xref: /freebsd/sys/kern/uipc_socket.c (revision b4a58fbf640409a1e507d9f7b411c83a3f83a2f3)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1982, 1986, 1988, 1990, 1993
5  *	The Regents of the University of California.
6  * Copyright (c) 2004 The FreeBSD Foundation
7  * Copyright (c) 2004-2008 Robert N. M. Watson
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the University nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  *
34  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
35  */
36 
37 /*
38  * Comments on the socket life cycle:
39  *
40  * soalloc() sets of socket layer state for a socket, called only by
41  * socreate() and sonewconn().  Socket layer private.
42  *
43  * sodealloc() tears down socket layer state for a socket, called only by
44  * sofree() and sonewconn().  Socket layer private.
45  *
46  * pru_attach() associates protocol layer state with an allocated socket;
47  * called only once, may fail, aborting socket allocation.  This is called
48  * from socreate() and sonewconn().  Socket layer private.
49  *
50  * pru_detach() disassociates protocol layer state from an attached socket,
51  * and will be called exactly once for sockets in which pru_attach() has
52  * been successfully called.  If pru_attach() returned an error,
53  * pru_detach() will not be called.  Socket layer private.
54  *
55  * pru_abort() and pru_close() notify the protocol layer that the last
56  * consumer of a socket is starting to tear down the socket, and that the
57  * protocol should terminate the connection.  Historically, pru_abort() also
58  * detached protocol state from the socket state, but this is no longer the
59  * case.
60  *
61  * socreate() creates a socket and attaches protocol state.  This is a public
62  * interface that may be used by socket layer consumers to create new
63  * sockets.
64  *
65  * sonewconn() creates a socket and attaches protocol state.  This is a
66  * public interface  that may be used by protocols to create new sockets when
67  * a new connection is received and will be available for accept() on a
68  * listen socket.
69  *
70  * soclose() destroys a socket after possibly waiting for it to disconnect.
71  * This is a public interface that socket consumers should use to close and
72  * release a socket when done with it.
73  *
74  * soabort() destroys a socket without waiting for it to disconnect (used
75  * only for incoming connections that are already partially or fully
76  * connected).  This is used internally by the socket layer when clearing
77  * listen socket queues (due to overflow or close on the listen socket), but
78  * is also a public interface protocols may use to abort connections in
79  * their incomplete listen queues should they no longer be required.  Sockets
80  * placed in completed connection listen queues should not be aborted for
81  * reasons described in the comment above the soclose() implementation.  This
82  * is not a general purpose close routine, and except in the specific
83  * circumstances described here, should not be used.
84  *
85  * sofree() will free a socket and its protocol state if all references on
86  * the socket have been released, and is the public interface to attempt to
87  * free a socket when a reference is removed.  This is a socket layer private
88  * interface.
89  *
90  * NOTE: In addition to socreate() and soclose(), which provide a single
91  * socket reference to the consumer to be managed as required, there are two
92  * calls to explicitly manage socket references, soref(), and sorele().
93  * Currently, these are generally required only when transitioning a socket
94  * from a listen queue to a file descriptor, in order to prevent garbage
95  * collection of the socket at an untimely moment.  For a number of reasons,
96  * these interfaces are not preferred, and should be avoided.
97  *
98  * NOTE: With regard to VNETs the general rule is that callers do not set
99  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
100  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
101  * and sorflush(), which are usually called from a pre-set VNET context.
102  * sopoll() currently does not need a VNET context to be set.
103  */
104 
105 #include <sys/cdefs.h>
106 __FBSDID("$FreeBSD$");
107 
108 #include "opt_inet.h"
109 #include "opt_inet6.h"
110 #include "opt_kern_tls.h"
111 #include "opt_sctp.h"
112 
113 #include <sys/param.h>
114 #include <sys/systm.h>
115 #include <sys/capsicum.h>
116 #include <sys/fcntl.h>
117 #include <sys/limits.h>
118 #include <sys/lock.h>
119 #include <sys/mac.h>
120 #include <sys/malloc.h>
121 #include <sys/mbuf.h>
122 #include <sys/mutex.h>
123 #include <sys/domain.h>
124 #include <sys/file.h>			/* for struct knote */
125 #include <sys/hhook.h>
126 #include <sys/kernel.h>
127 #include <sys/khelp.h>
128 #include <sys/ktls.h>
129 #include <sys/event.h>
130 #include <sys/eventhandler.h>
131 #include <sys/poll.h>
132 #include <sys/proc.h>
133 #include <sys/protosw.h>
134 #include <sys/sbuf.h>
135 #include <sys/socket.h>
136 #include <sys/socketvar.h>
137 #include <sys/resourcevar.h>
138 #include <net/route.h>
139 #include <sys/signalvar.h>
140 #include <sys/stat.h>
141 #include <sys/sx.h>
142 #include <sys/sysctl.h>
143 #include <sys/taskqueue.h>
144 #include <sys/uio.h>
145 #include <sys/un.h>
146 #include <sys/unpcb.h>
147 #include <sys/jail.h>
148 #include <sys/syslog.h>
149 #include <netinet/in.h>
150 #include <netinet/in_pcb.h>
151 #include <netinet/tcp.h>
152 
153 #include <net/vnet.h>
154 
155 #include <security/mac/mac_framework.h>
156 
157 #include <vm/uma.h>
158 
159 #ifdef COMPAT_FREEBSD32
160 #include <sys/mount.h>
161 #include <sys/sysent.h>
162 #include <compat/freebsd32/freebsd32.h>
163 #endif
164 
165 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
166 		    int flags);
167 static void	so_rdknl_lock(void *);
168 static void	so_rdknl_unlock(void *);
169 static void	so_rdknl_assert_lock(void *, int);
170 static void	so_wrknl_lock(void *);
171 static void	so_wrknl_unlock(void *);
172 static void	so_wrknl_assert_lock(void *, int);
173 
174 static void	filt_sordetach(struct knote *kn);
175 static int	filt_soread(struct knote *kn, long hint);
176 static void	filt_sowdetach(struct knote *kn);
177 static int	filt_sowrite(struct knote *kn, long hint);
178 static int	filt_soempty(struct knote *kn, long hint);
179 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
180 fo_kqfilter_t	soo_kqfilter;
181 
182 static struct filterops soread_filtops = {
183 	.f_isfd = 1,
184 	.f_detach = filt_sordetach,
185 	.f_event = filt_soread,
186 };
187 static struct filterops sowrite_filtops = {
188 	.f_isfd = 1,
189 	.f_detach = filt_sowdetach,
190 	.f_event = filt_sowrite,
191 };
192 static struct filterops soempty_filtops = {
193 	.f_isfd = 1,
194 	.f_detach = filt_sowdetach,
195 	.f_event = filt_soempty,
196 };
197 
198 so_gen_t	so_gencnt;	/* generation count for sockets */
199 
200 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
201 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
202 
203 #define	VNET_SO_ASSERT(so)						\
204 	VNET_ASSERT(curvnet != NULL,					\
205 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
206 
207 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
208 #define	V_socket_hhh		VNET(socket_hhh)
209 
210 /*
211  * Limit on the number of connections in the listen queue waiting
212  * for accept(2).
213  * NB: The original sysctl somaxconn is still available but hidden
214  * to prevent confusion about the actual purpose of this number.
215  */
216 static u_int somaxconn = SOMAXCONN;
217 
218 static int
219 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
220 {
221 	int error;
222 	int val;
223 
224 	val = somaxconn;
225 	error = sysctl_handle_int(oidp, &val, 0, req);
226 	if (error || !req->newptr )
227 		return (error);
228 
229 	/*
230 	 * The purpose of the UINT_MAX / 3 limit, is so that the formula
231 	 *   3 * so_qlimit / 2
232 	 * below, will not overflow.
233          */
234 
235 	if (val < 1 || val > UINT_MAX / 3)
236 		return (EINVAL);
237 
238 	somaxconn = val;
239 	return (0);
240 }
241 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue,
242     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, 0, sizeof(int),
243     sysctl_somaxconn, "I",
244     "Maximum listen socket pending connection accept queue size");
245 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
246     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP | CTLFLAG_NEEDGIANT, 0,
247     sizeof(int), sysctl_somaxconn, "I",
248     "Maximum listen socket pending connection accept queue size (compat)");
249 
250 static int numopensockets;
251 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
252     &numopensockets, 0, "Number of open sockets");
253 
254 /*
255  * accept_mtx locks down per-socket fields relating to accept queues.  See
256  * socketvar.h for an annotation of the protected fields of struct socket.
257  */
258 struct mtx accept_mtx;
259 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
260 
261 /*
262  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
263  * so_gencnt field.
264  */
265 static struct mtx so_global_mtx;
266 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
267 
268 /*
269  * General IPC sysctl name space, used by sockets and a variety of other IPC
270  * types.
271  */
272 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
273     "IPC");
274 
275 /*
276  * Initialize the socket subsystem and set up the socket
277  * memory allocator.
278  */
279 static uma_zone_t socket_zone;
280 int	maxsockets;
281 
282 static void
283 socket_zone_change(void *tag)
284 {
285 
286 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
287 }
288 
289 static void
290 socket_hhook_register(int subtype)
291 {
292 
293 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
294 	    &V_socket_hhh[subtype],
295 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
296 		printf("%s: WARNING: unable to register hook\n", __func__);
297 }
298 
299 static void
300 socket_hhook_deregister(int subtype)
301 {
302 
303 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
304 		printf("%s: WARNING: unable to deregister hook\n", __func__);
305 }
306 
307 static void
308 socket_init(void *tag)
309 {
310 
311 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
312 	    NULL, NULL, UMA_ALIGN_PTR, 0);
313 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
314 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
315 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
316 	    EVENTHANDLER_PRI_FIRST);
317 }
318 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
319 
320 static void
321 socket_vnet_init(const void *unused __unused)
322 {
323 	int i;
324 
325 	/* We expect a contiguous range */
326 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
327 		socket_hhook_register(i);
328 }
329 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
330     socket_vnet_init, NULL);
331 
332 static void
333 socket_vnet_uninit(const void *unused __unused)
334 {
335 	int i;
336 
337 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
338 		socket_hhook_deregister(i);
339 }
340 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
341     socket_vnet_uninit, NULL);
342 
343 /*
344  * Initialise maxsockets.  This SYSINIT must be run after
345  * tunable_mbinit().
346  */
347 static void
348 init_maxsockets(void *ignored)
349 {
350 
351 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
352 	maxsockets = imax(maxsockets, maxfiles);
353 }
354 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
355 
356 /*
357  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
358  * of the change so that they can update their dependent limits as required.
359  */
360 static int
361 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
362 {
363 	int error, newmaxsockets;
364 
365 	newmaxsockets = maxsockets;
366 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
367 	if (error == 0 && req->newptr) {
368 		if (newmaxsockets > maxsockets &&
369 		    newmaxsockets <= maxfiles) {
370 			maxsockets = newmaxsockets;
371 			EVENTHANDLER_INVOKE(maxsockets_change);
372 		} else
373 			error = EINVAL;
374 	}
375 	return (error);
376 }
377 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets,
378     CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_NEEDGIANT, &maxsockets, 0,
379     sysctl_maxsockets, "IU",
380     "Maximum number of sockets available");
381 
382 /*
383  * Socket operation routines.  These routines are called by the routines in
384  * sys_socket.c or from a system process, and implement the semantics of
385  * socket operations by switching out to the protocol specific routines.
386  */
387 
388 /*
389  * Get a socket structure from our zone, and initialize it.  Note that it
390  * would probably be better to allocate socket and PCB at the same time, but
391  * I'm not convinced that all the protocols can be easily modified to do
392  * this.
393  *
394  * soalloc() returns a socket with a ref count of 0.
395  */
396 static struct socket *
397 soalloc(struct vnet *vnet)
398 {
399 	struct socket *so;
400 
401 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
402 	if (so == NULL)
403 		return (NULL);
404 #ifdef MAC
405 	if (mac_socket_init(so, M_NOWAIT) != 0) {
406 		uma_zfree(socket_zone, so);
407 		return (NULL);
408 	}
409 #endif
410 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
411 		uma_zfree(socket_zone, so);
412 		return (NULL);
413 	}
414 
415 	/*
416 	 * The socket locking protocol allows to lock 2 sockets at a time,
417 	 * however, the first one must be a listening socket.  WITNESS lacks
418 	 * a feature to change class of an existing lock, so we use DUPOK.
419 	 */
420 	mtx_init(&so->so_lock, "socket", NULL, MTX_DEF | MTX_DUPOK);
421 	so->so_snd.sb_mtx = &so->so_snd_mtx;
422 	so->so_rcv.sb_mtx = &so->so_rcv_mtx;
423 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
424 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
425 	so->so_rcv.sb_sel = &so->so_rdsel;
426 	so->so_snd.sb_sel = &so->so_wrsel;
427 	sx_init(&so->so_snd_sx, "so_snd_sx");
428 	sx_init(&so->so_rcv_sx, "so_rcv_sx");
429 	TAILQ_INIT(&so->so_snd.sb_aiojobq);
430 	TAILQ_INIT(&so->so_rcv.sb_aiojobq);
431 	TASK_INIT(&so->so_snd.sb_aiotask, 0, soaio_snd, so);
432 	TASK_INIT(&so->so_rcv.sb_aiotask, 0, soaio_rcv, so);
433 #ifdef VIMAGE
434 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
435 	    __func__, __LINE__, so));
436 	so->so_vnet = vnet;
437 #endif
438 	/* We shouldn't need the so_global_mtx */
439 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
440 		/* Do we need more comprehensive error returns? */
441 		uma_zfree(socket_zone, so);
442 		return (NULL);
443 	}
444 	mtx_lock(&so_global_mtx);
445 	so->so_gencnt = ++so_gencnt;
446 	++numopensockets;
447 #ifdef VIMAGE
448 	vnet->vnet_sockcnt++;
449 #endif
450 	mtx_unlock(&so_global_mtx);
451 
452 	return (so);
453 }
454 
455 /*
456  * Free the storage associated with a socket at the socket layer, tear down
457  * locks, labels, etc.  All protocol state is assumed already to have been
458  * torn down (and possibly never set up) by the caller.
459  */
460 static void
461 sodealloc(struct socket *so)
462 {
463 
464 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
465 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
466 
467 	mtx_lock(&so_global_mtx);
468 	so->so_gencnt = ++so_gencnt;
469 	--numopensockets;	/* Could be below, but faster here. */
470 #ifdef VIMAGE
471 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
472 	    __func__, __LINE__, so));
473 	so->so_vnet->vnet_sockcnt--;
474 #endif
475 	mtx_unlock(&so_global_mtx);
476 #ifdef MAC
477 	mac_socket_destroy(so);
478 #endif
479 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
480 
481 	khelp_destroy_osd(&so->osd);
482 	if (SOLISTENING(so)) {
483 		if (so->sol_accept_filter != NULL)
484 			accept_filt_setopt(so, NULL);
485 	} else {
486 		if (so->so_rcv.sb_hiwat)
487 			(void)chgsbsize(so->so_cred->cr_uidinfo,
488 			    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
489 		if (so->so_snd.sb_hiwat)
490 			(void)chgsbsize(so->so_cred->cr_uidinfo,
491 			    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
492 		sx_destroy(&so->so_snd_sx);
493 		sx_destroy(&so->so_rcv_sx);
494 		SOCKBUF_LOCK_DESTROY(&so->so_snd);
495 		SOCKBUF_LOCK_DESTROY(&so->so_rcv);
496 	}
497 	crfree(so->so_cred);
498 	mtx_destroy(&so->so_lock);
499 	uma_zfree(socket_zone, so);
500 }
501 
502 /*
503  * socreate returns a socket with a ref count of 1.  The socket should be
504  * closed with soclose().
505  */
506 int
507 socreate(int dom, struct socket **aso, int type, int proto,
508     struct ucred *cred, struct thread *td)
509 {
510 	struct protosw *prp;
511 	struct socket *so;
512 	int error;
513 
514 	if (proto)
515 		prp = pffindproto(dom, proto, type);
516 	else
517 		prp = pffindtype(dom, type);
518 
519 	if (prp == NULL) {
520 		/* No support for domain. */
521 		if (pffinddomain(dom) == NULL)
522 			return (EAFNOSUPPORT);
523 		/* No support for socket type. */
524 		if (proto == 0 && type != 0)
525 			return (EPROTOTYPE);
526 		return (EPROTONOSUPPORT);
527 	}
528 	if (prp->pr_usrreqs->pru_attach == NULL ||
529 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
530 		return (EPROTONOSUPPORT);
531 
532 	if (IN_CAPABILITY_MODE(td) && (prp->pr_flags & PR_CAPATTACH) == 0)
533 		return (ECAPMODE);
534 
535 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
536 		return (EPROTONOSUPPORT);
537 
538 	if (prp->pr_type != type)
539 		return (EPROTOTYPE);
540 	so = soalloc(CRED_TO_VNET(cred));
541 	if (so == NULL)
542 		return (ENOBUFS);
543 
544 	so->so_type = type;
545 	so->so_cred = crhold(cred);
546 	if ((prp->pr_domain->dom_family == PF_INET) ||
547 	    (prp->pr_domain->dom_family == PF_INET6) ||
548 	    (prp->pr_domain->dom_family == PF_ROUTE))
549 		so->so_fibnum = td->td_proc->p_fibnum;
550 	else
551 		so->so_fibnum = 0;
552 	so->so_proto = prp;
553 #ifdef MAC
554 	mac_socket_create(cred, so);
555 #endif
556 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
557 	    so_rdknl_assert_lock);
558 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
559 	    so_wrknl_assert_lock);
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_usrreqs->pru_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 struct timeval overinterval = { 60, 0 };
583 SYSCTL_TIMEVAL_SEC(_kern_ipc, OID_AUTO, sooverinterval, CTLFLAG_RW,
584     &overinterval,
585     "Delay in seconds between warnings for listen socket overflows");
586 
587 /*
588  * When an attempt at a new connection is noted on a socket which accepts
589  * connections, sonewconn is called.  If the connection is possible (subject
590  * to space constraints, etc.) then we allocate a new structure, properly
591  * linked into the data structure of the original socket, and return this.
592  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
593  *
594  * Note: the ref count on the socket is 0 on return.
595  */
596 struct socket *
597 sonewconn(struct socket *head, int connstatus)
598 {
599 	struct sbuf descrsb;
600 	struct socket *so;
601 	int len, overcount;
602 	u_int qlen;
603 	const char localprefix[] = "local:";
604 	char descrbuf[SUNPATHLEN + sizeof(localprefix)];
605 #if defined(INET6)
606 	char addrbuf[INET6_ADDRSTRLEN];
607 #elif defined(INET)
608 	char addrbuf[INET_ADDRSTRLEN];
609 #endif
610 	bool dolog, over;
611 
612 	SOLISTEN_LOCK(head);
613 	over = (head->sol_qlen > 3 * head->sol_qlimit / 2);
614 #ifdef REGRESSION
615 	if (regression_sonewconn_earlytest && over) {
616 #else
617 	if (over) {
618 #endif
619 		head->sol_overcount++;
620 		dolog = !!ratecheck(&head->sol_lastover, &overinterval);
621 
622 		/*
623 		 * If we're going to log, copy the overflow count and queue
624 		 * length from the listen socket before dropping the lock.
625 		 * Also, reset the overflow count.
626 		 */
627 		if (dolog) {
628 			overcount = head->sol_overcount;
629 			head->sol_overcount = 0;
630 			qlen = head->sol_qlen;
631 		}
632 		SOLISTEN_UNLOCK(head);
633 
634 		if (dolog) {
635 			/*
636 			 * Try to print something descriptive about the
637 			 * socket for the error message.
638 			 */
639 			sbuf_new(&descrsb, descrbuf, sizeof(descrbuf),
640 			    SBUF_FIXEDLEN);
641 			switch (head->so_proto->pr_domain->dom_family) {
642 #if defined(INET) || defined(INET6)
643 #ifdef INET
644 			case AF_INET:
645 #endif
646 #ifdef INET6
647 			case AF_INET6:
648 				if (head->so_proto->pr_domain->dom_family ==
649 				    AF_INET6 ||
650 				    (sotoinpcb(head)->inp_inc.inc_flags &
651 				    INC_ISIPV6)) {
652 					ip6_sprintf(addrbuf,
653 					    &sotoinpcb(head)->inp_inc.inc6_laddr);
654 					sbuf_printf(&descrsb, "[%s]", addrbuf);
655 				} else
656 #endif
657 				{
658 #ifdef INET
659 					inet_ntoa_r(
660 					    sotoinpcb(head)->inp_inc.inc_laddr,
661 					    addrbuf);
662 					sbuf_cat(&descrsb, addrbuf);
663 #endif
664 				}
665 				sbuf_printf(&descrsb, ":%hu (proto %u)",
666 				    ntohs(sotoinpcb(head)->inp_inc.inc_lport),
667 				    head->so_proto->pr_protocol);
668 				break;
669 #endif /* INET || INET6 */
670 			case AF_UNIX:
671 				sbuf_cat(&descrsb, localprefix);
672 				if (sotounpcb(head)->unp_addr != NULL)
673 					len =
674 					    sotounpcb(head)->unp_addr->sun_len -
675 					    offsetof(struct sockaddr_un,
676 					    sun_path);
677 				else
678 					len = 0;
679 				if (len > 0)
680 					sbuf_bcat(&descrsb,
681 					    sotounpcb(head)->unp_addr->sun_path,
682 					    len);
683 				else
684 					sbuf_cat(&descrsb, "(unknown)");
685 				break;
686 			}
687 
688 			/*
689 			 * If we can't print something more specific, at least
690 			 * print the domain name.
691 			 */
692 			if (sbuf_finish(&descrsb) != 0 ||
693 			    sbuf_len(&descrsb) <= 0) {
694 				sbuf_clear(&descrsb);
695 				sbuf_cat(&descrsb,
696 				    head->so_proto->pr_domain->dom_name ?:
697 				    "unknown");
698 				sbuf_finish(&descrsb);
699 			}
700 			KASSERT(sbuf_len(&descrsb) > 0,
701 			    ("%s: sbuf creation failed", __func__));
702 			log(LOG_DEBUG,
703 			    "%s: pcb %p (%s): Listen queue overflow: "
704 			    "%i already in queue awaiting acceptance "
705 			    "(%d occurrences)\n",
706 			    __func__, head->so_pcb, sbuf_data(&descrsb),
707 			    qlen, overcount);
708 			sbuf_delete(&descrsb);
709 
710 			overcount = 0;
711 		}
712 
713 		return (NULL);
714 	}
715 	SOLISTEN_UNLOCK(head);
716 	VNET_ASSERT(head->so_vnet != NULL, ("%s: so %p vnet is NULL",
717 	    __func__, head));
718 	so = soalloc(head->so_vnet);
719 	if (so == NULL) {
720 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
721 		    "limit reached or out of memory\n",
722 		    __func__, head->so_pcb);
723 		return (NULL);
724 	}
725 	so->so_listen = head;
726 	so->so_type = head->so_type;
727 	so->so_options = head->so_options & ~SO_ACCEPTCONN;
728 	so->so_linger = head->so_linger;
729 	so->so_state = head->so_state | SS_NOFDREF;
730 	so->so_fibnum = head->so_fibnum;
731 	so->so_proto = head->so_proto;
732 	so->so_cred = crhold(head->so_cred);
733 #ifdef MAC
734 	mac_socket_newconn(head, so);
735 #endif
736 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
737 	    so_rdknl_assert_lock);
738 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
739 	    so_wrknl_assert_lock);
740 	VNET_SO_ASSERT(head);
741 	if (soreserve(so, head->sol_sbsnd_hiwat, head->sol_sbrcv_hiwat)) {
742 		sodealloc(so);
743 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
744 		    __func__, head->so_pcb);
745 		return (NULL);
746 	}
747 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
748 		sodealloc(so);
749 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
750 		    __func__, head->so_pcb);
751 		return (NULL);
752 	}
753 	so->so_rcv.sb_lowat = head->sol_sbrcv_lowat;
754 	so->so_snd.sb_lowat = head->sol_sbsnd_lowat;
755 	so->so_rcv.sb_timeo = head->sol_sbrcv_timeo;
756 	so->so_snd.sb_timeo = head->sol_sbsnd_timeo;
757 	so->so_rcv.sb_flags |= head->sol_sbrcv_flags & SB_AUTOSIZE;
758 	so->so_snd.sb_flags |= head->sol_sbsnd_flags & SB_AUTOSIZE;
759 
760 	SOLISTEN_LOCK(head);
761 	if (head->sol_accept_filter != NULL)
762 		connstatus = 0;
763 	so->so_state |= connstatus;
764 	soref(head); /* A socket on (in)complete queue refs head. */
765 	if (connstatus) {
766 		TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
767 		so->so_qstate = SQ_COMP;
768 		head->sol_qlen++;
769 		solisten_wakeup(head);	/* unlocks */
770 	} else {
771 		/*
772 		 * Keep removing sockets from the head until there's room for
773 		 * us to insert on the tail.  In pre-locking revisions, this
774 		 * was a simple if(), but as we could be racing with other
775 		 * threads and soabort() requires dropping locks, we must
776 		 * loop waiting for the condition to be true.
777 		 */
778 		while (head->sol_incqlen > head->sol_qlimit) {
779 			struct socket *sp;
780 
781 			sp = TAILQ_FIRST(&head->sol_incomp);
782 			TAILQ_REMOVE(&head->sol_incomp, sp, so_list);
783 			head->sol_incqlen--;
784 			SOCK_LOCK(sp);
785 			sp->so_qstate = SQ_NONE;
786 			sp->so_listen = NULL;
787 			SOCK_UNLOCK(sp);
788 			sorele(head);	/* does SOLISTEN_UNLOCK, head stays */
789 			soabort(sp);
790 			SOLISTEN_LOCK(head);
791 		}
792 		TAILQ_INSERT_TAIL(&head->sol_incomp, so, so_list);
793 		so->so_qstate = SQ_INCOMP;
794 		head->sol_incqlen++;
795 		SOLISTEN_UNLOCK(head);
796 	}
797 	return (so);
798 }
799 
800 #if defined(SCTP) || defined(SCTP_SUPPORT)
801 /*
802  * Socket part of sctp_peeloff().  Detach a new socket from an
803  * association.  The new socket is returned with a reference.
804  */
805 struct socket *
806 sopeeloff(struct socket *head)
807 {
808 	struct socket *so;
809 
810 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
811 	    __func__, __LINE__, head));
812 	so = soalloc(head->so_vnet);
813 	if (so == NULL) {
814 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
815 		    "limit reached or out of memory\n",
816 		    __func__, head->so_pcb);
817 		return (NULL);
818 	}
819 	so->so_type = head->so_type;
820 	so->so_options = head->so_options;
821 	so->so_linger = head->so_linger;
822 	so->so_state = (head->so_state & SS_NBIO) | SS_ISCONNECTED;
823 	so->so_fibnum = head->so_fibnum;
824 	so->so_proto = head->so_proto;
825 	so->so_cred = crhold(head->so_cred);
826 #ifdef MAC
827 	mac_socket_newconn(head, so);
828 #endif
829 	knlist_init(&so->so_rdsel.si_note, so, so_rdknl_lock, so_rdknl_unlock,
830 	    so_rdknl_assert_lock);
831 	knlist_init(&so->so_wrsel.si_note, so, so_wrknl_lock, so_wrknl_unlock,
832 	    so_wrknl_assert_lock);
833 	VNET_SO_ASSERT(head);
834 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
835 		sodealloc(so);
836 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
837 		    __func__, head->so_pcb);
838 		return (NULL);
839 	}
840 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
841 		sodealloc(so);
842 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
843 		    __func__, head->so_pcb);
844 		return (NULL);
845 	}
846 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
847 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
848 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
849 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
850 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
851 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
852 
853 	soref(so);
854 
855 	return (so);
856 }
857 #endif	/* SCTP */
858 
859 int
860 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
861 {
862 	int error;
863 
864 	CURVNET_SET(so->so_vnet);
865 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
866 	CURVNET_RESTORE();
867 	return (error);
868 }
869 
870 int
871 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
872 {
873 	int error;
874 
875 	CURVNET_SET(so->so_vnet);
876 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
877 	CURVNET_RESTORE();
878 	return (error);
879 }
880 
881 /*
882  * solisten() transitions a socket from a non-listening state to a listening
883  * state, but can also be used to update the listen queue depth on an
884  * existing listen socket.  The protocol will call back into the sockets
885  * layer using solisten_proto_check() and solisten_proto() to check and set
886  * socket-layer listen state.  Call backs are used so that the protocol can
887  * acquire both protocol and socket layer locks in whatever order is required
888  * by the protocol.
889  *
890  * Protocol implementors are advised to hold the socket lock across the
891  * socket-layer test and set to avoid races at the socket layer.
892  */
893 int
894 solisten(struct socket *so, int backlog, struct thread *td)
895 {
896 	int error;
897 
898 	CURVNET_SET(so->so_vnet);
899 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
900 	CURVNET_RESTORE();
901 	return (error);
902 }
903 
904 /*
905  * Prepare for a call to solisten_proto().  Acquire all socket buffer locks in
906  * order to interlock with socket I/O.
907  */
908 int
909 solisten_proto_check(struct socket *so)
910 {
911 	SOCK_LOCK_ASSERT(so);
912 
913 	if ((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
914 	    SS_ISDISCONNECTING)) != 0)
915 		return (EINVAL);
916 
917 	/*
918 	 * Sleeping is not permitted here, so simply fail if userspace is
919 	 * attempting to transmit or receive on the socket.  This kind of
920 	 * transient failure is not ideal, but it should occur only if userspace
921 	 * is misusing the socket interfaces.
922 	 */
923 	if (!sx_try_xlock(&so->so_snd_sx))
924 		return (EAGAIN);
925 	if (!sx_try_xlock(&so->so_rcv_sx)) {
926 		sx_xunlock(&so->so_snd_sx);
927 		return (EAGAIN);
928 	}
929 	mtx_lock(&so->so_snd_mtx);
930 	mtx_lock(&so->so_rcv_mtx);
931 
932 	/* Interlock with soo_aio_queue(). */
933 	if ((so->so_snd.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0 ||
934 	   (so->so_rcv.sb_flags & (SB_AIO | SB_AIO_RUNNING)) != 0) {
935 		solisten_proto_abort(so);
936 		return (EINVAL);
937 	}
938 	return (0);
939 }
940 
941 /*
942  * Undo the setup done by solisten_proto_check().
943  */
944 void
945 solisten_proto_abort(struct socket *so)
946 {
947 	mtx_unlock(&so->so_snd_mtx);
948 	mtx_unlock(&so->so_rcv_mtx);
949 	sx_xunlock(&so->so_snd_sx);
950 	sx_xunlock(&so->so_rcv_sx);
951 }
952 
953 void
954 solisten_proto(struct socket *so, int backlog)
955 {
956 	int sbrcv_lowat, sbsnd_lowat;
957 	u_int sbrcv_hiwat, sbsnd_hiwat;
958 	short sbrcv_flags, sbsnd_flags;
959 	sbintime_t sbrcv_timeo, sbsnd_timeo;
960 
961 	SOCK_LOCK_ASSERT(so);
962 	KASSERT((so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
963 	    SS_ISDISCONNECTING)) == 0,
964 	    ("%s: bad socket state %p", __func__, so));
965 
966 	if (SOLISTENING(so))
967 		goto listening;
968 
969 	/*
970 	 * Change this socket to listening state.
971 	 */
972 	sbrcv_lowat = so->so_rcv.sb_lowat;
973 	sbsnd_lowat = so->so_snd.sb_lowat;
974 	sbrcv_hiwat = so->so_rcv.sb_hiwat;
975 	sbsnd_hiwat = so->so_snd.sb_hiwat;
976 	sbrcv_flags = so->so_rcv.sb_flags;
977 	sbsnd_flags = so->so_snd.sb_flags;
978 	sbrcv_timeo = so->so_rcv.sb_timeo;
979 	sbsnd_timeo = so->so_snd.sb_timeo;
980 
981 	sbdestroy(&so->so_snd, so);
982 	sbdestroy(&so->so_rcv, so);
983 
984 #ifdef INVARIANTS
985 	bzero(&so->so_rcv,
986 	    sizeof(struct socket) - offsetof(struct socket, so_rcv));
987 #endif
988 
989 	so->sol_sbrcv_lowat = sbrcv_lowat;
990 	so->sol_sbsnd_lowat = sbsnd_lowat;
991 	so->sol_sbrcv_hiwat = sbrcv_hiwat;
992 	so->sol_sbsnd_hiwat = sbsnd_hiwat;
993 	so->sol_sbrcv_flags = sbrcv_flags;
994 	so->sol_sbsnd_flags = sbsnd_flags;
995 	so->sol_sbrcv_timeo = sbrcv_timeo;
996 	so->sol_sbsnd_timeo = sbsnd_timeo;
997 
998 	so->sol_qlen = so->sol_incqlen = 0;
999 	TAILQ_INIT(&so->sol_incomp);
1000 	TAILQ_INIT(&so->sol_comp);
1001 
1002 	so->sol_accept_filter = NULL;
1003 	so->sol_accept_filter_arg = NULL;
1004 	so->sol_accept_filter_str = NULL;
1005 
1006 	so->sol_upcall = NULL;
1007 	so->sol_upcallarg = NULL;
1008 
1009 	so->so_options |= SO_ACCEPTCONN;
1010 
1011 listening:
1012 	if (backlog < 0 || backlog > somaxconn)
1013 		backlog = somaxconn;
1014 	so->sol_qlimit = backlog;
1015 
1016 	mtx_unlock(&so->so_snd_mtx);
1017 	mtx_unlock(&so->so_rcv_mtx);
1018 	sx_xunlock(&so->so_snd_sx);
1019 	sx_xunlock(&so->so_rcv_sx);
1020 }
1021 
1022 /*
1023  * Wakeup listeners/subsystems once we have a complete connection.
1024  * Enters with lock, returns unlocked.
1025  */
1026 void
1027 solisten_wakeup(struct socket *sol)
1028 {
1029 
1030 	if (sol->sol_upcall != NULL)
1031 		(void )sol->sol_upcall(sol, sol->sol_upcallarg, M_NOWAIT);
1032 	else {
1033 		selwakeuppri(&sol->so_rdsel, PSOCK);
1034 		KNOTE_LOCKED(&sol->so_rdsel.si_note, 0);
1035 	}
1036 	SOLISTEN_UNLOCK(sol);
1037 	wakeup_one(&sol->sol_comp);
1038 	if ((sol->so_state & SS_ASYNC) && sol->so_sigio != NULL)
1039 		pgsigio(&sol->so_sigio, SIGIO, 0);
1040 }
1041 
1042 /*
1043  * Return single connection off a listening socket queue.  Main consumer of
1044  * the function is kern_accept4().  Some modules, that do their own accept
1045  * management also use the function.
1046  *
1047  * Listening socket must be locked on entry and is returned unlocked on
1048  * return.
1049  * The flags argument is set of accept4(2) flags and ACCEPT4_INHERIT.
1050  */
1051 int
1052 solisten_dequeue(struct socket *head, struct socket **ret, int flags)
1053 {
1054 	struct socket *so;
1055 	int error;
1056 
1057 	SOLISTEN_LOCK_ASSERT(head);
1058 
1059 	while (!(head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp) &&
1060 	    head->so_error == 0) {
1061 		error = msleep(&head->sol_comp, SOCK_MTX(head), PSOCK | PCATCH,
1062 		    "accept", 0);
1063 		if (error != 0) {
1064 			SOLISTEN_UNLOCK(head);
1065 			return (error);
1066 		}
1067 	}
1068 	if (head->so_error) {
1069 		error = head->so_error;
1070 		head->so_error = 0;
1071 	} else if ((head->so_state & SS_NBIO) && TAILQ_EMPTY(&head->sol_comp))
1072 		error = EWOULDBLOCK;
1073 	else
1074 		error = 0;
1075 	if (error) {
1076 		SOLISTEN_UNLOCK(head);
1077 		return (error);
1078 	}
1079 	so = TAILQ_FIRST(&head->sol_comp);
1080 	SOCK_LOCK(so);
1081 	KASSERT(so->so_qstate == SQ_COMP,
1082 	    ("%s: so %p not SQ_COMP", __func__, so));
1083 	soref(so);
1084 	head->sol_qlen--;
1085 	so->so_qstate = SQ_NONE;
1086 	so->so_listen = NULL;
1087 	TAILQ_REMOVE(&head->sol_comp, so, so_list);
1088 	if (flags & ACCEPT4_INHERIT)
1089 		so->so_state |= (head->so_state & SS_NBIO);
1090 	else
1091 		so->so_state |= (flags & SOCK_NONBLOCK) ? SS_NBIO : 0;
1092 	SOCK_UNLOCK(so);
1093 	sorele(head);
1094 
1095 	*ret = so;
1096 	return (0);
1097 }
1098 
1099 /*
1100  * Evaluate the reference count and named references on a socket; if no
1101  * references remain, free it.  This should be called whenever a reference is
1102  * released, such as in sorele(), but also when named reference flags are
1103  * cleared in socket or protocol code.
1104  *
1105  * sofree() will free the socket if:
1106  *
1107  * - There are no outstanding file descriptor references or related consumers
1108  *   (so_count == 0).
1109  *
1110  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
1111  *
1112  * - The protocol does not have an outstanding strong reference on the socket
1113  *   (SS_PROTOREF).
1114  *
1115  * - The socket is not in a completed connection queue, so a process has been
1116  *   notified that it is present.  If it is removed, the user process may
1117  *   block in accept() despite select() saying the socket was ready.
1118  */
1119 void
1120 sofree(struct socket *so)
1121 {
1122 	struct protosw *pr = so->so_proto;
1123 	bool last __diagused;
1124 
1125 	SOCK_LOCK_ASSERT(so);
1126 
1127 	if ((so->so_state & (SS_NOFDREF | SS_PROTOREF)) != SS_NOFDREF ||
1128 	    refcount_load(&so->so_count) != 0 || so->so_qstate == SQ_COMP) {
1129 		SOCK_UNLOCK(so);
1130 		return;
1131 	}
1132 
1133 	if (!SOLISTENING(so) && so->so_qstate == SQ_INCOMP) {
1134 		struct socket *sol;
1135 
1136 		sol = so->so_listen;
1137 		KASSERT(sol, ("%s: so %p on incomp of NULL", __func__, so));
1138 
1139 		/*
1140 		 * To solve race between close of a listening socket and
1141 		 * a socket on its incomplete queue, we need to lock both.
1142 		 * The order is first listening socket, then regular.
1143 		 * Since we don't have SS_NOFDREF neither SS_PROTOREF, this
1144 		 * function and the listening socket are the only pointers
1145 		 * to so.  To preserve so and sol, we reference both and then
1146 		 * relock.
1147 		 * After relock the socket may not move to so_comp since it
1148 		 * doesn't have PCB already, but it may be removed from
1149 		 * so_incomp. If that happens, we share responsiblity on
1150 		 * freeing the socket, but soclose() has already removed
1151 		 * it from queue.
1152 		 */
1153 		soref(sol);
1154 		soref(so);
1155 		SOCK_UNLOCK(so);
1156 		SOLISTEN_LOCK(sol);
1157 		SOCK_LOCK(so);
1158 		if (so->so_qstate == SQ_INCOMP) {
1159 			KASSERT(so->so_listen == sol,
1160 			    ("%s: so %p migrated out of sol %p",
1161 			    __func__, so, sol));
1162 			TAILQ_REMOVE(&sol->sol_incomp, so, so_list);
1163 			sol->sol_incqlen--;
1164 			last = refcount_release(&sol->so_count);
1165 			KASSERT(!last, ("%s: released last reference for %p",
1166 			    __func__, sol));
1167 			so->so_qstate = SQ_NONE;
1168 			so->so_listen = NULL;
1169 		} else
1170 			KASSERT(so->so_listen == NULL,
1171 			    ("%s: so %p not on (in)comp with so_listen",
1172 			    __func__, so));
1173 		sorele(sol);
1174 		KASSERT(refcount_load(&so->so_count) == 1,
1175 		    ("%s: so %p count %u", __func__, so, so->so_count));
1176 		so->so_count = 0;
1177 	}
1178 	if (SOLISTENING(so))
1179 		so->so_error = ECONNABORTED;
1180 	SOCK_UNLOCK(so);
1181 
1182 	if (so->so_dtor != NULL)
1183 		so->so_dtor(so);
1184 
1185 	VNET_SO_ASSERT(so);
1186 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
1187 		(*pr->pr_domain->dom_dispose)(so);
1188 	if (pr->pr_usrreqs->pru_detach != NULL)
1189 		(*pr->pr_usrreqs->pru_detach)(so);
1190 
1191 	/*
1192 	 * From this point on, we assume that no other references to this
1193 	 * socket exist anywhere else in the stack.  Therefore, no locks need
1194 	 * to be acquired or held.
1195 	 *
1196 	 * We used to do a lot of socket buffer and socket locking here, as
1197 	 * well as invoke sorflush() and perform wakeups.  The direct call to
1198 	 * dom_dispose() and sbdestroy() are an inlining of what was
1199 	 * necessary from sorflush().
1200 	 *
1201 	 * Notice that the socket buffer and kqueue state are torn down
1202 	 * before calling pru_detach.  This means that protocols shold not
1203 	 * assume they can perform socket wakeups, etc, in their detach code.
1204 	 */
1205 	if (!SOLISTENING(so)) {
1206 		sbdestroy(&so->so_snd, so);
1207 		sbdestroy(&so->so_rcv, so);
1208 	}
1209 	seldrain(&so->so_rdsel);
1210 	seldrain(&so->so_wrsel);
1211 	knlist_destroy(&so->so_rdsel.si_note);
1212 	knlist_destroy(&so->so_wrsel.si_note);
1213 	sodealloc(so);
1214 }
1215 
1216 /*
1217  * Close a socket on last file table reference removal.  Initiate disconnect
1218  * if connected.  Free socket when disconnect complete.
1219  *
1220  * This function will sorele() the socket.  Note that soclose() may be called
1221  * prior to the ref count reaching zero.  The actual socket structure will
1222  * not be freed until the ref count reaches zero.
1223  */
1224 int
1225 soclose(struct socket *so)
1226 {
1227 	struct accept_queue lqueue;
1228 	int error = 0;
1229 	bool listening, last __diagused;
1230 
1231 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1232 
1233 	CURVNET_SET(so->so_vnet);
1234 	funsetown(&so->so_sigio);
1235 	if (so->so_state & SS_ISCONNECTED) {
1236 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1237 			error = sodisconnect(so);
1238 			if (error) {
1239 				if (error == ENOTCONN)
1240 					error = 0;
1241 				goto drop;
1242 			}
1243 		}
1244 
1245 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1246 			if ((so->so_state & SS_ISDISCONNECTING) &&
1247 			    (so->so_state & SS_NBIO))
1248 				goto drop;
1249 			while (so->so_state & SS_ISCONNECTED) {
1250 				error = tsleep(&so->so_timeo,
1251 				    PSOCK | PCATCH, "soclos",
1252 				    so->so_linger * hz);
1253 				if (error)
1254 					break;
1255 			}
1256 		}
1257 	}
1258 
1259 drop:
1260 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1261 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1262 
1263 	SOCK_LOCK(so);
1264 	if ((listening = SOLISTENING(so))) {
1265 		struct socket *sp;
1266 
1267 		TAILQ_INIT(&lqueue);
1268 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1269 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1270 
1271 		so->sol_qlen = so->sol_incqlen = 0;
1272 
1273 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1274 			SOCK_LOCK(sp);
1275 			sp->so_qstate = SQ_NONE;
1276 			sp->so_listen = NULL;
1277 			SOCK_UNLOCK(sp);
1278 			last = refcount_release(&so->so_count);
1279 			KASSERT(!last, ("%s: released last reference for %p",
1280 			    __func__, so));
1281 		}
1282 	}
1283 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1284 	so->so_state |= SS_NOFDREF;
1285 	sorele(so);
1286 	if (listening) {
1287 		struct socket *sp, *tsp;
1288 
1289 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1290 			SOCK_LOCK(sp);
1291 			if (refcount_load(&sp->so_count) == 0) {
1292 				SOCK_UNLOCK(sp);
1293 				soabort(sp);
1294 			} else {
1295 				/* See the handling of queued sockets
1296 				   in sofree(). */
1297 				SOCK_UNLOCK(sp);
1298 			}
1299 		}
1300 	}
1301 	CURVNET_RESTORE();
1302 	return (error);
1303 }
1304 
1305 /*
1306  * soabort() is used to abruptly tear down a connection, such as when a
1307  * resource limit is reached (listen queue depth exceeded), or if a listen
1308  * socket is closed while there are sockets waiting to be accepted.
1309  *
1310  * This interface is tricky, because it is called on an unreferenced socket,
1311  * and must be called only by a thread that has actually removed the socket
1312  * from the listen queue it was on, or races with other threads are risked.
1313  *
1314  * This interface will call into the protocol code, so must not be called
1315  * with any socket locks held.  Protocols do call it while holding their own
1316  * recursible protocol mutexes, but this is something that should be subject
1317  * to review in the future.
1318  */
1319 void
1320 soabort(struct socket *so)
1321 {
1322 
1323 	/*
1324 	 * In as much as is possible, assert that no references to this
1325 	 * socket are held.  This is not quite the same as asserting that the
1326 	 * current thread is responsible for arranging for no references, but
1327 	 * is as close as we can get for now.
1328 	 */
1329 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1330 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1331 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1332 	VNET_SO_ASSERT(so);
1333 
1334 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1335 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1336 	SOCK_LOCK(so);
1337 	sofree(so);
1338 }
1339 
1340 int
1341 soaccept(struct socket *so, struct sockaddr **nam)
1342 {
1343 	int error;
1344 
1345 	SOCK_LOCK(so);
1346 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1347 	so->so_state &= ~SS_NOFDREF;
1348 	SOCK_UNLOCK(so);
1349 
1350 	CURVNET_SET(so->so_vnet);
1351 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1352 	CURVNET_RESTORE();
1353 	return (error);
1354 }
1355 
1356 int
1357 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1358 {
1359 
1360 	return (soconnectat(AT_FDCWD, so, nam, td));
1361 }
1362 
1363 int
1364 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1365 {
1366 	int error;
1367 
1368 	CURVNET_SET(so->so_vnet);
1369 	/*
1370 	 * If protocol is connection-based, can only connect once.
1371 	 * Otherwise, if connected, try to disconnect first.  This allows
1372 	 * user to disconnect by connecting to, e.g., a null address.
1373 	 */
1374 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1375 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1376 	    (error = sodisconnect(so)))) {
1377 		error = EISCONN;
1378 	} else {
1379 		/*
1380 		 * Prevent accumulated error from previous connection from
1381 		 * biting us.
1382 		 */
1383 		so->so_error = 0;
1384 		if (fd == AT_FDCWD) {
1385 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1386 			    nam, td);
1387 		} else {
1388 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1389 			    so, nam, td);
1390 		}
1391 	}
1392 	CURVNET_RESTORE();
1393 
1394 	return (error);
1395 }
1396 
1397 int
1398 soconnect2(struct socket *so1, struct socket *so2)
1399 {
1400 	int error;
1401 
1402 	CURVNET_SET(so1->so_vnet);
1403 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1404 	CURVNET_RESTORE();
1405 	return (error);
1406 }
1407 
1408 int
1409 sodisconnect(struct socket *so)
1410 {
1411 	int error;
1412 
1413 	if ((so->so_state & SS_ISCONNECTED) == 0)
1414 		return (ENOTCONN);
1415 	if (so->so_state & SS_ISDISCONNECTING)
1416 		return (EALREADY);
1417 	VNET_SO_ASSERT(so);
1418 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1419 	return (error);
1420 }
1421 
1422 int
1423 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1424     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1425 {
1426 	long space;
1427 	ssize_t resid;
1428 	int clen = 0, error, dontroute;
1429 
1430 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1431 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1432 	    ("sosend_dgram: !PR_ATOMIC"));
1433 
1434 	if (uio != NULL)
1435 		resid = uio->uio_resid;
1436 	else
1437 		resid = top->m_pkthdr.len;
1438 	/*
1439 	 * In theory resid should be unsigned.  However, space must be
1440 	 * signed, as it might be less than 0 if we over-committed, and we
1441 	 * must use a signed comparison of space and resid.  On the other
1442 	 * hand, a negative resid causes us to loop sending 0-length
1443 	 * segments to the protocol.
1444 	 */
1445 	if (resid < 0) {
1446 		error = EINVAL;
1447 		goto out;
1448 	}
1449 
1450 	dontroute =
1451 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1452 	if (td != NULL)
1453 		td->td_ru.ru_msgsnd++;
1454 	if (control != NULL)
1455 		clen = control->m_len;
1456 
1457 	SOCKBUF_LOCK(&so->so_snd);
1458 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1459 		SOCKBUF_UNLOCK(&so->so_snd);
1460 		error = EPIPE;
1461 		goto out;
1462 	}
1463 	if (so->so_error) {
1464 		error = so->so_error;
1465 		so->so_error = 0;
1466 		SOCKBUF_UNLOCK(&so->so_snd);
1467 		goto out;
1468 	}
1469 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1470 		/*
1471 		 * `sendto' and `sendmsg' is allowed on a connection-based
1472 		 * socket if it supports implied connect.  Return ENOTCONN if
1473 		 * not connected and no address is supplied.
1474 		 */
1475 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1476 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1477 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1478 			    !(resid == 0 && clen != 0)) {
1479 				SOCKBUF_UNLOCK(&so->so_snd);
1480 				error = ENOTCONN;
1481 				goto out;
1482 			}
1483 		} else if (addr == NULL) {
1484 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1485 				error = ENOTCONN;
1486 			else
1487 				error = EDESTADDRREQ;
1488 			SOCKBUF_UNLOCK(&so->so_snd);
1489 			goto out;
1490 		}
1491 	}
1492 
1493 	/*
1494 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1495 	 * problem and need fixing.
1496 	 */
1497 	space = sbspace(&so->so_snd);
1498 	if (flags & MSG_OOB)
1499 		space += 1024;
1500 	space -= clen;
1501 	SOCKBUF_UNLOCK(&so->so_snd);
1502 	if (resid > space) {
1503 		error = EMSGSIZE;
1504 		goto out;
1505 	}
1506 	if (uio == NULL) {
1507 		resid = 0;
1508 		if (flags & MSG_EOR)
1509 			top->m_flags |= M_EOR;
1510 	} else {
1511 		/*
1512 		 * Copy the data from userland into a mbuf chain.
1513 		 * If no data is to be copied in, a single empty mbuf
1514 		 * is returned.
1515 		 */
1516 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1517 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1518 		if (top == NULL) {
1519 			error = EFAULT;	/* only possible error */
1520 			goto out;
1521 		}
1522 		space -= resid - uio->uio_resid;
1523 		resid = uio->uio_resid;
1524 	}
1525 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1526 	/*
1527 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1528 	 * than with.
1529 	 */
1530 	if (dontroute) {
1531 		SOCK_LOCK(so);
1532 		so->so_options |= SO_DONTROUTE;
1533 		SOCK_UNLOCK(so);
1534 	}
1535 	/*
1536 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1537 	 * of date.  We could have received a reset packet in an interrupt or
1538 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1539 	 * probably recheck again inside the locking protection here, but
1540 	 * there are probably other places that this also happens.  We must
1541 	 * rethink this.
1542 	 */
1543 	VNET_SO_ASSERT(so);
1544 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1545 	    (flags & MSG_OOB) ? PRUS_OOB :
1546 	/*
1547 	 * If the user set MSG_EOF, the protocol understands this flag and
1548 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1549 	 */
1550 	    ((flags & MSG_EOF) &&
1551 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1552 	     (resid <= 0)) ?
1553 		PRUS_EOF :
1554 		/* If there is more to send set PRUS_MORETOCOME */
1555 		(flags & MSG_MORETOCOME) ||
1556 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1557 		top, addr, control, td);
1558 	if (dontroute) {
1559 		SOCK_LOCK(so);
1560 		so->so_options &= ~SO_DONTROUTE;
1561 		SOCK_UNLOCK(so);
1562 	}
1563 	clen = 0;
1564 	control = NULL;
1565 	top = NULL;
1566 out:
1567 	if (top != NULL)
1568 		m_freem(top);
1569 	if (control != NULL)
1570 		m_freem(control);
1571 	return (error);
1572 }
1573 
1574 /*
1575  * Send on a socket.  If send must go all at once and message is larger than
1576  * send buffering, then hard error.  Lock against other senders.  If must go
1577  * all at once and not enough room now, then inform user that this would
1578  * block and do nothing.  Otherwise, if nonblocking, send as much as
1579  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1580  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1581  * in mbuf chain must be small enough to send all at once.
1582  *
1583  * Returns nonzero on error, timeout or signal; callers must check for short
1584  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1585  * on return.
1586  */
1587 int
1588 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1589     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1590 {
1591 	long space;
1592 	ssize_t resid;
1593 	int clen = 0, error, dontroute;
1594 	int atomic = sosendallatonce(so) || top;
1595 	int pru_flag;
1596 #ifdef KERN_TLS
1597 	struct ktls_session *tls;
1598 	int tls_enq_cnt, tls_pruflag;
1599 	uint8_t tls_rtype;
1600 
1601 	tls = NULL;
1602 	tls_rtype = TLS_RLTYPE_APP;
1603 #endif
1604 	if (uio != NULL)
1605 		resid = uio->uio_resid;
1606 	else if ((top->m_flags & M_PKTHDR) != 0)
1607 		resid = top->m_pkthdr.len;
1608 	else
1609 		resid = m_length(top, NULL);
1610 	/*
1611 	 * In theory resid should be unsigned.  However, space must be
1612 	 * signed, as it might be less than 0 if we over-committed, and we
1613 	 * must use a signed comparison of space and resid.  On the other
1614 	 * hand, a negative resid causes us to loop sending 0-length
1615 	 * segments to the protocol.
1616 	 *
1617 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1618 	 * type sockets since that's an error.
1619 	 */
1620 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1621 		error = EINVAL;
1622 		goto out;
1623 	}
1624 
1625 	dontroute =
1626 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1627 	    (so->so_proto->pr_flags & PR_ATOMIC);
1628 	if (td != NULL)
1629 		td->td_ru.ru_msgsnd++;
1630 	if (control != NULL)
1631 		clen = control->m_len;
1632 
1633 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1634 	if (error)
1635 		goto out;
1636 
1637 #ifdef KERN_TLS
1638 	tls_pruflag = 0;
1639 	tls = ktls_hold(so->so_snd.sb_tls_info);
1640 	if (tls != NULL) {
1641 		if (tls->mode == TCP_TLS_MODE_SW)
1642 			tls_pruflag = PRUS_NOTREADY;
1643 
1644 		if (control != NULL) {
1645 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1646 
1647 			if (clen >= sizeof(*cm) &&
1648 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1649 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1650 				clen = 0;
1651 				m_freem(control);
1652 				control = NULL;
1653 				atomic = 1;
1654 			}
1655 		}
1656 	}
1657 #endif
1658 
1659 restart:
1660 	do {
1661 		SOCKBUF_LOCK(&so->so_snd);
1662 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1663 			SOCKBUF_UNLOCK(&so->so_snd);
1664 			error = EPIPE;
1665 			goto release;
1666 		}
1667 		if (so->so_error) {
1668 			error = so->so_error;
1669 			so->so_error = 0;
1670 			SOCKBUF_UNLOCK(&so->so_snd);
1671 			goto release;
1672 		}
1673 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1674 			/*
1675 			 * `sendto' and `sendmsg' is allowed on a connection-
1676 			 * based socket if it supports implied connect.
1677 			 * Return ENOTCONN if not connected and no address is
1678 			 * supplied.
1679 			 */
1680 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1681 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1682 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1683 				    !(resid == 0 && clen != 0)) {
1684 					SOCKBUF_UNLOCK(&so->so_snd);
1685 					error = ENOTCONN;
1686 					goto release;
1687 				}
1688 			} else if (addr == NULL) {
1689 				SOCKBUF_UNLOCK(&so->so_snd);
1690 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1691 					error = ENOTCONN;
1692 				else
1693 					error = EDESTADDRREQ;
1694 				goto release;
1695 			}
1696 		}
1697 		space = sbspace(&so->so_snd);
1698 		if (flags & MSG_OOB)
1699 			space += 1024;
1700 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1701 		    clen > so->so_snd.sb_hiwat) {
1702 			SOCKBUF_UNLOCK(&so->so_snd);
1703 			error = EMSGSIZE;
1704 			goto release;
1705 		}
1706 		if (space < resid + clen &&
1707 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1708 			if ((so->so_state & SS_NBIO) ||
1709 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1710 				SOCKBUF_UNLOCK(&so->so_snd);
1711 				error = EWOULDBLOCK;
1712 				goto release;
1713 			}
1714 			error = sbwait(&so->so_snd);
1715 			SOCKBUF_UNLOCK(&so->so_snd);
1716 			if (error)
1717 				goto release;
1718 			goto restart;
1719 		}
1720 		SOCKBUF_UNLOCK(&so->so_snd);
1721 		space -= clen;
1722 		do {
1723 			if (uio == NULL) {
1724 				resid = 0;
1725 				if (flags & MSG_EOR)
1726 					top->m_flags |= M_EOR;
1727 #ifdef KERN_TLS
1728 				if (tls != NULL) {
1729 					ktls_frame(top, tls, &tls_enq_cnt,
1730 					    tls_rtype);
1731 					tls_rtype = TLS_RLTYPE_APP;
1732 				}
1733 #endif
1734 			} else {
1735 				/*
1736 				 * Copy the data from userland into a mbuf
1737 				 * chain.  If resid is 0, which can happen
1738 				 * only if we have control to send, then
1739 				 * a single empty mbuf is returned.  This
1740 				 * is a workaround to prevent protocol send
1741 				 * methods to panic.
1742 				 */
1743 #ifdef KERN_TLS
1744 				if (tls != NULL) {
1745 					top = m_uiotombuf(uio, M_WAITOK, space,
1746 					    tls->params.max_frame_len,
1747 					    M_EXTPG |
1748 					    ((flags & MSG_EOR) ? M_EOR : 0));
1749 					if (top != NULL) {
1750 						ktls_frame(top, tls,
1751 						    &tls_enq_cnt, tls_rtype);
1752 					}
1753 					tls_rtype = TLS_RLTYPE_APP;
1754 				} else
1755 #endif
1756 					top = m_uiotombuf(uio, M_WAITOK, space,
1757 					    (atomic ? max_hdr : 0),
1758 					    (atomic ? M_PKTHDR : 0) |
1759 					    ((flags & MSG_EOR) ? M_EOR : 0));
1760 				if (top == NULL) {
1761 					error = EFAULT; /* only possible error */
1762 					goto release;
1763 				}
1764 				space -= resid - uio->uio_resid;
1765 				resid = uio->uio_resid;
1766 			}
1767 			if (dontroute) {
1768 				SOCK_LOCK(so);
1769 				so->so_options |= SO_DONTROUTE;
1770 				SOCK_UNLOCK(so);
1771 			}
1772 			/*
1773 			 * XXX all the SBS_CANTSENDMORE checks previously
1774 			 * done could be out of date.  We could have received
1775 			 * a reset packet in an interrupt or maybe we slept
1776 			 * while doing page faults in uiomove() etc.  We
1777 			 * could probably recheck again inside the locking
1778 			 * protection here, but there are probably other
1779 			 * places that this also happens.  We must rethink
1780 			 * this.
1781 			 */
1782 			VNET_SO_ASSERT(so);
1783 
1784 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1785 			/*
1786 			 * If the user set MSG_EOF, the protocol understands
1787 			 * this flag and nothing left to send then use
1788 			 * PRU_SEND_EOF instead of PRU_SEND.
1789 			 */
1790 			    ((flags & MSG_EOF) &&
1791 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1792 			     (resid <= 0)) ?
1793 				PRUS_EOF :
1794 			/* If there is more to send set PRUS_MORETOCOME. */
1795 			    (flags & MSG_MORETOCOME) ||
1796 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1797 
1798 #ifdef KERN_TLS
1799 			pru_flag |= tls_pruflag;
1800 #endif
1801 
1802 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1803 			    pru_flag, top, addr, control, td);
1804 
1805 			if (dontroute) {
1806 				SOCK_LOCK(so);
1807 				so->so_options &= ~SO_DONTROUTE;
1808 				SOCK_UNLOCK(so);
1809 			}
1810 
1811 #ifdef KERN_TLS
1812 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1813 				if (error != 0) {
1814 					m_freem(top);
1815 					top = NULL;
1816 				} else {
1817 					soref(so);
1818 					ktls_enqueue(top, so, tls_enq_cnt);
1819 				}
1820 			}
1821 #endif
1822 			clen = 0;
1823 			control = NULL;
1824 			top = NULL;
1825 			if (error)
1826 				goto release;
1827 		} while (resid && space > 0);
1828 	} while (resid);
1829 
1830 release:
1831 	SOCK_IO_SEND_UNLOCK(so);
1832 out:
1833 #ifdef KERN_TLS
1834 	if (tls != NULL)
1835 		ktls_free(tls);
1836 #endif
1837 	if (top != NULL)
1838 		m_freem(top);
1839 	if (control != NULL)
1840 		m_freem(control);
1841 	return (error);
1842 }
1843 
1844 int
1845 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1846     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1847 {
1848 	int error;
1849 
1850 	CURVNET_SET(so->so_vnet);
1851 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1852 	    top, control, flags, td);
1853 	CURVNET_RESTORE();
1854 	return (error);
1855 }
1856 
1857 /*
1858  * The part of soreceive() that implements reading non-inline out-of-band
1859  * data from a socket.  For more complete comments, see soreceive(), from
1860  * which this code originated.
1861  *
1862  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1863  * unable to return an mbuf chain to the caller.
1864  */
1865 static int
1866 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1867 {
1868 	struct protosw *pr = so->so_proto;
1869 	struct mbuf *m;
1870 	int error;
1871 
1872 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1873 	VNET_SO_ASSERT(so);
1874 
1875 	m = m_get(M_WAITOK, MT_DATA);
1876 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1877 	if (error)
1878 		goto bad;
1879 	do {
1880 		error = uiomove(mtod(m, void *),
1881 		    (int) min(uio->uio_resid, m->m_len), uio);
1882 		m = m_free(m);
1883 	} while (uio->uio_resid && error == 0 && m);
1884 bad:
1885 	if (m != NULL)
1886 		m_freem(m);
1887 	return (error);
1888 }
1889 
1890 /*
1891  * Following replacement or removal of the first mbuf on the first mbuf chain
1892  * of a socket buffer, push necessary state changes back into the socket
1893  * buffer so that other consumers see the values consistently.  'nextrecord'
1894  * is the callers locally stored value of the original value of
1895  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1896  * NOTE: 'nextrecord' may be NULL.
1897  */
1898 static __inline void
1899 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1900 {
1901 
1902 	SOCKBUF_LOCK_ASSERT(sb);
1903 	/*
1904 	 * First, update for the new value of nextrecord.  If necessary, make
1905 	 * it the first record.
1906 	 */
1907 	if (sb->sb_mb != NULL)
1908 		sb->sb_mb->m_nextpkt = nextrecord;
1909 	else
1910 		sb->sb_mb = nextrecord;
1911 
1912 	/*
1913 	 * Now update any dependent socket buffer fields to reflect the new
1914 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1915 	 * addition of a second clause that takes care of the case where
1916 	 * sb_mb has been updated, but remains the last record.
1917 	 */
1918 	if (sb->sb_mb == NULL) {
1919 		sb->sb_mbtail = NULL;
1920 		sb->sb_lastrecord = NULL;
1921 	} else if (sb->sb_mb->m_nextpkt == NULL)
1922 		sb->sb_lastrecord = sb->sb_mb;
1923 }
1924 
1925 /*
1926  * Implement receive operations on a socket.  We depend on the way that
1927  * records are added to the sockbuf by sbappend.  In particular, each record
1928  * (mbufs linked through m_next) must begin with an address if the protocol
1929  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1930  * data, and then zero or more mbufs of data.  In order to allow parallelism
1931  * between network receive and copying to user space, as well as avoid
1932  * sleeping with a mutex held, we release the socket buffer mutex during the
1933  * user space copy.  Although the sockbuf is locked, new data may still be
1934  * appended, and thus we must maintain consistency of the sockbuf during that
1935  * time.
1936  *
1937  * The caller may receive the data as a single mbuf chain by supplying an
1938  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1939  * the count in uio_resid.
1940  */
1941 int
1942 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1943     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1944 {
1945 	struct mbuf *m, **mp;
1946 	int flags, error, offset;
1947 	ssize_t len;
1948 	struct protosw *pr = so->so_proto;
1949 	struct mbuf *nextrecord;
1950 	int moff, type = 0;
1951 	ssize_t orig_resid = uio->uio_resid;
1952 
1953 	mp = mp0;
1954 	if (psa != NULL)
1955 		*psa = NULL;
1956 	if (controlp != NULL)
1957 		*controlp = NULL;
1958 	if (flagsp != NULL)
1959 		flags = *flagsp &~ MSG_EOR;
1960 	else
1961 		flags = 0;
1962 	if (flags & MSG_OOB)
1963 		return (soreceive_rcvoob(so, uio, flags));
1964 	if (mp != NULL)
1965 		*mp = NULL;
1966 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1967 	    && uio->uio_resid) {
1968 		VNET_SO_ASSERT(so);
1969 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1970 	}
1971 
1972 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1973 	if (error)
1974 		return (error);
1975 
1976 restart:
1977 	SOCKBUF_LOCK(&so->so_rcv);
1978 	m = so->so_rcv.sb_mb;
1979 	/*
1980 	 * If we have less data than requested, block awaiting more (subject
1981 	 * to any timeout) if:
1982 	 *   1. the current count is less than the low water mark, or
1983 	 *   2. MSG_DONTWAIT is not set
1984 	 */
1985 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1986 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1987 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1988 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1989 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1990 		    ("receive: m == %p sbavail == %u",
1991 		    m, sbavail(&so->so_rcv)));
1992 		if (so->so_error || so->so_rerror) {
1993 			if (m != NULL)
1994 				goto dontblock;
1995 			if (so->so_error)
1996 				error = so->so_error;
1997 			else
1998 				error = so->so_rerror;
1999 			if ((flags & MSG_PEEK) == 0) {
2000 				if (so->so_error)
2001 					so->so_error = 0;
2002 				else
2003 					so->so_rerror = 0;
2004 			}
2005 			SOCKBUF_UNLOCK(&so->so_rcv);
2006 			goto release;
2007 		}
2008 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2009 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2010 			if (m != NULL)
2011 				goto dontblock;
2012 #ifdef KERN_TLS
2013 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2014 			    so->so_rcv.sb_tlscc == 0) {
2015 #else
2016 			else {
2017 #endif
2018 				SOCKBUF_UNLOCK(&so->so_rcv);
2019 				goto release;
2020 			}
2021 		}
2022 		for (; m != NULL; m = m->m_next)
2023 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2024 				m = so->so_rcv.sb_mb;
2025 				goto dontblock;
2026 			}
2027 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2028 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2029 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2030 			SOCKBUF_UNLOCK(&so->so_rcv);
2031 			error = ENOTCONN;
2032 			goto release;
2033 		}
2034 		if (uio->uio_resid == 0) {
2035 			SOCKBUF_UNLOCK(&so->so_rcv);
2036 			goto release;
2037 		}
2038 		if ((so->so_state & SS_NBIO) ||
2039 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2040 			SOCKBUF_UNLOCK(&so->so_rcv);
2041 			error = EWOULDBLOCK;
2042 			goto release;
2043 		}
2044 		SBLASTRECORDCHK(&so->so_rcv);
2045 		SBLASTMBUFCHK(&so->so_rcv);
2046 		error = sbwait(&so->so_rcv);
2047 		SOCKBUF_UNLOCK(&so->so_rcv);
2048 		if (error)
2049 			goto release;
2050 		goto restart;
2051 	}
2052 dontblock:
2053 	/*
2054 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2055 	 * pointer to the next record in the socket buffer.  We must keep the
2056 	 * various socket buffer pointers and local stack versions of the
2057 	 * pointers in sync, pushing out modifications before dropping the
2058 	 * socket buffer mutex, and re-reading them when picking it up.
2059 	 *
2060 	 * Otherwise, we will race with the network stack appending new data
2061 	 * or records onto the socket buffer by using inconsistent/stale
2062 	 * versions of the field, possibly resulting in socket buffer
2063 	 * corruption.
2064 	 *
2065 	 * By holding the high-level sblock(), we prevent simultaneous
2066 	 * readers from pulling off the front of the socket buffer.
2067 	 */
2068 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2069 	if (uio->uio_td)
2070 		uio->uio_td->td_ru.ru_msgrcv++;
2071 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2072 	SBLASTRECORDCHK(&so->so_rcv);
2073 	SBLASTMBUFCHK(&so->so_rcv);
2074 	nextrecord = m->m_nextpkt;
2075 	if (pr->pr_flags & PR_ADDR) {
2076 		KASSERT(m->m_type == MT_SONAME,
2077 		    ("m->m_type == %d", m->m_type));
2078 		orig_resid = 0;
2079 		if (psa != NULL)
2080 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2081 			    M_NOWAIT);
2082 		if (flags & MSG_PEEK) {
2083 			m = m->m_next;
2084 		} else {
2085 			sbfree(&so->so_rcv, m);
2086 			so->so_rcv.sb_mb = m_free(m);
2087 			m = so->so_rcv.sb_mb;
2088 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2089 		}
2090 	}
2091 
2092 	/*
2093 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2094 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2095 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2096 	 * perform externalization (or freeing if controlp == NULL).
2097 	 */
2098 	if (m != NULL && m->m_type == MT_CONTROL) {
2099 		struct mbuf *cm = NULL, *cmn;
2100 		struct mbuf **cme = &cm;
2101 #ifdef KERN_TLS
2102 		struct cmsghdr *cmsg;
2103 		struct tls_get_record tgr;
2104 
2105 		/*
2106 		 * For MSG_TLSAPPDATA, check for a non-application data
2107 		 * record.  If found, return ENXIO without removing
2108 		 * it from the receive queue.  This allows a subsequent
2109 		 * call without MSG_TLSAPPDATA to receive it.
2110 		 * Note that, for TLS, there should only be a single
2111 		 * control mbuf with the TLS_GET_RECORD message in it.
2112 		 */
2113 		if (flags & MSG_TLSAPPDATA) {
2114 			cmsg = mtod(m, struct cmsghdr *);
2115 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2116 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2117 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2118 				/* This will need to change for TLS 1.3. */
2119 				if (tgr.tls_type != TLS_RLTYPE_APP) {
2120 					SOCKBUF_UNLOCK(&so->so_rcv);
2121 					error = ENXIO;
2122 					goto release;
2123 				}
2124 			}
2125 		}
2126 #endif
2127 
2128 		do {
2129 			if (flags & MSG_PEEK) {
2130 				if (controlp != NULL) {
2131 					*controlp = m_copym(m, 0, m->m_len,
2132 					    M_NOWAIT);
2133 					controlp = &(*controlp)->m_next;
2134 				}
2135 				m = m->m_next;
2136 			} else {
2137 				sbfree(&so->so_rcv, m);
2138 				so->so_rcv.sb_mb = m->m_next;
2139 				m->m_next = NULL;
2140 				*cme = m;
2141 				cme = &(*cme)->m_next;
2142 				m = so->so_rcv.sb_mb;
2143 			}
2144 		} while (m != NULL && m->m_type == MT_CONTROL);
2145 		if ((flags & MSG_PEEK) == 0)
2146 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2147 		while (cm != NULL) {
2148 			cmn = cm->m_next;
2149 			cm->m_next = NULL;
2150 			if (pr->pr_domain->dom_externalize != NULL) {
2151 				SOCKBUF_UNLOCK(&so->so_rcv);
2152 				VNET_SO_ASSERT(so);
2153 				error = (*pr->pr_domain->dom_externalize)
2154 				    (cm, controlp, flags);
2155 				SOCKBUF_LOCK(&so->so_rcv);
2156 			} else if (controlp != NULL)
2157 				*controlp = cm;
2158 			else
2159 				m_freem(cm);
2160 			if (controlp != NULL) {
2161 				orig_resid = 0;
2162 				while (*controlp != NULL)
2163 					controlp = &(*controlp)->m_next;
2164 			}
2165 			cm = cmn;
2166 		}
2167 		if (m != NULL)
2168 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2169 		else
2170 			nextrecord = so->so_rcv.sb_mb;
2171 		orig_resid = 0;
2172 	}
2173 	if (m != NULL) {
2174 		if ((flags & MSG_PEEK) == 0) {
2175 			KASSERT(m->m_nextpkt == nextrecord,
2176 			    ("soreceive: post-control, nextrecord !sync"));
2177 			if (nextrecord == NULL) {
2178 				KASSERT(so->so_rcv.sb_mb == m,
2179 				    ("soreceive: post-control, sb_mb!=m"));
2180 				KASSERT(so->so_rcv.sb_lastrecord == m,
2181 				    ("soreceive: post-control, lastrecord!=m"));
2182 			}
2183 		}
2184 		type = m->m_type;
2185 		if (type == MT_OOBDATA)
2186 			flags |= MSG_OOB;
2187 	} else {
2188 		if ((flags & MSG_PEEK) == 0) {
2189 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2190 			    ("soreceive: sb_mb != nextrecord"));
2191 			if (so->so_rcv.sb_mb == NULL) {
2192 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2193 				    ("soreceive: sb_lastercord != NULL"));
2194 			}
2195 		}
2196 	}
2197 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2198 	SBLASTRECORDCHK(&so->so_rcv);
2199 	SBLASTMBUFCHK(&so->so_rcv);
2200 
2201 	/*
2202 	 * Now continue to read any data mbufs off of the head of the socket
2203 	 * buffer until the read request is satisfied.  Note that 'type' is
2204 	 * used to store the type of any mbuf reads that have happened so far
2205 	 * such that soreceive() can stop reading if the type changes, which
2206 	 * causes soreceive() to return only one of regular data and inline
2207 	 * out-of-band data in a single socket receive operation.
2208 	 */
2209 	moff = 0;
2210 	offset = 0;
2211 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2212 	    && error == 0) {
2213 		/*
2214 		 * If the type of mbuf has changed since the last mbuf
2215 		 * examined ('type'), end the receive operation.
2216 		 */
2217 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2218 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2219 			if (type != m->m_type)
2220 				break;
2221 		} else if (type == MT_OOBDATA)
2222 			break;
2223 		else
2224 		    KASSERT(m->m_type == MT_DATA,
2225 			("m->m_type == %d", m->m_type));
2226 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2227 		len = uio->uio_resid;
2228 		if (so->so_oobmark && len > so->so_oobmark - offset)
2229 			len = so->so_oobmark - offset;
2230 		if (len > m->m_len - moff)
2231 			len = m->m_len - moff;
2232 		/*
2233 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2234 		 * them out via the uio, then free.  Sockbuf must be
2235 		 * consistent here (points to current mbuf, it points to next
2236 		 * record) when we drop priority; we must note any additions
2237 		 * to the sockbuf when we block interrupts again.
2238 		 */
2239 		if (mp == NULL) {
2240 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2241 			SBLASTRECORDCHK(&so->so_rcv);
2242 			SBLASTMBUFCHK(&so->so_rcv);
2243 			SOCKBUF_UNLOCK(&so->so_rcv);
2244 			if ((m->m_flags & M_EXTPG) != 0)
2245 				error = m_unmapped_uiomove(m, moff, uio,
2246 				    (int)len);
2247 			else
2248 				error = uiomove(mtod(m, char *) + moff,
2249 				    (int)len, uio);
2250 			SOCKBUF_LOCK(&so->so_rcv);
2251 			if (error) {
2252 				/*
2253 				 * The MT_SONAME mbuf has already been removed
2254 				 * from the record, so it is necessary to
2255 				 * remove the data mbufs, if any, to preserve
2256 				 * the invariant in the case of PR_ADDR that
2257 				 * requires MT_SONAME mbufs at the head of
2258 				 * each record.
2259 				 */
2260 				if (pr->pr_flags & PR_ATOMIC &&
2261 				    ((flags & MSG_PEEK) == 0))
2262 					(void)sbdroprecord_locked(&so->so_rcv);
2263 				SOCKBUF_UNLOCK(&so->so_rcv);
2264 				goto release;
2265 			}
2266 		} else
2267 			uio->uio_resid -= len;
2268 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2269 		if (len == m->m_len - moff) {
2270 			if (m->m_flags & M_EOR)
2271 				flags |= MSG_EOR;
2272 			if (flags & MSG_PEEK) {
2273 				m = m->m_next;
2274 				moff = 0;
2275 			} else {
2276 				nextrecord = m->m_nextpkt;
2277 				sbfree(&so->so_rcv, m);
2278 				if (mp != NULL) {
2279 					m->m_nextpkt = NULL;
2280 					*mp = m;
2281 					mp = &m->m_next;
2282 					so->so_rcv.sb_mb = m = m->m_next;
2283 					*mp = NULL;
2284 				} else {
2285 					so->so_rcv.sb_mb = m_free(m);
2286 					m = so->so_rcv.sb_mb;
2287 				}
2288 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2289 				SBLASTRECORDCHK(&so->so_rcv);
2290 				SBLASTMBUFCHK(&so->so_rcv);
2291 			}
2292 		} else {
2293 			if (flags & MSG_PEEK)
2294 				moff += len;
2295 			else {
2296 				if (mp != NULL) {
2297 					if (flags & MSG_DONTWAIT) {
2298 						*mp = m_copym(m, 0, len,
2299 						    M_NOWAIT);
2300 						if (*mp == NULL) {
2301 							/*
2302 							 * m_copym() couldn't
2303 							 * allocate an mbuf.
2304 							 * Adjust uio_resid back
2305 							 * (it was adjusted
2306 							 * down by len bytes,
2307 							 * which we didn't end
2308 							 * up "copying" over).
2309 							 */
2310 							uio->uio_resid += len;
2311 							break;
2312 						}
2313 					} else {
2314 						SOCKBUF_UNLOCK(&so->so_rcv);
2315 						*mp = m_copym(m, 0, len,
2316 						    M_WAITOK);
2317 						SOCKBUF_LOCK(&so->so_rcv);
2318 					}
2319 				}
2320 				sbcut_locked(&so->so_rcv, len);
2321 			}
2322 		}
2323 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2324 		if (so->so_oobmark) {
2325 			if ((flags & MSG_PEEK) == 0) {
2326 				so->so_oobmark -= len;
2327 				if (so->so_oobmark == 0) {
2328 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2329 					break;
2330 				}
2331 			} else {
2332 				offset += len;
2333 				if (offset == so->so_oobmark)
2334 					break;
2335 			}
2336 		}
2337 		if (flags & MSG_EOR)
2338 			break;
2339 		/*
2340 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2341 		 * must not quit until "uio->uio_resid == 0" or an error
2342 		 * termination.  If a signal/timeout occurs, return with a
2343 		 * short count but without error.  Keep sockbuf locked
2344 		 * against other readers.
2345 		 */
2346 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2347 		    !sosendallatonce(so) && nextrecord == NULL) {
2348 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2349 			if (so->so_error || so->so_rerror ||
2350 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2351 				break;
2352 			/*
2353 			 * Notify the protocol that some data has been
2354 			 * drained before blocking.
2355 			 */
2356 			if (pr->pr_flags & PR_WANTRCVD) {
2357 				SOCKBUF_UNLOCK(&so->so_rcv);
2358 				VNET_SO_ASSERT(so);
2359 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2360 				SOCKBUF_LOCK(&so->so_rcv);
2361 			}
2362 			SBLASTRECORDCHK(&so->so_rcv);
2363 			SBLASTMBUFCHK(&so->so_rcv);
2364 			/*
2365 			 * We could receive some data while was notifying
2366 			 * the protocol. Skip blocking in this case.
2367 			 */
2368 			if (so->so_rcv.sb_mb == NULL) {
2369 				error = sbwait(&so->so_rcv);
2370 				if (error) {
2371 					SOCKBUF_UNLOCK(&so->so_rcv);
2372 					goto release;
2373 				}
2374 			}
2375 			m = so->so_rcv.sb_mb;
2376 			if (m != NULL)
2377 				nextrecord = m->m_nextpkt;
2378 		}
2379 	}
2380 
2381 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2382 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2383 		flags |= MSG_TRUNC;
2384 		if ((flags & MSG_PEEK) == 0)
2385 			(void) sbdroprecord_locked(&so->so_rcv);
2386 	}
2387 	if ((flags & MSG_PEEK) == 0) {
2388 		if (m == NULL) {
2389 			/*
2390 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2391 			 * part makes sure sb_lastrecord is up-to-date if
2392 			 * there is still data in the socket buffer.
2393 			 */
2394 			so->so_rcv.sb_mb = nextrecord;
2395 			if (so->so_rcv.sb_mb == NULL) {
2396 				so->so_rcv.sb_mbtail = NULL;
2397 				so->so_rcv.sb_lastrecord = NULL;
2398 			} else if (nextrecord->m_nextpkt == NULL)
2399 				so->so_rcv.sb_lastrecord = nextrecord;
2400 		}
2401 		SBLASTRECORDCHK(&so->so_rcv);
2402 		SBLASTMBUFCHK(&so->so_rcv);
2403 		/*
2404 		 * If soreceive() is being done from the socket callback,
2405 		 * then don't need to generate ACK to peer to update window,
2406 		 * since ACK will be generated on return to TCP.
2407 		 */
2408 		if (!(flags & MSG_SOCALLBCK) &&
2409 		    (pr->pr_flags & PR_WANTRCVD)) {
2410 			SOCKBUF_UNLOCK(&so->so_rcv);
2411 			VNET_SO_ASSERT(so);
2412 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2413 			SOCKBUF_LOCK(&so->so_rcv);
2414 		}
2415 	}
2416 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2417 	if (orig_resid == uio->uio_resid && orig_resid &&
2418 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2419 		SOCKBUF_UNLOCK(&so->so_rcv);
2420 		goto restart;
2421 	}
2422 	SOCKBUF_UNLOCK(&so->so_rcv);
2423 
2424 	if (flagsp != NULL)
2425 		*flagsp |= flags;
2426 release:
2427 	SOCK_IO_RECV_UNLOCK(so);
2428 	return (error);
2429 }
2430 
2431 /*
2432  * Optimized version of soreceive() for stream (TCP) sockets.
2433  */
2434 int
2435 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2436     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2437 {
2438 	int len = 0, error = 0, flags, oresid;
2439 	struct sockbuf *sb;
2440 	struct mbuf *m, *n = NULL;
2441 
2442 	/* We only do stream sockets. */
2443 	if (so->so_type != SOCK_STREAM)
2444 		return (EINVAL);
2445 	if (psa != NULL)
2446 		*psa = NULL;
2447 	if (flagsp != NULL)
2448 		flags = *flagsp &~ MSG_EOR;
2449 	else
2450 		flags = 0;
2451 	if (controlp != NULL)
2452 		*controlp = NULL;
2453 	if (flags & MSG_OOB)
2454 		return (soreceive_rcvoob(so, uio, flags));
2455 	if (mp0 != NULL)
2456 		*mp0 = NULL;
2457 
2458 	sb = &so->so_rcv;
2459 
2460 #ifdef KERN_TLS
2461 	/*
2462 	 * KTLS store TLS records as records with a control message to
2463 	 * describe the framing.
2464 	 *
2465 	 * We check once here before acquiring locks to optimize the
2466 	 * common case.
2467 	 */
2468 	if (sb->sb_tls_info != NULL)
2469 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2470 		    flagsp));
2471 #endif
2472 
2473 	/* Prevent other readers from entering the socket. */
2474 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2475 	if (error)
2476 		return (error);
2477 	SOCKBUF_LOCK(sb);
2478 
2479 #ifdef KERN_TLS
2480 	if (sb->sb_tls_info != NULL) {
2481 		SOCKBUF_UNLOCK(sb);
2482 		SOCK_IO_RECV_UNLOCK(so);
2483 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2484 		    flagsp));
2485 	}
2486 #endif
2487 
2488 	/* Easy one, no space to copyout anything. */
2489 	if (uio->uio_resid == 0) {
2490 		error = EINVAL;
2491 		goto out;
2492 	}
2493 	oresid = uio->uio_resid;
2494 
2495 	/* We will never ever get anything unless we are or were connected. */
2496 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2497 		error = ENOTCONN;
2498 		goto out;
2499 	}
2500 
2501 restart:
2502 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2503 
2504 	/* Abort if socket has reported problems. */
2505 	if (so->so_error) {
2506 		if (sbavail(sb) > 0)
2507 			goto deliver;
2508 		if (oresid > uio->uio_resid)
2509 			goto out;
2510 		error = so->so_error;
2511 		if (!(flags & MSG_PEEK))
2512 			so->so_error = 0;
2513 		goto out;
2514 	}
2515 
2516 	/* Door is closed.  Deliver what is left, if any. */
2517 	if (sb->sb_state & SBS_CANTRCVMORE) {
2518 		if (sbavail(sb) > 0)
2519 			goto deliver;
2520 		else
2521 			goto out;
2522 	}
2523 
2524 	/* Socket buffer is empty and we shall not block. */
2525 	if (sbavail(sb) == 0 &&
2526 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2527 		error = EAGAIN;
2528 		goto out;
2529 	}
2530 
2531 	/* Socket buffer got some data that we shall deliver now. */
2532 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2533 	    ((so->so_state & SS_NBIO) ||
2534 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2535 	     sbavail(sb) >= sb->sb_lowat ||
2536 	     sbavail(sb) >= uio->uio_resid ||
2537 	     sbavail(sb) >= sb->sb_hiwat) ) {
2538 		goto deliver;
2539 	}
2540 
2541 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2542 	if ((flags & MSG_WAITALL) &&
2543 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2544 		goto deliver;
2545 
2546 	/*
2547 	 * Wait and block until (more) data comes in.
2548 	 * NB: Drops the sockbuf lock during wait.
2549 	 */
2550 	error = sbwait(sb);
2551 	if (error)
2552 		goto out;
2553 	goto restart;
2554 
2555 deliver:
2556 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2557 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2558 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2559 
2560 	/* Statistics. */
2561 	if (uio->uio_td)
2562 		uio->uio_td->td_ru.ru_msgrcv++;
2563 
2564 	/* Fill uio until full or current end of socket buffer is reached. */
2565 	len = min(uio->uio_resid, sbavail(sb));
2566 	if (mp0 != NULL) {
2567 		/* Dequeue as many mbufs as possible. */
2568 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2569 			if (*mp0 == NULL)
2570 				*mp0 = sb->sb_mb;
2571 			else
2572 				m_cat(*mp0, sb->sb_mb);
2573 			for (m = sb->sb_mb;
2574 			     m != NULL && m->m_len <= len;
2575 			     m = m->m_next) {
2576 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2577 				    ("%s: m %p not available", __func__, m));
2578 				len -= m->m_len;
2579 				uio->uio_resid -= m->m_len;
2580 				sbfree(sb, m);
2581 				n = m;
2582 			}
2583 			n->m_next = NULL;
2584 			sb->sb_mb = m;
2585 			sb->sb_lastrecord = sb->sb_mb;
2586 			if (sb->sb_mb == NULL)
2587 				SB_EMPTY_FIXUP(sb);
2588 		}
2589 		/* Copy the remainder. */
2590 		if (len > 0) {
2591 			KASSERT(sb->sb_mb != NULL,
2592 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2593 
2594 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2595 			if (m == NULL)
2596 				len = 0;	/* Don't flush data from sockbuf. */
2597 			else
2598 				uio->uio_resid -= len;
2599 			if (*mp0 != NULL)
2600 				m_cat(*mp0, m);
2601 			else
2602 				*mp0 = m;
2603 			if (*mp0 == NULL) {
2604 				error = ENOBUFS;
2605 				goto out;
2606 			}
2607 		}
2608 	} else {
2609 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2610 		SOCKBUF_UNLOCK(sb);
2611 		error = m_mbuftouio(uio, sb->sb_mb, len);
2612 		SOCKBUF_LOCK(sb);
2613 		if (error)
2614 			goto out;
2615 	}
2616 	SBLASTRECORDCHK(sb);
2617 	SBLASTMBUFCHK(sb);
2618 
2619 	/*
2620 	 * Remove the delivered data from the socket buffer unless we
2621 	 * were only peeking.
2622 	 */
2623 	if (!(flags & MSG_PEEK)) {
2624 		if (len > 0)
2625 			sbdrop_locked(sb, len);
2626 
2627 		/* Notify protocol that we drained some data. */
2628 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2629 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2630 		     !(flags & MSG_SOCALLBCK))) {
2631 			SOCKBUF_UNLOCK(sb);
2632 			VNET_SO_ASSERT(so);
2633 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2634 			SOCKBUF_LOCK(sb);
2635 		}
2636 	}
2637 
2638 	/*
2639 	 * For MSG_WAITALL we may have to loop again and wait for
2640 	 * more data to come in.
2641 	 */
2642 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2643 		goto restart;
2644 out:
2645 	SBLASTRECORDCHK(sb);
2646 	SBLASTMBUFCHK(sb);
2647 	SOCKBUF_UNLOCK(sb);
2648 	SOCK_IO_RECV_UNLOCK(so);
2649 	return (error);
2650 }
2651 
2652 /*
2653  * Optimized version of soreceive() for simple datagram cases from userspace.
2654  * Unlike in the stream case, we're able to drop a datagram if copyout()
2655  * fails, and because we handle datagrams atomically, we don't need to use a
2656  * sleep lock to prevent I/O interlacing.
2657  */
2658 int
2659 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2660     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2661 {
2662 	struct mbuf *m, *m2;
2663 	int flags, error;
2664 	ssize_t len;
2665 	struct protosw *pr = so->so_proto;
2666 	struct mbuf *nextrecord;
2667 
2668 	if (psa != NULL)
2669 		*psa = NULL;
2670 	if (controlp != NULL)
2671 		*controlp = NULL;
2672 	if (flagsp != NULL)
2673 		flags = *flagsp &~ MSG_EOR;
2674 	else
2675 		flags = 0;
2676 
2677 	/*
2678 	 * For any complicated cases, fall back to the full
2679 	 * soreceive_generic().
2680 	 */
2681 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2682 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2683 		    flagsp));
2684 
2685 	/*
2686 	 * Enforce restrictions on use.
2687 	 */
2688 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2689 	    ("soreceive_dgram: wantrcvd"));
2690 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2691 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2692 	    ("soreceive_dgram: SBS_RCVATMARK"));
2693 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2694 	    ("soreceive_dgram: P_CONNREQUIRED"));
2695 
2696 	/*
2697 	 * Loop blocking while waiting for a datagram.
2698 	 */
2699 	SOCKBUF_LOCK(&so->so_rcv);
2700 	while ((m = so->so_rcv.sb_mb) == NULL) {
2701 		KASSERT(sbavail(&so->so_rcv) == 0,
2702 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2703 		    sbavail(&so->so_rcv)));
2704 		if (so->so_error) {
2705 			error = so->so_error;
2706 			so->so_error = 0;
2707 			SOCKBUF_UNLOCK(&so->so_rcv);
2708 			return (error);
2709 		}
2710 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2711 		    uio->uio_resid == 0) {
2712 			SOCKBUF_UNLOCK(&so->so_rcv);
2713 			return (0);
2714 		}
2715 		if ((so->so_state & SS_NBIO) ||
2716 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2717 			SOCKBUF_UNLOCK(&so->so_rcv);
2718 			return (EWOULDBLOCK);
2719 		}
2720 		SBLASTRECORDCHK(&so->so_rcv);
2721 		SBLASTMBUFCHK(&so->so_rcv);
2722 		error = sbwait(&so->so_rcv);
2723 		if (error) {
2724 			SOCKBUF_UNLOCK(&so->so_rcv);
2725 			return (error);
2726 		}
2727 	}
2728 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2729 
2730 	if (uio->uio_td)
2731 		uio->uio_td->td_ru.ru_msgrcv++;
2732 	SBLASTRECORDCHK(&so->so_rcv);
2733 	SBLASTMBUFCHK(&so->so_rcv);
2734 	nextrecord = m->m_nextpkt;
2735 	if (nextrecord == NULL) {
2736 		KASSERT(so->so_rcv.sb_lastrecord == m,
2737 		    ("soreceive_dgram: lastrecord != m"));
2738 	}
2739 
2740 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2741 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2742 
2743 	/*
2744 	 * Pull 'm' and its chain off the front of the packet queue.
2745 	 */
2746 	so->so_rcv.sb_mb = NULL;
2747 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2748 
2749 	/*
2750 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2751 	 */
2752 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2753 		sbfree(&so->so_rcv, m2);
2754 
2755 	/*
2756 	 * Do a few last checks before we let go of the lock.
2757 	 */
2758 	SBLASTRECORDCHK(&so->so_rcv);
2759 	SBLASTMBUFCHK(&so->so_rcv);
2760 	SOCKBUF_UNLOCK(&so->so_rcv);
2761 
2762 	if (pr->pr_flags & PR_ADDR) {
2763 		KASSERT(m->m_type == MT_SONAME,
2764 		    ("m->m_type == %d", m->m_type));
2765 		if (psa != NULL)
2766 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2767 			    M_NOWAIT);
2768 		m = m_free(m);
2769 	}
2770 	if (m == NULL) {
2771 		/* XXXRW: Can this happen? */
2772 		return (0);
2773 	}
2774 
2775 	/*
2776 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2777 	 * queue.
2778 	 *
2779 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2780 	 * in the first mbuf chain on the socket buffer.  We call into the
2781 	 * protocol to perform externalization (or freeing if controlp ==
2782 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2783 	 * MT_DATA mbufs.
2784 	 */
2785 	if (m->m_type == MT_CONTROL) {
2786 		struct mbuf *cm = NULL, *cmn;
2787 		struct mbuf **cme = &cm;
2788 
2789 		do {
2790 			m2 = m->m_next;
2791 			m->m_next = NULL;
2792 			*cme = m;
2793 			cme = &(*cme)->m_next;
2794 			m = m2;
2795 		} while (m != NULL && m->m_type == MT_CONTROL);
2796 		while (cm != NULL) {
2797 			cmn = cm->m_next;
2798 			cm->m_next = NULL;
2799 			if (pr->pr_domain->dom_externalize != NULL) {
2800 				error = (*pr->pr_domain->dom_externalize)
2801 				    (cm, controlp, flags);
2802 			} else if (controlp != NULL)
2803 				*controlp = cm;
2804 			else
2805 				m_freem(cm);
2806 			if (controlp != NULL) {
2807 				while (*controlp != NULL)
2808 					controlp = &(*controlp)->m_next;
2809 			}
2810 			cm = cmn;
2811 		}
2812 	}
2813 	KASSERT(m == NULL || m->m_type == MT_DATA,
2814 	    ("soreceive_dgram: !data"));
2815 	while (m != NULL && uio->uio_resid > 0) {
2816 		len = uio->uio_resid;
2817 		if (len > m->m_len)
2818 			len = m->m_len;
2819 		error = uiomove(mtod(m, char *), (int)len, uio);
2820 		if (error) {
2821 			m_freem(m);
2822 			return (error);
2823 		}
2824 		if (len == m->m_len)
2825 			m = m_free(m);
2826 		else {
2827 			m->m_data += len;
2828 			m->m_len -= len;
2829 		}
2830 	}
2831 	if (m != NULL) {
2832 		flags |= MSG_TRUNC;
2833 		m_freem(m);
2834 	}
2835 	if (flagsp != NULL)
2836 		*flagsp |= flags;
2837 	return (0);
2838 }
2839 
2840 int
2841 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2842     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2843 {
2844 	int error;
2845 
2846 	CURVNET_SET(so->so_vnet);
2847 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2848 	    mp0, controlp, flagsp));
2849 	CURVNET_RESTORE();
2850 	return (error);
2851 }
2852 
2853 int
2854 soshutdown(struct socket *so, int how)
2855 {
2856 	struct protosw *pr;
2857 	int error, soerror_enotconn;
2858 
2859 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2860 		return (EINVAL);
2861 
2862 	soerror_enotconn = 0;
2863 	SOCK_LOCK(so);
2864 	if ((so->so_state &
2865 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2866 		/*
2867 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2868 		 * invoked on a datagram sockets, however historically we would
2869 		 * actually tear socket down. This is known to be leveraged by
2870 		 * some applications to unblock process waiting in recvXXX(2)
2871 		 * by other process that it shares that socket with. Try to meet
2872 		 * both backward-compatibility and POSIX requirements by forcing
2873 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2874 		 */
2875 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2876 			SOCK_UNLOCK(so);
2877 			return (ENOTCONN);
2878 		}
2879 		soerror_enotconn = 1;
2880 	}
2881 
2882 	if (SOLISTENING(so)) {
2883 		if (how != SHUT_WR) {
2884 			so->so_error = ECONNABORTED;
2885 			solisten_wakeup(so);	/* unlocks so */
2886 		} else {
2887 			SOCK_UNLOCK(so);
2888 		}
2889 		goto done;
2890 	}
2891 	SOCK_UNLOCK(so);
2892 
2893 	CURVNET_SET(so->so_vnet);
2894 	pr = so->so_proto;
2895 	if (pr->pr_usrreqs->pru_flush != NULL)
2896 		(*pr->pr_usrreqs->pru_flush)(so, how);
2897 	if (how != SHUT_WR)
2898 		sorflush(so);
2899 	if (how != SHUT_RD) {
2900 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2901 		wakeup(&so->so_timeo);
2902 		CURVNET_RESTORE();
2903 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2904 	}
2905 	wakeup(&so->so_timeo);
2906 	CURVNET_RESTORE();
2907 
2908 done:
2909 	return (soerror_enotconn ? ENOTCONN : 0);
2910 }
2911 
2912 void
2913 sorflush(struct socket *so)
2914 {
2915 	struct socket aso;
2916 	struct protosw *pr;
2917 	int error;
2918 
2919 	VNET_SO_ASSERT(so);
2920 
2921 	/*
2922 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2923 	 * held, we make a partial copy of the socket buffer and clear the
2924 	 * original.  The new socket buffer copy won't have initialized locks so
2925 	 * we can only call routines that won't use or assert those locks.
2926 	 * Ideally calling socantrcvmore() would prevent data from being added
2927 	 * to the buffer, but currently it merely prevents buffered data from
2928 	 * being read by userspace.  We make this effort to free buffered data
2929 	 * nonetheless.
2930 	 *
2931 	 * Dislodge threads currently blocked in receive and wait to acquire
2932 	 * a lock against other simultaneous readers before clearing the
2933 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2934 	 * despite any existing socket disposition on interruptable waiting.
2935 	 */
2936 	socantrcvmore(so);
2937 
2938 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2939 	if (error != 0) {
2940 		KASSERT(SOLISTENING(so),
2941 		    ("%s: soiolock(%p) failed", __func__, so));
2942 		return;
2943 	}
2944 
2945 	SOCK_RECVBUF_LOCK(so);
2946 	bzero(&aso, sizeof(aso));
2947 	aso.so_pcb = so->so_pcb;
2948 	bcopy(&so->so_rcv.sb_startzero, &aso.so_rcv.sb_startzero,
2949 	    offsetof(struct sockbuf, sb_endzero) -
2950 	    offsetof(struct sockbuf, sb_startzero));
2951 	bzero(&so->so_rcv.sb_startzero,
2952 	    offsetof(struct sockbuf, sb_endzero) -
2953 	    offsetof(struct sockbuf, sb_startzero));
2954 	SOCK_RECVBUF_UNLOCK(so);
2955 	SOCK_IO_RECV_UNLOCK(so);
2956 
2957 	/*
2958 	 * Dispose of special rights and flush the copied socket.  Don't call
2959 	 * any unsafe routines (that rely on locks being initialized) on aso.
2960 	 */
2961 	pr = so->so_proto;
2962 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2963 		(*pr->pr_domain->dom_dispose)(&aso);
2964 	sbrelease_internal(&aso.so_rcv, so);
2965 }
2966 
2967 /*
2968  * Wrapper for Socket established helper hook.
2969  * Parameters: socket, context of the hook point, hook id.
2970  */
2971 static int inline
2972 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2973 {
2974 	struct socket_hhook_data hhook_data = {
2975 		.so = so,
2976 		.hctx = hctx,
2977 		.m = NULL,
2978 		.status = 0
2979 	};
2980 
2981 	CURVNET_SET(so->so_vnet);
2982 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2983 	CURVNET_RESTORE();
2984 
2985 	/* Ugly but needed, since hhooks return void for now */
2986 	return (hhook_data.status);
2987 }
2988 
2989 /*
2990  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2991  * additional variant to handle the case where the option value needs to be
2992  * some kind of integer, but not a specific size.  In addition to their use
2993  * here, these functions are also called by the protocol-level pr_ctloutput()
2994  * routines.
2995  */
2996 int
2997 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2998 {
2999 	size_t	valsize;
3000 
3001 	/*
3002 	 * If the user gives us more than we wanted, we ignore it, but if we
3003 	 * don't get the minimum length the caller wants, we return EINVAL.
3004 	 * On success, sopt->sopt_valsize is set to however much we actually
3005 	 * retrieved.
3006 	 */
3007 	if ((valsize = sopt->sopt_valsize) < minlen)
3008 		return EINVAL;
3009 	if (valsize > len)
3010 		sopt->sopt_valsize = valsize = len;
3011 
3012 	if (sopt->sopt_td != NULL)
3013 		return (copyin(sopt->sopt_val, buf, valsize));
3014 
3015 	bcopy(sopt->sopt_val, buf, valsize);
3016 	return (0);
3017 }
3018 
3019 /*
3020  * Kernel version of setsockopt(2).
3021  *
3022  * XXX: optlen is size_t, not socklen_t
3023  */
3024 int
3025 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3026     size_t optlen)
3027 {
3028 	struct sockopt sopt;
3029 
3030 	sopt.sopt_level = level;
3031 	sopt.sopt_name = optname;
3032 	sopt.sopt_dir = SOPT_SET;
3033 	sopt.sopt_val = optval;
3034 	sopt.sopt_valsize = optlen;
3035 	sopt.sopt_td = NULL;
3036 	return (sosetopt(so, &sopt));
3037 }
3038 
3039 int
3040 sosetopt(struct socket *so, struct sockopt *sopt)
3041 {
3042 	int	error, optval;
3043 	struct	linger l;
3044 	struct	timeval tv;
3045 	sbintime_t val;
3046 	uint32_t val32;
3047 #ifdef MAC
3048 	struct mac extmac;
3049 #endif
3050 
3051 	CURVNET_SET(so->so_vnet);
3052 	error = 0;
3053 	if (sopt->sopt_level != SOL_SOCKET) {
3054 		if (so->so_proto->pr_ctloutput != NULL)
3055 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3056 		else
3057 			error = ENOPROTOOPT;
3058 	} else {
3059 		switch (sopt->sopt_name) {
3060 		case SO_ACCEPTFILTER:
3061 			error = accept_filt_setopt(so, sopt);
3062 			if (error)
3063 				goto bad;
3064 			break;
3065 
3066 		case SO_LINGER:
3067 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3068 			if (error)
3069 				goto bad;
3070 			if (l.l_linger < 0 ||
3071 			    l.l_linger > USHRT_MAX ||
3072 			    l.l_linger > (INT_MAX / hz)) {
3073 				error = EDOM;
3074 				goto bad;
3075 			}
3076 			SOCK_LOCK(so);
3077 			so->so_linger = l.l_linger;
3078 			if (l.l_onoff)
3079 				so->so_options |= SO_LINGER;
3080 			else
3081 				so->so_options &= ~SO_LINGER;
3082 			SOCK_UNLOCK(so);
3083 			break;
3084 
3085 		case SO_DEBUG:
3086 		case SO_KEEPALIVE:
3087 		case SO_DONTROUTE:
3088 		case SO_USELOOPBACK:
3089 		case SO_BROADCAST:
3090 		case SO_REUSEADDR:
3091 		case SO_REUSEPORT:
3092 		case SO_REUSEPORT_LB:
3093 		case SO_OOBINLINE:
3094 		case SO_TIMESTAMP:
3095 		case SO_BINTIME:
3096 		case SO_NOSIGPIPE:
3097 		case SO_NO_DDP:
3098 		case SO_NO_OFFLOAD:
3099 		case SO_RERROR:
3100 			error = sooptcopyin(sopt, &optval, sizeof optval,
3101 			    sizeof optval);
3102 			if (error)
3103 				goto bad;
3104 			SOCK_LOCK(so);
3105 			if (optval)
3106 				so->so_options |= sopt->sopt_name;
3107 			else
3108 				so->so_options &= ~sopt->sopt_name;
3109 			SOCK_UNLOCK(so);
3110 			break;
3111 
3112 		case SO_SETFIB:
3113 			error = sooptcopyin(sopt, &optval, sizeof optval,
3114 			    sizeof optval);
3115 			if (error)
3116 				goto bad;
3117 
3118 			if (optval < 0 || optval >= rt_numfibs) {
3119 				error = EINVAL;
3120 				goto bad;
3121 			}
3122 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3123 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3124 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3125 				so->so_fibnum = optval;
3126 			else
3127 				so->so_fibnum = 0;
3128 			break;
3129 
3130 		case SO_USER_COOKIE:
3131 			error = sooptcopyin(sopt, &val32, sizeof val32,
3132 			    sizeof val32);
3133 			if (error)
3134 				goto bad;
3135 			so->so_user_cookie = val32;
3136 			break;
3137 
3138 		case SO_SNDBUF:
3139 		case SO_RCVBUF:
3140 		case SO_SNDLOWAT:
3141 		case SO_RCVLOWAT:
3142 			error = sooptcopyin(sopt, &optval, sizeof optval,
3143 			    sizeof optval);
3144 			if (error)
3145 				goto bad;
3146 
3147 			/*
3148 			 * Values < 1 make no sense for any of these options,
3149 			 * so disallow them.
3150 			 */
3151 			if (optval < 1) {
3152 				error = EINVAL;
3153 				goto bad;
3154 			}
3155 
3156 			error = sbsetopt(so, sopt->sopt_name, optval);
3157 			break;
3158 
3159 		case SO_SNDTIMEO:
3160 		case SO_RCVTIMEO:
3161 #ifdef COMPAT_FREEBSD32
3162 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3163 				struct timeval32 tv32;
3164 
3165 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3166 				    sizeof tv32);
3167 				CP(tv32, tv, tv_sec);
3168 				CP(tv32, tv, tv_usec);
3169 			} else
3170 #endif
3171 				error = sooptcopyin(sopt, &tv, sizeof tv,
3172 				    sizeof tv);
3173 			if (error)
3174 				goto bad;
3175 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3176 			    tv.tv_usec >= 1000000) {
3177 				error = EDOM;
3178 				goto bad;
3179 			}
3180 			if (tv.tv_sec > INT32_MAX)
3181 				val = SBT_MAX;
3182 			else
3183 				val = tvtosbt(tv);
3184 			switch (sopt->sopt_name) {
3185 			case SO_SNDTIMEO:
3186 				so->so_snd.sb_timeo = val;
3187 				break;
3188 			case SO_RCVTIMEO:
3189 				so->so_rcv.sb_timeo = val;
3190 				break;
3191 			}
3192 			break;
3193 
3194 		case SO_LABEL:
3195 #ifdef MAC
3196 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3197 			    sizeof extmac);
3198 			if (error)
3199 				goto bad;
3200 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3201 			    so, &extmac);
3202 #else
3203 			error = EOPNOTSUPP;
3204 #endif
3205 			break;
3206 
3207 		case SO_TS_CLOCK:
3208 			error = sooptcopyin(sopt, &optval, sizeof optval,
3209 			    sizeof optval);
3210 			if (error)
3211 				goto bad;
3212 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3213 				error = EINVAL;
3214 				goto bad;
3215 			}
3216 			so->so_ts_clock = optval;
3217 			break;
3218 
3219 		case SO_MAX_PACING_RATE:
3220 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3221 			    sizeof(val32));
3222 			if (error)
3223 				goto bad;
3224 			so->so_max_pacing_rate = val32;
3225 			break;
3226 
3227 		default:
3228 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3229 				error = hhook_run_socket(so, sopt,
3230 				    HHOOK_SOCKET_OPT);
3231 			else
3232 				error = ENOPROTOOPT;
3233 			break;
3234 		}
3235 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3236 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3237 	}
3238 bad:
3239 	CURVNET_RESTORE();
3240 	return (error);
3241 }
3242 
3243 /*
3244  * Helper routine for getsockopt.
3245  */
3246 int
3247 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3248 {
3249 	int	error;
3250 	size_t	valsize;
3251 
3252 	error = 0;
3253 
3254 	/*
3255 	 * Documented get behavior is that we always return a value, possibly
3256 	 * truncated to fit in the user's buffer.  Traditional behavior is
3257 	 * that we always tell the user precisely how much we copied, rather
3258 	 * than something useful like the total amount we had available for
3259 	 * her.  Note that this interface is not idempotent; the entire
3260 	 * answer must be generated ahead of time.
3261 	 */
3262 	valsize = min(len, sopt->sopt_valsize);
3263 	sopt->sopt_valsize = valsize;
3264 	if (sopt->sopt_val != NULL) {
3265 		if (sopt->sopt_td != NULL)
3266 			error = copyout(buf, sopt->sopt_val, valsize);
3267 		else
3268 			bcopy(buf, sopt->sopt_val, valsize);
3269 	}
3270 	return (error);
3271 }
3272 
3273 int
3274 sogetopt(struct socket *so, struct sockopt *sopt)
3275 {
3276 	int	error, optval;
3277 	struct	linger l;
3278 	struct	timeval tv;
3279 #ifdef MAC
3280 	struct mac extmac;
3281 #endif
3282 
3283 	CURVNET_SET(so->so_vnet);
3284 	error = 0;
3285 	if (sopt->sopt_level != SOL_SOCKET) {
3286 		if (so->so_proto->pr_ctloutput != NULL)
3287 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3288 		else
3289 			error = ENOPROTOOPT;
3290 		CURVNET_RESTORE();
3291 		return (error);
3292 	} else {
3293 		switch (sopt->sopt_name) {
3294 		case SO_ACCEPTFILTER:
3295 			error = accept_filt_getopt(so, sopt);
3296 			break;
3297 
3298 		case SO_LINGER:
3299 			SOCK_LOCK(so);
3300 			l.l_onoff = so->so_options & SO_LINGER;
3301 			l.l_linger = so->so_linger;
3302 			SOCK_UNLOCK(so);
3303 			error = sooptcopyout(sopt, &l, sizeof l);
3304 			break;
3305 
3306 		case SO_USELOOPBACK:
3307 		case SO_DONTROUTE:
3308 		case SO_DEBUG:
3309 		case SO_KEEPALIVE:
3310 		case SO_REUSEADDR:
3311 		case SO_REUSEPORT:
3312 		case SO_REUSEPORT_LB:
3313 		case SO_BROADCAST:
3314 		case SO_OOBINLINE:
3315 		case SO_ACCEPTCONN:
3316 		case SO_TIMESTAMP:
3317 		case SO_BINTIME:
3318 		case SO_NOSIGPIPE:
3319 		case SO_NO_DDP:
3320 		case SO_NO_OFFLOAD:
3321 		case SO_RERROR:
3322 			optval = so->so_options & sopt->sopt_name;
3323 integer:
3324 			error = sooptcopyout(sopt, &optval, sizeof optval);
3325 			break;
3326 
3327 		case SO_DOMAIN:
3328 			optval = so->so_proto->pr_domain->dom_family;
3329 			goto integer;
3330 
3331 		case SO_TYPE:
3332 			optval = so->so_type;
3333 			goto integer;
3334 
3335 		case SO_PROTOCOL:
3336 			optval = so->so_proto->pr_protocol;
3337 			goto integer;
3338 
3339 		case SO_ERROR:
3340 			SOCK_LOCK(so);
3341 			if (so->so_error) {
3342 				optval = so->so_error;
3343 				so->so_error = 0;
3344 			} else {
3345 				optval = so->so_rerror;
3346 				so->so_rerror = 0;
3347 			}
3348 			SOCK_UNLOCK(so);
3349 			goto integer;
3350 
3351 		case SO_SNDBUF:
3352 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3353 			    so->so_snd.sb_hiwat;
3354 			goto integer;
3355 
3356 		case SO_RCVBUF:
3357 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3358 			    so->so_rcv.sb_hiwat;
3359 			goto integer;
3360 
3361 		case SO_SNDLOWAT:
3362 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3363 			    so->so_snd.sb_lowat;
3364 			goto integer;
3365 
3366 		case SO_RCVLOWAT:
3367 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3368 			    so->so_rcv.sb_lowat;
3369 			goto integer;
3370 
3371 		case SO_SNDTIMEO:
3372 		case SO_RCVTIMEO:
3373 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3374 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3375 #ifdef COMPAT_FREEBSD32
3376 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3377 				struct timeval32 tv32;
3378 
3379 				CP(tv, tv32, tv_sec);
3380 				CP(tv, tv32, tv_usec);
3381 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3382 			} else
3383 #endif
3384 				error = sooptcopyout(sopt, &tv, sizeof tv);
3385 			break;
3386 
3387 		case SO_LABEL:
3388 #ifdef MAC
3389 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3390 			    sizeof(extmac));
3391 			if (error)
3392 				goto bad;
3393 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3394 			    so, &extmac);
3395 			if (error)
3396 				goto bad;
3397 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3398 #else
3399 			error = EOPNOTSUPP;
3400 #endif
3401 			break;
3402 
3403 		case SO_PEERLABEL:
3404 #ifdef MAC
3405 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3406 			    sizeof(extmac));
3407 			if (error)
3408 				goto bad;
3409 			error = mac_getsockopt_peerlabel(
3410 			    sopt->sopt_td->td_ucred, so, &extmac);
3411 			if (error)
3412 				goto bad;
3413 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3414 #else
3415 			error = EOPNOTSUPP;
3416 #endif
3417 			break;
3418 
3419 		case SO_LISTENQLIMIT:
3420 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3421 			goto integer;
3422 
3423 		case SO_LISTENQLEN:
3424 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3425 			goto integer;
3426 
3427 		case SO_LISTENINCQLEN:
3428 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3429 			goto integer;
3430 
3431 		case SO_TS_CLOCK:
3432 			optval = so->so_ts_clock;
3433 			goto integer;
3434 
3435 		case SO_MAX_PACING_RATE:
3436 			optval = so->so_max_pacing_rate;
3437 			goto integer;
3438 
3439 		default:
3440 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3441 				error = hhook_run_socket(so, sopt,
3442 				    HHOOK_SOCKET_OPT);
3443 			else
3444 				error = ENOPROTOOPT;
3445 			break;
3446 		}
3447 	}
3448 #ifdef MAC
3449 bad:
3450 #endif
3451 	CURVNET_RESTORE();
3452 	return (error);
3453 }
3454 
3455 int
3456 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3457 {
3458 	struct mbuf *m, *m_prev;
3459 	int sopt_size = sopt->sopt_valsize;
3460 
3461 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3462 	if (m == NULL)
3463 		return ENOBUFS;
3464 	if (sopt_size > MLEN) {
3465 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3466 		if ((m->m_flags & M_EXT) == 0) {
3467 			m_free(m);
3468 			return ENOBUFS;
3469 		}
3470 		m->m_len = min(MCLBYTES, sopt_size);
3471 	} else {
3472 		m->m_len = min(MLEN, sopt_size);
3473 	}
3474 	sopt_size -= m->m_len;
3475 	*mp = m;
3476 	m_prev = m;
3477 
3478 	while (sopt_size) {
3479 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3480 		if (m == NULL) {
3481 			m_freem(*mp);
3482 			return ENOBUFS;
3483 		}
3484 		if (sopt_size > MLEN) {
3485 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3486 			    M_NOWAIT);
3487 			if ((m->m_flags & M_EXT) == 0) {
3488 				m_freem(m);
3489 				m_freem(*mp);
3490 				return ENOBUFS;
3491 			}
3492 			m->m_len = min(MCLBYTES, sopt_size);
3493 		} else {
3494 			m->m_len = min(MLEN, sopt_size);
3495 		}
3496 		sopt_size -= m->m_len;
3497 		m_prev->m_next = m;
3498 		m_prev = m;
3499 	}
3500 	return (0);
3501 }
3502 
3503 int
3504 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3505 {
3506 	struct mbuf *m0 = m;
3507 
3508 	if (sopt->sopt_val == NULL)
3509 		return (0);
3510 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3511 		if (sopt->sopt_td != NULL) {
3512 			int error;
3513 
3514 			error = copyin(sopt->sopt_val, mtod(m, char *),
3515 			    m->m_len);
3516 			if (error != 0) {
3517 				m_freem(m0);
3518 				return(error);
3519 			}
3520 		} else
3521 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3522 		sopt->sopt_valsize -= m->m_len;
3523 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3524 		m = m->m_next;
3525 	}
3526 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3527 		panic("ip6_sooptmcopyin");
3528 	return (0);
3529 }
3530 
3531 int
3532 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3533 {
3534 	struct mbuf *m0 = m;
3535 	size_t valsize = 0;
3536 
3537 	if (sopt->sopt_val == NULL)
3538 		return (0);
3539 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3540 		if (sopt->sopt_td != NULL) {
3541 			int error;
3542 
3543 			error = copyout(mtod(m, char *), sopt->sopt_val,
3544 			    m->m_len);
3545 			if (error != 0) {
3546 				m_freem(m0);
3547 				return(error);
3548 			}
3549 		} else
3550 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3551 		sopt->sopt_valsize -= m->m_len;
3552 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3553 		valsize += m->m_len;
3554 		m = m->m_next;
3555 	}
3556 	if (m != NULL) {
3557 		/* enough soopt buffer should be given from user-land */
3558 		m_freem(m0);
3559 		return(EINVAL);
3560 	}
3561 	sopt->sopt_valsize = valsize;
3562 	return (0);
3563 }
3564 
3565 /*
3566  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3567  * out-of-band data, which will then notify socket consumers.
3568  */
3569 void
3570 sohasoutofband(struct socket *so)
3571 {
3572 
3573 	if (so->so_sigio != NULL)
3574 		pgsigio(&so->so_sigio, SIGURG, 0);
3575 	selwakeuppri(&so->so_rdsel, PSOCK);
3576 }
3577 
3578 int
3579 sopoll(struct socket *so, int events, struct ucred *active_cred,
3580     struct thread *td)
3581 {
3582 
3583 	/*
3584 	 * We do not need to set or assert curvnet as long as everyone uses
3585 	 * sopoll_generic().
3586 	 */
3587 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3588 	    td));
3589 }
3590 
3591 int
3592 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3593     struct thread *td)
3594 {
3595 	int revents;
3596 
3597 	SOCK_LOCK(so);
3598 	if (SOLISTENING(so)) {
3599 		if (!(events & (POLLIN | POLLRDNORM)))
3600 			revents = 0;
3601 		else if (!TAILQ_EMPTY(&so->sol_comp))
3602 			revents = events & (POLLIN | POLLRDNORM);
3603 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3604 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3605 		else {
3606 			selrecord(td, &so->so_rdsel);
3607 			revents = 0;
3608 		}
3609 	} else {
3610 		revents = 0;
3611 		SOCKBUF_LOCK(&so->so_snd);
3612 		SOCKBUF_LOCK(&so->so_rcv);
3613 		if (events & (POLLIN | POLLRDNORM))
3614 			if (soreadabledata(so))
3615 				revents |= events & (POLLIN | POLLRDNORM);
3616 		if (events & (POLLOUT | POLLWRNORM))
3617 			if (sowriteable(so))
3618 				revents |= events & (POLLOUT | POLLWRNORM);
3619 		if (events & (POLLPRI | POLLRDBAND))
3620 			if (so->so_oobmark ||
3621 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3622 				revents |= events & (POLLPRI | POLLRDBAND);
3623 		if ((events & POLLINIGNEOF) == 0) {
3624 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3625 				revents |= events & (POLLIN | POLLRDNORM);
3626 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3627 					revents |= POLLHUP;
3628 			}
3629 		}
3630 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3631 			revents |= events & POLLRDHUP;
3632 		if (revents == 0) {
3633 			if (events &
3634 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3635 				selrecord(td, &so->so_rdsel);
3636 				so->so_rcv.sb_flags |= SB_SEL;
3637 			}
3638 			if (events & (POLLOUT | POLLWRNORM)) {
3639 				selrecord(td, &so->so_wrsel);
3640 				so->so_snd.sb_flags |= SB_SEL;
3641 			}
3642 		}
3643 		SOCKBUF_UNLOCK(&so->so_rcv);
3644 		SOCKBUF_UNLOCK(&so->so_snd);
3645 	}
3646 	SOCK_UNLOCK(so);
3647 	return (revents);
3648 }
3649 
3650 int
3651 soo_kqfilter(struct file *fp, struct knote *kn)
3652 {
3653 	struct socket *so = kn->kn_fp->f_data;
3654 	struct sockbuf *sb;
3655 	struct knlist *knl;
3656 
3657 	switch (kn->kn_filter) {
3658 	case EVFILT_READ:
3659 		kn->kn_fop = &soread_filtops;
3660 		knl = &so->so_rdsel.si_note;
3661 		sb = &so->so_rcv;
3662 		break;
3663 	case EVFILT_WRITE:
3664 		kn->kn_fop = &sowrite_filtops;
3665 		knl = &so->so_wrsel.si_note;
3666 		sb = &so->so_snd;
3667 		break;
3668 	case EVFILT_EMPTY:
3669 		kn->kn_fop = &soempty_filtops;
3670 		knl = &so->so_wrsel.si_note;
3671 		sb = &so->so_snd;
3672 		break;
3673 	default:
3674 		return (EINVAL);
3675 	}
3676 
3677 	SOCK_LOCK(so);
3678 	if (SOLISTENING(so)) {
3679 		knlist_add(knl, kn, 1);
3680 	} else {
3681 		SOCKBUF_LOCK(sb);
3682 		knlist_add(knl, kn, 1);
3683 		sb->sb_flags |= SB_KNOTE;
3684 		SOCKBUF_UNLOCK(sb);
3685 	}
3686 	SOCK_UNLOCK(so);
3687 	return (0);
3688 }
3689 
3690 /*
3691  * Some routines that return EOPNOTSUPP for entry points that are not
3692  * supported by a protocol.  Fill in as needed.
3693  */
3694 int
3695 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3696 {
3697 
3698 	return EOPNOTSUPP;
3699 }
3700 
3701 int
3702 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3703 {
3704 
3705 	return EOPNOTSUPP;
3706 }
3707 
3708 int
3709 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3710 {
3711 
3712 	return EOPNOTSUPP;
3713 }
3714 
3715 int
3716 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3717 {
3718 
3719 	return EOPNOTSUPP;
3720 }
3721 
3722 int
3723 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3724     struct thread *td)
3725 {
3726 
3727 	return EOPNOTSUPP;
3728 }
3729 
3730 int
3731 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3732 {
3733 
3734 	return EOPNOTSUPP;
3735 }
3736 
3737 int
3738 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3739     struct thread *td)
3740 {
3741 
3742 	return EOPNOTSUPP;
3743 }
3744 
3745 int
3746 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3747 {
3748 
3749 	return EOPNOTSUPP;
3750 }
3751 
3752 int
3753 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3754     struct ifnet *ifp, struct thread *td)
3755 {
3756 
3757 	return EOPNOTSUPP;
3758 }
3759 
3760 int
3761 pru_disconnect_notsupp(struct socket *so)
3762 {
3763 
3764 	return EOPNOTSUPP;
3765 }
3766 
3767 int
3768 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3769 {
3770 
3771 	return EOPNOTSUPP;
3772 }
3773 
3774 int
3775 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3776 {
3777 
3778 	return EOPNOTSUPP;
3779 }
3780 
3781 int
3782 pru_rcvd_notsupp(struct socket *so, int flags)
3783 {
3784 
3785 	return EOPNOTSUPP;
3786 }
3787 
3788 int
3789 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3790 {
3791 
3792 	return EOPNOTSUPP;
3793 }
3794 
3795 int
3796 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3797     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3798 {
3799 
3800 	if (control != NULL)
3801 		m_freem(control);
3802 	if ((flags & PRUS_NOTREADY) == 0)
3803 		m_freem(m);
3804 	return (EOPNOTSUPP);
3805 }
3806 
3807 int
3808 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3809 {
3810 
3811 	return (EOPNOTSUPP);
3812 }
3813 
3814 /*
3815  * This isn't really a ``null'' operation, but it's the default one and
3816  * doesn't do anything destructive.
3817  */
3818 int
3819 pru_sense_null(struct socket *so, struct stat *sb)
3820 {
3821 
3822 	sb->st_blksize = so->so_snd.sb_hiwat;
3823 	return 0;
3824 }
3825 
3826 int
3827 pru_shutdown_notsupp(struct socket *so)
3828 {
3829 
3830 	return EOPNOTSUPP;
3831 }
3832 
3833 int
3834 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3835 {
3836 
3837 	return EOPNOTSUPP;
3838 }
3839 
3840 int
3841 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3842     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3843 {
3844 
3845 	return EOPNOTSUPP;
3846 }
3847 
3848 int
3849 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3850     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3851 {
3852 
3853 	return EOPNOTSUPP;
3854 }
3855 
3856 int
3857 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3858     struct thread *td)
3859 {
3860 
3861 	return EOPNOTSUPP;
3862 }
3863 
3864 static void
3865 filt_sordetach(struct knote *kn)
3866 {
3867 	struct socket *so = kn->kn_fp->f_data;
3868 
3869 	so_rdknl_lock(so);
3870 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3871 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3872 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3873 	so_rdknl_unlock(so);
3874 }
3875 
3876 /*ARGSUSED*/
3877 static int
3878 filt_soread(struct knote *kn, long hint)
3879 {
3880 	struct socket *so;
3881 
3882 	so = kn->kn_fp->f_data;
3883 
3884 	if (SOLISTENING(so)) {
3885 		SOCK_LOCK_ASSERT(so);
3886 		kn->kn_data = so->sol_qlen;
3887 		if (so->so_error) {
3888 			kn->kn_flags |= EV_EOF;
3889 			kn->kn_fflags = so->so_error;
3890 			return (1);
3891 		}
3892 		return (!TAILQ_EMPTY(&so->sol_comp));
3893 	}
3894 
3895 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3896 
3897 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3898 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3899 		kn->kn_flags |= EV_EOF;
3900 		kn->kn_fflags = so->so_error;
3901 		return (1);
3902 	} else if (so->so_error || so->so_rerror)
3903 		return (1);
3904 
3905 	if (kn->kn_sfflags & NOTE_LOWAT) {
3906 		if (kn->kn_data >= kn->kn_sdata)
3907 			return (1);
3908 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3909 		return (1);
3910 
3911 	/* This hook returning non-zero indicates an event, not error */
3912 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3913 }
3914 
3915 static void
3916 filt_sowdetach(struct knote *kn)
3917 {
3918 	struct socket *so = kn->kn_fp->f_data;
3919 
3920 	so_wrknl_lock(so);
3921 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3922 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3923 		so->so_snd.sb_flags &= ~SB_KNOTE;
3924 	so_wrknl_unlock(so);
3925 }
3926 
3927 /*ARGSUSED*/
3928 static int
3929 filt_sowrite(struct knote *kn, long hint)
3930 {
3931 	struct socket *so;
3932 
3933 	so = kn->kn_fp->f_data;
3934 
3935 	if (SOLISTENING(so))
3936 		return (0);
3937 
3938 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3939 	kn->kn_data = sbspace(&so->so_snd);
3940 
3941 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3942 
3943 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3944 		kn->kn_flags |= EV_EOF;
3945 		kn->kn_fflags = so->so_error;
3946 		return (1);
3947 	} else if (so->so_error)	/* temporary udp error */
3948 		return (1);
3949 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3950 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3951 		return (0);
3952 	else if (kn->kn_sfflags & NOTE_LOWAT)
3953 		return (kn->kn_data >= kn->kn_sdata);
3954 	else
3955 		return (kn->kn_data >= so->so_snd.sb_lowat);
3956 }
3957 
3958 static int
3959 filt_soempty(struct knote *kn, long hint)
3960 {
3961 	struct socket *so;
3962 
3963 	so = kn->kn_fp->f_data;
3964 
3965 	if (SOLISTENING(so))
3966 		return (1);
3967 
3968 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3969 	kn->kn_data = sbused(&so->so_snd);
3970 
3971 	if (kn->kn_data == 0)
3972 		return (1);
3973 	else
3974 		return (0);
3975 }
3976 
3977 int
3978 socheckuid(struct socket *so, uid_t uid)
3979 {
3980 
3981 	if (so == NULL)
3982 		return (EPERM);
3983 	if (so->so_cred->cr_uid != uid)
3984 		return (EPERM);
3985 	return (0);
3986 }
3987 
3988 /*
3989  * These functions are used by protocols to notify the socket layer (and its
3990  * consumers) of state changes in the sockets driven by protocol-side events.
3991  */
3992 
3993 /*
3994  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3995  *
3996  * Normal sequence from the active (originating) side is that
3997  * soisconnecting() is called during processing of connect() call, resulting
3998  * in an eventual call to soisconnected() if/when the connection is
3999  * established.  When the connection is torn down soisdisconnecting() is
4000  * called during processing of disconnect() call, and soisdisconnected() is
4001  * called when the connection to the peer is totally severed.  The semantics
4002  * of these routines are such that connectionless protocols can call
4003  * soisconnected() and soisdisconnected() only, bypassing the in-progress
4004  * calls when setting up a ``connection'' takes no time.
4005  *
4006  * From the passive side, a socket is created with two queues of sockets:
4007  * so_incomp for connections in progress and so_comp for connections already
4008  * made and awaiting user acceptance.  As a protocol is preparing incoming
4009  * connections, it creates a socket structure queued on so_incomp by calling
4010  * sonewconn().  When the connection is established, soisconnected() is
4011  * called, and transfers the socket structure to so_comp, making it available
4012  * to accept().
4013  *
4014  * If a socket is closed with sockets on either so_incomp or so_comp, these
4015  * sockets are dropped.
4016  *
4017  * If higher-level protocols are implemented in the kernel, the wakeups done
4018  * here will sometimes cause software-interrupt process scheduling.
4019  */
4020 void
4021 soisconnecting(struct socket *so)
4022 {
4023 
4024 	SOCK_LOCK(so);
4025 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4026 	so->so_state |= SS_ISCONNECTING;
4027 	SOCK_UNLOCK(so);
4028 }
4029 
4030 void
4031 soisconnected(struct socket *so)
4032 {
4033 	bool last __diagused;
4034 
4035 	SOCK_LOCK(so);
4036 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
4037 	so->so_state |= SS_ISCONNECTED;
4038 
4039 	if (so->so_qstate == SQ_INCOMP) {
4040 		struct socket *head = so->so_listen;
4041 		int ret;
4042 
4043 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4044 		/*
4045 		 * Promoting a socket from incomplete queue to complete, we
4046 		 * need to go through reverse order of locking.  We first do
4047 		 * trylock, and if that doesn't succeed, we go the hard way
4048 		 * leaving a reference and rechecking consistency after proper
4049 		 * locking.
4050 		 */
4051 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4052 			soref(head);
4053 			SOCK_UNLOCK(so);
4054 			SOLISTEN_LOCK(head);
4055 			SOCK_LOCK(so);
4056 			if (__predict_false(head != so->so_listen)) {
4057 				/*
4058 				 * The socket went off the listen queue,
4059 				 * should be lost race to close(2) of sol.
4060 				 * The socket is about to soabort().
4061 				 */
4062 				SOCK_UNLOCK(so);
4063 				sorele(head);
4064 				return;
4065 			}
4066 			last = refcount_release(&head->so_count);
4067 			KASSERT(!last, ("%s: released last reference for %p",
4068 			    __func__, head));
4069 		}
4070 again:
4071 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4072 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4073 			head->sol_incqlen--;
4074 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4075 			head->sol_qlen++;
4076 			so->so_qstate = SQ_COMP;
4077 			SOCK_UNLOCK(so);
4078 			solisten_wakeup(head);	/* unlocks */
4079 		} else {
4080 			SOCKBUF_LOCK(&so->so_rcv);
4081 			soupcall_set(so, SO_RCV,
4082 			    head->sol_accept_filter->accf_callback,
4083 			    head->sol_accept_filter_arg);
4084 			so->so_options &= ~SO_ACCEPTFILTER;
4085 			ret = head->sol_accept_filter->accf_callback(so,
4086 			    head->sol_accept_filter_arg, M_NOWAIT);
4087 			if (ret == SU_ISCONNECTED) {
4088 				soupcall_clear(so, SO_RCV);
4089 				SOCKBUF_UNLOCK(&so->so_rcv);
4090 				goto again;
4091 			}
4092 			SOCKBUF_UNLOCK(&so->so_rcv);
4093 			SOCK_UNLOCK(so);
4094 			SOLISTEN_UNLOCK(head);
4095 		}
4096 		return;
4097 	}
4098 	SOCK_UNLOCK(so);
4099 	wakeup(&so->so_timeo);
4100 	sorwakeup(so);
4101 	sowwakeup(so);
4102 }
4103 
4104 void
4105 soisdisconnecting(struct socket *so)
4106 {
4107 
4108 	SOCK_LOCK(so);
4109 	so->so_state &= ~SS_ISCONNECTING;
4110 	so->so_state |= SS_ISDISCONNECTING;
4111 
4112 	if (!SOLISTENING(so)) {
4113 		SOCKBUF_LOCK(&so->so_rcv);
4114 		socantrcvmore_locked(so);
4115 		SOCKBUF_LOCK(&so->so_snd);
4116 		socantsendmore_locked(so);
4117 	}
4118 	SOCK_UNLOCK(so);
4119 	wakeup(&so->so_timeo);
4120 }
4121 
4122 void
4123 soisdisconnected(struct socket *so)
4124 {
4125 
4126 	SOCK_LOCK(so);
4127 
4128 	/*
4129 	 * There is at least one reader of so_state that does not
4130 	 * acquire socket lock, namely soreceive_generic().  Ensure
4131 	 * that it never sees all flags that track connection status
4132 	 * cleared, by ordering the update with a barrier semantic of
4133 	 * our release thread fence.
4134 	 */
4135 	so->so_state |= SS_ISDISCONNECTED;
4136 	atomic_thread_fence_rel();
4137 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4138 
4139 	if (!SOLISTENING(so)) {
4140 		SOCK_UNLOCK(so);
4141 		SOCKBUF_LOCK(&so->so_rcv);
4142 		socantrcvmore_locked(so);
4143 		SOCKBUF_LOCK(&so->so_snd);
4144 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4145 		socantsendmore_locked(so);
4146 	} else
4147 		SOCK_UNLOCK(so);
4148 	wakeup(&so->so_timeo);
4149 }
4150 
4151 int
4152 soiolock(struct socket *so, struct sx *sx, int flags)
4153 {
4154 	int error;
4155 
4156 	KASSERT((flags & SBL_VALID) == flags,
4157 	    ("soiolock: invalid flags %#x", flags));
4158 
4159 	if ((flags & SBL_WAIT) != 0) {
4160 		if ((flags & SBL_NOINTR) != 0) {
4161 			sx_xlock(sx);
4162 		} else {
4163 			error = sx_xlock_sig(sx);
4164 			if (error != 0)
4165 				return (error);
4166 		}
4167 	} else if (!sx_try_xlock(sx)) {
4168 		return (EWOULDBLOCK);
4169 	}
4170 
4171 	if (__predict_false(SOLISTENING(so))) {
4172 		sx_xunlock(sx);
4173 		return (ENOTCONN);
4174 	}
4175 	return (0);
4176 }
4177 
4178 void
4179 soiounlock(struct sx *sx)
4180 {
4181 	sx_xunlock(sx);
4182 }
4183 
4184 /*
4185  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4186  */
4187 struct sockaddr *
4188 sodupsockaddr(const struct sockaddr *sa, int mflags)
4189 {
4190 	struct sockaddr *sa2;
4191 
4192 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4193 	if (sa2)
4194 		bcopy(sa, sa2, sa->sa_len);
4195 	return sa2;
4196 }
4197 
4198 /*
4199  * Register per-socket destructor.
4200  */
4201 void
4202 sodtor_set(struct socket *so, so_dtor_t *func)
4203 {
4204 
4205 	SOCK_LOCK_ASSERT(so);
4206 	so->so_dtor = func;
4207 }
4208 
4209 /*
4210  * Register per-socket buffer upcalls.
4211  */
4212 void
4213 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4214 {
4215 	struct sockbuf *sb;
4216 
4217 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4218 
4219 	switch (which) {
4220 	case SO_RCV:
4221 		sb = &so->so_rcv;
4222 		break;
4223 	case SO_SND:
4224 		sb = &so->so_snd;
4225 		break;
4226 	default:
4227 		panic("soupcall_set: bad which");
4228 	}
4229 	SOCKBUF_LOCK_ASSERT(sb);
4230 	sb->sb_upcall = func;
4231 	sb->sb_upcallarg = arg;
4232 	sb->sb_flags |= SB_UPCALL;
4233 }
4234 
4235 void
4236 soupcall_clear(struct socket *so, int which)
4237 {
4238 	struct sockbuf *sb;
4239 
4240 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4241 
4242 	switch (which) {
4243 	case SO_RCV:
4244 		sb = &so->so_rcv;
4245 		break;
4246 	case SO_SND:
4247 		sb = &so->so_snd;
4248 		break;
4249 	default:
4250 		panic("soupcall_clear: bad which");
4251 	}
4252 	SOCKBUF_LOCK_ASSERT(sb);
4253 	KASSERT(sb->sb_upcall != NULL,
4254 	    ("%s: so %p no upcall to clear", __func__, so));
4255 	sb->sb_upcall = NULL;
4256 	sb->sb_upcallarg = NULL;
4257 	sb->sb_flags &= ~SB_UPCALL;
4258 }
4259 
4260 void
4261 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4262 {
4263 
4264 	SOLISTEN_LOCK_ASSERT(so);
4265 	so->sol_upcall = func;
4266 	so->sol_upcallarg = arg;
4267 }
4268 
4269 static void
4270 so_rdknl_lock(void *arg)
4271 {
4272 	struct socket *so = arg;
4273 
4274 	if (SOLISTENING(so))
4275 		SOCK_LOCK(so);
4276 	else
4277 		SOCKBUF_LOCK(&so->so_rcv);
4278 }
4279 
4280 static void
4281 so_rdknl_unlock(void *arg)
4282 {
4283 	struct socket *so = arg;
4284 
4285 	if (SOLISTENING(so))
4286 		SOCK_UNLOCK(so);
4287 	else
4288 		SOCKBUF_UNLOCK(&so->so_rcv);
4289 }
4290 
4291 static void
4292 so_rdknl_assert_lock(void *arg, int what)
4293 {
4294 	struct socket *so = arg;
4295 
4296 	if (what == LA_LOCKED) {
4297 		if (SOLISTENING(so))
4298 			SOCK_LOCK_ASSERT(so);
4299 		else
4300 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4301 	} else {
4302 		if (SOLISTENING(so))
4303 			SOCK_UNLOCK_ASSERT(so);
4304 		else
4305 			SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4306 	}
4307 }
4308 
4309 static void
4310 so_wrknl_lock(void *arg)
4311 {
4312 	struct socket *so = arg;
4313 
4314 	if (SOLISTENING(so))
4315 		SOCK_LOCK(so);
4316 	else
4317 		SOCKBUF_LOCK(&so->so_snd);
4318 }
4319 
4320 static void
4321 so_wrknl_unlock(void *arg)
4322 {
4323 	struct socket *so = arg;
4324 
4325 	if (SOLISTENING(so))
4326 		SOCK_UNLOCK(so);
4327 	else
4328 		SOCKBUF_UNLOCK(&so->so_snd);
4329 }
4330 
4331 static void
4332 so_wrknl_assert_lock(void *arg, int what)
4333 {
4334 	struct socket *so = arg;
4335 
4336 	if (what == LA_LOCKED) {
4337 		if (SOLISTENING(so))
4338 			SOCK_LOCK_ASSERT(so);
4339 		else
4340 			SOCKBUF_LOCK_ASSERT(&so->so_snd);
4341 	} else {
4342 		if (SOLISTENING(so))
4343 			SOCK_UNLOCK_ASSERT(so);
4344 		else
4345 			SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4346 	}
4347 }
4348 
4349 /*
4350  * Create an external-format (``xsocket'') structure using the information in
4351  * the kernel-format socket structure pointed to by so.  This is done to
4352  * reduce the spew of irrelevant information over this interface, to isolate
4353  * user code from changes in the kernel structure, and potentially to provide
4354  * information-hiding if we decide that some of this information should be
4355  * hidden from users.
4356  */
4357 void
4358 sotoxsocket(struct socket *so, struct xsocket *xso)
4359 {
4360 
4361 	bzero(xso, sizeof(*xso));
4362 	xso->xso_len = sizeof *xso;
4363 	xso->xso_so = (uintptr_t)so;
4364 	xso->so_type = so->so_type;
4365 	xso->so_options = so->so_options;
4366 	xso->so_linger = so->so_linger;
4367 	xso->so_state = so->so_state;
4368 	xso->so_pcb = (uintptr_t)so->so_pcb;
4369 	xso->xso_protocol = so->so_proto->pr_protocol;
4370 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4371 	xso->so_timeo = so->so_timeo;
4372 	xso->so_error = so->so_error;
4373 	xso->so_uid = so->so_cred->cr_uid;
4374 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4375 	if (SOLISTENING(so)) {
4376 		xso->so_qlen = so->sol_qlen;
4377 		xso->so_incqlen = so->sol_incqlen;
4378 		xso->so_qlimit = so->sol_qlimit;
4379 		xso->so_oobmark = 0;
4380 	} else {
4381 		xso->so_state |= so->so_qstate;
4382 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4383 		xso->so_oobmark = so->so_oobmark;
4384 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4385 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4386 	}
4387 }
4388 
4389 struct sockbuf *
4390 so_sockbuf_rcv(struct socket *so)
4391 {
4392 
4393 	return (&so->so_rcv);
4394 }
4395 
4396 struct sockbuf *
4397 so_sockbuf_snd(struct socket *so)
4398 {
4399 
4400 	return (&so->so_snd);
4401 }
4402 
4403 int
4404 so_state_get(const struct socket *so)
4405 {
4406 
4407 	return (so->so_state);
4408 }
4409 
4410 void
4411 so_state_set(struct socket *so, int val)
4412 {
4413 
4414 	so->so_state = val;
4415 }
4416 
4417 int
4418 so_options_get(const struct socket *so)
4419 {
4420 
4421 	return (so->so_options);
4422 }
4423 
4424 void
4425 so_options_set(struct socket *so, int val)
4426 {
4427 
4428 	so->so_options = val;
4429 }
4430 
4431 int
4432 so_error_get(const struct socket *so)
4433 {
4434 
4435 	return (so->so_error);
4436 }
4437 
4438 void
4439 so_error_set(struct socket *so, int val)
4440 {
4441 
4442 	so->so_error = val;
4443 }
4444 
4445 int
4446 so_linger_get(const struct socket *so)
4447 {
4448 
4449 	return (so->so_linger);
4450 }
4451 
4452 void
4453 so_linger_set(struct socket *so, int val)
4454 {
4455 
4456 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4457 	    ("%s: val %d out of range", __func__, val));
4458 
4459 	so->so_linger = val;
4460 }
4461 
4462 struct protosw *
4463 so_protosw_get(const struct socket *so)
4464 {
4465 
4466 	return (so->so_proto);
4467 }
4468 
4469 void
4470 so_protosw_set(struct socket *so, struct protosw *val)
4471 {
4472 
4473 	so->so_proto = val;
4474 }
4475 
4476 void
4477 so_sorwakeup(struct socket *so)
4478 {
4479 
4480 	sorwakeup(so);
4481 }
4482 
4483 void
4484 so_sowwakeup(struct socket *so)
4485 {
4486 
4487 	sowwakeup(so);
4488 }
4489 
4490 void
4491 so_sorwakeup_locked(struct socket *so)
4492 {
4493 
4494 	sorwakeup_locked(so);
4495 }
4496 
4497 void
4498 so_sowwakeup_locked(struct socket *so)
4499 {
4500 
4501 	sowwakeup_locked(so);
4502 }
4503 
4504 void
4505 so_lock(struct socket *so)
4506 {
4507 
4508 	SOCK_LOCK(so);
4509 }
4510 
4511 void
4512 so_unlock(struct socket *so)
4513 {
4514 
4515 	SOCK_UNLOCK(so);
4516 }
4517