid == 0) {
2822 			SOCKBUF_UNLOCK(&so->so_rcv);
2823 			return (0);
2824 		}
2825 		if ((so->so_state & SS_NBIO) ||
2826 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2827 			SOCKBUF_UNLOCK(&so->so_rcv);
2828 			return (EWOULDBLOCK);
2829 		}
2830 		SBLASTRECORDCHK(&so->so_rcv);
2831 		SBLASTMBUFCHK(&so->so_rcv);
2832 		error = sbwait(so, SO_RCV);
2833 		if (error) {
2834 			SOCKBUF_UNLOCK(&so->so_rcv);
2835 			return (error);
2836 		}
2837 	}
2838 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2839 
2840 	if (uio->uio_td)
2841 		uio->uio_td->td_ru.ru_msgrcv++;
2842 	SBLASTRECORDCHK(&so->so_rcv);
2843 	SBLASTMBUFCHK(&so->so_rcv);
2844 	nextrecord = m->m_nextpkt;
2845 	if (nextrecord == NULL) {
2846 		KASSERT(so->so_rcv.sb_lastrecord == m,
2847 		    ("soreceive_dgram: lastrecord != m"));
2848 	}
2849 
2850 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2851 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2852 
2853 	/*
2854 	 * Pull 'm' and its chain off the front of the packet queue.
2855 	 */
2856 	so->so_rcv.sb_mb = NULL;
2857 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2858 
2859 	/*
2860 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2861 	 */
2862 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2863 		sbfree(&so->so_rcv, m2);
2864 
2865 	/*
2866 	 * Do a few last checks before we let go of the lock.
2867 	 */
2868 	SBLASTRECORDCHK(&so->so_rcv);
2869 	SBLASTMBUFCHK(&so->so_rcv);
2870 	SOCKBUF_UNLOCK(&so->so_rcv);
2871 
2872 	if (pr->pr_flags & PR_ADDR) {
2873 		KASSERT(m->m_type == MT_SONAME,
2874 		    ("m->m_type == %d", m->m_type));
2875 		if (psa != NULL)
2876 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2877 			    M_NOWAIT);
2878 		m = m_free(m);
2879 	}
2880 	if (m == NULL) {
2881 		/* XXXRW: Can this happen? */
2882 		return (0);
2883 	}
2884 
2885 	/*
2886 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2887 	 * queue.
2888 	 *
2889 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2890 	 * in the first mbuf chain on the socket buffer.  We call into the
2891 	 * protocol to perform externalization (or freeing if controlp ==
2892 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2893 	 * MT_DATA mbufs.
2894 	 */
2895 	if (m->m_type == MT_CONTROL) {
2896 		struct mbuf *cm = NULL, *cmn;
2897 		struct mbuf **cme = &cm;
2898 
2899 		do {
2900 			m2 = m->m_next;
2901 			m->m_next = NULL;
2902 			*cme = m;
2903 			cme = &(*cme)->m_next;
2904 			m = m2;
2905 		} while (m != NULL && m->m_type == MT_CONTROL);
2906 		while (cm != NULL) {
2907 			cmn = cm->m_next;
2908 			cm->m_next = NULL;
2909 			if (pr->pr_domain->dom_externalize != NULL) {
2910 				error = (*pr->pr_domain->dom_externalize)
2911 				    (cm, controlp, flags);
2912 			} else if (controlp != NULL)
2913 				*controlp = cm;
2914 			else
2915 				m_freem(cm);
2916 			if (controlp != NULL) {
2917 				while (*controlp != NULL)
2918 					controlp = &(*controlp)->m_next;
2919 			}
2920 			cm = cmn;
2921 		}
2922 	}
2923 	KASSERT(m == NULL || m->m_type == MT_DATA,
2924 	    ("soreceive_dgram: !data"));
2925 	while (m != NULL && uio->uio_resid > 0) {
2926 		len = uio->uio_resid;
2927 		if (len > m->m_len)
2928 			len = m->m_len;
2929 		error = uiomove(mtod(m, char *), (int)len, uio);
2930 		if (error) {
2931 			m_freem(m);
2932 			return (error);
2933 		}
2934 		if (len == m->m_len)
2935 			m = m_free(m);
2936 		else {
2937 			m->m_data += len;
2938 			m->m_len -= len;
2939 		}
2940 	}
2941 	if (m != NULL) {
2942 		flags |= MSG_TRUNC;
2943 		m_freem(m);
2944 	}
2945 	if (flagsp != NULL)
2946 		*flagsp |= flags;
2947 	return (0);
2948 }
2949 
2950 int
2951 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2952     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2953 {
2954 	int error;
2955 
2956 	CURVNET_SET(so->so_vnet);
2957 	error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
2958 	CURVNET_RESTORE();
2959 	return (error);
2960 }
2961 
2962 int
2963 soshutdown(struct socket *so, enum shutdown_how how)
2964 {
2965 	int error;
2966 
2967 	CURVNET_SET(so->so_vnet);
2968 	error = so->so_proto->pr_shutdown(so, how);
2969 	CURVNET_RESTORE();
2970 
2971 	return (error);
2972 }
2973 
2974 /*
2975  * Wrapper for Socket established helper hook.
2976  * Parameters: socket, context of the hook point, hook id.
2977  */
2978 static int inline
2979 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2980 {
2981 	struct socket_hhook_data hhook_data = {
2982 		.so = so,
2983 		.hctx = hctx,
2984 		.m = NULL,
2985 		.status = 0
2986 	};
2987 
2988 	CURVNET_SET(so->so_vnet);
2989 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2990 	CURVNET_RESTORE();
2991 
2992 	/* Ugly but needed, since hhooks return void for now */
2993 	return (hhook_data.status);
2994 }
2995 
2996 /*
2997  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2998  * additional variant to handle the case where the option value needs to be
2999  * some kind of integer, but not a specific size.  In addition to their use
3000  * here, these functions are also called by the protocol-level pr_ctloutput()
3001  * routines.
3002  */
3003 int
3004 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3005 {
3006 	size_t	valsize;
3007 
3008 	/*
3009 	 * If the user gives us more than we wanted, we ignore it, but if we
3010 	 * don't get the minimum length the caller wants, we return EINVAL.
3011 	 * On success, sopt->sopt_valsize is set to however much we actually
3012 	 * retrieved.
3013 	 */
3014 	if ((valsize = sopt->sopt_valsize) < minlen)
3015 		return EINVAL;
3016 	if (valsize > len)
3017 		sopt->sopt_valsize = valsize = len;
3018 
3019 	if (sopt->sopt_td != NULL)
3020 		return (copyin(sopt->sopt_val, buf, valsize));
3021 
3022 	bcopy(sopt->sopt_val, buf, valsize);
3023 	return (0);
3024 }
3025 
3026 /*
3027  * Kernel version of setsockopt(2).
3028  *
3029  * XXX: optlen is size_t, not socklen_t
3030  */
3031 int
3032 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3033     size_t optlen)
3034 {
3035 	struct sockopt sopt;
3036 
3037 	sopt.sopt_level = level;
3038 	sopt.sopt_name = optname;
3039 	sopt.sopt_dir = SOPT_SET;
3040 	sopt.sopt_val = optval;
3041 	sopt.sopt_valsize = optlen;
3042 	sopt.sopt_td = NULL;
3043 	return (sosetopt(so, &sopt));
3044 }
3045 
3046 int
3047 sosetopt(struct socket *so, struct sockopt *sopt)
3048 {
3049 	int	error, optval;
3050 	struct	linger l;
3051 	struct	timeval tv;
3052 	sbintime_t val, *valp;
3053 	uint32_t val32;
3054 #ifdef MAC
3055 	struct mac extmac;
3056 #endif
3057 
3058 	CURVNET_SET(so->so_vnet);
3059 	error = 0;
3060 	if (sopt->sopt_level != SOL_SOCKET) {
3061 		if (so->so_proto->pr_ctloutput != NULL)
3062 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3063 		else
3064 			error = ENOPROTOOPT;
3065 	} else {
3066 		switch (sopt->sopt_name) {
3067 		case SO_ACCEPTFILTER:
3068 			error = accept_filt_setopt(so, sopt);
3069 			if (error)
3070 				goto bad;
3071 			break;
3072 
3073 		case SO_LINGER:
3074 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3075 			if (error)
3076 				goto bad;
3077 			if (l.l_linger < 0 ||
3078 			    l.l_linger > USHRT_MAX ||
3079 			    l.l_linger > (INT_MAX / hz)) {
3080 				error = EDOM;
3081 				goto bad;
3082 			}
3083 			SOCK_LOCK(so);
3084 			so->so_linger = l.l_linger;
3085 			if (l.l_onoff)
3086 				so->so_options |= SO_LINGER;
3087 			else
3088 				so->so_options &= ~SO_LINGER;
3089 			SOCK_UNLOCK(so);
3090 			break;
3091 
3092 		case SO_DEBUG:
3093 		case SO_KEEPALIVE:
3094 		case SO_DONTROUTE:
3095 		case SO_USELOOPBACK:
3096 		case SO_BROADCAST:
3097 		case SO_REUSEADDR:
3098 		case SO_REUSEPORT:
3099 		case SO_REUSEPORT_LB:
3100 		case SO_OOBINLINE:
3101 		case SO_TIMESTAMP:
3102 		case SO_BINTIME:
3103 		case SO_NOSIGPIPE:
3104 		case SO_NO_DDP:
3105 		case SO_NO_OFFLOAD:
3106 		case SO_RERROR:
3107 			error = sooptcopyin(sopt, &optval, sizeof optval,
3108 			    sizeof optval);
3109 			if (error)
3110 				goto bad;
3111 			SOCK_LOCK(so);
3112 			if (optval)
3113 				so->so_options |= sopt->sopt_name;
3114 			else
3115 				so->so_options &= ~sopt->sopt_name;
3116 			SOCK_UNLOCK(so);
3117 			break;
3118 
3119 		case SO_SETFIB:
3120 			error = sooptcopyin(sopt, &optval, sizeof optval,
3121 			    sizeof optval);
3122 			if (error)
3123 				goto bad;
3124 
3125 			if (optval < 0 || optval >= rt_numfibs) {
3126 				error = EINVAL;
3127 				goto bad;
3128 			}
3129 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3130 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3131 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3132 				so->so_fibnum = optval;
3133 			else
3134 				so->so_fibnum = 0;
3135 			break;
3136 
3137 		case SO_USER_COOKIE:
3138 			error = sooptcopyin(sopt, &val32, sizeof val32,
3139 			    sizeof val32);
3140 			if (error)
3141 				goto bad;
3142 			so->so_user_cookie = val32;
3143 			break;
3144 
3145 		case SO_SNDBUF:
3146 		case SO_RCVBUF:
3147 		case SO_SNDLOWAT:
3148 		case SO_RCVLOWAT:
3149 			error = so->so_proto->pr_setsbopt(so, sopt);
3150 			if (error)
3151 				goto bad;
3152 			break;
3153 
3154 		case SO_SNDTIMEO:
3155 		case SO_RCVTIMEO:
3156 #ifdef COMPAT_FREEBSD32
3157 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3158 				struct timeval32 tv32;
3159 
3160 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3161 				    sizeof tv32);
3162 				CP(tv32, tv, tv_sec);
3163 				CP(tv32, tv, tv_usec);
3164 			} else
3165 #endif
3166 				error = sooptcopyin(sopt, &tv, sizeof tv,
3167 				    sizeof tv);
3168 			if (error)
3169 				goto bad;
3170 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3171 			    tv.tv_usec >= 1000000) {
3172 				error = EDOM;
3173 				goto bad;
3174 			}
3175 			if (tv.tv_sec > INT32_MAX)
3176 				val = SBT_MAX;
3177 			else
3178 				val = tvtosbt(tv);
3179 			SOCK_LOCK(so);
3180 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3181 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3182 			    &so->so_snd.sb_timeo) :
3183 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3184 			    &so->so_rcv.sb_timeo);
3185 			*valp = val;
3186 			SOCK_UNLOCK(so);
3187 			break;
3188 
3189 		case SO_LABEL:
3190 #ifdef MAC
3191 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3192 			    sizeof extmac);
3193 			if (error)
3194 				goto bad;
3195 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3196 			    so, &extmac);
3197 #else
3198 			error = EOPNOTSUPP;
3199 #endif
3200 			break;
3201 
3202 		case SO_TS_CLOCK:
3203 			error = sooptcopyin(sopt, &optval, sizeof optval,
3204 			    sizeof optval);
3205 			if (error)
3206 				goto bad;
3207 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3208 				error = EINVAL;
3209 				goto bad;
3210 			}
3211 			so->so_ts_clock = optval;
3212 			break;
3213 
3214 		case SO_MAX_PACING_RATE:
3215 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3216 			    sizeof(val32));
3217 			if (error)
3218 				goto bad;
3219 			so->so_max_pacing_rate = val32;
3220 			break;
3221 
3222 		default:
3223 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3224 				error = hhook_run_socket(so, sopt,
3225 				    HHOOK_SOCKET_OPT);
3226 			else
3227 				error = ENOPROTOOPT;
3228 			break;
3229 		}
3230 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3231 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3232 	}
3233 bad:
3234 	CURVNET_RESTORE();
3235 	return (error);
3236 }
3237 
3238 /*
3239  * Helper routine for getsockopt.
3240  */
3241 int
3242 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3243 {
3244 	int	error;
3245 	size_t	valsize;
3246 
3247 	error = 0;
3248 
3249 	/*
3250 	 * Documented get behavior is that we always return a value, possibly
3251 	 * truncated to fit in the user's buffer.  Traditional behavior is
3252 	 * that we always tell the user precisely how much we copied, rather
3253 	 * than something useful like the total amount we had available for
3254 	 * her.  Note that this interface is not idempotent; the entire
3255 	 * answer must be generated ahead of time.
3256 	 */
3257 	valsize = min(len, sopt->sopt_valsize);
3258 	sopt->sopt_valsize = valsize;
3259 	if (sopt->sopt_val != NULL) {
3260 		if (sopt->sopt_td != NULL)
3261 			error = copyout(buf, sopt->sopt_val, valsize);
3262 		else
3263 			bcopy(buf, sopt->sopt_val, valsize);
3264 	}
3265 	return (error);
3266 }
3267 
3268 int
3269 sogetopt(struct socket *so, struct sockopt *sopt)
3270 {
3271 	int	error, optval;
3272 	struct	linger l;
3273 	struct	timeval tv;
3274 #ifdef MAC
3275 	struct mac extmac;
3276 #endif
3277 
3278 	CURVNET_SET(so->so_vnet);
3279 	error = 0;
3280 	if (sopt->sopt_level != SOL_SOCKET) {
3281 		if (so->so_proto->pr_ctloutput != NULL)
3282 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3283 		else
3284 			error = ENOPROTOOPT;
3285 		CURVNET_RESTORE();
3286 		return (error);
3287 	} else {
3288 		switch (sopt->sopt_name) {
3289 		case SO_ACCEPTFILTER:
3290 			error = accept_filt_getopt(so, sopt);
3291 			break;
3292 
3293 		case SO_LINGER:
3294 			SOCK_LOCK(so);
3295 			l.l_onoff = so->so_options & SO_LINGER;
3296 			l.l_linger = so->so_linger;
3297 			SOCK_UNLOCK(so);
3298 			error = sooptcopyout(sopt, &l, sizeof l);
3299 			break;
3300 
3301 		case SO_USELOOPBACK:
3302 		case SO_DONTROUTE:
3303 		case SO_DEBUG:
3304 		case SO_KEEPALIVE:
3305 		case SO_REUSEADDR:
3306 		case SO_REUSEPORT:
3307 		case SO_REUSEPORT_LB:
3308 		case SO_BROADCAST:
3309 		case SO_OOBINLINE:
3310 		case SO_ACCEPTCONN:
3311 		case SO_TIMESTAMP:
3312 		case SO_BINTIME:
3313 		case SO_NOSIGPIPE:
3314 		case SO_NO_DDP:
3315 		case SO_NO_OFFLOAD:
3316 		case SO_RERROR:
3317 			optval = so->so_options & sopt->sopt_name;
3318 integer:
3319 			error = sooptcopyout(sopt, &optval, sizeof optval);
3320 			break;
3321 
3322 		case SO_DOMAIN:
3323 			optval = so->so_proto->pr_domain->dom_family;
3324 			goto integer;
3325 
3326 		case SO_TYPE:
3327 			optval = so->so_type;
3328 			goto integer;
3329 
3330 		case SO_PROTOCOL:
3331 			optval = so->so_proto->pr_protocol;
3332 			goto integer;
3333 
3334 		case SO_ERROR:
3335 			SOCK_LOCK(so);
3336 			if (so->so_error) {
3337 				optval = so->so_error;
3338 				so->so_error = 0;
3339 			} else {
3340 				optval = so->so_rerror;
3341 				so->so_rerror = 0;
3342 			}
3343 			SOCK_UNLOCK(so);
3344 			goto integer;
3345 
3346 		case SO_SNDBUF:
3347 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3348 			    so->so_snd.sb_hiwat;
3349 			goto integer;
3350 
3351 		case SO_RCVBUF:
3352 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3353 			    so->so_rcv.sb_hiwat;
3354 			goto integer;
3355 
3356 		case SO_SNDLOWAT:
3357 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3358 			    so->so_snd.sb_lowat;
3359 			goto integer;
3360 
3361 		case SO_RCVLOWAT:
3362 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3363 			    so->so_rcv.sb_lowat;
3364 			goto integer;
3365 
3366 		case SO_SNDTIMEO:
3367 		case SO_RCVTIMEO:
3368 			SOCK_LOCK(so);
3369 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3370 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3371 			    so->so_snd.sb_timeo) :
3372 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3373 			    so->so_rcv.sb_timeo));
3374 			SOCK_UNLOCK(so);
3375 #ifdef COMPAT_FREEBSD32
3376 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3377 				struct timeval32 tv32;
3378 
3379 				CP(tv, tv32, tv_sec);
3380 				CP(tv, tv32, tv_usec);
3381 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3382 			} else
3383 #endif
3384 				error = sooptcopyout(sopt, &tv, sizeof tv);
3385 			break;
3386 
3387 		case SO_LABEL:
3388 #ifdef MAC
3389 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3390 			    sizeof(extmac));
3391 			if (error)
3392 				goto bad;
3393 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3394 			    so, &extmac);
3395 			if (error)
3396 				goto bad;
3397 			/* Don't copy out extmac, it is unchanged. */
3398 #else
3399 			error = EOPNOTSUPP;
3400 #endif
3401 			break;
3402 
3403 		case SO_PEERLABEL:
3404 #ifdef MAC
3405 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3406 			    sizeof(extmac));
3407 			if (error)
3408 				goto bad;
3409 			error = mac_getsockopt_peerlabel(
3410 			    sopt->sopt_td->td_ucred, so, &extmac);
3411 			if (error)
3412 				goto bad;
3413 			/* Don't copy out extmac, it is unchanged. */
3414 #else
3415 			error = EOPNOTSUPP;
3416 #endif
3417 			break;
3418 
3419 		case SO_LISTENQLIMIT:
3420 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3421 			goto integer;
3422 
3423 		case SO_LISTENQLEN:
3424 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3425 			goto integer;
3426 
3427 		case SO_LISTENINCQLEN:
3428 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3429 			goto integer;
3430 
3431 		case SO_TS_CLOCK:
3432 			optval = so->so_ts_clock;
3433 			goto integer;
3434 
3435 		case SO_MAX_PACING_RATE:
3436 			optval = so->so_max_pacing_rate;
3437 			goto integer;
3438 
3439 		default:
3440 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3441 				error = hhook_run_socket(so, sopt,
3442 				    HHOOK_SOCKET_OPT);
3443 			else
3444 				error = ENOPROTOOPT;
3445 			break;
3446 		}
3447 	}
3448 #ifdef MAC
3449 bad:
3450 #endif
3451 	CURVNET_RESTORE();
3452 	return (error);
3453 }
3454 
3455 int
3456 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3457 {
3458 	struct mbuf *m, *m_prev;
3459 	int sopt_size = sopt->sopt_valsize;
3460 
3461 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3462 	if (m == NULL)
3463 		return ENOBUFS;
3464 	if (sopt_size > MLEN) {
3465 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3466 		if ((m->m_flags & M_EXT) == 0) {
3467 			m_free(m);
3468 			return ENOBUFS;
3469 		}
3470 		m->m_len = min(MCLBYTES, sopt_size);
3471 	} else {
3472 		m->m_len = min(MLEN, sopt_size);
3473 	}
3474 	sopt_size -= m->m_len;
3475 	*mp = m;
3476 	m_prev = m;
3477 
3478 	while (sopt_size) {
3479 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3480 		if (m == NULL) {
3481 			m_freem(*mp);
3482 			return ENOBUFS;
3483 		}
3484 		if (sopt_size > MLEN) {
3485 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3486 			    M_NOWAIT);
3487 			if ((m->m_flags & M_EXT) == 0) {
3488 				m_freem(m);
3489 				m_freem(*mp);
3490 				return ENOBUFS;
3491 			}
3492 			m->m_len = min(MCLBYTES, sopt_size);
3493 		} else {
3494 			m->m_len = min(MLEN, sopt_size);
3495 		}
3496 		sopt_size -= m->m_len;
3497 		m_prev->m_next = m;
3498 		m_prev = m;
3499 	}
3500 	return (0);
3501 }
3502 
3503 int
3504 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3505 {
3506 	struct mbuf *m0 = m;
3507 
3508 	if (sopt->sopt_val == NULL)
3509 		return (0);
3510 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3511 		if (sopt->sopt_td != NULL) {
3512 			int error;
3513 
3514 			error = copyin(sopt->sopt_val, mtod(m, char *),
3515 			    m->m_len);
3516 			if (error != 0) {
3517 				m_freem(m0);
3518 				return(error);
3519 			}
3520 		} else
3521 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3522 		sopt->sopt_valsize -= m->m_len;
3523 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3524 		m = m->m_next;
3525 	}
3526 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3527 		panic("ip6_sooptmcopyin");
3528 	return (0);
3529 }
3530 
3531 int
3532 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3533 {
3534 	struct mbuf *m0 = m;
3535 	size_t valsize = 0;
3536 
3537 	if (sopt->sopt_val == NULL)
3538 		return (0);
3539 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3540 		if (sopt->sopt_td != NULL) {
3541 			int error;
3542 
3543 			error = copyout(mtod(m, char *), sopt->sopt_val,
3544 			    m->m_len);
3545 			if (error != 0) {
3546 				m_freem(m0);
3547 				return(error);
3548 			}
3549 		} else
3550 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3551 		sopt->sopt_valsize -= m->m_len;
3552 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3553 		valsize += m->m_len;
3554 		m = m->m_next;
3555 	}
3556 	if (m != NULL) {
3557 		/* enough soopt buffer should be given from user-land */
3558 		m_freem(m0);
3559 		return(EINVAL);
3560 	}
3561 	sopt->sopt_valsize = valsize;
3562 	return (0);
3563 }
3564 
3565 /*
3566  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3567  * out-of-band data, which will then notify socket consumers.
3568  */
3569 void
3570 sohasoutofband(struct socket *so)
3571 {
3572 
3573 	if (so->so_sigio != NULL)
3574 		pgsigio(&so->so_sigio, SIGURG, 0);
3575 	selwakeuppri(&so->so_rdsel, PSOCK);
3576 }
3577 
3578 int
3579 sopoll(struct socket *so, int events, struct ucred *active_cred,
3580     struct thread *td)
3581 {
3582 
3583 	/*
3584 	 * We do not need to set or assert curvnet as long as everyone uses
3585 	 * sopoll_generic().
3586 	 */
3587 	return (so->so_proto->pr_sopoll(so, events, active_cred, td));
3588 }
3589 
3590 int
3591 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3592     struct thread *td)
3593 {
3594 	int revents;
3595 
3596 	SOCK_LOCK(so);
3597 	if (SOLISTENING(so)) {
3598 		if (!(events & (POLLIN | POLLRDNORM)))
3599 			revents = 0;
3600 		else if (!TAILQ_EMPTY(&so->sol_comp))
3601 			revents = events & (POLLIN | POLLRDNORM);
3602 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3603 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3604 		else {
3605 			selrecord(td, &so->so_rdsel);
3606 			revents = 0;
3607 		}
3608 	} else {
3609 		revents = 0;
3610 		SOCK_SENDBUF_LOCK(so);
3611 		SOCK_RECVBUF_LOCK(so);
3612 		if (events & (POLLIN | POLLRDNORM))
3613 			if (soreadabledata(so))
3614 				revents |= events & (POLLIN | POLLRDNORM);
3615 		if (events & (POLLOUT | POLLWRNORM))
3616 			if (sowriteable(so))
3617 				revents |= events & (POLLOUT | POLLWRNORM);
3618 		if (events & (POLLPRI | POLLRDBAND))
3619 			if (so->so_oobmark ||
3620 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3621 				revents |= events & (POLLPRI | POLLRDBAND);
3622 		if ((events & POLLINIGNEOF) == 0) {
3623 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3624 				revents |= events & (POLLIN | POLLRDNORM);
3625 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3626 					revents |= POLLHUP;
3627 			}
3628 		}
3629 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3630 			revents |= events & POLLRDHUP;
3631 		if (revents == 0) {
3632 			if (events &
3633 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3634 				selrecord(td, &so->so_rdsel);
3635 				so->so_rcv.sb_flags |= SB_SEL;
3636 			}
3637 			if (events & (POLLOUT | POLLWRNORM)) {
3638 				selrecord(td, &so->so_wrsel);
3639 				so->so_snd.sb_flags |= SB_SEL;
3640 			}
3641 		}
3642 		SOCK_RECVBUF_UNLOCK(so);
3643 		SOCK_SENDBUF_UNLOCK(so);
3644 	}
3645 	SOCK_UNLOCK(so);
3646 	return (revents);
3647 }
3648 
3649 int
3650 soo_kqfilter(struct file *fp, struct knote *kn)
3651 {
3652 	struct socket *so = kn->kn_fp->f_data;
3653 	struct sockbuf *sb;
3654 	sb_which which;
3655 	struct knlist *knl;
3656 
3657 	switch (kn->kn_filter) {
3658 	case EVFILT_READ:
3659 		kn->kn_fop = &soread_filtops;
3660 		knl = &so->so_rdsel.si_note;
3661 		sb = &so->so_rcv;
3662 		which = SO_RCV;
3663 		break;
3664 	case EVFILT_WRITE:
3665 		kn->kn_fop = &sowrite_filtops;
3666 		knl = &so->so_wrsel.si_note;
3667 		sb = &so->so_snd;
3668 		which = SO_SND;
3669 		break;
3670 	case EVFILT_EMPTY:
3671 		kn->kn_fop = &soempty_filtops;
3672 		knl = &so->so_wrsel.si_note;
3673 		sb = &so->so_snd;
3674 		which = SO_SND;
3675 		break;
3676 	default:
3677 		return (EINVAL);
3678 	}
3679 
3680 	SOCK_LOCK(so);
3681 	if (SOLISTENING(so)) {
3682 		knlist_add(knl, kn, 1);
3683 	} else {
3684 		SOCK_BUF_LOCK(so, which);
3685 		knlist_add(knl, kn, 1);
3686 		sb->sb_flags |= SB_KNOTE;
3687 		SOCK_BUF_UNLOCK(so, which);
3688 	}
3689 	SOCK_UNLOCK(so);
3690 	return (0);
3691 }
3692 
3693 static void
3694 filt_sordetach(struct knote *kn)
3695 {
3696 	struct socket *so = kn->kn_fp->f_data;
3697 
3698 	so_rdknl_lock(so);
3699 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3700 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3701 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3702 	so_rdknl_unlock(so);
3703 }
3704 
3705 /*ARGSUSED*/
3706 static int
3707 filt_soread(struct knote *kn, long hint)
3708 {
3709 	struct socket *so;
3710 
3711 	so = kn->kn_fp->f_data;
3712 
3713 	if (SOLISTENING(so)) {
3714 		SOCK_LOCK_ASSERT(so);
3715 		kn->kn_data = so->sol_qlen;
3716 		if (so->so_error) {
3717 			kn->kn_flags |= EV_EOF;
3718 			kn->kn_fflags = so->so_error;
3719 			return (1);
3720 		}
3721 		return (!TAILQ_EMPTY(&so->sol_comp));
3722 	}
3723 
3724 	SOCK_RECVBUF_LOCK_ASSERT(so);
3725 
3726 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3727 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3728 		kn->kn_flags |= EV_EOF;
3729 		kn->kn_fflags = so->so_error;
3730 		return (1);
3731 	} else if (so->so_error || so->so_rerror)
3732 		return (1);
3733 
3734 	if (kn->kn_sfflags & NOTE_LOWAT) {
3735 		if (kn->kn_data >= kn->kn_sdata)
3736 			return (1);
3737 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3738 		return (1);
3739 
3740 	/* This hook returning non-zero indicates an event, not error */
3741 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3742 }
3743 
3744 static void
3745 filt_sowdetach(struct knote *kn)
3746 {
3747 	struct socket *so = kn->kn_fp->f_data;
3748 
3749 	so_wrknl_lock(so);
3750 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3751 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3752 		so->so_snd.sb_flags &= ~SB_KNOTE;
3753 	so_wrknl_unlock(so);
3754 }
3755 
3756 /*ARGSUSED*/
3757 static int
3758 filt_sowrite(struct knote *kn, long hint)
3759 {
3760 	struct socket *so;
3761 
3762 	so = kn->kn_fp->f_data;
3763 
3764 	if (SOLISTENING(so))
3765 		return (0);
3766 
3767 	SOCK_SENDBUF_LOCK_ASSERT(so);
3768 	kn->kn_data = sbspace(&so->so_snd);
3769 
3770 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3771 
3772 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3773 		kn->kn_flags |= EV_EOF;
3774 		kn->kn_fflags = so->so_error;
3775 		return (1);
3776 	} else if (so->so_error)	/* temporary udp error */
3777 		return (1);
3778 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3779 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3780 		return (0);
3781 	else if (kn->kn_sfflags & NOTE_LOWAT)
3782 		return (kn->kn_data >= kn->kn_sdata);
3783 	else
3784 		return (kn->kn_data >= so->so_snd.sb_lowat);
3785 }
3786 
3787 static int
3788 filt_soempty(struct knote *kn, long hint)
3789 {
3790 	struct socket *so;
3791 
3792 	so = kn->kn_fp->f_data;
3793 
3794 	if (SOLISTENING(so))
3795 		return (1);
3796 
3797 	SOCK_SENDBUF_LOCK_ASSERT(so);
3798 	kn->kn_data = sbused(&so->so_snd);
3799 
3800 	if (kn->kn_data == 0)
3801 		return (1);
3802 	else
3803 		return (0);
3804 }
3805 
3806 int
3807 socheckuid(struct socket *so, uid_t uid)
3808 {
3809 
3810 	if (so == NULL)
3811 		return (EPERM);
3812 	if (so->so_cred->cr_uid != uid)
3813 		return (EPERM);
3814 	return (0);
3815 }
3816 
3817 /*
3818  * These functions are used by protocols to notify the socket layer (and its
3819  * consumers) of state changes in the sockets driven by protocol-side events.
3820  */
3821 
3822 /*
3823  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3824  *
3825  * Normal sequence from the active (originating) side is that
3826  * soisconnecting() is called during processing of connect() call, resulting
3827  * in an eventual call to soisconnected() if/when the connection is
3828  * established.  When the connection is torn down soisdisconnecting() is
3829  * called during processing of disconnect() call, and soisdisconnected() is
3830  * called when the connection to the peer is totally severed.  The semantics
3831  * of these routines are such that connectionless protocols can call
3832  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3833  * calls when setting up a ``connection'' takes no time.
3834  *
3835  * From the passive side, a socket is created with two queues of sockets:
3836  * so_incomp for connections in progress and so_comp for connections already
3837  * made and awaiting user acceptance.  As a protocol is preparing incoming
3838  * connections, it creates a socket structure queued on so_incomp by calling
3839  * sonewconn().  When the connection is established, soisconnected() is
3840  * called, and transfers the socket structure to so_comp, making it available
3841  * to accept().
3842  *
3843  * If a socket is closed with sockets on either so_incomp or so_comp, these
3844  * sockets are dropped.
3845  *
3846  * If higher-level protocols are implemented in the kernel, the wakeups done
3847  * here will sometimes cause software-interrupt process scheduling.
3848  */
3849 void
3850 soisconnecting(struct socket *so)
3851 {
3852 
3853 	SOCK_LOCK(so);
3854 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3855 	so->so_state |= SS_ISCONNECTING;
3856 	SOCK_UNLOCK(so);
3857 }
3858 
3859 void
3860 soisconnected(struct socket *so)
3861 {
3862 	bool last __diagused;
3863 
3864 	SOCK_LOCK(so);
3865 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3866 	so->so_state |= SS_ISCONNECTED;
3867 
3868 	if (so->so_qstate == SQ_INCOMP) {
3869 		struct socket *head = so->so_listen;
3870 		int ret;
3871 
3872 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3873 		/*
3874 		 * Promoting a socket from incomplete queue to complete, we
3875 		 * need to go through reverse order of locking.  We first do
3876 		 * trylock, and if that doesn't succeed, we go the hard way
3877 		 * leaving a reference and rechecking consistency after proper
3878 		 * locking.
3879 		 */
3880 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3881 			soref(head);
3882 			SOCK_UNLOCK(so);
3883 			SOLISTEN_LOCK(head);
3884 			SOCK_LOCK(so);
3885 			if (__predict_false(head != so->so_listen)) {
3886 				/*
3887 				 * The socket went off the listen queue,
3888 				 * should be lost race to close(2) of sol.
3889 				 * The socket is about to soabort().
3890 				 */
3891 				SOCK_UNLOCK(so);
3892 				sorele_locked(head);
3893 				return;
3894 			}
3895 			last = refcount_release(&head->so_count);
3896 			KASSERT(!last, ("%s: released last reference for %p",
3897 			    __func__, head));
3898 		}
3899 again:
3900 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3901 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3902 			head->sol_incqlen--;
3903 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3904 			head->sol_qlen++;
3905 			so->so_qstate = SQ_COMP;
3906 			SOCK_UNLOCK(so);
3907 			solisten_wakeup(head);	/* unlocks */
3908 		} else {
3909 			SOCK_RECVBUF_LOCK(so);
3910 			soupcall_set(so, SO_RCV,
3911 			    head->sol_accept_filter->accf_callback,
3912 			    head->sol_accept_filter_arg);
3913 			so->so_options &= ~SO_ACCEPTFILTER;
3914 			ret = head->sol_accept_filter->accf_callback(so,
3915 			    head->sol_accept_filter_arg, M_NOWAIT);
3916 			if (ret == SU_ISCONNECTED) {
3917 				soupcall_clear(so, SO_RCV);
3918 				SOCK_RECVBUF_UNLOCK(so);
3919 				goto again;
3920 			}
3921 			SOCK_RECVBUF_UNLOCK(so);
3922 			SOCK_UNLOCK(so);
3923 			SOLISTEN_UNLOCK(head);
3924 		}
3925 		return;
3926 	}
3927 	SOCK_UNLOCK(so);
3928 	wakeup(&so->so_timeo);
3929 	sorwakeup(so);
3930 	sowwakeup(so);
3931 }
3932 
3933 void
3934 soisdisconnecting(struct socket *so)
3935 {
3936 
3937 	SOCK_LOCK(so);
3938 	so->so_state &= ~SS_ISCONNECTING;
3939 	so->so_state |= SS_ISDISCONNECTING;
3940 
3941 	if (!SOLISTENING(so)) {
3942 		SOCK_RECVBUF_LOCK(so);
3943 		socantrcvmore_locked(so);
3944 		SOCK_SENDBUF_LOCK(so);
3945 		socantsendmore_locked(so);
3946 	}
3947 	SOCK_UNLOCK(so);
3948 	wakeup(&so->so_timeo);
3949 }
3950 
3951 void
3952 soisdisconnected(struct socket *so)
3953 {
3954 
3955 	SOCK_LOCK(so);
3956 
3957 	/*
3958 	 * There is at least one reader of so_state that does not
3959 	 * acquire socket lock, namely soreceive_generic().  Ensure
3960 	 * that it never sees all flags that track connection status
3961 	 * cleared, by ordering the update with a barrier semantic of
3962 	 * our release thread fence.
3963 	 */
3964 	so->so_state |= SS_ISDISCONNECTED;
3965 	atomic_thread_fence_rel();
3966 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3967 
3968 	if (!SOLISTENING(so)) {
3969 		SOCK_UNLOCK(so);
3970 		SOCK_RECVBUF_LOCK(so);
3971 		socantrcvmore_locked(so);
3972 		SOCK_SENDBUF_LOCK(so);
3973 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3974 		socantsendmore_locked(so);
3975 	} else
3976 		SOCK_UNLOCK(so);
3977 	wakeup(&so->so_timeo);
3978 }
3979 
3980 int
3981 soiolock(struct socket *so, struct sx *sx, int flags)
3982 {
3983 	int error;
3984 
3985 	KASSERT((flags & SBL_VALID) == flags,
3986 	    ("soiolock: invalid flags %#x", flags));
3987 
3988 	if ((flags & SBL_WAIT) != 0) {
3989 		if ((flags & SBL_NOINTR) != 0) {
3990 			sx_xlock(sx);
3991 		} else {
3992 			error = sx_xlock_sig(sx);
3993 			if (error != 0)
3994 				return (error);
3995 		}
3996 	} else if (!sx_try_xlock(sx)) {
3997 		return (EWOULDBLOCK);
3998 	}
3999 
4000 	if (__predict_false(SOLISTENING(so))) {
4001 		sx_xunlock(sx);
4002 		return (ENOTCONN);
4003 	}
4004 	return (0);
4005 }
4006 
4007 void
4008 soiounlock(struct sx *sx)
4009 {
4010 	sx_xunlock(sx);
4011 }
4012 
4013 /*
4014  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4015  */
4016 struct sockaddr *
4017 sodupsockaddr(const struct sockaddr *sa, int mflags)
4018 {
4019 	struct sockaddr *sa2;
4020 
4021 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4022 	if (sa2)
4023 		bcopy(sa, sa2, sa->sa_len);
4024 	return sa2;
4025 }
4026 
4027 /*
4028  * Register per-socket destructor.
4029  */
4030 void
4031 sodtor_set(struct socket *so, so_dtor_t *func)
4032 {
4033 
4034 	SOCK_LOCK_ASSERT(so);
4035 	so->so_dtor = func;
4036 }
4037 
4038 /*
4039  * Register per-socket buffer upcalls.
4040  */
4041 void
4042 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4043 {
4044 	struct sockbuf *sb;
4045 
4046 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4047 
4048 	switch (which) {
4049 	case SO_RCV:
4050 		sb = &so->so_rcv;
4051 		break;
4052 	case SO_SND:
4053 		sb = &so->so_snd;
4054 		break;
4055 	}
4056 	SOCK_BUF_LOCK_ASSERT(so, which);
4057 	sb->sb_upcall = func;
4058 	sb->sb_upcallarg = arg;
4059 	sb->sb_flags |= SB_UPCALL;
4060 }
4061 
4062 void
4063 soupcall_clear(struct socket *so, sb_which which)
4064 {
4065 	struct sockbuf *sb;
4066 
4067 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4068 
4069 	switch (which) {
4070 	case SO_RCV:
4071 		sb = &so->so_rcv;
4072 		break;
4073 	case SO_SND:
4074 		sb = &so->so_snd;
4075 		break;
4076 	}
4077 	SOCK_BUF_LOCK_ASSERT(so, which);
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 retry:
4100 	if (SOLISTENING(so)) {
4101 		SOLISTEN_LOCK(so);
4102 	} else {
4103 		SOCK_RECVBUF_LOCK(so);
4104 		if (__predict_false(SOLISTENING(so))) {
4105 			SOCK_RECVBUF_UNLOCK(so);
4106 			goto retry;
4107 		}
4108 	}
4109 }
4110 
4111 static void
4112 so_rdknl_unlock(void *arg)
4113 {
4114 	struct socket *so = arg;
4115 
4116 	if (SOLISTENING(so))
4117 		SOLISTEN_UNLOCK(so);
4118 	else
4119 		SOCK_RECVBUF_UNLOCK(so);
4120 }
4121 
4122 static void
4123 so_rdknl_assert_lock(void *arg, int what)
4124 {
4125 	struct socket *so = arg;
4126 
4127 	if (what == LA_LOCKED) {
4128 		if (SOLISTENING(so))
4129 			SOLISTEN_LOCK_ASSERT(so);
4130 		else
4131 			SOCK_RECVBUF_LOCK_ASSERT(so);
4132 	} else {
4133 		if (SOLISTENING(so))
4134 			SOLISTEN_UNLOCK_ASSERT(so);
4135 		else
4136 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4137 	}
4138 }
4139 
4140 static void
4141 so_wrknl_lock(void *arg)
4142 {
4143 	struct socket *so = arg;
4144 
4145 retry:
4146 	if (SOLISTENING(so)) {
4147 		SOLISTEN_LOCK(so);
4148 	} else {
4149 		SOCK_SENDBUF_LOCK(so);
4150 		if (__predict_false(SOLISTENING(so))) {
4151 			SOCK_SENDBUF_UNLOCK(so);
4152 			goto retry;
4153 		}
4154 	}
4155 }
4156 
4157 static void
4158 so_wrknl_unlock(void *arg)
4159 {
4160 	struct socket *so = arg;
4161 
4162 	if (SOLISTENING(so))
4163 		SOLISTEN_UNLOCK(so);
4164 	else
4165 		SOCK_SENDBUF_UNLOCK(so);
4166 }
4167 
4168 static void
4169 so_wrknl_assert_lock(void *arg, int what)
4170 {
4171 	struct socket *so = arg;
4172 
4173 	if (what == LA_LOCKED) {
4174 		if (SOLISTENING(so))
4175 			SOLISTEN_LOCK_ASSERT(so);
4176 		else
4177 			SOCK_SENDBUF_LOCK_ASSERT(so);
4178 	} else {
4179 		if (SOLISTENING(so))
4180 			SOLISTEN_UNLOCK_ASSERT(so);
4181 		else
4182 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
4183 	}
4184 }
4185 
4186 /*
4187  * Create an external-format (``xsocket'') structure using the information in
4188  * the kernel-format socket structure pointed to by so.  This is done to
4189  * reduce the spew of irrelevant information over this interface, to isolate
4190  * user code from changes in the kernel structure, and potentially to provide
4191  * information-hiding if we decide that some of this information should be
4192  * hidden from users.
4193  */
4194 void
4195 sotoxsocket(struct socket *so, struct xsocket *xso)
4196 {
4197 
4198 	bzero(xso, sizeof(*xso));
4199 	xso->xso_len = sizeof *xso;
4200 	xso->xso_so = (uintptr_t)so;
4201 	xso->so_type = so->so_type;
4202 	xso->so_options = so->so_options;
4203 	xso->so_linger = so->so_linger;
4204 	xso->so_state = so->so_state;
4205 	xso->so_pcb = (uintptr_t)so->so_pcb;
4206 	xso->xso_protocol = so->so_proto->pr_protocol;
4207 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4208 	xso->so_timeo = so->so_timeo;
4209 	xso->so_error = so->so_error;
4210 	xso->so_uid = so->so_cred->cr_uid;
4211 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4212 	if (SOLISTENING(so)) {
4213 		xso->so_qlen = so->sol_qlen;
4214 		xso->so_incqlen = so->sol_incqlen;
4215 		xso->so_qlimit = so->sol_qlimit;
4216 		xso->so_oobmark = 0;
4217 	} else {
4218 		xso->so_state |= so->so_qstate;
4219 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4220 		xso->so_oobmark = so->so_oobmark;
4221 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4222 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4223 	}
4224 }
4225 
4226 struct sockbuf *
4227 so_sockbuf_rcv(struct socket *so)
4228 {
4229 
4230 	return (&so->so_rcv);
4231 }
4232 
4233 struct sockbuf *
4234 so_sockbuf_snd(struct socket *so)
4235 {
4236 
4237 	return (&so->so_snd);
4238 }
4239 
4240 int
4241 so_state_get(const struct socket *so)
4242 {
4243 
4244 	return (so->so_state);
4245 }
4246 
4247 void
4248 so_state_set(struct socket *so, int val)
4249 {
4250 
4251 	so->so_state = val;
4252 }
4253 
4254 int
4255 so_options_get(const struct socket *so)
4256 {
4257 
4258 	return (so->so_options);
4259 }
4260 
4261 void
4262 so_options_set(struct socket *so, int val)
4263 {
4264 
4265 	so->so_options = val;
4266 }
4267 
4268 int
4269 so_error_get(const struct socket *so)
4270 {
4271 
4272 	return (so->so_error);
4273 }
4274 
4275 void
4276 so_error_set(struct socket *so, int val)
4277 {
4278 
4279 	so->so_error = val;
4280 }
4281 
4282 int
4283 so_linger_get(const struct socket *so)
4284 {
4285 
4286 	return (so->so_linger);
4287 }
4288 
4289 void
4290 so_linger_set(struct socket *so, int val)
4291 {
4292 
4293 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4294 	    ("%s: val %d out of range", __func__, val));
4295 
4296 	so->so_linger = val;
4297 }
4298 
4299 struct protosw *
4300 so_protosw_get(const struct socket *so)
4301 {
4302 
4303 	return (so->so_proto);
4304 }
4305 
4306 void
4307 so_protosw_set(struct socket *so, struct protosw *val)
4308 {
4309 
4310 	so->so_proto = val;
4311 }
4312 
4313 void
4314 so_sorwakeup(struct socket *so)
4315 {
4316 
4317 	sorwakeup(so);
4318 }
4319 
4320 void
4321 so_sowwakeup(struct socket *so)
4322 {
4323 
4324 	sowwakeup(so);
4325 }
4326 
4327 void
4328 so_sorwakeup_locked(struct socket *so)
4329 {
4330 
4331 	sorwakeup_locked(so);
4332 }
4333 
4334 void
4335 so_sowwakeup_locked(struct socket *so)
4336 {
4337 
4338 	sowwakeup_locked(so);
4339 }
4340 
4341 void
4342 so_lock(struct socket *so)
4343 {
4344 
4345 	SOCK_LOCK(so);
4346 }
4347 
4348 void
4349 so_unlock(struct socket *so)
4350 {
4351 
4352 	SOCK_UNLOCK(so);
4353 }
4354