xref: /freebsd/sys/kern/uipc_socket.c (revision aa1a8ff2d6dbc51ef058f46f3db5a8bb77967145)
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(), sorele(), sonewconn() and sorflush(), which are usually called
99  * from a pre-set VNET context.  sopoll() currently does not need a VNET
100  * 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_ktrace.h"
108 #include "opt_sctp.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/capsicum.h>
113 #include <sys/fcntl.h>
114 #include <sys/limits.h>
115 #include <sys/lock.h>
116 #include <sys/mac.h>
117 #include <sys/malloc.h>
118 #include <sys/mbuf.h>
119 #include <sys/mutex.h>
120 #include <sys/domain.h>
121 #include <sys/file.h>			/* for struct knote */
122 #include <sys/hhook.h>
123 #include <sys/kernel.h>
124 #include <sys/khelp.h>
125 #include <sys/ktls.h>
126 #include <sys/event.h>
127 #include <sys/eventhandler.h>
128 #include <sys/poll.h>
129 #include <sys/proc.h>
130 #include <sys/protosw.h>
131 #include <sys/sbuf.h>
132 #include <sys/socket.h>
133 #include <sys/socketvar.h>
134 #include <sys/resourcevar.h>
135 #include <net/route.h>
136 #include <sys/signalvar.h>
137 #include <sys/stat.h>
138 #include <sys/sx.h>
139 #include <sys/sysctl.h>
140 #include <sys/taskqueue.h>
141 #include <sys/uio.h>
142 #include <sys/un.h>
143 #include <sys/unpcb.h>
144 #include <sys/jail.h>
145 #include <sys/syslog.h>
146 #include <netinet/in.h>
147 #include <netinet/in_pcb.h>
148 #include <netinet/tcp.h>
149 
150 #include <net/vnet.h>
151 
152 #include <security/mac/mac_framework.h>
153 
154 #include <vm/uma.h>
155 
156 #ifdef COMPAT_FREEBSD32
157 #include <sys/mount.h>
158 #include <sys/sysent.h>
159 #include <compat/freebsd32/freebsd32.h>
160 #endif
161 
162 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
163 		    int flags);
164 static void	so_rdknl_lock(void *);
165 static void	so_rdknl_unlock(void *);
166 static void	so_rdknl_assert_lock(void *, int);
167 static void	so_wrknl_lock(void *);
168 static void	so_wrknl_unlock(void *);
169 static void	so_wrknl_assert_lock(void *, int);
170 
171 static void	filt_sordetach(struct knote *kn);
172 static int	filt_soread(struct knote *kn, long hint);
173 static void	filt_sowdetach(struct knote *kn);
174 static int	filt_sowrite(struct knote *kn, long hint);
175 static int	filt_soempty(struct knote *kn, long hint);
176 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
177 fo_kqfilter_t	soo_kqfilter;
178 
179 static struct filterops soread_filtops = {
180 	.f_isfd = 1,
181 	.f_detach = filt_sordetach,
182 	.f_event = filt_soread,
183 };
184 static struct filterops sowrite_filtops = {
185 	.f_isfd = 1,
186 	.f_detach = filt_sowdetach,
187 	.f_event = filt_sowrite,
188 };
189 static struct filterops soempty_filtops = {
190 	.f_isfd = 1,
191 	.f_detach = filt_sowdetach,
192 	.f_event = filt_soempty,
193 };
194 
195 so_gen_t	so_gencnt;	/* generation count for sockets */
196 
197 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
198 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
199 
200 #define	VNET_SO_ASSERT(so)						\
201 	VNET_ASSERT(curvnet != NULL,					\
202 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
203 
204 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
205 #define	V_socket_hhh		VNET(socket_hhh)
206 
207 /*
208  * Limit on the number of connections in the listen queue waiting
209  * for accept(2).
210  * NB: The original sysctl somaxconn is still available but hidden
211  * to prevent confusion about the actual purpose of this number.
212  */
213 static u_int somaxconn = SOMAXCONN;
214 
215 static int
216 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
217 {
218 	int error;
219 	int val;
220 
221 	val = somaxconn;
222 	error = sysctl_handle_int(oidp, &val, 0, req);
223 	if (error || !req->newptr )
224 		return (error);
225 
226 	/*
227 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
228 	 *   3 * so_qlimit / 2
229 	 * below, will not overflow.
230          */
231 
232 	if (val < 1 || val > UINT_MAX / 3)
233 		return (EINVAL);
234 
235 	somaxconn = val;
236 	return (0);
237 }
238 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
239     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, 0, sizeof(int),
240     sysctl_somaxconn, "I",
241     "Maximum listen socket pending connection accept queue size");
242 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
243     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_MPSAFE, 0,
244     sizeof(int), sysctl_somaxconn, "I",
245     "Maximum listen socket pending connection accept queue size (compat)");
246 
247 static int numopensockets;
248 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
249     &numopensockets, 0, "Number of open sockets");
250 
251 /*
252  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
253  * so_gencnt field.
254  */
255 static struct mtx so_global_mtx;
256 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
257 
258 /*
259  * General IPC sysctl name space, used by sockets and a variety of other IPC
260  * types.
261  */
262 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
263     "IPC");
264 
265 /*
266  * Initialize the socket subsystem and set up the socket
267  * memory allocator.
268  */
269 static uma_zone_t socket_zone;
270 int	maxsockets;
271 
272 static void
273 socket_zone_change(void *tag)
274 {
275 
276 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
277 }
278 
279 static void
280 socket_hhook_register(int subtype)
281 {
282 
283 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
284 	    &V_socket_hhh[subtype],
285 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
286 		printf("%s: WARNING: unable to register hook\n", __func__);
287 }
288 
289 static void
290 socket_hhook_deregister(int subtype)
291 {
292 
293 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
294 		printf("%s: WARNING: unable to deregister hook\n", __func__);
295 }
296 
297 static void
298 socket_init(void *tag)
299 {
300 
301 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
302 	    NULL, NULL, UMA_ALIGN_PTR, 0);
303 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
304 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
305 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
306 	    EVENTHANDLER_PRI_FIRST);
307 }
308 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
309 
310 static void
311 socket_vnet_init(const void *unused __unused)
312 {
313 	int i;
314 
315 	/* We expect a contiguous range */
316 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
317 		socket_hhook_register(i);
318 }
319 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
320     socket_vnet_init, NULL);
321 
322 static void
323 socket_vnet_uninit(const void *unused __unused)
324 {
325 	int i;
326 
327 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
328 		socket_hhook_deregister(i);
329 }
330 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
331     socket_vnet_uninit, NULL);
332 
333 /*
334  * Initialise maxsockets.  This SYSINIT must be run after
335  * tunable_mbinit().
336  */
337 static void
338 init_maxsockets(void *ignored)
339 {
340 
341 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
342 	maxsockets = imax(maxsockets, maxfiles);
343 }
344 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
345 
346 /*
347  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
348  * of the change so that they can update their dependent limits as required.
349  */
350 static int
351 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
352 {
353 	int error, newmaxsockets;
354 
355 	newmaxsockets = maxsockets;
356 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
357 	if (error == 0 && req->newptr && newmaxsockets != maxsockets) {
358 		if (newmaxsockets > maxsockets &&
359 		    newmaxsockets <= maxfiles) {
360 			maxsockets = newmaxsockets;
361 			EVENTHANDLER_INVOKE(maxsockets_change);
362 		} else
363 			error = EINVAL;
364 	}
365 	return (error);
366 }
367 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
368     CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_NOFETCH | CTLFLAG_MPSAFE,
369     &maxsockets, 0, sysctl_maxsockets, "IU",
370     "Maximum number of sockets available");
371 
372 /*
373  * Socket operation routines.  These routines are called by the routines in
374  * sys_socket.c or from a system process, and implement the semantics of
375  * socket operations by switching out to the protocol specific routines.
376  */
377 
378 /*
379  * Get a socket structure from our zone, and initialize it.  Note that it
380  * would probably be better to allocate socket and PCB at the same time, but
381  * I'm not convinced that all the protocols can be easily modified to do
382  * this.
383  *
384  * soalloc() returns a socket with a ref count of 0.
385  */
386 static struct socket *
387 soalloc(struct vnet *vnet)
388 {
389 	struct socket *so;
390 
391 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
392 	if (so == NULL)
393 		return (NULL);
394 #ifdef MAC
395 	if (mac_socket_init(so, M_NOWAIT) != 0) {
396 		uma_zfree(socket_zone, so);
397 		return (NULL);
398 	}
399 #endif
400 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
401 		uma_zfree(socket_zone, so);
402 		return (NULL);
403 	}
404 
405 	/*
406 	 * The socket locking protocol allows to lock 2 sockets at a time,
407 	 * however, the first one must be a listening socket.  WITNESS lacks
408 	 * a feature to change class of an existing lock, so we use DUPOK.
409 	 */
410 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
411 	mtx_init(&so->so_snd_mtx, "so_snd", NULL, MTX_DEF);
412 	mtx_init(&so->so_rcv_mtx, "so_rcv", NULL, MTX_DEF);
413 	so->so_rcv.sb_sel = &so->so_rdsel;
414 	so->so_snd.sb_sel = &so->so_wrsel;
415 	sx_init(&so->so_snd_sx, "so_snd_sx");
416 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
417 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
418 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
419 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
420 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
421 #ifdef VIMAGE
422 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
423 	    __func__, __LINE__, so));
424 	so->so_vnet = vnet;
425 #endif
426 	/* We shouldn't need the so_global_mtx */
427 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
428 		/* Do we need more comprehensive error returns? */
429 		uma_zfree(socket_zone, so);
430 		return (NULL);
431 	}
432 	mtx_lock(&so_global_mtx);
433 	so->so_gencnt = ++so_gencnt;
434 	++numopensockets;
435 #ifdef VIMAGE
436 	vnet->vnet_sockcnt++;
437 #endif
438 	mtx_unlock(&so_global_mtx);
439 
440 	return (so);
441 }
442 
443 /*
444  * Free the storage associated with a socket at the socket layer, tear down
445  * locks, labels, etc.  All protocol state is assumed already to have been
446  * torn down (and possibly never set up) by the caller.
447  */
448 void
449 sodealloc(struct socket *so)
450 {
451 
452 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
453 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
454 
455 	mtx_lock(&so_global_mtx);
456 	so->so_gencnt = ++so_gencnt;
457 	--numopensockets;	/* Could be below, but faster here. */
458 #ifdef VIMAGE
459 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
460 	    __func__, __LINE__, so));
461 	so->so_vnet->vnet_sockcnt--;
462 #endif
463 	mtx_unlock(&so_global_mtx);
464 #ifdef MAC
465 	mac_socket_destroy(so);
466 #endif
467 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
468 
469 	khelp_destroy_osd(&so->osd);
470 	if (SOLISTENING(so)) {
471 		if (so->sol_accept_filter != NULL)
472 			accept_filt_setopt(so, NULL);
473 	} else {
474 		if (so->so_rcv.sb_hiwat)
475 			(void)chgsbsize(so->so_cred->cr_uidinfo,
476 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
477 		if (so->so_snd.sb_hiwat)
478 			(void)chgsbsize(so->so_cred->cr_uidinfo,
479 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
480 		sx_destroy(&so->so_snd_sx);
481 		sx_destroy(&so->so_rcv_sx);
482 		mtx_destroy(&so->so_snd_mtx);
483 		mtx_destroy(&so->so_rcv_mtx);
484 	}
485 	crfree(so->so_cred);
486 	mtx_destroy(&so->so_lock);
487 	uma_zfree(socket_zone, so);
488 }
489 
490 /*
491  * socreate returns a socket with a ref count of 1 and a file descriptor
492  * reference.  The socket should be closed with soclose().
493  */
494 int
495 socreate(int dom, struct socket **aso, int type, int proto,
496     struct ucred *cred, struct thread *td)
497 {
498 	struct protosw *prp;
499 	struct socket *so;
500 	int error;
501 
502 	/*
503 	 * XXX: divert(4) historically abused PF_INET.  Keep this compatibility
504 	 * shim until all applications have been updated.
505 	 */
506 	if (__predict_false(dom == PF_INET && type == SOCK_RAW &&
507 	    proto == IPPROTO_DIVERT)) {
508 		dom = PF_DIVERT;
509 		printf("%s uses obsolete way to create divert(4) socket\n",
510 		    td->td_proc->p_comm);
511 	}
512 
513 	prp = pffindproto(dom, type, proto);
514 	if (prp == NULL) {
515 		/* No support for domain. */
516 		if (pffinddomain(dom) == NULL)
517 			return (EAFNOSUPPORT);
518 		/* No support for socket type. */
519 		if (proto == 0 && type != 0)
520 			return (EPROTOTYPE);
521 		return (EPROTONOSUPPORT);
522 	}
523 
524 	MPASS(prp->pr_attach);
525 
526 	if ((prp->pr_flags & PR_CAPATTACH) == 0) {
527 		if (CAP_TRACING(td))
528 			ktrcapfail(CAPFAIL_PROTO, &proto);
529 		if (IN_CAPABILITY_MODE(td))
530 			return (ECAPMODE);
531 	}
532 
533 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
534 		return (EPROTONOSUPPORT);
535 
536 	so = soalloc(CRED_TO_VNET(cred));
537 	if (so == NULL)
538 		return (ENOBUFS);
539 
540 	so->so_type = type;
541 	so->so_cred = crhold(cred);
542 	if ((prp->pr_domain->dom_family == PF_INET) ||
543 	    (prp->pr_domain->dom_family == PF_INET6) ||
544 	    (prp->pr_domain->dom_family == PF_ROUTE))
545 		so->so_fibnum = td->td_proc->p_fibnum;
546 	else
547 		so->so_fibnum = 0;
548 	so->so_proto = prp;
549 #ifdef MAC
550 	mac_socket_create(cred, so);
551 #endif
552 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
553 	    so_rdknl_assert_lock);
554 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
555 	    so_wrknl_assert_lock);
556 	if ((prp->pr_flags & PR_SOCKBUF) == 0) {
557 		so->so_snd.sb_mtx = &so->so_snd_mtx;
558 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
559 	}
560 	/*
561 	 * Auto-sizing of socket buffers is managed by the protocols and
562 	 * the appropriate flags must be set in the pru_attach function.
563 	 */
564 	CURVNET_SET(so->so_vnet);
565 	error = prp->pr_attach(so, proto, td);
566 	CURVNET_RESTORE();
567 	if (error) {
568 		sodealloc(so);
569 		return (error);
570 	}
571 	soref(so);
572 	*aso = so;
573 	return (0);
574 }
575 
576 #ifdef REGRESSION
577 static int regression_sonewconn_earlytest = 1;
578 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
579     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
580 #endif
581 
582 static int sooverprio = LOG_DEBUG;
583 SYSCTL_INT(_kern_ipc, OID_AUTO, sooverprio, CTLFLAG_RW,
584     &sooverprio, 0, "Log priority for listen socket overflows: 0..7 or -1 to disable");
585 
586 static struct timeval overinterval = { 60, 0 };
587 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
588     &overinterval,
589     "Delay in seconds between warnings for listen socket overflows");
590 
591 /*
592  * When an attempt at a new connection is noted on a socket which supports
593  * accept(2), the protocol has two options:
594  * 1) Call legacy sonewconn() function, which would call protocol attach
595  *    method, same as used for socket(2).
596  * 2) Call solisten_clone(), do attach that is specific to a cloned connection,
597  *    and then call solisten_enqueue().
598  *
599  * Note: the ref count on the socket is 0 on return.
600  */
601 struct socket *
602 solisten_clone(struct socket *head)
603 {
604 	struct sbuf descrsb;
605 	struct socket *so;
606 	int len, overcount;
607 	u_int qlen;
608 	const char localprefix[] = "local:";
609 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
610 #if defined(INET6)
611 	char addrbuf[INET6_ADDRSTRLEN];
612 #elif defined(INET)
613 	char addrbuf[INET_ADDRSTRLEN];
614 #endif
615 	bool dolog, over;
616 
617 	SOLISTEN_LOCK(head);
618 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
619 #ifdef REGRESSION
620 	if (regression_sonewconn_earlytest && over) {
621 #else
622 	if (over) {
623 #endif
624 		head->sol_overcount++;
625 		dolog = (sooverprio >= 0) &&
626 			!!ratecheck(&head->sol_lastover, &overinterval);
627 
628 		/*
629 		 * If we're going to log, copy the overflow count and queue
630 		 * length from the listen socket before dropping the lock.
631 		 * Also, reset the overflow count.
632 		 */
633 		if (dolog) {
634 			overcount = head->sol_overcount;
635 			head->sol_overcount = 0;
636 			qlen = head->sol_qlen;
637 		}
638 		SOLISTEN_UNLOCK(head);
639 
640 		if (dolog) {
641 			/*
642 			 * Try to print something descriptive about the
643 			 * socket for the error message.
644 			 */
645 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
646 			    SBUF_FIXEDLEN);
647 			switch (head->so_proto->pr_domain->dom_family) {
648 #if defined(INET) || defined(INET6)
649 #ifdef INET
650 			case AF_INET:
651 #endif
652 #ifdef INET6
653 			case AF_INET6:
654 				if (head->so_proto->pr_domain->dom_family ==
655 				    AF_INET6 ||
656 				    (sotoinpcb(head)->inp_inc.inc_flags &
657 				    INC_ISIPV6)) {
658 					ip6_sprintf(addrbuf,
659 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
660 					sbuf_printf(&descrsb, "[%s]", addrbuf);
661 				} else
662 #endif
663 				{
664 #ifdef INET
665 					inet_ntoa_r(
666 					    sotoinpcb(head)->inp_inc.inc_laddr,
667 					    addrbuf);
668 					sbuf_cat(&descrsb, addrbuf);
669 #endif
670 				}
671 				sbuf_printf(&descrsb, ":%hu (proto %u)",
672 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
673 				    head->so_proto->pr_protocol);
674 				break;
675 #endif /* INET || INET6 */
676 			case AF_UNIX:
677 				sbuf_cat(&descrsb, localprefix);
678 				if (sotounpcb(head)->unp_addr != NULL)
679 					len =
680 					    sotounpcb(head)->unp_addr->sun_len -
681 					    offsetof(struct sockaddr_un,
682 					    sun_path);
683 				else
684 					len = 0;
685 				if (len > 0)
686 					sbuf_bcat(&descrsb,
687 					    sotounpcb(head)->unp_addr->sun_path,
688 					    len);
689 				else
690 					sbuf_cat(&descrsb, "(unknown)");
691 				break;
692 			}
693 
694 			/*
695 			 * If we can't print something more specific, at least
696 			 * print the domain name.
697 			 */
698 			if (sbuf_finish(&descrsb) != 0 ||
699 			    sbuf_len(&descrsb) <= 0) {
700 				sbuf_clear(&descrsb);
701 				sbuf_cat(&descrsb,
702 				    head->so_proto->pr_domain->dom_name ?:
703 				    "unknown");
704 				sbuf_finish(&descrsb);
705 			}
706 			KASSERT(sbuf_len(&descrsb) > 0,
707 			    ("%s: sbuf creation failed", __func__));
708 			/*
709 			 * Preserve the historic listen queue overflow log
710 			 * message, that starts with "sonewconn:".  It has
711 			 * been known to sysadmins for years and also test
712 			 * sys/kern/sonewconn_overflow checks for it.
713 			 */
714 			if (head->so_cred == 0) {
715 				log(LOG_PRI(sooverprio),
716 				    "sonewconn: pcb %p (%s): "
717 				    "Listen queue overflow: %i already in "
718 				    "queue awaiting acceptance (%d "
719 				    "occurrences)\n", head->so_pcb,
720 				    sbuf_data(&descrsb),
721 			    	qlen, overcount);
722 			} else {
723 				log(LOG_PRI(sooverprio),
724 				    "sonewconn: pcb %p (%s): "
725 				    "Listen queue overflow: "
726 				    "%i already in queue awaiting acceptance "
727 				    "(%d occurrences), euid %d, rgid %d, jail %s\n",
728 				    head->so_pcb, sbuf_data(&descrsb), qlen,
729 				    overcount, head->so_cred->cr_uid,
730 				    head->so_cred->cr_rgid,
731 				    head->so_cred->cr_prison ?
732 					head->so_cred->cr_prison->pr_name :
733 					"not_jailed");
734 			}
735 			sbuf_delete(&descrsb);
736 
737 			overcount = 0;
738 		}
739 
740 		return (NULL);
741 	}
742 	SOLISTEN_UNLOCK(head);
743 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
744 	    __func__, head));
745 	so = soalloc(head->so_vnet);
746 	if (so == NULL) {
747 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
748 		    "limit reached or out of memory\n",
749 		    __func__, head->so_pcb);
750 		return (NULL);
751 	}
752 	so->so_listen = head;
753 	so->so_type = head->so_type;
754 	/*
755 	 * POSIX is ambiguous on what options an accept(2)ed socket should
756 	 * inherit from the listener.  Words "create a new socket" may be
757 	 * interpreted as not inheriting anything.  Best programming practice
758 	 * for application developers is to not rely on such inheritance.
759 	 * FreeBSD had historically inherited all so_options excluding
760 	 * SO_ACCEPTCONN, which virtually means all SOL_SOCKET level options,
761 	 * including those completely irrelevant to a new born socket.  For
762 	 * compatibility with older versions we will inherit a list of
763 	 * meaningful options.
764 	 * The crucial bit to inherit is SO_ACCEPTFILTER.  We need it present
765 	 * in the child socket for soisconnected() promoting socket from the
766 	 * incomplete queue to complete.  It will be cleared before the child
767 	 * gets available to accept(2).
768 	 */
769 	so->so_options = head->so_options & (SO_ACCEPTFILTER | SO_KEEPALIVE |
770 	    SO_DONTROUTE | SO_LINGER | SO_OOBINLINE | SO_NOSIGPIPE);
771 	so->so_linger = head->so_linger;
772 	so->so_state = head->so_state;
773 	so->so_fibnum = head->so_fibnum;
774 	so->so_proto = head->so_proto;
775 	so->so_cred = crhold(head->so_cred);
776 	if (V_socket_hhh[HHOOK_SOCKET_NEWCONN]->hhh_nhooks > 0) {
777 		if (hhook_run_socket(so, head, HHOOK_SOCKET_NEWCONN)) {
778 			sodealloc(so);
779 			log(LOG_DEBUG, "%s: hhook run failed\n", __func__);
780 			return (NULL);
781 		}
782 	}
783 #ifdef MAC
784 	mac_socket_newconn(head, so);
785 #endif
786 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
787 	    so_rdknl_assert_lock);
788 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
789 	    so_wrknl_assert_lock);
790 	VNET_SO_ASSERT(head);
791 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
792 		sodealloc(so);
793 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
794 		    __func__, head->so_pcb);
795 		return (NULL);
796 	}
797 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
798 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
799 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
800 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
801 	so->so_rcv.sb_flags = head->sol_sbrcv_flags & SB_AUTOSIZE;
802 	so->so_snd.sb_flags = head->sol_sbsnd_flags & SB_AUTOSIZE;
803 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
804 		so->so_snd.sb_mtx = &so->so_snd_mtx;
805 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
806 	}
807 
808 	return (so);
809 }
810 
811 /* Connstatus may be 0 or SS_ISCONNECTED. */
812 struct socket *
813 sonewconn(struct socket *head, int connstatus)
814 {
815 	struct socket *so;
816 
817 	if ((so = solisten_clone(head)) == NULL)
818 		return (NULL);
819 
820 	if (so->so_proto->pr_attach(so, 0, NULL) != 0) {
821 		sodealloc(so);
822 		log(LOG_DEBUG, "%s: pcb %p: pr_attach() failed\n",
823 		    __func__, head->so_pcb);
824 		return (NULL);
825 	}
826 
827 	(void)solisten_enqueue(so, connstatus);
828 
829 	return (so);
830 }
831 
832 /*
833  * Enqueue socket cloned by solisten_clone() to the listen queue of the
834  * listener it has been cloned from.
835  *
836  * Return 'true' if socket landed on complete queue, otherwise 'false'.
837  */
838 bool
839 solisten_enqueue(struct socket *so, int connstatus)
840 {
841 	struct socket *head = so->so_listen;
842 
843 	MPASS(refcount_load(&so->so_count) == 0);
844 	refcount_init(&so->so_count, 1);
845 
846 	SOLISTEN_LOCK(head);
847 	if (head->sol_accept_filter != NULL)
848 		connstatus = 0;
849 	so->so_state |= connstatus;
850 	soref(head); /* A socket on (in)complete queue refs head. */
851 	if (connstatus) {
852 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
853 		so->so_qstate = SQ_COMP;
854 		head->sol_qlen++;
855 		solisten_wakeup(head);	/* unlocks */
856 		return (true);
857 	} else {
858 		/*
859 		 * Keep removing sockets from the head until there's room for
860 		 * us to insert on the tail.  In pre-locking revisions, this
861 		 * was a simple if(), but as we could be racing with other
862 		 * threads and soabort() requires dropping locks, we must
863 		 * loop waiting for the condition to be true.
864 		 */
865 		while (head->sol_incqlen > head->sol_qlimit) {
866 			struct socket *sp;
867 
868 			sp = TAILQ_FIRST(&head->sol_incomp);
869 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
870 			head->sol_incqlen--;
871 			SOCK_LOCK(sp);
872 			sp->so_qstate = SQ_NONE;
873 			sp->so_listen = NULL;
874 			SOCK_UNLOCK(sp);
875 			sorele_locked(head);	/* does SOLISTEN_UNLOCK, head stays */
876 			soabort(sp);
877 			SOLISTEN_LOCK(head);
878 		}
879 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
880 		so->so_qstate = SQ_INCOMP;
881 		head->sol_incqlen++;
882 		SOLISTEN_UNLOCK(head);
883 		return (false);
884 	}
885 }
886 
887 #if defined(SCTP) || defined(SCTP_SUPPORT)
888 /*
889  * Socket part of sctp_peeloff().  Detach a new socket from an
890  * association.  The new socket is returned with a reference.
891  *
892  * XXXGL: reduce copy-paste with solisten_clone().
893  */
894 struct socket *
895 sopeeloff(struct socket *head)
896 {
897 	struct socket *so;
898 
899 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
900 	    __func__, __LINE__, head));
901 	so = soalloc(head->so_vnet);
902 	if (so == NULL) {
903 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
904 		    "limit reached or out of memory\n",
905 		    __func__, head->so_pcb);
906 		return (NULL);
907 	}
908 	so->so_type = head->so_type;
909 	so->so_options = head->so_options;
910 	so->so_linger = head->so_linger;
911 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
912 	so->so_fibnum = head->so_fibnum;
913 	so->so_proto = head->so_proto;
914 	so->so_cred = crhold(head->so_cred);
915 #ifdef MAC
916 	mac_socket_newconn(head, so);
917 #endif
918 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
919 	    so_rdknl_assert_lock);
920 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
921 	    so_wrknl_assert_lock);
922 	VNET_SO_ASSERT(head);
923 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
924 		sodealloc(so);
925 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
926 		    __func__, head->so_pcb);
927 		return (NULL);
928 	}
929 	if ((*so->so_proto->pr_attach)(so, 0, NULL)) {
930 		sodealloc(so);
931 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
932 		    __func__, head->so_pcb);
933 		return (NULL);
934 	}
935 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
936 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
937 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
938 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
939 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
940 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
941 	if ((so->so_proto->pr_flags & PR_SOCKBUF) == 0) {
942 		so->so_snd.sb_mtx = &so->so_snd_mtx;
943 		so->so_rcv.sb_mtx = &so->so_rcv_mtx;
944 	}
945 
946 	soref(so);
947 
948 	return (so);
949 }
950 #endif	/* SCTP */
951 
952 int
953 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
954 {
955 	int error;
956 
957 	CURVNET_SET(so->so_vnet);
958 	error = so->so_proto->pr_bind(so, nam, td);
959 	CURVNET_RESTORE();
960 	return (error);
961 }
962 
963 int
964 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
965 {
966 	int error;
967 
968 	CURVNET_SET(so->so_vnet);
969 	error = so->so_proto->pr_bindat(fd, so, nam, td);
970 	CURVNET_RESTORE();
971 	return (error);
972 }
973 
974 /*
975  * solisten() transitions a socket from a non-listening state to a listening
976  * state, but can also be used to update the listen queue depth on an
977  * existing listen socket.  The protocol will call back into the sockets
978  * layer using solisten_proto_check() and solisten_proto() to check and set
979  * socket-layer listen state.  Call backs are used so that the protocol can
980  * acquire both protocol and socket layer locks in whatever order is required
981  * by the protocol.
982  *
983  * Protocol implementors are advised to hold the socket lock across the
984  * socket-layer test and set to avoid races at the socket layer.
985  */
986 int
987 solisten(struct socket *so, int backlog, struct thread *td)
988 {
989 	int error;
990 
991 	CURVNET_SET(so->so_vnet);
992 	error = so->so_proto->pr_listen(so, backlog, td);
993 	CURVNET_RESTORE();
994 	return (error);
995 }
996 
997 /*
998  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
999  * order to interlock with socket I/O.
1000  */
1001 int
1002 solisten_proto_check(struct socket *so)
1003 {
1004 	SOCK_LOCK_ASSERT(so);
1005 
1006 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1007 	    SS_ISDISCONNECTING)) != 0)
1008 		return (EINVAL);
1009 
1010 	/*
1011 	 * Sleeping is not permitted here, so simply fail if userspace is
1012 	 * attempting to transmit or receive on the socket.  This kind of
1013 	 * transient failure is not ideal, but it should occur only if userspace
1014 	 * is misusing the socket interfaces.
1015 	 */
1016 	if (!sx_try_xlock(&so->so_snd_sx))
1017 		return (EAGAIN);
1018 	if (!sx_try_xlock(&so->so_rcv_sx)) {
1019 		sx_xunlock(&so->so_snd_sx);
1020 		return (EAGAIN);
1021 	}
1022 	mtx_lock(&so->so_snd_mtx);
1023 	mtx_lock(&so->so_rcv_mtx);
1024 
1025 	/* Interlock with soo_aio_queue() and KTLS. */
1026 	if (!SOLISTENING(so)) {
1027 		bool ktls;
1028 
1029 #ifdef KERN_TLS
1030 		ktls = so->so_snd.sb_tls_info != NULL ||
1031 		    so->so_rcv.sb_tls_info != NULL;
1032 #else
1033 		ktls = false;
1034 #endif
1035 		if (ktls ||
1036 		    (so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
1037 		    (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
1038 			solisten_proto_abort(so);
1039 			return (EINVAL);
1040 		}
1041 	}
1042 
1043 	return (0);
1044 }
1045 
1046 /*
1047  * Undo the setup done by solisten_proto_check().
1048  */
1049 void
1050 solisten_proto_abort(struct socket *so)
1051 {
1052 	mtx_unlock(&so->so_snd_mtx);
1053 	mtx_unlock(&so->so_rcv_mtx);
1054 	sx_xunlock(&so->so_snd_sx);
1055 	sx_xunlock(&so->so_rcv_sx);
1056 }
1057 
1058 void
1059 solisten_proto(struct socket *so, int backlog)
1060 {
1061 	int sbrcv_lowat, sbsnd_lowat;
1062 	u_int sbrcv_hiwat, sbsnd_hiwat;
1063 	short sbrcv_flags, sbsnd_flags;
1064 	sbintime_t sbrcv_timeo, sbsnd_timeo;
1065 
1066 	SOCK_LOCK_ASSERT(so);
1067 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
1068 	    SS_ISDISCONNECTING)) == 0,
1069 	    ("%s: bad socket state %p", __func__, so));
1070 
1071 	if (SOLISTENING(so))
1072 		goto listening;
1073 
1074 	/*
1075 	 * Change this socket to listening state.
1076 	 */
1077 	sbrcv_lowat = so->so_rcv.sb_lowat;
1078 	sbsnd_lowat = so->so_snd.sb_lowat;
1079 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
1080 	sbsnd_hiwat = so->so_snd.sb_hiwat;
1081 	sbrcv_flags = so->so_rcv.sb_flags;
1082 	sbsnd_flags = so->so_snd.sb_flags;
1083 	sbrcv_timeo = so->so_rcv.sb_timeo;
1084 	sbsnd_timeo = so->so_snd.sb_timeo;
1085 
1086 	if (!(so->so_proto->pr_flags & PR_SOCKBUF)) {
1087 		sbdestroy(so, SO_SND);
1088 		sbdestroy(so, SO_RCV);
1089 	}
1090 
1091 #ifdef INVARIANTS
1092 	bzero(&so->so_rcv,
1093 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
1094 #endif
1095 
1096 	so->sol_sbrcv_lowat = sbrcv_lowat;
1097 	so->sol_sbsnd_lowat = sbsnd_lowat;
1098 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
1099 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
1100 	so->sol_sbrcv_flags = sbrcv_flags;
1101 	so->sol_sbsnd_flags = sbsnd_flags;
1102 	so->sol_sbrcv_timeo = sbrcv_timeo;
1103 	so->sol_sbsnd_timeo = sbsnd_timeo;
1104 
1105 	so->sol_qlen = so->sol_incqlen = 0;
1106 	TAILQ_INIT(&so->sol_incomp);
1107 	TAILQ_INIT(&so->sol_comp);
1108 
1109 	so->sol_accept_filter = NULL;
1110 	so->sol_accept_filter_arg = NULL;
1111 	so->sol_accept_filter_str = NULL;
1112 
1113 	so->sol_upcall = NULL;
1114 	so->sol_upcallarg = NULL;
1115 
1116 	so->so_options |= SO_ACCEPTCONN;
1117 
1118 listening:
1119 	if (backlog < 0 || backlog > somaxconn)
1120 		backlog = somaxconn;
1121 	so->sol_qlimit = backlog;
1122 
1123 	mtx_unlock(&so->so_snd_mtx);
1124 	mtx_unlock(&so->so_rcv_mtx);
1125 	sx_xunlock(&so->so_snd_sx);
1126 	sx_xunlock(&so->so_rcv_sx);
1127 }
1128 
1129 /*
1130  * Wakeup listeners/subsystems once we have a complete connection.
1131  * Enters with lock, returns unlocked.
1132  */
1133 void
1134 solisten_wakeup(struct socket *sol)
1135 {
1136 
1137 	if (sol->sol_upcall != NULL)
1138 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1139 	else {
1140 		selwakeuppri(&sol->so_rdsel, PSOCK);
1141 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1142 	}
1143 	SOLISTEN_UNLOCK(sol);
1144 	wakeup_one(&sol->sol_comp);
1145 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1146 		pgsigio(&sol->so_sigio, SIGIO, 0);
1147 }
1148 
1149 /*
1150  * Return single connection off a listening socket queue.  Main consumer of
1151  * the function is kern_accept4().  Some modules, that do their own accept
1152  * management also use the function.  The socket reference held by the
1153  * listen queue is handed to the caller.
1154  *
1155  * Listening socket must be locked on entry and is returned unlocked on
1156  * return.
1157  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1158  */
1159 int
1160 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1161 {
1162 	struct socket *so;
1163 	int error;
1164 
1165 	SOLISTEN_LOCK_ASSERT(head);
1166 
1167 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1168 	    head->so_error == 0) {
1169 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1170 		    "accept", 0);
1171 		if (error != 0) {
1172 			SOLISTEN_UNLOCK(head);
1173 			return (error);
1174 		}
1175 	}
1176 	if (head->so_error) {
1177 		error = head->so_error;
1178 		head->so_error = 0;
1179 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1180 		error = EWOULDBLOCK;
1181 	else
1182 		error = 0;
1183 	if (error) {
1184 		SOLISTEN_UNLOCK(head);
1185 		return (error);
1186 	}
1187 	so = TAILQ_FIRST(&head->sol_comp);
1188 	SOCK_LOCK(so);
1189 	KASSERT(so->so_qstate == SQ_COMP,
1190 	    ("%s: so %p not SQ_COMP", __func__, so));
1191 	head->sol_qlen--;
1192 	so->so_qstate = SQ_NONE;
1193 	so->so_listen = NULL;
1194 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1195 	if (flags & ACCEPT4_INHERIT)
1196 		so->so_state |= (head->so_state & SS_NBIO);
1197 	else
1198 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1199 	SOCK_UNLOCK(so);
1200 	sorele_locked(head);
1201 
1202 	*ret = so;
1203 	return (0);
1204 }
1205 
1206 /*
1207  * Free socket upon release of the very last reference.
1208  */
1209 static void
1210 sofree(struct socket *so)
1211 {
1212 	struct protosw *pr = so->so_proto;
1213 
1214 	SOCK_LOCK_ASSERT(so);
1215 	KASSERT(refcount_load(&so->so_count) == 0,
1216 	    ("%s: so %p has references", __func__, so));
1217 	KASSERT(SOLISTENING(so) || so->so_qstate == SQ_NONE,
1218 	    ("%s: so %p is on listen queue", __func__, so));
1219 
1220 	SOCK_UNLOCK(so);
1221 
1222 	if (so->so_dtor != NULL)
1223 		so->so_dtor(so);
1224 
1225 	VNET_SO_ASSERT(so);
1226 	if (pr->pr_detach != NULL)
1227 		pr->pr_detach(so);
1228 
1229 	/*
1230 	 * From this point on, we assume that no other references to this
1231 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1232 	 * to be acquired or held.
1233 	 */
1234 	if (!(pr->pr_flags & PR_SOCKBUF) && !SOLISTENING(so)) {
1235 		sbdestroy(so, SO_SND);
1236 		sbdestroy(so, SO_RCV);
1237 	}
1238 	seldrain(&so->so_rdsel);
1239 	seldrain(&so->so_wrsel);
1240 	knlist_destroy(&so->so_rdsel.si_note);
1241 	knlist_destroy(&so->so_wrsel.si_note);
1242 	sodealloc(so);
1243 }
1244 
1245 /*
1246  * Release a reference on a socket while holding the socket lock.
1247  * Unlocks the socket lock before returning.
1248  */
1249 void
1250 sorele_locked(struct socket *so)
1251 {
1252 	SOCK_LOCK_ASSERT(so);
1253 	if (refcount_release(&so->so_count))
1254 		sofree(so);
1255 	else
1256 		SOCK_UNLOCK(so);
1257 }
1258 
1259 /*
1260  * Close a socket on last file table reference removal.  Initiate disconnect
1261  * if connected.  Free socket when disconnect complete.
1262  *
1263  * This function will sorele() the socket.  Note that soclose() may be called
1264  * prior to the ref count reaching zero.  The actual socket structure will
1265  * not be freed until the ref count reaches zero.
1266  */
1267 int
1268 soclose(struct socket *so)
1269 {
1270 	struct accept_queue lqueue;
1271 	int error = 0;
1272 	bool listening, last __diagused;
1273 
1274 	CURVNET_SET(so->so_vnet);
1275 	funsetown(&so->so_sigio);
1276 	if (so->so_state & SS_ISCONNECTED) {
1277 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1278 			error = sodisconnect(so);
1279 			if (error) {
1280 				if (error == ENOTCONN)
1281 					error = 0;
1282 				goto drop;
1283 			}
1284 		}
1285 
1286 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1287 			if ((so->so_state & SS_ISDISCONNECTING) &&
1288 			    (so->so_state & SS_NBIO))
1289 				goto drop;
1290 			while (so->so_state & SS_ISCONNECTED) {
1291 				error = tsleep(&so->so_timeo,
1292 				    PSOCK | PCATCH, "soclos",
1293 				    so->so_linger * hz);
1294 				if (error)
1295 					break;
1296 			}
1297 		}
1298 	}
1299 
1300 drop:
1301 	if (so->so_proto->pr_close != NULL)
1302 		so->so_proto->pr_close(so);
1303 
1304 	SOCK_LOCK(so);
1305 	if ((listening = SOLISTENING(so))) {
1306 		struct socket *sp;
1307 
1308 		TAILQ_INIT(&lqueue);
1309 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1310 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1311 
1312 		so->sol_qlen = so->sol_incqlen = 0;
1313 
1314 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1315 			SOCK_LOCK(sp);
1316 			sp->so_qstate = SQ_NONE;
1317 			sp->so_listen = NULL;
1318 			SOCK_UNLOCK(sp);
1319 			last = refcount_release(&so->so_count);
1320 			KASSERT(!last, ("%s: released last reference for %p",
1321 			    __func__, so));
1322 		}
1323 	}
1324 	sorele_locked(so);
1325 	if (listening) {
1326 		struct socket *sp, *tsp;
1327 
1328 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp)
1329 			soabort(sp);
1330 	}
1331 	CURVNET_RESTORE();
1332 	return (error);
1333 }
1334 
1335 /*
1336  * soabort() is used to abruptly tear down a connection, such as when a
1337  * resource limit is reached (listen queue depth exceeded), or if a listen
1338  * socket is closed while there are sockets waiting to be accepted.
1339  *
1340  * This interface is tricky, because it is called on an unreferenced socket,
1341  * and must be called only by a thread that has actually removed the socket
1342  * from the listen queue it was on.  Likely this thread holds the last
1343  * reference on the socket and soabort() will proceed with sofree().  But
1344  * it might be not the last, as the sockets on the listen queues are seen
1345  * from the protocol side.
1346  *
1347  * This interface will call into the protocol code, so must not be called
1348  * with any socket locks held.  Protocols do call it while holding their own
1349  * recursible protocol mutexes, but this is something that should be subject
1350  * to review in the future.
1351  *
1352  * Usually socket should have a single reference left, but this is not a
1353  * requirement.  In the past, when we have had named references for file
1354  * descriptor and protocol, we asserted that none of them are being held.
1355  */
1356 void
1357 soabort(struct socket *so)
1358 {
1359 
1360 	VNET_SO_ASSERT(so);
1361 
1362 	if (so->so_proto->pr_abort != NULL)
1363 		so->so_proto->pr_abort(so);
1364 	SOCK_LOCK(so);
1365 	sorele_locked(so);
1366 }
1367 
1368 int
1369 soaccept(struct socket *so, struct sockaddr *sa)
1370 {
1371 #ifdef INVARIANTS
1372 	u_char len = sa->sa_len;
1373 #endif
1374 	int error;
1375 
1376 	CURVNET_SET(so->so_vnet);
1377 	error = so->so_proto->pr_accept(so, sa);
1378 	KASSERT(sa->sa_len <= len,
1379 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1380 	CURVNET_RESTORE();
1381 	return (error);
1382 }
1383 
1384 int
1385 sopeeraddr(struct socket *so, struct sockaddr *sa)
1386 {
1387 #ifdef INVARIANTS
1388 	u_char len = sa->sa_len;
1389 #endif
1390 	int error;
1391 
1392 	CURVNET_SET(so->so_vnet);
1393 	error = so->so_proto->pr_peeraddr(so, sa);
1394 	KASSERT(sa->sa_len <= len,
1395 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1396 	CURVNET_RESTORE();
1397 
1398 	return (error);
1399 }
1400 
1401 int
1402 sosockaddr(struct socket *so, struct sockaddr *sa)
1403 {
1404 #ifdef INVARIANTS
1405 	u_char len = sa->sa_len;
1406 #endif
1407 	int error;
1408 
1409 	CURVNET_SET(so->so_vnet);
1410 	error = so->so_proto->pr_sockaddr(so, sa);
1411 	KASSERT(sa->sa_len <= len,
1412 	    ("%s: protocol %p sockaddr overflow", __func__, so->so_proto));
1413 	CURVNET_RESTORE();
1414 
1415 	return (error);
1416 }
1417 
1418 int
1419 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1420 {
1421 
1422 	return (soconnectat(AT_FDCWD, so, nam, td));
1423 }
1424 
1425 int
1426 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1427 {
1428 	int error;
1429 
1430 	CURVNET_SET(so->so_vnet);
1431 
1432 	/*
1433 	 * If protocol is connection-based, can only connect once.
1434 	 * Otherwise, if connected, try to disconnect first.  This allows
1435 	 * user to disconnect by connecting to, e.g., a null address.
1436 	 *
1437 	 * Note, this check is racy and may need to be re-evaluated at the
1438 	 * protocol layer.
1439 	 */
1440 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1441 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1442 	    (error = sodisconnect(so)))) {
1443 		error = EISCONN;
1444 	} else {
1445 		/*
1446 		 * Prevent accumulated error from previous connection from
1447 		 * biting us.
1448 		 */
1449 		so->so_error = 0;
1450 		if (fd == AT_FDCWD) {
1451 			error = so->so_proto->pr_connect(so, nam, td);
1452 		} else {
1453 			error = so->so_proto->pr_connectat(fd, so, nam, td);
1454 		}
1455 	}
1456 	CURVNET_RESTORE();
1457 
1458 	return (error);
1459 }
1460 
1461 int
1462 soconnect2(struct socket *so1, struct socket *so2)
1463 {
1464 	int error;
1465 
1466 	CURVNET_SET(so1->so_vnet);
1467 	error = so1->so_proto->pr_connect2(so1, so2);
1468 	CURVNET_RESTORE();
1469 	return (error);
1470 }
1471 
1472 int
1473 sodisconnect(struct socket *so)
1474 {
1475 	int error;
1476 
1477 	if ((so->so_state & SS_ISCONNECTED) == 0)
1478 		return (ENOTCONN);
1479 	if (so->so_state & SS_ISDISCONNECTING)
1480 		return (EALREADY);
1481 	VNET_SO_ASSERT(so);
1482 	error = so->so_proto->pr_disconnect(so);
1483 	return (error);
1484 }
1485 
1486 int
1487 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1488     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1489 {
1490 	long space;
1491 	ssize_t resid;
1492 	int clen = 0, error, dontroute;
1493 
1494 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1495 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1496 	    ("sosend_dgram: !PR_ATOMIC"));
1497 
1498 	if (uio != NULL)
1499 		resid = uio->uio_resid;
1500 	else
1501 		resid = top->m_pkthdr.len;
1502 	/*
1503 	 * In theory resid should be unsigned.  However, space must be
1504 	 * signed, as it might be less than 0 if we over-committed, and we
1505 	 * must use a signed comparison of space and resid.  On the other
1506 	 * hand, a negative resid causes us to loop sending 0-length
1507 	 * segments to the protocol.
1508 	 */
1509 	if (resid < 0) {
1510 		error = EINVAL;
1511 		goto out;
1512 	}
1513 
1514 	dontroute =
1515 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1516 	if (td != NULL)
1517 		td->td_ru.ru_msgsnd++;
1518 	if (control != NULL)
1519 		clen = control->m_len;
1520 
1521 	SOCKBUF_LOCK(&so->so_snd);
1522 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1523 		SOCKBUF_UNLOCK(&so->so_snd);
1524 		error = EPIPE;
1525 		goto out;
1526 	}
1527 	if (so->so_error) {
1528 		error = so->so_error;
1529 		so->so_error = 0;
1530 		SOCKBUF_UNLOCK(&so->so_snd);
1531 		goto out;
1532 	}
1533 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1534 		/*
1535 		 * `sendto' and `sendmsg' is allowed on a connection-based
1536 		 * socket if it supports implied connect.  Return ENOTCONN if
1537 		 * not connected and no address is supplied.
1538 		 */
1539 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1540 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1541 			if (!(resid == 0 && clen != 0)) {
1542 				SOCKBUF_UNLOCK(&so->so_snd);
1543 				error = ENOTCONN;
1544 				goto out;
1545 			}
1546 		} else if (addr == NULL) {
1547 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1548 				error = ENOTCONN;
1549 			else
1550 				error = EDESTADDRREQ;
1551 			SOCKBUF_UNLOCK(&so->so_snd);
1552 			goto out;
1553 		}
1554 	}
1555 
1556 	/*
1557 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1558 	 * problem and need fixing.
1559 	 */
1560 	space = sbspace(&so->so_snd);
1561 	if (flags & MSG_OOB)
1562 		space += 1024;
1563 	space -= clen;
1564 	SOCKBUF_UNLOCK(&so->so_snd);
1565 	if (resid > space) {
1566 		error = EMSGSIZE;
1567 		goto out;
1568 	}
1569 	if (uio == NULL) {
1570 		resid = 0;
1571 		if (flags & MSG_EOR)
1572 			top->m_flags |= M_EOR;
1573 	} else {
1574 		/*
1575 		 * Copy the data from userland into a mbuf chain.
1576 		 * If no data is to be copied in, a single empty mbuf
1577 		 * is returned.
1578 		 */
1579 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1580 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1581 		if (top == NULL) {
1582 			error = EFAULT;	/* only possible error */
1583 			goto out;
1584 		}
1585 		space -= resid - uio->uio_resid;
1586 		resid = uio->uio_resid;
1587 	}
1588 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1589 	/*
1590 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1591 	 * than with.
1592 	 */
1593 	if (dontroute) {
1594 		SOCK_LOCK(so);
1595 		so->so_options |= SO_DONTROUTE;
1596 		SOCK_UNLOCK(so);
1597 	}
1598 	/*
1599 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1600 	 * of date.  We could have received a reset packet in an interrupt or
1601 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1602 	 * probably recheck again inside the locking protection here, but
1603 	 * there are probably other places that this also happens.  We must
1604 	 * rethink this.
1605 	 */
1606 	VNET_SO_ASSERT(so);
1607 	error = so->so_proto->pr_send(so, (flags & MSG_OOB) ? PRUS_OOB :
1608 	/*
1609 	 * If the user set MSG_EOF, the protocol understands this flag and
1610 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1611 	 */
1612 	    ((flags & MSG_EOF) &&
1613 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1614 	     (resid <= 0)) ?
1615 		PRUS_EOF :
1616 		/* If there is more to send set PRUS_MORETOCOME */
1617 		(flags & MSG_MORETOCOME) ||
1618 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1619 		top, addr, control, td);
1620 	if (dontroute) {
1621 		SOCK_LOCK(so);
1622 		so->so_options &= ~SO_DONTROUTE;
1623 		SOCK_UNLOCK(so);
1624 	}
1625 	clen = 0;
1626 	control = NULL;
1627 	top = NULL;
1628 out:
1629 	if (top != NULL)
1630 		m_freem(top);
1631 	if (control != NULL)
1632 		m_freem(control);
1633 	return (error);
1634 }
1635 
1636 /*
1637  * Send on a socket.  If send must go all at once and message is larger than
1638  * send buffering, then hard error.  Lock against other senders.  If must go
1639  * all at once and not enough room now, then inform user that this would
1640  * block and do nothing.  Otherwise, if nonblocking, send as much as
1641  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1642  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1643  * in mbuf chain must be small enough to send all at once.
1644  *
1645  * Returns nonzero on error, timeout or signal; callers must check for short
1646  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1647  * on return.
1648  */
1649 int
1650 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1651     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1652 {
1653 	long space;
1654 	ssize_t resid;
1655 	int clen = 0, error, dontroute;
1656 	int atomic = sosendallatonce(so) || top;
1657 	int pr_send_flag;
1658 #ifdef KERN_TLS
1659 	struct ktls_session *tls;
1660 	int tls_enq_cnt, tls_send_flag;
1661 	uint8_t tls_rtype;
1662 
1663 	tls = NULL;
1664 	tls_rtype = TLS_RLTYPE_APP;
1665 #endif
1666 	if (uio != NULL)
1667 		resid = uio->uio_resid;
1668 	else if ((top->m_flags & M_PKTHDR) != 0)
1669 		resid = top->m_pkthdr.len;
1670 	else
1671 		resid = m_length(top, NULL);
1672 	/*
1673 	 * In theory resid should be unsigned.  However, space must be
1674 	 * signed, as it might be less than 0 if we over-committed, and we
1675 	 * must use a signed comparison of space and resid.  On the other
1676 	 * hand, a negative resid causes us to loop sending 0-length
1677 	 * segments to the protocol.
1678 	 *
1679 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1680 	 * type sockets since that's an error.
1681 	 */
1682 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1683 		error = EINVAL;
1684 		goto out;
1685 	}
1686 
1687 	dontroute =
1688 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1689 	    (so->so_proto->pr_flags & PR_ATOMIC);
1690 	if (td != NULL)
1691 		td->td_ru.ru_msgsnd++;
1692 	if (control != NULL)
1693 		clen = control->m_len;
1694 
1695 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1696 	if (error)
1697 		goto out;
1698 
1699 #ifdef KERN_TLS
1700 	tls_send_flag = 0;
1701 	tls = ktls_hold(so->so_snd.sb_tls_info);
1702 	if (tls != NULL) {
1703 		if (tls->mode == TCP_TLS_MODE_SW)
1704 			tls_send_flag = PRUS_NOTREADY;
1705 
1706 		if (control != NULL) {
1707 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1708 
1709 			if (clen >= sizeof(*cm) &&
1710 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1711 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1712 				clen = 0;
1713 				m_freem(control);
1714 				control = NULL;
1715 				atomic = 1;
1716 			}
1717 		}
1718 
1719 		if (resid == 0 && !ktls_permit_empty_frames(tls)) {
1720 			error = EINVAL;
1721 			goto release;
1722 		}
1723 	}
1724 #endif
1725 
1726 restart:
1727 	do {
1728 		SOCKBUF_LOCK(&so->so_snd);
1729 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1730 			SOCKBUF_UNLOCK(&so->so_snd);
1731 			error = EPIPE;
1732 			goto release;
1733 		}
1734 		if (so->so_error) {
1735 			error = so->so_error;
1736 			so->so_error = 0;
1737 			SOCKBUF_UNLOCK(&so->so_snd);
1738 			goto release;
1739 		}
1740 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1741 			/*
1742 			 * `sendto' and `sendmsg' is allowed on a connection-
1743 			 * based socket if it supports implied connect.
1744 			 * Return ENOTCONN if not connected and no address is
1745 			 * supplied.
1746 			 */
1747 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1748 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1749 				if (!(resid == 0 && clen != 0)) {
1750 					SOCKBUF_UNLOCK(&so->so_snd);
1751 					error = ENOTCONN;
1752 					goto release;
1753 				}
1754 			} else if (addr == NULL) {
1755 				SOCKBUF_UNLOCK(&so->so_snd);
1756 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1757 					error = ENOTCONN;
1758 				else
1759 					error = EDESTADDRREQ;
1760 				goto release;
1761 			}
1762 		}
1763 		space = sbspace(&so->so_snd);
1764 		if (flags & MSG_OOB)
1765 			space += 1024;
1766 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1767 		    clen > so->so_snd.sb_hiwat) {
1768 			SOCKBUF_UNLOCK(&so->so_snd);
1769 			error = EMSGSIZE;
1770 			goto release;
1771 		}
1772 		if (space < resid + clen &&
1773 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1774 			if ((so->so_state & SS_NBIO) ||
1775 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1776 				SOCKBUF_UNLOCK(&so->so_snd);
1777 				error = EWOULDBLOCK;
1778 				goto release;
1779 			}
1780 			error = sbwait(so, SO_SND);
1781 			SOCKBUF_UNLOCK(&so->so_snd);
1782 			if (error)
1783 				goto release;
1784 			goto restart;
1785 		}
1786 		SOCKBUF_UNLOCK(&so->so_snd);
1787 		space -= clen;
1788 		do {
1789 			if (uio == NULL) {
1790 				resid = 0;
1791 				if (flags & MSG_EOR)
1792 					top->m_flags |= M_EOR;
1793 #ifdef KERN_TLS
1794 				if (tls != NULL) {
1795 					ktls_frame(top, tls, &tls_enq_cnt,
1796 					    tls_rtype);
1797 					tls_rtype = TLS_RLTYPE_APP;
1798 				}
1799 #endif
1800 			} else {
1801 				/*
1802 				 * Copy the data from userland into a mbuf
1803 				 * chain.  If resid is 0, which can happen
1804 				 * only if we have control to send, then
1805 				 * a single empty mbuf is returned.  This
1806 				 * is a workaround to prevent protocol send
1807 				 * methods to panic.
1808 				 */
1809 #ifdef KERN_TLS
1810 				if (tls != NULL) {
1811 					top = m_uiotombuf(uio, M_WAITOK, space,
1812 					    tls->params.max_frame_len,
1813 					    M_EXTPG |
1814 					    ((flags & MSG_EOR) ? M_EOR : 0));
1815 					if (top != NULL) {
1816 						ktls_frame(top, tls,
1817 						    &tls_enq_cnt, tls_rtype);
1818 					}
1819 					tls_rtype = TLS_RLTYPE_APP;
1820 				} else
1821 #endif
1822 					top = m_uiotombuf(uio, M_WAITOK, space,
1823 					    (atomic ? max_hdr : 0),
1824 					    (atomic ? M_PKTHDR : 0) |
1825 					    ((flags & MSG_EOR) ? M_EOR : 0));
1826 				if (top == NULL) {
1827 					error = EFAULT; /* only possible error */
1828 					goto release;
1829 				}
1830 				space -= resid - uio->uio_resid;
1831 				resid = uio->uio_resid;
1832 			}
1833 			if (dontroute) {
1834 				SOCK_LOCK(so);
1835 				so->so_options |= SO_DONTROUTE;
1836 				SOCK_UNLOCK(so);
1837 			}
1838 			/*
1839 			 * XXX all the SBS_CANTSENDMORE checks previously
1840 			 * done could be out of date.  We could have received
1841 			 * a reset packet in an interrupt or maybe we slept
1842 			 * while doing page faults in uiomove() etc.  We
1843 			 * could probably recheck again inside the locking
1844 			 * protection here, but there are probably other
1845 			 * places that this also happens.  We must rethink
1846 			 * this.
1847 			 */
1848 			VNET_SO_ASSERT(so);
1849 
1850 			pr_send_flag = (flags & MSG_OOB) ? PRUS_OOB :
1851 			/*
1852 			 * If the user set MSG_EOF, the protocol understands
1853 			 * this flag and nothing left to send then use
1854 			 * PRU_SEND_EOF instead of PRU_SEND.
1855 			 */
1856 			    ((flags & MSG_EOF) &&
1857 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1858 			     (resid <= 0)) ?
1859 				PRUS_EOF :
1860 			/* If there is more to send set PRUS_MORETOCOME. */
1861 			    (flags & MSG_MORETOCOME) ||
1862 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1863 
1864 #ifdef KERN_TLS
1865 			pr_send_flag |= tls_send_flag;
1866 #endif
1867 
1868 			error = so->so_proto->pr_send(so, pr_send_flag, top,
1869 			    addr, control, td);
1870 
1871 			if (dontroute) {
1872 				SOCK_LOCK(so);
1873 				so->so_options &= ~SO_DONTROUTE;
1874 				SOCK_UNLOCK(so);
1875 			}
1876 
1877 #ifdef KERN_TLS
1878 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1879 				if (error != 0) {
1880 					m_freem(top);
1881 					top = NULL;
1882 				} else {
1883 					soref(so);
1884 					ktls_enqueue(top, so, tls_enq_cnt);
1885 				}
1886 			}
1887 #endif
1888 			clen = 0;
1889 			control = NULL;
1890 			top = NULL;
1891 			if (error)
1892 				goto release;
1893 		} while (resid && space > 0);
1894 	} while (resid);
1895 
1896 release:
1897 	SOCK_IO_SEND_UNLOCK(so);
1898 out:
1899 #ifdef KERN_TLS
1900 	if (tls != NULL)
1901 		ktls_free(tls);
1902 #endif
1903 	if (top != NULL)
1904 		m_freem(top);
1905 	if (control != NULL)
1906 		m_freem(control);
1907 	return (error);
1908 }
1909 
1910 /*
1911  * Send to a socket from a kernel thread.
1912  *
1913  * XXXGL: in almost all cases uio is NULL and the mbuf is supplied.
1914  * Exception is nfs/bootp_subr.c.  It is arguable that the VNET context needs
1915  * to be set at all.  This function should just boil down to a static inline
1916  * calling the protocol method.
1917  */
1918 int
1919 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1920     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1921 {
1922 	int error;
1923 
1924 	CURVNET_SET(so->so_vnet);
1925 	error = so->so_proto->pr_sosend(so, addr, uio,
1926 	    top, control, flags, td);
1927 	CURVNET_RESTORE();
1928 	return (error);
1929 }
1930 
1931 /*
1932  * send(2), write(2) or aio_write(2) on a socket.
1933  */
1934 int
1935 sousrsend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1936     struct mbuf *control, int flags, struct proc *userproc)
1937 {
1938 	struct thread *td;
1939 	ssize_t len;
1940 	int error;
1941 
1942 	td = uio->uio_td;
1943 	len = uio->uio_resid;
1944 	CURVNET_SET(so->so_vnet);
1945 	error = so->so_proto->pr_sosend(so, addr, uio, NULL, control, flags,
1946 	    td);
1947 	CURVNET_RESTORE();
1948 	if (error != 0) {
1949 		/*
1950 		 * Clear transient errors for stream protocols if they made
1951 		 * some progress.  Make exclusion for aio(4) that would
1952 		 * schedule a new write in case of EWOULDBLOCK and clear
1953 		 * error itself.  See soaio_process_job().
1954 		 */
1955 		if (uio->uio_resid != len &&
1956 		    (so->so_proto->pr_flags & PR_ATOMIC) == 0 &&
1957 		    userproc == NULL &&
1958 		    (error == ERESTART || error == EINTR ||
1959 		    error == EWOULDBLOCK))
1960 			error = 0;
1961 		/* Generation of SIGPIPE can be controlled per socket. */
1962 		if (error == EPIPE && (so->so_options & SO_NOSIGPIPE) == 0 &&
1963 		    (flags & MSG_NOSIGNAL) == 0) {
1964 			if (userproc != NULL) {
1965 				/* aio(4) job */
1966 				PROC_LOCK(userproc);
1967 				kern_psignal(userproc, SIGPIPE);
1968 				PROC_UNLOCK(userproc);
1969 			} else {
1970 				PROC_LOCK(td->td_proc);
1971 				tdsignal(td, SIGPIPE);
1972 				PROC_UNLOCK(td->td_proc);
1973 			}
1974 		}
1975 	}
1976 	return (error);
1977 }
1978 
1979 /*
1980  * The part of soreceive() that implements reading non-inline out-of-band
1981  * data from a socket.  For more complete comments, see soreceive(), from
1982  * which this code originated.
1983  *
1984  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1985  * unable to return an mbuf chain to the caller.
1986  */
1987 static int
1988 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1989 {
1990 	struct protosw *pr = so->so_proto;
1991 	struct mbuf *m;
1992 	int error;
1993 
1994 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1995 	VNET_SO_ASSERT(so);
1996 
1997 	m = m_get(M_WAITOK, MT_DATA);
1998 	error = pr->pr_rcvoob(so, m, flags & MSG_PEEK);
1999 	if (error)
2000 		goto bad;
2001 	do {
2002 		error = uiomove(mtod(m, void *),
2003 		    (int) min(uio->uio_resid, m->m_len), uio);
2004 		m = m_free(m);
2005 	} while (uio->uio_resid && error == 0 && m);
2006 bad:
2007 	if (m != NULL)
2008 		m_freem(m);
2009 	return (error);
2010 }
2011 
2012 /*
2013  * Following replacement or removal of the first mbuf on the first mbuf chain
2014  * of a socket buffer, push necessary state changes back into the socket
2015  * buffer so that other consumers see the values consistently.  'nextrecord'
2016  * is the callers locally stored value of the original value of
2017  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
2018  * NOTE: 'nextrecord' may be NULL.
2019  */
2020 static __inline void
2021 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
2022 {
2023 
2024 	SOCKBUF_LOCK_ASSERT(sb);
2025 	/*
2026 	 * First, update for the new value of nextrecord.  If necessary, make
2027 	 * it the first record.
2028 	 */
2029 	if (sb->sb_mb != NULL)
2030 		sb->sb_mb->m_nextpkt = nextrecord;
2031 	else
2032 		sb->sb_mb = nextrecord;
2033 
2034 	/*
2035 	 * Now update any dependent socket buffer fields to reflect the new
2036 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
2037 	 * addition of a second clause that takes care of the case where
2038 	 * sb_mb has been updated, but remains the last record.
2039 	 */
2040 	if (sb->sb_mb == NULL) {
2041 		sb->sb_mbtail = NULL;
2042 		sb->sb_lastrecord = NULL;
2043 	} else if (sb->sb_mb->m_nextpkt == NULL)
2044 		sb->sb_lastrecord = sb->sb_mb;
2045 }
2046 
2047 /*
2048  * Implement receive operations on a socket.  We depend on the way that
2049  * records are added to the sockbuf by sbappend.  In particular, each record
2050  * (mbufs linked through m_next) must begin with an address if the protocol
2051  * so specifies, followed by an optional mbuf or mbufs containing ancillary
2052  * data, and then zero or more mbufs of data.  In order to allow parallelism
2053  * between network receive and copying to user space, as well as avoid
2054  * sleeping with a mutex held, we release the socket buffer mutex during the
2055  * user space copy.  Although the sockbuf is locked, new data may still be
2056  * appended, and thus we must maintain consistency of the sockbuf during that
2057  * time.
2058  *
2059  * The caller may receive the data as a single mbuf chain by supplying an
2060  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
2061  * the count in uio_resid.
2062  */
2063 int
2064 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
2065     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2066 {
2067 	struct mbuf *m, **mp;
2068 	int flags, error, offset;
2069 	ssize_t len;
2070 	struct protosw *pr = so->so_proto;
2071 	struct mbuf *nextrecord;
2072 	int moff, type = 0;
2073 	ssize_t orig_resid = uio->uio_resid;
2074 	bool report_real_len = false;
2075 
2076 	mp = mp0;
2077 	if (psa != NULL)
2078 		*psa = NULL;
2079 	if (controlp != NULL)
2080 		*controlp = NULL;
2081 	if (flagsp != NULL) {
2082 		report_real_len = *flagsp & MSG_TRUNC;
2083 		*flagsp &= ~MSG_TRUNC;
2084 		flags = *flagsp &~ MSG_EOR;
2085 	} else
2086 		flags = 0;
2087 	if (flags & MSG_OOB)
2088 		return (soreceive_rcvoob(so, uio, flags));
2089 	if (mp != NULL)
2090 		*mp = NULL;
2091 
2092 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2093 	if (error)
2094 		return (error);
2095 
2096 restart:
2097 	SOCKBUF_LOCK(&so->so_rcv);
2098 	m = so->so_rcv.sb_mb;
2099 	/*
2100 	 * If we have less data than requested, block awaiting more (subject
2101 	 * to any timeout) if:
2102 	 *   1. the current count is less than the low water mark, or
2103 	 *   2. MSG_DONTWAIT is not set
2104 	 */
2105 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2106 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
2107 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2108 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2109 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
2110 		    ("receive: m == %p sbavail == %u",
2111 		    m, sbavail(&so->so_rcv)));
2112 		if (so->so_error || so->so_rerror) {
2113 			if (m != NULL)
2114 				goto dontblock;
2115 			if (so->so_error)
2116 				error = so->so_error;
2117 			else
2118 				error = so->so_rerror;
2119 			if ((flags & MSG_PEEK) == 0) {
2120 				if (so->so_error)
2121 					so->so_error = 0;
2122 				else
2123 					so->so_rerror = 0;
2124 			}
2125 			SOCKBUF_UNLOCK(&so->so_rcv);
2126 			goto release;
2127 		}
2128 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2129 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2130 			if (m != NULL)
2131 				goto dontblock;
2132 #ifdef KERN_TLS
2133 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2134 			    so->so_rcv.sb_tlscc == 0) {
2135 #else
2136 			else {
2137 #endif
2138 				SOCKBUF_UNLOCK(&so->so_rcv);
2139 				goto release;
2140 			}
2141 		}
2142 		for (; m != NULL; m = m->m_next)
2143 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2144 				m = so->so_rcv.sb_mb;
2145 				goto dontblock;
2146 			}
2147 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2148 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2149 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2150 			SOCKBUF_UNLOCK(&so->so_rcv);
2151 			error = ENOTCONN;
2152 			goto release;
2153 		}
2154 		if (uio->uio_resid == 0 && !report_real_len) {
2155 			SOCKBUF_UNLOCK(&so->so_rcv);
2156 			goto release;
2157 		}
2158 		if ((so->so_state & SS_NBIO) ||
2159 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2160 			SOCKBUF_UNLOCK(&so->so_rcv);
2161 			error = EWOULDBLOCK;
2162 			goto release;
2163 		}
2164 		SBLASTRECORDCHK(&so->so_rcv);
2165 		SBLASTMBUFCHK(&so->so_rcv);
2166 		error = sbwait(so, SO_RCV);
2167 		SOCKBUF_UNLOCK(&so->so_rcv);
2168 		if (error)
2169 			goto release;
2170 		goto restart;
2171 	}
2172 dontblock:
2173 	/*
2174 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2175 	 * pointer to the next record in the socket buffer.  We must keep the
2176 	 * various socket buffer pointers and local stack versions of the
2177 	 * pointers in sync, pushing out modifications before dropping the
2178 	 * socket buffer mutex, and re-reading them when picking it up.
2179 	 *
2180 	 * Otherwise, we will race with the network stack appending new data
2181 	 * or records onto the socket buffer by using inconsistent/stale
2182 	 * versions of the field, possibly resulting in socket buffer
2183 	 * corruption.
2184 	 *
2185 	 * By holding the high-level sblock(), we prevent simultaneous
2186 	 * readers from pulling off the front of the socket buffer.
2187 	 */
2188 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2189 	if (uio->uio_td)
2190 		uio->uio_td->td_ru.ru_msgrcv++;
2191 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2192 	SBLASTRECORDCHK(&so->so_rcv);
2193 	SBLASTMBUFCHK(&so->so_rcv);
2194 	nextrecord = m->m_nextpkt;
2195 	if (pr->pr_flags & PR_ADDR) {
2196 		KASSERT(m->m_type == MT_SONAME,
2197 		    ("m->m_type == %d", m->m_type));
2198 		orig_resid = 0;
2199 		if (psa != NULL)
2200 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2201 			    M_NOWAIT);
2202 		if (flags & MSG_PEEK) {
2203 			m = m->m_next;
2204 		} else {
2205 			sbfree(&so->so_rcv, m);
2206 			so->so_rcv.sb_mb = m_free(m);
2207 			m = so->so_rcv.sb_mb;
2208 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2209 		}
2210 	}
2211 
2212 	/*
2213 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2214 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2215 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2216 	 * perform externalization (or freeing if controlp == NULL).
2217 	 */
2218 	if (m != NULL && m->m_type == MT_CONTROL) {
2219 		struct mbuf *cm = NULL, *cmn;
2220 		struct mbuf **cme = &cm;
2221 #ifdef KERN_TLS
2222 		struct cmsghdr *cmsg;
2223 		struct tls_get_record tgr;
2224 
2225 		/*
2226 		 * For MSG_TLSAPPDATA, check for an alert record.
2227 		 * If found, return ENXIO without removing
2228 		 * it from the receive queue.  This allows a subsequent
2229 		 * call without MSG_TLSAPPDATA to receive it.
2230 		 * Note that, for TLS, there should only be a single
2231 		 * control mbuf with the TLS_GET_RECORD message in it.
2232 		 */
2233 		if (flags & MSG_TLSAPPDATA) {
2234 			cmsg = mtod(m, struct cmsghdr *);
2235 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2236 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2237 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2238 				if (__predict_false(tgr.tls_type ==
2239 				    TLS_RLTYPE_ALERT)) {
2240 					SOCKBUF_UNLOCK(&so->so_rcv);
2241 					error = ENXIO;
2242 					goto release;
2243 				}
2244 			}
2245 		}
2246 #endif
2247 
2248 		do {
2249 			if (flags & MSG_PEEK) {
2250 				if (controlp != NULL) {
2251 					*controlp = m_copym(m, 0, m->m_len,
2252 					    M_NOWAIT);
2253 					controlp = &(*controlp)->m_next;
2254 				}
2255 				m = m->m_next;
2256 			} else {
2257 				sbfree(&so->so_rcv, m);
2258 				so->so_rcv.sb_mb = m->m_next;
2259 				m->m_next = NULL;
2260 				*cme = m;
2261 				cme = &(*cme)->m_next;
2262 				m = so->so_rcv.sb_mb;
2263 			}
2264 		} while (m != NULL && m->m_type == MT_CONTROL);
2265 		if ((flags & MSG_PEEK) == 0)
2266 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2267 		while (cm != NULL) {
2268 			cmn = cm->m_next;
2269 			cm->m_next = NULL;
2270 			if (pr->pr_domain->dom_externalize != NULL) {
2271 				SOCKBUF_UNLOCK(&so->so_rcv);
2272 				VNET_SO_ASSERT(so);
2273 				error = (*pr->pr_domain->dom_externalize)
2274 				    (cm, controlp, flags);
2275 				SOCKBUF_LOCK(&so->so_rcv);
2276 			} else if (controlp != NULL)
2277 				*controlp = cm;
2278 			else
2279 				m_freem(cm);
2280 			if (controlp != NULL) {
2281 				while (*controlp != NULL)
2282 					controlp = &(*controlp)->m_next;
2283 			}
2284 			cm = cmn;
2285 		}
2286 		if (m != NULL)
2287 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2288 		else
2289 			nextrecord = so->so_rcv.sb_mb;
2290 		orig_resid = 0;
2291 	}
2292 	if (m != NULL) {
2293 		if ((flags & MSG_PEEK) == 0) {
2294 			KASSERT(m->m_nextpkt == nextrecord,
2295 			    ("soreceive: post-control, nextrecord !sync"));
2296 			if (nextrecord == NULL) {
2297 				KASSERT(so->so_rcv.sb_mb == m,
2298 				    ("soreceive: post-control, sb_mb!=m"));
2299 				KASSERT(so->so_rcv.sb_lastrecord == m,
2300 				    ("soreceive: post-control, lastrecord!=m"));
2301 			}
2302 		}
2303 		type = m->m_type;
2304 		if (type == MT_OOBDATA)
2305 			flags |= MSG_OOB;
2306 	} else {
2307 		if ((flags & MSG_PEEK) == 0) {
2308 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2309 			    ("soreceive: sb_mb != nextrecord"));
2310 			if (so->so_rcv.sb_mb == NULL) {
2311 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2312 				    ("soreceive: sb_lastercord != NULL"));
2313 			}
2314 		}
2315 	}
2316 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2317 	SBLASTRECORDCHK(&so->so_rcv);
2318 	SBLASTMBUFCHK(&so->so_rcv);
2319 
2320 	/*
2321 	 * Now continue to read any data mbufs off of the head of the socket
2322 	 * buffer until the read request is satisfied.  Note that 'type' is
2323 	 * used to store the type of any mbuf reads that have happened so far
2324 	 * such that soreceive() can stop reading if the type changes, which
2325 	 * causes soreceive() to return only one of regular data and inline
2326 	 * out-of-band data in a single socket receive operation.
2327 	 */
2328 	moff = 0;
2329 	offset = 0;
2330 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2331 	    && error == 0) {
2332 		/*
2333 		 * If the type of mbuf has changed since the last mbuf
2334 		 * examined ('type'), end the receive operation.
2335 		 */
2336 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2337 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2338 			if (type != m->m_type)
2339 				break;
2340 		} else if (type == MT_OOBDATA)
2341 			break;
2342 		else
2343 		    KASSERT(m->m_type == MT_DATA,
2344 			("m->m_type == %d", m->m_type));
2345 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2346 		len = uio->uio_resid;
2347 		if (so->so_oobmark && len > so->so_oobmark - offset)
2348 			len = so->so_oobmark - offset;
2349 		if (len > m->m_len - moff)
2350 			len = m->m_len - moff;
2351 		/*
2352 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2353 		 * them out via the uio, then free.  Sockbuf must be
2354 		 * consistent here (points to current mbuf, it points to next
2355 		 * record) when we drop priority; we must note any additions
2356 		 * to the sockbuf when we block interrupts again.
2357 		 */
2358 		if (mp == NULL) {
2359 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2360 			SBLASTRECORDCHK(&so->so_rcv);
2361 			SBLASTMBUFCHK(&so->so_rcv);
2362 			SOCKBUF_UNLOCK(&so->so_rcv);
2363 			if ((m->m_flags & M_EXTPG) != 0)
2364 				error = m_unmapped_uiomove(m, moff, uio,
2365 				    (int)len);
2366 			else
2367 				error = uiomove(mtod(m, char *) + moff,
2368 				    (int)len, uio);
2369 			SOCKBUF_LOCK(&so->so_rcv);
2370 			if (error) {
2371 				/*
2372 				 * The MT_SONAME mbuf has already been removed
2373 				 * from the record, so it is necessary to
2374 				 * remove the data mbufs, if any, to preserve
2375 				 * the invariant in the case of PR_ADDR that
2376 				 * requires MT_SONAME mbufs at the head of
2377 				 * each record.
2378 				 */
2379 				if (pr->pr_flags & PR_ATOMIC &&
2380 				    ((flags & MSG_PEEK) == 0))
2381 					(void)sbdroprecord_locked(&so->so_rcv);
2382 				SOCKBUF_UNLOCK(&so->so_rcv);
2383 				goto release;
2384 			}
2385 		} else
2386 			uio->uio_resid -= len;
2387 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2388 		if (len == m->m_len - moff) {
2389 			if (m->m_flags & M_EOR)
2390 				flags |= MSG_EOR;
2391 			if (flags & MSG_PEEK) {
2392 				m = m->m_next;
2393 				moff = 0;
2394 			} else {
2395 				nextrecord = m->m_nextpkt;
2396 				sbfree(&so->so_rcv, m);
2397 				if (mp != NULL) {
2398 					m->m_nextpkt = NULL;
2399 					*mp = m;
2400 					mp = &m->m_next;
2401 					so->so_rcv.sb_mb = m = m->m_next;
2402 					*mp = NULL;
2403 				} else {
2404 					so->so_rcv.sb_mb = m_free(m);
2405 					m = so->so_rcv.sb_mb;
2406 				}
2407 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2408 				SBLASTRECORDCHK(&so->so_rcv);
2409 				SBLASTMBUFCHK(&so->so_rcv);
2410 			}
2411 		} else {
2412 			if (flags & MSG_PEEK)
2413 				moff += len;
2414 			else {
2415 				if (mp != NULL) {
2416 					if (flags & MSG_DONTWAIT) {
2417 						*mp = m_copym(m, 0, len,
2418 						    M_NOWAIT);
2419 						if (*mp == NULL) {
2420 							/*
2421 							 * m_copym() couldn't
2422 							 * allocate an mbuf.
2423 							 * Adjust uio_resid back
2424 							 * (it was adjusted
2425 							 * down by len bytes,
2426 							 * which we didn't end
2427 							 * up "copying" over).
2428 							 */
2429 							uio->uio_resid += len;
2430 							break;
2431 						}
2432 					} else {
2433 						SOCKBUF_UNLOCK(&so->so_rcv);
2434 						*mp = m_copym(m, 0, len,
2435 						    M_WAITOK);
2436 						SOCKBUF_LOCK(&so->so_rcv);
2437 					}
2438 				}
2439 				sbcut_locked(&so->so_rcv, len);
2440 			}
2441 		}
2442 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2443 		if (so->so_oobmark) {
2444 			if ((flags & MSG_PEEK) == 0) {
2445 				so->so_oobmark -= len;
2446 				if (so->so_oobmark == 0) {
2447 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2448 					break;
2449 				}
2450 			} else {
2451 				offset += len;
2452 				if (offset == so->so_oobmark)
2453 					break;
2454 			}
2455 		}
2456 		if (flags & MSG_EOR)
2457 			break;
2458 		/*
2459 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2460 		 * must not quit until "uio->uio_resid == 0" or an error
2461 		 * termination.  If a signal/timeout occurs, return with a
2462 		 * short count but without error.  Keep sockbuf locked
2463 		 * against other readers.
2464 		 */
2465 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2466 		    !sosendallatonce(so) && nextrecord == NULL) {
2467 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2468 			if (so->so_error || so->so_rerror ||
2469 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2470 				break;
2471 			/*
2472 			 * Notify the protocol that some data has been
2473 			 * drained before blocking.
2474 			 */
2475 			if (pr->pr_flags & PR_WANTRCVD) {
2476 				SOCKBUF_UNLOCK(&so->so_rcv);
2477 				VNET_SO_ASSERT(so);
2478 				pr->pr_rcvd(so, flags);
2479 				SOCKBUF_LOCK(&so->so_rcv);
2480 				if (__predict_false(so->so_rcv.sb_mb == NULL &&
2481 				    (so->so_error || so->so_rerror ||
2482 				    so->so_rcv.sb_state & SBS_CANTRCVMORE)))
2483 					break;
2484 			}
2485 			SBLASTRECORDCHK(&so->so_rcv);
2486 			SBLASTMBUFCHK(&so->so_rcv);
2487 			/*
2488 			 * We could receive some data while was notifying
2489 			 * the protocol. Skip blocking in this case.
2490 			 */
2491 			if (so->so_rcv.sb_mb == NULL) {
2492 				error = sbwait(so, SO_RCV);
2493 				if (error) {
2494 					SOCKBUF_UNLOCK(&so->so_rcv);
2495 					goto release;
2496 				}
2497 			}
2498 			m = so->so_rcv.sb_mb;
2499 			if (m != NULL)
2500 				nextrecord = m->m_nextpkt;
2501 		}
2502 	}
2503 
2504 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2505 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2506 		if (report_real_len)
2507 			uio->uio_resid -= m_length(m, NULL) - moff;
2508 		flags |= MSG_TRUNC;
2509 		if ((flags & MSG_PEEK) == 0)
2510 			(void) sbdroprecord_locked(&so->so_rcv);
2511 	}
2512 	if ((flags & MSG_PEEK) == 0) {
2513 		if (m == NULL) {
2514 			/*
2515 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2516 			 * part makes sure sb_lastrecord is up-to-date if
2517 			 * there is still data in the socket buffer.
2518 			 */
2519 			so->so_rcv.sb_mb = nextrecord;
2520 			if (so->so_rcv.sb_mb == NULL) {
2521 				so->so_rcv.sb_mbtail = NULL;
2522 				so->so_rcv.sb_lastrecord = NULL;
2523 			} else if (nextrecord->m_nextpkt == NULL)
2524 				so->so_rcv.sb_lastrecord = nextrecord;
2525 		}
2526 		SBLASTRECORDCHK(&so->so_rcv);
2527 		SBLASTMBUFCHK(&so->so_rcv);
2528 		/*
2529 		 * If soreceive() is being done from the socket callback,
2530 		 * then don't need to generate ACK to peer to update window,
2531 		 * since ACK will be generated on return to TCP.
2532 		 */
2533 		if (!(flags & MSG_SOCALLBCK) &&
2534 		    (pr->pr_flags & PR_WANTRCVD)) {
2535 			SOCKBUF_UNLOCK(&so->so_rcv);
2536 			VNET_SO_ASSERT(so);
2537 			pr->pr_rcvd(so, flags);
2538 			SOCKBUF_LOCK(&so->so_rcv);
2539 		}
2540 	}
2541 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2542 	if (orig_resid == uio->uio_resid && orig_resid &&
2543 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2544 		SOCKBUF_UNLOCK(&so->so_rcv);
2545 		goto restart;
2546 	}
2547 	SOCKBUF_UNLOCK(&so->so_rcv);
2548 
2549 	if (flagsp != NULL)
2550 		*flagsp |= flags;
2551 release:
2552 	SOCK_IO_RECV_UNLOCK(so);
2553 	return (error);
2554 }
2555 
2556 /*
2557  * Optimized version of soreceive() for stream (TCP) sockets.
2558  */
2559 int
2560 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2561     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2562 {
2563 	int len = 0, error = 0, flags, oresid;
2564 	struct sockbuf *sb;
2565 	struct mbuf *m, *n = NULL;
2566 
2567 	/* We only do stream sockets. */
2568 	if (so->so_type != SOCK_STREAM)
2569 		return (EINVAL);
2570 	if (psa != NULL)
2571 		*psa = NULL;
2572 	if (flagsp != NULL)
2573 		flags = *flagsp &~ MSG_EOR;
2574 	else
2575 		flags = 0;
2576 	if (controlp != NULL)
2577 		*controlp = NULL;
2578 	if (flags & MSG_OOB)
2579 		return (soreceive_rcvoob(so, uio, flags));
2580 	if (mp0 != NULL)
2581 		*mp0 = NULL;
2582 
2583 	sb = &so->so_rcv;
2584 
2585 #ifdef KERN_TLS
2586 	/*
2587 	 * KTLS store TLS records as records with a control message to
2588 	 * describe the framing.
2589 	 *
2590 	 * We check once here before acquiring locks to optimize the
2591 	 * common case.
2592 	 */
2593 	if (sb->sb_tls_info != NULL)
2594 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2595 		    flagsp));
2596 #endif
2597 
2598 	/* Prevent other readers from entering the socket. */
2599 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2600 	if (error)
2601 		return (error);
2602 	SOCKBUF_LOCK(sb);
2603 
2604 #ifdef KERN_TLS
2605 	if (sb->sb_tls_info != NULL) {
2606 		SOCKBUF_UNLOCK(sb);
2607 		SOCK_IO_RECV_UNLOCK(so);
2608 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2609 		    flagsp));
2610 	}
2611 #endif
2612 
2613 	/* Easy one, no space to copyout anything. */
2614 	if (uio->uio_resid == 0) {
2615 		error = EINVAL;
2616 		goto out;
2617 	}
2618 	oresid = uio->uio_resid;
2619 
2620 	/* We will never ever get anything unless we are or were connected. */
2621 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2622 		error = ENOTCONN;
2623 		goto out;
2624 	}
2625 
2626 restart:
2627 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2628 
2629 	/* Abort if socket has reported problems. */
2630 	if (so->so_error) {
2631 		if (sbavail(sb) > 0)
2632 			goto deliver;
2633 		if (oresid > uio->uio_resid)
2634 			goto out;
2635 		error = so->so_error;
2636 		if (!(flags & MSG_PEEK))
2637 			so->so_error = 0;
2638 		goto out;
2639 	}
2640 
2641 	/* Door is closed.  Deliver what is left, if any. */
2642 	if (sb->sb_state & SBS_CANTRCVMORE) {
2643 		if (sbavail(sb) > 0)
2644 			goto deliver;
2645 		else
2646 			goto out;
2647 	}
2648 
2649 	/* Socket buffer is empty and we shall not block. */
2650 	if (sbavail(sb) == 0 &&
2651 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2652 		error = EAGAIN;
2653 		goto out;
2654 	}
2655 
2656 	/* Socket buffer got some data that we shall deliver now. */
2657 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2658 	    ((so->so_state & SS_NBIO) ||
2659 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2660 	     sbavail(sb) >= sb->sb_lowat ||
2661 	     sbavail(sb) >= uio->uio_resid ||
2662 	     sbavail(sb) >= sb->sb_hiwat) ) {
2663 		goto deliver;
2664 	}
2665 
2666 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2667 	if ((flags & MSG_WAITALL) &&
2668 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2669 		goto deliver;
2670 
2671 	/*
2672 	 * Wait and block until (more) data comes in.
2673 	 * NB: Drops the sockbuf lock during wait.
2674 	 */
2675 	error = sbwait(so, SO_RCV);
2676 	if (error)
2677 		goto out;
2678 	goto restart;
2679 
2680 deliver:
2681 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2682 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2683 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2684 
2685 	/* Statistics. */
2686 	if (uio->uio_td)
2687 		uio->uio_td->td_ru.ru_msgrcv++;
2688 
2689 	/* Fill uio until full or current end of socket buffer is reached. */
2690 	len = min(uio->uio_resid, sbavail(sb));
2691 	if (mp0 != NULL) {
2692 		/* Dequeue as many mbufs as possible. */
2693 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2694 			if (*mp0 == NULL)
2695 				*mp0 = sb->sb_mb;
2696 			else
2697 				m_cat(*mp0, sb->sb_mb);
2698 			for (m = sb->sb_mb;
2699 			     m != NULL && m->m_len <= len;
2700 			     m = m->m_next) {
2701 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2702 				    ("%s: m %p not available", __func__, m));
2703 				len -= m->m_len;
2704 				uio->uio_resid -= m->m_len;
2705 				sbfree(sb, m);
2706 				n = m;
2707 			}
2708 			n->m_next = NULL;
2709 			sb->sb_mb = m;
2710 			sb->sb_lastrecord = sb->sb_mb;
2711 			if (sb->sb_mb == NULL)
2712 				SB_EMPTY_FIXUP(sb);
2713 		}
2714 		/* Copy the remainder. */
2715 		if (len > 0) {
2716 			KASSERT(sb->sb_mb != NULL,
2717 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2718 
2719 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2720 			if (m == NULL)
2721 				len = 0;	/* Don't flush data from sockbuf. */
2722 			else
2723 				uio->uio_resid -= len;
2724 			if (*mp0 != NULL)
2725 				m_cat(*mp0, m);
2726 			else
2727 				*mp0 = m;
2728 			if (*mp0 == NULL) {
2729 				error = ENOBUFS;
2730 				goto out;
2731 			}
2732 		}
2733 	} else {
2734 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2735 		SOCKBUF_UNLOCK(sb);
2736 		error = m_mbuftouio(uio, sb->sb_mb, len);
2737 		SOCKBUF_LOCK(sb);
2738 		if (error)
2739 			goto out;
2740 	}
2741 	SBLASTRECORDCHK(sb);
2742 	SBLASTMBUFCHK(sb);
2743 
2744 	/*
2745 	 * Remove the delivered data from the socket buffer unless we
2746 	 * were only peeking.
2747 	 */
2748 	if (!(flags & MSG_PEEK)) {
2749 		if (len > 0)
2750 			sbdrop_locked(sb, len);
2751 
2752 		/* Notify protocol that we drained some data. */
2753 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2754 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2755 		     !(flags & MSG_SOCALLBCK))) {
2756 			SOCKBUF_UNLOCK(sb);
2757 			VNET_SO_ASSERT(so);
2758 			so->so_proto->pr_rcvd(so, flags);
2759 			SOCKBUF_LOCK(sb);
2760 		}
2761 	}
2762 
2763 	/*
2764 	 * For MSG_WAITALL we may have to loop again and wait for
2765 	 * more data to come in.
2766 	 */
2767 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2768 		goto restart;
2769 out:
2770 	SBLASTRECORDCHK(sb);
2771 	SBLASTMBUFCHK(sb);
2772 	SOCKBUF_UNLOCK(sb);
2773 	SOCK_IO_RECV_UNLOCK(so);
2774 	return (error);
2775 }
2776 
2777 /*
2778  * Optimized version of soreceive() for simple datagram cases from userspace.
2779  * Unlike in the stream case, we're able to drop a datagram if copyout()
2780  * fails, and because we handle datagrams atomically, we don't need to use a
2781  * sleep lock to prevent I/O interlacing.
2782  */
2783 int
2784 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2785     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2786 {
2787 	struct mbuf *m, *m2;
2788 	int flags, error;
2789 	ssize_t len;
2790 	struct protosw *pr = so->so_proto;
2791 	struct mbuf *nextrecord;
2792 
2793 	if (psa != NULL)
2794 		*psa = NULL;
2795 	if (controlp != NULL)
2796 		*controlp = NULL;
2797 	if (flagsp != NULL)
2798 		flags = *flagsp &~ MSG_EOR;
2799 	else
2800 		flags = 0;
2801 
2802 	/*
2803 	 * For any complicated cases, fall back to the full
2804 	 * soreceive_generic().
2805 	 */
2806 	if (mp0 != NULL || (flags & (MSG_PEEK | MSG_OOB | MSG_TRUNC)))
2807 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2808 		    flagsp));
2809 
2810 	/*
2811 	 * Enforce restrictions on use.
2812 	 */
2813 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2814 	    ("soreceive_dgram: wantrcvd"));
2815 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2816 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2817 	    ("soreceive_dgram: SBS_RCVATMARK"));
2818 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2819 	    ("soreceive_dgram: P_CONNREQUIRED"));
2820 
2821 	/*
2822 	 * Loop blocking while waiting for a datagram.
2823 	 */
2824 	SOCKBUF_LOCK(&so->so_rcv);
2825 	while ((m = so->so_rcv.sb_mb) == NULL) {
2826 		KASSERT(sbavail(&so->so_rcv) == 0,
2827 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2828 		    sbavail(&so->so_rcv)));
2829 		if (so->so_error) {
2830 			error = so->so_error;
2831 			so->so_error = 0;
2832 			SOCKBUF_UNLOCK(&so->so_rcv);
2833 			return (error);
2834 		}
2835 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2836 		    uio->uio_resid == 0) {
2837 			SOCKBUF_UNLOCK(&so->so_rcv);
2838 			return (0);
2839 		}
2840 		if ((so->so_state & SS_NBIO) ||
2841 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2842 			SOCKBUF_UNLOCK(&so->so_rcv);
2843 			return (EWOULDBLOCK);
2844 		}
2845 		SBLASTRECORDCHK(&so->so_rcv);
2846 		SBLASTMBUFCHK(&so->so_rcv);
2847 		error = sbwait(so, SO_RCV);
2848 		if (error) {
2849 			SOCKBUF_UNLOCK(&so->so_rcv);
2850 			return (error);
2851 		}
2852 	}
2853 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2854 
2855 	if (uio->uio_td)
2856 		uio->uio_td->td_ru.ru_msgrcv++;
2857 	SBLASTRECORDCHK(&so->so_rcv);
2858 	SBLASTMBUFCHK(&so->so_rcv);
2859 	nextrecord = m->m_nextpkt;
2860 	if (nextrecord == NULL) {
2861 		KASSERT(so->so_rcv.sb_lastrecord == m,
2862 		    ("soreceive_dgram: lastrecord != m"));
2863 	}
2864 
2865 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2866 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2867 
2868 	/*
2869 	 * Pull 'm' and its chain off the front of the packet queue.
2870 	 */
2871 	so->so_rcv.sb_mb = NULL;
2872 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2873 
2874 	/*
2875 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2876 	 */
2877 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2878 		sbfree(&so->so_rcv, m2);
2879 
2880 	/*
2881 	 * Do a few last checks before we let go of the lock.
2882 	 */
2883 	SBLASTRECORDCHK(&so->so_rcv);
2884 	SBLASTMBUFCHK(&so->so_rcv);
2885 	SOCKBUF_UNLOCK(&so->so_rcv);
2886 
2887 	if (pr->pr_flags & PR_ADDR) {
2888 		KASSERT(m->m_type == MT_SONAME,
2889 		    ("m->m_type == %d", m->m_type));
2890 		if (psa != NULL)
2891 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2892 			    M_WAITOK);
2893 		m = m_free(m);
2894 	}
2895 	KASSERT(m, ("%s: no data or control after soname", __func__));
2896 
2897 	/*
2898 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2899 	 * queue.
2900 	 *
2901 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2902 	 * in the first mbuf chain on the socket buffer.  We call into the
2903 	 * protocol to perform externalization (or freeing if controlp ==
2904 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2905 	 * MT_DATA mbufs.
2906 	 */
2907 	if (m->m_type == MT_CONTROL) {
2908 		struct mbuf *cm = NULL, *cmn;
2909 		struct mbuf **cme = &cm;
2910 
2911 		do {
2912 			m2 = m->m_next;
2913 			m->m_next = NULL;
2914 			*cme = m;
2915 			cme = &(*cme)->m_next;
2916 			m = m2;
2917 		} while (m != NULL && m->m_type == MT_CONTROL);
2918 		while (cm != NULL) {
2919 			cmn = cm->m_next;
2920 			cm->m_next = NULL;
2921 			if (pr->pr_domain->dom_externalize != NULL) {
2922 				error = (*pr->pr_domain->dom_externalize)
2923 				    (cm, controlp, flags);
2924 			} else if (controlp != NULL)
2925 				*controlp = cm;
2926 			else
2927 				m_freem(cm);
2928 			if (controlp != NULL) {
2929 				while (*controlp != NULL)
2930 					controlp = &(*controlp)->m_next;
2931 			}
2932 			cm = cmn;
2933 		}
2934 	}
2935 	KASSERT(m == NULL || m->m_type == MT_DATA,
2936 	    ("soreceive_dgram: !data"));
2937 	while (m != NULL && uio->uio_resid > 0) {
2938 		len = uio->uio_resid;
2939 		if (len > m->m_len)
2940 			len = m->m_len;
2941 		error = uiomove(mtod(m, char *), (int)len, uio);
2942 		if (error) {
2943 			m_freem(m);
2944 			return (error);
2945 		}
2946 		if (len == m->m_len)
2947 			m = m_free(m);
2948 		else {
2949 			m->m_data += len;
2950 			m->m_len -= len;
2951 		}
2952 	}
2953 	if (m != NULL) {
2954 		flags |= MSG_TRUNC;
2955 		m_freem(m);
2956 	}
2957 	if (flagsp != NULL)
2958 		*flagsp |= flags;
2959 	return (0);
2960 }
2961 
2962 int
2963 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2964     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2965 {
2966 	int error;
2967 
2968 	CURVNET_SET(so->so_vnet);
2969 	error = so->so_proto->pr_soreceive(so, psa, uio, mp0, controlp, flagsp);
2970 	CURVNET_RESTORE();
2971 	return (error);
2972 }
2973 
2974 int
2975 soshutdown(struct socket *so, enum shutdown_how how)
2976 {
2977 	int error;
2978 
2979 	CURVNET_SET(so->so_vnet);
2980 	error = so->so_proto->pr_shutdown(so, how);
2981 	CURVNET_RESTORE();
2982 
2983 	return (error);
2984 }
2985 
2986 /*
2987  * Used by several pr_shutdown implementations that use generic socket buffers.
2988  */
2989 void
2990 sorflush(struct socket *so)
2991 {
2992 	int error;
2993 
2994 	VNET_SO_ASSERT(so);
2995 
2996 	/*
2997 	 * Dislodge threads currently blocked in receive and wait to acquire
2998 	 * a lock against other simultaneous readers before clearing the
2999 	 * socket buffer.  Don't let our acquire be interrupted by a signal
3000 	 * despite any existing socket disposition on interruptable waiting.
3001 	 *
3002 	 * The SOCK_IO_RECV_LOCK() is important here as there some pr_soreceive
3003 	 * methods that read the top of the socket buffer without acquisition
3004 	 * of the socket buffer mutex, assuming that top of the buffer
3005 	 * exclusively belongs to the read(2) syscall.  This is handy when
3006 	 * performing MSG_PEEK.
3007 	 */
3008 	socantrcvmore(so);
3009 
3010 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
3011 	if (error != 0) {
3012 		KASSERT(SOLISTENING(so),
3013 		    ("%s: soiolock(%p) failed", __func__, so));
3014 		return;
3015 	}
3016 
3017 	sbrelease(so, SO_RCV);
3018 	SOCK_IO_RECV_UNLOCK(so);
3019 
3020 }
3021 
3022 /*
3023  * Wrapper for Socket established helper hook.
3024  * Parameters: socket, context of the hook point, hook id.
3025  */
3026 static int inline
3027 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
3028 {
3029 	struct socket_hhook_data hhook_data = {
3030 		.so = so,
3031 		.hctx = hctx,
3032 		.m = NULL,
3033 		.status = 0
3034 	};
3035 
3036 	CURVNET_SET(so->so_vnet);
3037 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
3038 	CURVNET_RESTORE();
3039 
3040 	/* Ugly but needed, since hhooks return void for now */
3041 	return (hhook_data.status);
3042 }
3043 
3044 /*
3045  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
3046  * additional variant to handle the case where the option value needs to be
3047  * some kind of integer, but not a specific size.  In addition to their use
3048  * here, these functions are also called by the protocol-level pr_ctloutput()
3049  * routines.
3050  */
3051 int
3052 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3053 {
3054 	size_t	valsize;
3055 
3056 	/*
3057 	 * If the user gives us more than we wanted, we ignore it, but if we
3058 	 * don't get the minimum length the caller wants, we return EINVAL.
3059 	 * On success, sopt->sopt_valsize is set to however much we actually
3060 	 * retrieved.
3061 	 */
3062 	if ((valsize = sopt->sopt_valsize) < minlen)
3063 		return EINVAL;
3064 	if (valsize > len)
3065 		sopt->sopt_valsize = valsize = len;
3066 
3067 	if (sopt->sopt_td != NULL)
3068 		return (copyin(sopt->sopt_val, buf, valsize));
3069 
3070 	bcopy(sopt->sopt_val, buf, valsize);
3071 	return (0);
3072 }
3073 
3074 /*
3075  * Kernel version of setsockopt(2).
3076  *
3077  * XXX: optlen is size_t, not socklen_t
3078  */
3079 int
3080 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3081     size_t optlen)
3082 {
3083 	struct sockopt sopt;
3084 
3085 	sopt.sopt_level = level;
3086 	sopt.sopt_name = optname;
3087 	sopt.sopt_dir = SOPT_SET;
3088 	sopt.sopt_val = optval;
3089 	sopt.sopt_valsize = optlen;
3090 	sopt.sopt_td = NULL;
3091 	return (sosetopt(so, &sopt));
3092 }
3093 
3094 int
3095 sosetopt(struct socket *so, struct sockopt *sopt)
3096 {
3097 	int	error, optval;
3098 	struct	linger l;
3099 	struct	timeval tv;
3100 	sbintime_t val, *valp;
3101 	uint32_t val32;
3102 #ifdef MAC
3103 	struct mac extmac;
3104 #endif
3105 
3106 	CURVNET_SET(so->so_vnet);
3107 	error = 0;
3108 	if (sopt->sopt_level != SOL_SOCKET) {
3109 		if (so->so_proto->pr_ctloutput != NULL)
3110 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3111 		else
3112 			error = ENOPROTOOPT;
3113 	} else {
3114 		switch (sopt->sopt_name) {
3115 		case SO_ACCEPTFILTER:
3116 			error = accept_filt_setopt(so, sopt);
3117 			if (error)
3118 				goto bad;
3119 			break;
3120 
3121 		case SO_LINGER:
3122 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3123 			if (error)
3124 				goto bad;
3125 			if (l.l_linger < 0 ||
3126 			    l.l_linger > USHRT_MAX ||
3127 			    l.l_linger > (INT_MAX / hz)) {
3128 				error = EDOM;
3129 				goto bad;
3130 			}
3131 			SOCK_LOCK(so);
3132 			so->so_linger = l.l_linger;
3133 			if (l.l_onoff)
3134 				so->so_options |= SO_LINGER;
3135 			else
3136 				so->so_options &= ~SO_LINGER;
3137 			SOCK_UNLOCK(so);
3138 			break;
3139 
3140 		case SO_DEBUG:
3141 		case SO_KEEPALIVE:
3142 		case SO_DONTROUTE:
3143 		case SO_USELOOPBACK:
3144 		case SO_BROADCAST:
3145 		case SO_REUSEADDR:
3146 		case SO_REUSEPORT:
3147 		case SO_REUSEPORT_LB:
3148 		case SO_OOBINLINE:
3149 		case SO_TIMESTAMP:
3150 		case SO_BINTIME:
3151 		case SO_NOSIGPIPE:
3152 		case SO_NO_DDP:
3153 		case SO_NO_OFFLOAD:
3154 		case SO_RERROR:
3155 			error = sooptcopyin(sopt, &optval, sizeof optval,
3156 			    sizeof optval);
3157 			if (error)
3158 				goto bad;
3159 			SOCK_LOCK(so);
3160 			if (optval)
3161 				so->so_options |= sopt->sopt_name;
3162 			else
3163 				so->so_options &= ~sopt->sopt_name;
3164 			SOCK_UNLOCK(so);
3165 			break;
3166 
3167 		case SO_SETFIB:
3168 			error = sooptcopyin(sopt, &optval, sizeof optval,
3169 			    sizeof optval);
3170 			if (error)
3171 				goto bad;
3172 
3173 			if (optval < 0 || optval >= rt_numfibs) {
3174 				error = EINVAL;
3175 				goto bad;
3176 			}
3177 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3178 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3179 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3180 				so->so_fibnum = optval;
3181 			else
3182 				so->so_fibnum = 0;
3183 			break;
3184 
3185 		case SO_USER_COOKIE:
3186 			error = sooptcopyin(sopt, &val32, sizeof val32,
3187 			    sizeof val32);
3188 			if (error)
3189 				goto bad;
3190 			so->so_user_cookie = val32;
3191 			break;
3192 
3193 		case SO_SNDBUF:
3194 		case SO_RCVBUF:
3195 		case SO_SNDLOWAT:
3196 		case SO_RCVLOWAT:
3197 			error = so->so_proto->pr_setsbopt(so, sopt);
3198 			if (error)
3199 				goto bad;
3200 			break;
3201 
3202 		case SO_SNDTIMEO:
3203 		case SO_RCVTIMEO:
3204 #ifdef COMPAT_FREEBSD32
3205 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3206 				struct timeval32 tv32;
3207 
3208 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3209 				    sizeof tv32);
3210 				CP(tv32, tv, tv_sec);
3211 				CP(tv32, tv, tv_usec);
3212 			} else
3213 #endif
3214 				error = sooptcopyin(sopt, &tv, sizeof tv,
3215 				    sizeof tv);
3216 			if (error)
3217 				goto bad;
3218 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3219 			    tv.tv_usec >= 1000000) {
3220 				error = EDOM;
3221 				goto bad;
3222 			}
3223 			if (tv.tv_sec > INT32_MAX)
3224 				val = SBT_MAX;
3225 			else
3226 				val = tvtosbt(tv);
3227 			SOCK_LOCK(so);
3228 			valp = sopt->sopt_name == SO_SNDTIMEO ?
3229 			    (SOLISTENING(so) ? &so->sol_sbsnd_timeo :
3230 			    &so->so_snd.sb_timeo) :
3231 			    (SOLISTENING(so) ? &so->sol_sbrcv_timeo :
3232 			    &so->so_rcv.sb_timeo);
3233 			*valp = val;
3234 			SOCK_UNLOCK(so);
3235 			break;
3236 
3237 		case SO_LABEL:
3238 #ifdef MAC
3239 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3240 			    sizeof extmac);
3241 			if (error)
3242 				goto bad;
3243 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3244 			    so, &extmac);
3245 #else
3246 			error = EOPNOTSUPP;
3247 #endif
3248 			break;
3249 
3250 		case SO_TS_CLOCK:
3251 			error = sooptcopyin(sopt, &optval, sizeof optval,
3252 			    sizeof optval);
3253 			if (error)
3254 				goto bad;
3255 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3256 				error = EINVAL;
3257 				goto bad;
3258 			}
3259 			so->so_ts_clock = optval;
3260 			break;
3261 
3262 		case SO_MAX_PACING_RATE:
3263 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3264 			    sizeof(val32));
3265 			if (error)
3266 				goto bad;
3267 			so->so_max_pacing_rate = val32;
3268 			break;
3269 
3270 		default:
3271 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3272 				error = hhook_run_socket(so, sopt,
3273 				    HHOOK_SOCKET_OPT);
3274 			else
3275 				error = ENOPROTOOPT;
3276 			break;
3277 		}
3278 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3279 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3280 	}
3281 bad:
3282 	CURVNET_RESTORE();
3283 	return (error);
3284 }
3285 
3286 /*
3287  * Helper routine for getsockopt.
3288  */
3289 int
3290 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3291 {
3292 	int	error;
3293 	size_t	valsize;
3294 
3295 	error = 0;
3296 
3297 	/*
3298 	 * Documented get behavior is that we always return a value, possibly
3299 	 * truncated to fit in the user's buffer.  Traditional behavior is
3300 	 * that we always tell the user precisely how much we copied, rather
3301 	 * than something useful like the total amount we had available for
3302 	 * her.  Note that this interface is not idempotent; the entire
3303 	 * answer must be generated ahead of time.
3304 	 */
3305 	valsize = min(len, sopt->sopt_valsize);
3306 	sopt->sopt_valsize = valsize;
3307 	if (sopt->sopt_val != NULL) {
3308 		if (sopt->sopt_td != NULL)
3309 			error = copyout(buf, sopt->sopt_val, valsize);
3310 		else
3311 			bcopy(buf, sopt->sopt_val, valsize);
3312 	}
3313 	return (error);
3314 }
3315 
3316 int
3317 sogetopt(struct socket *so, struct sockopt *sopt)
3318 {
3319 	int	error, optval;
3320 	struct	linger l;
3321 	struct	timeval tv;
3322 #ifdef MAC
3323 	struct mac extmac;
3324 #endif
3325 
3326 	CURVNET_SET(so->so_vnet);
3327 	error = 0;
3328 	if (sopt->sopt_level != SOL_SOCKET) {
3329 		if (so->so_proto->pr_ctloutput != NULL)
3330 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3331 		else
3332 			error = ENOPROTOOPT;
3333 		CURVNET_RESTORE();
3334 		return (error);
3335 	} else {
3336 		switch (sopt->sopt_name) {
3337 		case SO_ACCEPTFILTER:
3338 			error = accept_filt_getopt(so, sopt);
3339 			break;
3340 
3341 		case SO_LINGER:
3342 			SOCK_LOCK(so);
3343 			l.l_onoff = so->so_options & SO_LINGER;
3344 			l.l_linger = so->so_linger;
3345 			SOCK_UNLOCK(so);
3346 			error = sooptcopyout(sopt, &l, sizeof l);
3347 			break;
3348 
3349 		case SO_USELOOPBACK:
3350 		case SO_DONTROUTE:
3351 		case SO_DEBUG:
3352 		case SO_KEEPALIVE:
3353 		case SO_REUSEADDR:
3354 		case SO_REUSEPORT:
3355 		case SO_REUSEPORT_LB:
3356 		case SO_BROADCAST:
3357 		case SO_OOBINLINE:
3358 		case SO_ACCEPTCONN:
3359 		case SO_TIMESTAMP:
3360 		case SO_BINTIME:
3361 		case SO_NOSIGPIPE:
3362 		case SO_NO_DDP:
3363 		case SO_NO_OFFLOAD:
3364 		case SO_RERROR:
3365 			optval = so->so_options & sopt->sopt_name;
3366 integer:
3367 			error = sooptcopyout(sopt, &optval, sizeof optval);
3368 			break;
3369 
3370 		case SO_DOMAIN:
3371 			optval = so->so_proto->pr_domain->dom_family;
3372 			goto integer;
3373 
3374 		case SO_TYPE:
3375 			optval = so->so_type;
3376 			goto integer;
3377 
3378 		case SO_PROTOCOL:
3379 			optval = so->so_proto->pr_protocol;
3380 			goto integer;
3381 
3382 		case SO_ERROR:
3383 			SOCK_LOCK(so);
3384 			if (so->so_error) {
3385 				optval = so->so_error;
3386 				so->so_error = 0;
3387 			} else {
3388 				optval = so->so_rerror;
3389 				so->so_rerror = 0;
3390 			}
3391 			SOCK_UNLOCK(so);
3392 			goto integer;
3393 
3394 		case SO_SNDBUF:
3395 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3396 			    so->so_snd.sb_hiwat;
3397 			goto integer;
3398 
3399 		case SO_RCVBUF:
3400 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3401 			    so->so_rcv.sb_hiwat;
3402 			goto integer;
3403 
3404 		case SO_SNDLOWAT:
3405 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3406 			    so->so_snd.sb_lowat;
3407 			goto integer;
3408 
3409 		case SO_RCVLOWAT:
3410 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3411 			    so->so_rcv.sb_lowat;
3412 			goto integer;
3413 
3414 		case SO_SNDTIMEO:
3415 		case SO_RCVTIMEO:
3416 			SOCK_LOCK(so);
3417 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3418 			    (SOLISTENING(so) ? so->sol_sbsnd_timeo :
3419 			    so->so_snd.sb_timeo) :
3420 			    (SOLISTENING(so) ? so->sol_sbrcv_timeo :
3421 			    so->so_rcv.sb_timeo));
3422 			SOCK_UNLOCK(so);
3423 #ifdef COMPAT_FREEBSD32
3424 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3425 				struct timeval32 tv32;
3426 
3427 				CP(tv, tv32, tv_sec);
3428 				CP(tv, tv32, tv_usec);
3429 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3430 			} else
3431 #endif
3432 				error = sooptcopyout(sopt, &tv, sizeof tv);
3433 			break;
3434 
3435 		case SO_LABEL:
3436 #ifdef MAC
3437 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3438 			    sizeof(extmac));
3439 			if (error)
3440 				goto bad;
3441 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3442 			    so, &extmac);
3443 			if (error)
3444 				goto bad;
3445 			/* Don't copy out extmac, it is unchanged. */
3446 #else
3447 			error = EOPNOTSUPP;
3448 #endif
3449 			break;
3450 
3451 		case SO_PEERLABEL:
3452 #ifdef MAC
3453 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3454 			    sizeof(extmac));
3455 			if (error)
3456 				goto bad;
3457 			error = mac_getsockopt_peerlabel(
3458 			    sopt->sopt_td->td_ucred, so, &extmac);
3459 			if (error)
3460 				goto bad;
3461 			/* Don't copy out extmac, it is unchanged. */
3462 #else
3463 			error = EOPNOTSUPP;
3464 #endif
3465 			break;
3466 
3467 		case SO_LISTENQLIMIT:
3468 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3469 			goto integer;
3470 
3471 		case SO_LISTENQLEN:
3472 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3473 			goto integer;
3474 
3475 		case SO_LISTENINCQLEN:
3476 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3477 			goto integer;
3478 
3479 		case SO_TS_CLOCK:
3480 			optval = so->so_ts_clock;
3481 			goto integer;
3482 
3483 		case SO_MAX_PACING_RATE:
3484 			optval = so->so_max_pacing_rate;
3485 			goto integer;
3486 
3487 		default:
3488 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3489 				error = hhook_run_socket(so, sopt,
3490 				    HHOOK_SOCKET_OPT);
3491 			else
3492 				error = ENOPROTOOPT;
3493 			break;
3494 		}
3495 	}
3496 #ifdef MAC
3497 bad:
3498 #endif
3499 	CURVNET_RESTORE();
3500 	return (error);
3501 }
3502 
3503 int
3504 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3505 {
3506 	struct mbuf *m, *m_prev;
3507 	int sopt_size = sopt->sopt_valsize;
3508 
3509 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3510 	if (m == NULL)
3511 		return ENOBUFS;
3512 	if (sopt_size > MLEN) {
3513 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3514 		if ((m->m_flags & M_EXT) == 0) {
3515 			m_free(m);
3516 			return ENOBUFS;
3517 		}
3518 		m->m_len = min(MCLBYTES, sopt_size);
3519 	} else {
3520 		m->m_len = min(MLEN, sopt_size);
3521 	}
3522 	sopt_size -= m->m_len;
3523 	*mp = m;
3524 	m_prev = m;
3525 
3526 	while (sopt_size) {
3527 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3528 		if (m == NULL) {
3529 			m_freem(*mp);
3530 			return ENOBUFS;
3531 		}
3532 		if (sopt_size > MLEN) {
3533 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3534 			    M_NOWAIT);
3535 			if ((m->m_flags & M_EXT) == 0) {
3536 				m_freem(m);
3537 				m_freem(*mp);
3538 				return ENOBUFS;
3539 			}
3540 			m->m_len = min(MCLBYTES, sopt_size);
3541 		} else {
3542 			m->m_len = min(MLEN, sopt_size);
3543 		}
3544 		sopt_size -= m->m_len;
3545 		m_prev->m_next = m;
3546 		m_prev = m;
3547 	}
3548 	return (0);
3549 }
3550 
3551 int
3552 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3553 {
3554 	struct mbuf *m0 = m;
3555 
3556 	if (sopt->sopt_val == NULL)
3557 		return (0);
3558 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3559 		if (sopt->sopt_td != NULL) {
3560 			int error;
3561 
3562 			error = copyin(sopt->sopt_val, mtod(m, char *),
3563 			    m->m_len);
3564 			if (error != 0) {
3565 				m_freem(m0);
3566 				return(error);
3567 			}
3568 		} else
3569 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3570 		sopt->sopt_valsize -= m->m_len;
3571 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3572 		m = m->m_next;
3573 	}
3574 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3575 		panic("ip6_sooptmcopyin");
3576 	return (0);
3577 }
3578 
3579 int
3580 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3581 {
3582 	struct mbuf *m0 = m;
3583 	size_t valsize = 0;
3584 
3585 	if (sopt->sopt_val == NULL)
3586 		return (0);
3587 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3588 		if (sopt->sopt_td != NULL) {
3589 			int error;
3590 
3591 			error = copyout(mtod(m, char *), sopt->sopt_val,
3592 			    m->m_len);
3593 			if (error != 0) {
3594 				m_freem(m0);
3595 				return(error);
3596 			}
3597 		} else
3598 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3599 		sopt->sopt_valsize -= m->m_len;
3600 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3601 		valsize += m->m_len;
3602 		m = m->m_next;
3603 	}
3604 	if (m != NULL) {
3605 		/* enough soopt buffer should be given from user-land */
3606 		m_freem(m0);
3607 		return(EINVAL);
3608 	}
3609 	sopt->sopt_valsize = valsize;
3610 	return (0);
3611 }
3612 
3613 /*
3614  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3615  * out-of-band data, which will then notify socket consumers.
3616  */
3617 void
3618 sohasoutofband(struct socket *so)
3619 {
3620 
3621 	if (so->so_sigio != NULL)
3622 		pgsigio(&so->so_sigio, SIGURG, 0);
3623 	selwakeuppri(&so->so_rdsel, PSOCK);
3624 }
3625 
3626 int
3627 sopoll(struct socket *so, int events, struct ucred *active_cred,
3628     struct thread *td)
3629 {
3630 
3631 	/*
3632 	 * We do not need to set or assert curvnet as long as everyone uses
3633 	 * sopoll_generic().
3634 	 */
3635 	return (so->so_proto->pr_sopoll(so, events, active_cred, td));
3636 }
3637 
3638 int
3639 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3640     struct thread *td)
3641 {
3642 	int revents;
3643 
3644 	SOCK_LOCK(so);
3645 	if (SOLISTENING(so)) {
3646 		if (!(events & (POLLIN | POLLRDNORM)))
3647 			revents = 0;
3648 		else if (!TAILQ_EMPTY(&so->sol_comp))
3649 			revents = events & (POLLIN | POLLRDNORM);
3650 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3651 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3652 		else {
3653 			selrecord(td, &so->so_rdsel);
3654 			revents = 0;
3655 		}
3656 	} else {
3657 		revents = 0;
3658 		SOCK_SENDBUF_LOCK(so);
3659 		SOCK_RECVBUF_LOCK(so);
3660 		if (events & (POLLIN | POLLRDNORM))
3661 			if (soreadabledata(so))
3662 				revents |= events & (POLLIN | POLLRDNORM);
3663 		if (events & (POLLOUT | POLLWRNORM))
3664 			if (sowriteable(so))
3665 				revents |= events & (POLLOUT | POLLWRNORM);
3666 		if (events & (POLLPRI | POLLRDBAND))
3667 			if (so->so_oobmark ||
3668 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3669 				revents |= events & (POLLPRI | POLLRDBAND);
3670 		if ((events & POLLINIGNEOF) == 0) {
3671 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3672 				revents |= events & (POLLIN | POLLRDNORM);
3673 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3674 					revents |= POLLHUP;
3675 			}
3676 		}
3677 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3678 			revents |= events & POLLRDHUP;
3679 		if (revents == 0) {
3680 			if (events &
3681 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3682 				selrecord(td, &so->so_rdsel);
3683 				so->so_rcv.sb_flags |= SB_SEL;
3684 			}
3685 			if (events & (POLLOUT | POLLWRNORM)) {
3686 				selrecord(td, &so->so_wrsel);
3687 				so->so_snd.sb_flags |= SB_SEL;
3688 			}
3689 		}
3690 		SOCK_RECVBUF_UNLOCK(so);
3691 		SOCK_SENDBUF_UNLOCK(so);
3692 	}
3693 	SOCK_UNLOCK(so);
3694 	return (revents);
3695 }
3696 
3697 int
3698 soo_kqfilter(struct file *fp, struct knote *kn)
3699 {
3700 	struct socket *so = kn->kn_fp->f_data;
3701 	struct sockbuf *sb;
3702 	sb_which which;
3703 	struct knlist *knl;
3704 
3705 	switch (kn->kn_filter) {
3706 	case EVFILT_READ:
3707 		kn->kn_fop = &soread_filtops;
3708 		knl = &so->so_rdsel.si_note;
3709 		sb = &so->so_rcv;
3710 		which = SO_RCV;
3711 		break;
3712 	case EVFILT_WRITE:
3713 		kn->kn_fop = &sowrite_filtops;
3714 		knl = &so->so_wrsel.si_note;
3715 		sb = &so->so_snd;
3716 		which = SO_SND;
3717 		break;
3718 	case EVFILT_EMPTY:
3719 		kn->kn_fop = &soempty_filtops;
3720 		knl = &so->so_wrsel.si_note;
3721 		sb = &so->so_snd;
3722 		which = SO_SND;
3723 		break;
3724 	default:
3725 		return (EINVAL);
3726 	}
3727 
3728 	SOCK_LOCK(so);
3729 	if (SOLISTENING(so)) {
3730 		knlist_add(knl, kn, 1);
3731 	} else {
3732 		SOCK_BUF_LOCK(so, which);
3733 		knlist_add(knl, kn, 1);
3734 		sb->sb_flags |= SB_KNOTE;
3735 		SOCK_BUF_UNLOCK(so, which);
3736 	}
3737 	SOCK_UNLOCK(so);
3738 	return (0);
3739 }
3740 
3741 static void
3742 filt_sordetach(struct knote *kn)
3743 {
3744 	struct socket *so = kn->kn_fp->f_data;
3745 
3746 	so_rdknl_lock(so);
3747 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3748 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3749 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3750 	so_rdknl_unlock(so);
3751 }
3752 
3753 /*ARGSUSED*/
3754 static int
3755 filt_soread(struct knote *kn, long hint)
3756 {
3757 	struct socket *so;
3758 
3759 	so = kn->kn_fp->f_data;
3760 
3761 	if (SOLISTENING(so)) {
3762 		SOCK_LOCK_ASSERT(so);
3763 		kn->kn_data = so->sol_qlen;
3764 		if (so->so_error) {
3765 			kn->kn_flags |= EV_EOF;
3766 			kn->kn_fflags = so->so_error;
3767 			return (1);
3768 		}
3769 		return (!TAILQ_EMPTY(&so->sol_comp));
3770 	}
3771 
3772 	SOCK_RECVBUF_LOCK_ASSERT(so);
3773 
3774 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3775 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3776 		kn->kn_flags |= EV_EOF;
3777 		kn->kn_fflags = so->so_error;
3778 		return (1);
3779 	} else if (so->so_error || so->so_rerror)
3780 		return (1);
3781 
3782 	if (kn->kn_sfflags & NOTE_LOWAT) {
3783 		if (kn->kn_data >= kn->kn_sdata)
3784 			return (1);
3785 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3786 		return (1);
3787 
3788 	/* This hook returning non-zero indicates an event, not error */
3789 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3790 }
3791 
3792 static void
3793 filt_sowdetach(struct knote *kn)
3794 {
3795 	struct socket *so = kn->kn_fp->f_data;
3796 
3797 	so_wrknl_lock(so);
3798 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3799 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3800 		so->so_snd.sb_flags &= ~SB_KNOTE;
3801 	so_wrknl_unlock(so);
3802 }
3803 
3804 /*ARGSUSED*/
3805 static int
3806 filt_sowrite(struct knote *kn, long hint)
3807 {
3808 	struct socket *so;
3809 
3810 	so = kn->kn_fp->f_data;
3811 
3812 	if (SOLISTENING(so))
3813 		return (0);
3814 
3815 	SOCK_SENDBUF_LOCK_ASSERT(so);
3816 	kn->kn_data = sbspace(&so->so_snd);
3817 
3818 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3819 
3820 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3821 		kn->kn_flags |= EV_EOF;
3822 		kn->kn_fflags = so->so_error;
3823 		return (1);
3824 	} else if (so->so_error)	/* temporary udp error */
3825 		return (1);
3826 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3827 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3828 		return (0);
3829 	else if (kn->kn_sfflags & NOTE_LOWAT)
3830 		return (kn->kn_data >= kn->kn_sdata);
3831 	else
3832 		return (kn->kn_data >= so->so_snd.sb_lowat);
3833 }
3834 
3835 static int
3836 filt_soempty(struct knote *kn, long hint)
3837 {
3838 	struct socket *so;
3839 
3840 	so = kn->kn_fp->f_data;
3841 
3842 	if (SOLISTENING(so))
3843 		return (1);
3844 
3845 	SOCK_SENDBUF_LOCK_ASSERT(so);
3846 	kn->kn_data = sbused(&so->so_snd);
3847 
3848 	if (kn->kn_data == 0)
3849 		return (1);
3850 	else
3851 		return (0);
3852 }
3853 
3854 int
3855 socheckuid(struct socket *so, uid_t uid)
3856 {
3857 
3858 	if (so == NULL)
3859 		return (EPERM);
3860 	if (so->so_cred->cr_uid != uid)
3861 		return (EPERM);
3862 	return (0);
3863 }
3864 
3865 /*
3866  * These functions are used by protocols to notify the socket layer (and its
3867  * consumers) of state changes in the sockets driven by protocol-side events.
3868  */
3869 
3870 /*
3871  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3872  *
3873  * Normal sequence from the active (originating) side is that
3874  * soisconnecting() is called during processing of connect() call, resulting
3875  * in an eventual call to soisconnected() if/when the connection is
3876  * established.  When the connection is torn down soisdisconnecting() is
3877  * called during processing of disconnect() call, and soisdisconnected() is
3878  * called when the connection to the peer is totally severed.  The semantics
3879  * of these routines are such that connectionless protocols can call
3880  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3881  * calls when setting up a ``connection'' takes no time.
3882  *
3883  * From the passive side, a socket is created with two queues of sockets:
3884  * so_incomp for connections in progress and so_comp for connections already
3885  * made and awaiting user acceptance.  As a protocol is preparing incoming
3886  * connections, it creates a socket structure queued on so_incomp by calling
3887  * sonewconn().  When the connection is established, soisconnected() is
3888  * called, and transfers the socket structure to so_comp, making it available
3889  * to accept().
3890  *
3891  * If a socket is closed with sockets on either so_incomp or so_comp, these
3892  * sockets are dropped.
3893  *
3894  * If higher-level protocols are implemented in the kernel, the wakeups done
3895  * here will sometimes cause software-interrupt process scheduling.
3896  */
3897 void
3898 soisconnecting(struct socket *so)
3899 {
3900 
3901 	SOCK_LOCK(so);
3902 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3903 	so->so_state |= SS_ISCONNECTING;
3904 	SOCK_UNLOCK(so);
3905 }
3906 
3907 void
3908 soisconnected(struct socket *so)
3909 {
3910 	bool last __diagused;
3911 
3912 	SOCK_LOCK(so);
3913 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
3914 	so->so_state |= SS_ISCONNECTED;
3915 
3916 	if (so->so_qstate == SQ_INCOMP) {
3917 		struct socket *head = so->so_listen;
3918 		int ret;
3919 
3920 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
3921 		/*
3922 		 * Promoting a socket from incomplete queue to complete, we
3923 		 * need to go through reverse order of locking.  We first do
3924 		 * trylock, and if that doesn't succeed, we go the hard way
3925 		 * leaving a reference and rechecking consistency after proper
3926 		 * locking.
3927 		 */
3928 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
3929 			soref(head);
3930 			SOCK_UNLOCK(so);
3931 			SOLISTEN_LOCK(head);
3932 			SOCK_LOCK(so);
3933 			if (__predict_false(head != so->so_listen)) {
3934 				/*
3935 				 * The socket went off the listen queue,
3936 				 * should be lost race to close(2) of sol.
3937 				 * The socket is about to soabort().
3938 				 */
3939 				SOCK_UNLOCK(so);
3940 				sorele_locked(head);
3941 				return;
3942 			}
3943 			last = refcount_release(&head->so_count);
3944 			KASSERT(!last, ("%s: released last reference for %p",
3945 			    __func__, head));
3946 		}
3947 again:
3948 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3949 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
3950 			head->sol_incqlen--;
3951 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
3952 			head->sol_qlen++;
3953 			so->so_qstate = SQ_COMP;
3954 			SOCK_UNLOCK(so);
3955 			solisten_wakeup(head);	/* unlocks */
3956 		} else {
3957 			SOCK_RECVBUF_LOCK(so);
3958 			soupcall_set(so, SO_RCV,
3959 			    head->sol_accept_filter->accf_callback,
3960 			    head->sol_accept_filter_arg);
3961 			so->so_options &= ~SO_ACCEPTFILTER;
3962 			ret = head->sol_accept_filter->accf_callback(so,
3963 			    head->sol_accept_filter_arg, M_NOWAIT);
3964 			if (ret == SU_ISCONNECTED) {
3965 				soupcall_clear(so, SO_RCV);
3966 				SOCK_RECVBUF_UNLOCK(so);
3967 				goto again;
3968 			}
3969 			SOCK_RECVBUF_UNLOCK(so);
3970 			SOCK_UNLOCK(so);
3971 			SOLISTEN_UNLOCK(head);
3972 		}
3973 		return;
3974 	}
3975 	SOCK_UNLOCK(so);
3976 	wakeup(&so->so_timeo);
3977 	sorwakeup(so);
3978 	sowwakeup(so);
3979 }
3980 
3981 void
3982 soisdisconnecting(struct socket *so)
3983 {
3984 
3985 	SOCK_LOCK(so);
3986 	so->so_state &= ~SS_ISCONNECTING;
3987 	so->so_state |= SS_ISDISCONNECTING;
3988 
3989 	if (!SOLISTENING(so)) {
3990 		SOCK_RECVBUF_LOCK(so);
3991 		socantrcvmore_locked(so);
3992 		SOCK_SENDBUF_LOCK(so);
3993 		socantsendmore_locked(so);
3994 	}
3995 	SOCK_UNLOCK(so);
3996 	wakeup(&so->so_timeo);
3997 }
3998 
3999 void
4000 soisdisconnected(struct socket *so)
4001 {
4002 
4003 	SOCK_LOCK(so);
4004 
4005 	/*
4006 	 * There is at least one reader of so_state that does not
4007 	 * acquire socket lock, namely soreceive_generic().  Ensure
4008 	 * that it never sees all flags that track connection status
4009 	 * cleared, by ordering the update with a barrier semantic of
4010 	 * our release thread fence.
4011 	 */
4012 	so->so_state |= SS_ISDISCONNECTED;
4013 	atomic_thread_fence_rel();
4014 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4015 
4016 	if (!SOLISTENING(so)) {
4017 		SOCK_UNLOCK(so);
4018 		SOCK_RECVBUF_LOCK(so);
4019 		socantrcvmore_locked(so);
4020 		SOCK_SENDBUF_LOCK(so);
4021 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4022 		socantsendmore_locked(so);
4023 	} else
4024 		SOCK_UNLOCK(so);
4025 	wakeup(&so->so_timeo);
4026 }
4027 
4028 int
4029 soiolock(struct socket *so, struct sx *sx, int flags)
4030 {
4031 	int error;
4032 
4033 	KASSERT((flags & SBL_VALID) == flags,
4034 	    ("soiolock: invalid flags %#x", flags));
4035 
4036 	if ((flags & SBL_WAIT) != 0) {
4037 		if ((flags & SBL_NOINTR) != 0) {
4038 			sx_xlock(sx);
4039 		} else {
4040 			error = sx_xlock_sig(sx);
4041 			if (error != 0)
4042 				return (error);
4043 		}
4044 	} else if (!sx_try_xlock(sx)) {
4045 		return (EWOULDBLOCK);
4046 	}
4047 
4048 	if (__predict_false(SOLISTENING(so))) {
4049 		sx_xunlock(sx);
4050 		return (ENOTCONN);
4051 	}
4052 	return (0);
4053 }
4054 
4055 void
4056 soiounlock(struct sx *sx)
4057 {
4058 	sx_xunlock(sx);
4059 }
4060 
4061 /*
4062  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4063  */
4064 struct sockaddr *
4065 sodupsockaddr(const struct sockaddr *sa, int mflags)
4066 {
4067 	struct sockaddr *sa2;
4068 
4069 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4070 	if (sa2)
4071 		bcopy(sa, sa2, sa->sa_len);
4072 	return sa2;
4073 }
4074 
4075 /*
4076  * Register per-socket destructor.
4077  */
4078 void
4079 sodtor_set(struct socket *so, so_dtor_t *func)
4080 {
4081 
4082 	SOCK_LOCK_ASSERT(so);
4083 	so->so_dtor = func;
4084 }
4085 
4086 /*
4087  * Register per-socket buffer upcalls.
4088  */
4089 void
4090 soupcall_set(struct socket *so, sb_which which, so_upcall_t func, void *arg)
4091 {
4092 	struct sockbuf *sb;
4093 
4094 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4095 
4096 	switch (which) {
4097 	case SO_RCV:
4098 		sb = &so->so_rcv;
4099 		break;
4100 	case SO_SND:
4101 		sb = &so->so_snd;
4102 		break;
4103 	}
4104 	SOCK_BUF_LOCK_ASSERT(so, which);
4105 	sb->sb_upcall = func;
4106 	sb->sb_upcallarg = arg;
4107 	sb->sb_flags |= SB_UPCALL;
4108 }
4109 
4110 void
4111 soupcall_clear(struct socket *so, sb_which which)
4112 {
4113 	struct sockbuf *sb;
4114 
4115 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4116 
4117 	switch (which) {
4118 	case SO_RCV:
4119 		sb = &so->so_rcv;
4120 		break;
4121 	case SO_SND:
4122 		sb = &so->so_snd;
4123 		break;
4124 	}
4125 	SOCK_BUF_LOCK_ASSERT(so, which);
4126 	KASSERT(sb->sb_upcall != NULL,
4127 	    ("%s: so %p no upcall to clear", __func__, so));
4128 	sb->sb_upcall = NULL;
4129 	sb->sb_upcallarg = NULL;
4130 	sb->sb_flags &= ~SB_UPCALL;
4131 }
4132 
4133 void
4134 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4135 {
4136 
4137 	SOLISTEN_LOCK_ASSERT(so);
4138 	so->sol_upcall = func;
4139 	so->sol_upcallarg = arg;
4140 }
4141 
4142 static void
4143 so_rdknl_lock(void *arg)
4144 {
4145 	struct socket *so = arg;
4146 
4147 retry:
4148 	if (SOLISTENING(so)) {
4149 		SOLISTEN_LOCK(so);
4150 	} else {
4151 		SOCK_RECVBUF_LOCK(so);
4152 		if (__predict_false(SOLISTENING(so))) {
4153 			SOCK_RECVBUF_UNLOCK(so);
4154 			goto retry;
4155 		}
4156 	}
4157 }
4158 
4159 static void
4160 so_rdknl_unlock(void *arg)
4161 {
4162 	struct socket *so = arg;
4163 
4164 	if (SOLISTENING(so))
4165 		SOLISTEN_UNLOCK(so);
4166 	else
4167 		SOCK_RECVBUF_UNLOCK(so);
4168 }
4169 
4170 static void
4171 so_rdknl_assert_lock(void *arg, int what)
4172 {
4173 	struct socket *so = arg;
4174 
4175 	if (what == LA_LOCKED) {
4176 		if (SOLISTENING(so))
4177 			SOLISTEN_LOCK_ASSERT(so);
4178 		else
4179 			SOCK_RECVBUF_LOCK_ASSERT(so);
4180 	} else {
4181 		if (SOLISTENING(so))
4182 			SOLISTEN_UNLOCK_ASSERT(so);
4183 		else
4184 			SOCK_RECVBUF_UNLOCK_ASSERT(so);
4185 	}
4186 }
4187 
4188 static void
4189 so_wrknl_lock(void *arg)
4190 {
4191 	struct socket *so = arg;
4192 
4193 retry:
4194 	if (SOLISTENING(so)) {
4195 		SOLISTEN_LOCK(so);
4196 	} else {
4197 		SOCK_SENDBUF_LOCK(so);
4198 		if (__predict_false(SOLISTENING(so))) {
4199 			SOCK_SENDBUF_UNLOCK(so);
4200 			goto retry;
4201 		}
4202 	}
4203 }
4204 
4205 static void
4206 so_wrknl_unlock(void *arg)
4207 {
4208 	struct socket *so = arg;
4209 
4210 	if (SOLISTENING(so))
4211 		SOLISTEN_UNLOCK(so);
4212 	else
4213 		SOCK_SENDBUF_UNLOCK(so);
4214 }
4215 
4216 static void
4217 so_wrknl_assert_lock(void *arg, int what)
4218 {
4219 	struct socket *so = arg;
4220 
4221 	if (what == LA_LOCKED) {
4222 		if (SOLISTENING(so))
4223 			SOLISTEN_LOCK_ASSERT(so);
4224 		else
4225 			SOCK_SENDBUF_LOCK_ASSERT(so);
4226 	} else {
4227 		if (SOLISTENING(so))
4228 			SOLISTEN_UNLOCK_ASSERT(so);
4229 		else
4230 			SOCK_SENDBUF_UNLOCK_ASSERT(so);
4231 	}
4232 }
4233 
4234 /*
4235  * Create an external-format (``xsocket'') structure using the information in
4236  * the kernel-format socket structure pointed to by so.  This is done to
4237  * reduce the spew of irrelevant information over this interface, to isolate
4238  * user code from changes in the kernel structure, and potentially to provide
4239  * information-hiding if we decide that some of this information should be
4240  * hidden from users.
4241  */
4242 void
4243 sotoxsocket(struct socket *so, struct xsocket *xso)
4244 {
4245 
4246 	bzero(xso, sizeof(*xso));
4247 	xso->xso_len = sizeof *xso;
4248 	xso->xso_so = (uintptr_t)so;
4249 	xso->so_type = so->so_type;
4250 	xso->so_options = so->so_options;
4251 	xso->so_linger = so->so_linger;
4252 	xso->so_state = so->so_state;
4253 	xso->so_pcb = (uintptr_t)so->so_pcb;
4254 	xso->xso_protocol = so->so_proto->pr_protocol;
4255 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4256 	xso->so_timeo = so->so_timeo;
4257 	xso->so_error = so->so_error;
4258 	xso->so_uid = so->so_cred->cr_uid;
4259 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4260 	if (SOLISTENING(so)) {
4261 		xso->so_qlen = so->sol_qlen;
4262 		xso->so_incqlen = so->sol_incqlen;
4263 		xso->so_qlimit = so->sol_qlimit;
4264 		xso->so_oobmark = 0;
4265 	} else {
4266 		xso->so_state |= so->so_qstate;
4267 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4268 		xso->so_oobmark = so->so_oobmark;
4269 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4270 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4271 	}
4272 }
4273 
4274 int
4275 so_options_get(const struct socket *so)
4276 {
4277 
4278 	return (so->so_options);
4279 }
4280 
4281 void
4282 so_options_set(struct socket *so, int val)
4283 {
4284 
4285 	so->so_options = val;
4286 }
4287 
4288 int
4289 so_error_get(const struct socket *so)
4290 {
4291 
4292 	return (so->so_error);
4293 }
4294 
4295 void
4296 so_error_set(struct socket *so, int val)
4297 {
4298 
4299 	so->so_error = val;
4300 }
4301