xref: /freebsd/sys/kern/uipc_socket.c (revision 24ccef81405eb25efc65f16b6e9a787f3a51151a)
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 	struct socket *sp, *tsp;
1229 	int error = 0;
1230 	bool last __diagused;
1231 
1232 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1233 
1234 	CURVNET_SET(so->so_vnet);
1235 	funsetown(&so->so_sigio);
1236 	if (so->so_state & SS_ISCONNECTED) {
1237 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1238 			error = sodisconnect(so);
1239 			if (error) {
1240 				if (error == ENOTCONN)
1241 					error = 0;
1242 				goto drop;
1243 			}
1244 		}
1245 
1246 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1247 			if ((so->so_state & SS_ISDISCONNECTING) &&
1248 			    (so->so_state & SS_NBIO))
1249 				goto drop;
1250 			while (so->so_state & SS_ISCONNECTED) {
1251 				error = tsleep(&so->so_timeo,
1252 				    PSOCK | PCATCH, "soclos",
1253 				    so->so_linger * hz);
1254 				if (error)
1255 					break;
1256 			}
1257 		}
1258 	}
1259 
1260 drop:
1261 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1262 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1263 
1264 	TAILQ_INIT(&lqueue);
1265 	SOCK_LOCK(so);
1266 	if (SOLISTENING(so)) {
1267 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1268 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1269 
1270 		so->sol_qlen = so->sol_incqlen = 0;
1271 
1272 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1273 			SOCK_LOCK(sp);
1274 			sp->so_qstate = SQ_NONE;
1275 			sp->so_listen = NULL;
1276 			SOCK_UNLOCK(sp);
1277 			last = refcount_release(&so->so_count);
1278 			KASSERT(!last, ("%s: released last reference for %p",
1279 			    __func__, so));
1280 		}
1281 	}
1282 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1283 	so->so_state |= SS_NOFDREF;
1284 	sorele(so);
1285 	TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1286 		SOCK_LOCK(sp);
1287 		if (refcount_load(&sp->so_count) == 0) {
1288 			SOCK_UNLOCK(sp);
1289 			soabort(sp);
1290 		} else {
1291 			/* See the handling of queued sockets in sofree(). */
1292 			SOCK_UNLOCK(sp);
1293 		}
1294 	}
1295 	CURVNET_RESTORE();
1296 	return (error);
1297 }
1298 
1299 /*
1300  * soabort() is used to abruptly tear down a connection, such as when a
1301  * resource limit is reached (listen queue depth exceeded), or if a listen
1302  * socket is closed while there are sockets waiting to be accepted.
1303  *
1304  * This interface is tricky, because it is called on an unreferenced socket,
1305  * and must be called only by a thread that has actually removed the socket
1306  * from the listen queue it was on, or races with other threads are risked.
1307  *
1308  * This interface will call into the protocol code, so must not be called
1309  * with any socket locks held.  Protocols do call it while holding their own
1310  * recursible protocol mutexes, but this is something that should be subject
1311  * to review in the future.
1312  */
1313 void
1314 soabort(struct socket *so)
1315 {
1316 
1317 	/*
1318 	 * In as much as is possible, assert that no references to this
1319 	 * socket are held.  This is not quite the same as asserting that the
1320 	 * current thread is responsible for arranging for no references, but
1321 	 * is as close as we can get for now.
1322 	 */
1323 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1324 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1325 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1326 	VNET_SO_ASSERT(so);
1327 
1328 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1329 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1330 	SOCK_LOCK(so);
1331 	sofree(so);
1332 }
1333 
1334 int
1335 soaccept(struct socket *so, struct sockaddr **nam)
1336 {
1337 	int error;
1338 
1339 	SOCK_LOCK(so);
1340 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1341 	so->so_state &= ~SS_NOFDREF;
1342 	SOCK_UNLOCK(so);
1343 
1344 	CURVNET_SET(so->so_vnet);
1345 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1346 	CURVNET_RESTORE();
1347 	return (error);
1348 }
1349 
1350 int
1351 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1352 {
1353 
1354 	return (soconnectat(AT_FDCWD, so, nam, td));
1355 }
1356 
1357 int
1358 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1359 {
1360 	int error;
1361 
1362 	CURVNET_SET(so->so_vnet);
1363 	/*
1364 	 * If protocol is connection-based, can only connect once.
1365 	 * Otherwise, if connected, try to disconnect first.  This allows
1366 	 * user to disconnect by connecting to, e.g., a null address.
1367 	 */
1368 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1369 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1370 	    (error = sodisconnect(so)))) {
1371 		error = EISCONN;
1372 	} else {
1373 		/*
1374 		 * Prevent accumulated error from previous connection from
1375 		 * biting us.
1376 		 */
1377 		so->so_error = 0;
1378 		if (fd == AT_FDCWD) {
1379 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1380 			    nam, td);
1381 		} else {
1382 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1383 			    so, nam, td);
1384 		}
1385 	}
1386 	CURVNET_RESTORE();
1387 
1388 	return (error);
1389 }
1390 
1391 int
1392 soconnect2(struct socket *so1, struct socket *so2)
1393 {
1394 	int error;
1395 
1396 	CURVNET_SET(so1->so_vnet);
1397 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1398 	CURVNET_RESTORE();
1399 	return (error);
1400 }
1401 
1402 int
1403 sodisconnect(struct socket *so)
1404 {
1405 	int error;
1406 
1407 	if ((so->so_state & SS_ISCONNECTED) == 0)
1408 		return (ENOTCONN);
1409 	if (so->so_state & SS_ISDISCONNECTING)
1410 		return (EALREADY);
1411 	VNET_SO_ASSERT(so);
1412 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1413 	return (error);
1414 }
1415 
1416 int
1417 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1418     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1419 {
1420 	long space;
1421 	ssize_t resid;
1422 	int clen = 0, error, dontroute;
1423 
1424 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1425 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1426 	    ("sosend_dgram: !PR_ATOMIC"));
1427 
1428 	if (uio != NULL)
1429 		resid = uio->uio_resid;
1430 	else
1431 		resid = top->m_pkthdr.len;
1432 	/*
1433 	 * In theory resid should be unsigned.  However, space must be
1434 	 * signed, as it might be less than 0 if we over-committed, and we
1435 	 * must use a signed comparison of space and resid.  On the other
1436 	 * hand, a negative resid causes us to loop sending 0-length
1437 	 * segments to the protocol.
1438 	 */
1439 	if (resid < 0) {
1440 		error = EINVAL;
1441 		goto out;
1442 	}
1443 
1444 	dontroute =
1445 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1446 	if (td != NULL)
1447 		td->td_ru.ru_msgsnd++;
1448 	if (control != NULL)
1449 		clen = control->m_len;
1450 
1451 	SOCKBUF_LOCK(&so->so_snd);
1452 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1453 		SOCKBUF_UNLOCK(&so->so_snd);
1454 		error = EPIPE;
1455 		goto out;
1456 	}
1457 	if (so->so_error) {
1458 		error = so->so_error;
1459 		so->so_error = 0;
1460 		SOCKBUF_UNLOCK(&so->so_snd);
1461 		goto out;
1462 	}
1463 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1464 		/*
1465 		 * `sendto' and `sendmsg' is allowed on a connection-based
1466 		 * socket if it supports implied connect.  Return ENOTCONN if
1467 		 * not connected and no address is supplied.
1468 		 */
1469 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1470 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1471 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1472 			    !(resid == 0 && clen != 0)) {
1473 				SOCKBUF_UNLOCK(&so->so_snd);
1474 				error = ENOTCONN;
1475 				goto out;
1476 			}
1477 		} else if (addr == NULL) {
1478 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1479 				error = ENOTCONN;
1480 			else
1481 				error = EDESTADDRREQ;
1482 			SOCKBUF_UNLOCK(&so->so_snd);
1483 			goto out;
1484 		}
1485 	}
1486 
1487 	/*
1488 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1489 	 * problem and need fixing.
1490 	 */
1491 	space = sbspace(&so->so_snd);
1492 	if (flags & MSG_OOB)
1493 		space += 1024;
1494 	space -= clen;
1495 	SOCKBUF_UNLOCK(&so->so_snd);
1496 	if (resid > space) {
1497 		error = EMSGSIZE;
1498 		goto out;
1499 	}
1500 	if (uio == NULL) {
1501 		resid = 0;
1502 		if (flags & MSG_EOR)
1503 			top->m_flags |= M_EOR;
1504 	} else {
1505 		/*
1506 		 * Copy the data from userland into a mbuf chain.
1507 		 * If no data is to be copied in, a single empty mbuf
1508 		 * is returned.
1509 		 */
1510 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1511 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1512 		if (top == NULL) {
1513 			error = EFAULT;	/* only possible error */
1514 			goto out;
1515 		}
1516 		space -= resid - uio->uio_resid;
1517 		resid = uio->uio_resid;
1518 	}
1519 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1520 	/*
1521 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1522 	 * than with.
1523 	 */
1524 	if (dontroute) {
1525 		SOCK_LOCK(so);
1526 		so->so_options |= SO_DONTROUTE;
1527 		SOCK_UNLOCK(so);
1528 	}
1529 	/*
1530 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1531 	 * of date.  We could have received a reset packet in an interrupt or
1532 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1533 	 * probably recheck again inside the locking protection here, but
1534 	 * there are probably other places that this also happens.  We must
1535 	 * rethink this.
1536 	 */
1537 	VNET_SO_ASSERT(so);
1538 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1539 	    (flags & MSG_OOB) ? PRUS_OOB :
1540 	/*
1541 	 * If the user set MSG_EOF, the protocol understands this flag and
1542 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1543 	 */
1544 	    ((flags & MSG_EOF) &&
1545 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1546 	     (resid <= 0)) ?
1547 		PRUS_EOF :
1548 		/* If there is more to send set PRUS_MORETOCOME */
1549 		(flags & MSG_MORETOCOME) ||
1550 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1551 		top, addr, control, td);
1552 	if (dontroute) {
1553 		SOCK_LOCK(so);
1554 		so->so_options &= ~SO_DONTROUTE;
1555 		SOCK_UNLOCK(so);
1556 	}
1557 	clen = 0;
1558 	control = NULL;
1559 	top = NULL;
1560 out:
1561 	if (top != NULL)
1562 		m_freem(top);
1563 	if (control != NULL)
1564 		m_freem(control);
1565 	return (error);
1566 }
1567 
1568 /*
1569  * Send on a socket.  If send must go all at once and message is larger than
1570  * send buffering, then hard error.  Lock against other senders.  If must go
1571  * all at once and not enough room now, then inform user that this would
1572  * block and do nothing.  Otherwise, if nonblocking, send as much as
1573  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1574  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1575  * in mbuf chain must be small enough to send all at once.
1576  *
1577  * Returns nonzero on error, timeout or signal; callers must check for short
1578  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1579  * on return.
1580  */
1581 int
1582 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1583     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1584 {
1585 	long space;
1586 	ssize_t resid;
1587 	int clen = 0, error, dontroute;
1588 	int atomic = sosendallatonce(so) || top;
1589 	int pru_flag;
1590 #ifdef KERN_TLS
1591 	struct ktls_session *tls;
1592 	int tls_enq_cnt, tls_pruflag;
1593 	uint8_t tls_rtype;
1594 
1595 	tls = NULL;
1596 	tls_rtype = TLS_RLTYPE_APP;
1597 #endif
1598 	if (uio != NULL)
1599 		resid = uio->uio_resid;
1600 	else if ((top->m_flags & M_PKTHDR) != 0)
1601 		resid = top->m_pkthdr.len;
1602 	else
1603 		resid = m_length(top, NULL);
1604 	/*
1605 	 * In theory resid should be unsigned.  However, space must be
1606 	 * signed, as it might be less than 0 if we over-committed, and we
1607 	 * must use a signed comparison of space and resid.  On the other
1608 	 * hand, a negative resid causes us to loop sending 0-length
1609 	 * segments to the protocol.
1610 	 *
1611 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1612 	 * type sockets since that's an error.
1613 	 */
1614 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1615 		error = EINVAL;
1616 		goto out;
1617 	}
1618 
1619 	dontroute =
1620 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1621 	    (so->so_proto->pr_flags & PR_ATOMIC);
1622 	if (td != NULL)
1623 		td->td_ru.ru_msgsnd++;
1624 	if (control != NULL)
1625 		clen = control->m_len;
1626 
1627 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1628 	if (error)
1629 		goto out;
1630 
1631 #ifdef KERN_TLS
1632 	tls_pruflag = 0;
1633 	tls = ktls_hold(so->so_snd.sb_tls_info);
1634 	if (tls != NULL) {
1635 		if (tls->mode == TCP_TLS_MODE_SW)
1636 			tls_pruflag = PRUS_NOTREADY;
1637 
1638 		if (control != NULL) {
1639 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1640 
1641 			if (clen >= sizeof(*cm) &&
1642 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1643 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1644 				clen = 0;
1645 				m_freem(control);
1646 				control = NULL;
1647 				atomic = 1;
1648 			}
1649 		}
1650 	}
1651 #endif
1652 
1653 restart:
1654 	do {
1655 		SOCKBUF_LOCK(&so->so_snd);
1656 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1657 			SOCKBUF_UNLOCK(&so->so_snd);
1658 			error = EPIPE;
1659 			goto release;
1660 		}
1661 		if (so->so_error) {
1662 			error = so->so_error;
1663 			so->so_error = 0;
1664 			SOCKBUF_UNLOCK(&so->so_snd);
1665 			goto release;
1666 		}
1667 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1668 			/*
1669 			 * `sendto' and `sendmsg' is allowed on a connection-
1670 			 * based socket if it supports implied connect.
1671 			 * Return ENOTCONN if not connected and no address is
1672 			 * supplied.
1673 			 */
1674 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1675 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1676 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1677 				    !(resid == 0 && clen != 0)) {
1678 					SOCKBUF_UNLOCK(&so->so_snd);
1679 					error = ENOTCONN;
1680 					goto release;
1681 				}
1682 			} else if (addr == NULL) {
1683 				SOCKBUF_UNLOCK(&so->so_snd);
1684 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1685 					error = ENOTCONN;
1686 				else
1687 					error = EDESTADDRREQ;
1688 				goto release;
1689 			}
1690 		}
1691 		space = sbspace(&so->so_snd);
1692 		if (flags & MSG_OOB)
1693 			space += 1024;
1694 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1695 		    clen > so->so_snd.sb_hiwat) {
1696 			SOCKBUF_UNLOCK(&so->so_snd);
1697 			error = EMSGSIZE;
1698 			goto release;
1699 		}
1700 		if (space < resid + clen &&
1701 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1702 			if ((so->so_state & SS_NBIO) ||
1703 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1704 				SOCKBUF_UNLOCK(&so->so_snd);
1705 				error = EWOULDBLOCK;
1706 				goto release;
1707 			}
1708 			error = sbwait(&so->so_snd);
1709 			SOCKBUF_UNLOCK(&so->so_snd);
1710 			if (error)
1711 				goto release;
1712 			goto restart;
1713 		}
1714 		SOCKBUF_UNLOCK(&so->so_snd);
1715 		space -= clen;
1716 		do {
1717 			if (uio == NULL) {
1718 				resid = 0;
1719 				if (flags & MSG_EOR)
1720 					top->m_flags |= M_EOR;
1721 #ifdef KERN_TLS
1722 				if (tls != NULL) {
1723 					ktls_frame(top, tls, &tls_enq_cnt,
1724 					    tls_rtype);
1725 					tls_rtype = TLS_RLTYPE_APP;
1726 				}
1727 #endif
1728 			} else {
1729 				/*
1730 				 * Copy the data from userland into a mbuf
1731 				 * chain.  If resid is 0, which can happen
1732 				 * only if we have control to send, then
1733 				 * a single empty mbuf is returned.  This
1734 				 * is a workaround to prevent protocol send
1735 				 * methods to panic.
1736 				 */
1737 #ifdef KERN_TLS
1738 				if (tls != NULL) {
1739 					top = m_uiotombuf(uio, M_WAITOK, space,
1740 					    tls->params.max_frame_len,
1741 					    M_EXTPG |
1742 					    ((flags & MSG_EOR) ? M_EOR : 0));
1743 					if (top != NULL) {
1744 						ktls_frame(top, tls,
1745 						    &tls_enq_cnt, tls_rtype);
1746 					}
1747 					tls_rtype = TLS_RLTYPE_APP;
1748 				} else
1749 #endif
1750 					top = m_uiotombuf(uio, M_WAITOK, space,
1751 					    (atomic ? max_hdr : 0),
1752 					    (atomic ? M_PKTHDR : 0) |
1753 					    ((flags & MSG_EOR) ? M_EOR : 0));
1754 				if (top == NULL) {
1755 					error = EFAULT; /* only possible error */
1756 					goto release;
1757 				}
1758 				space -= resid - uio->uio_resid;
1759 				resid = uio->uio_resid;
1760 			}
1761 			if (dontroute) {
1762 				SOCK_LOCK(so);
1763 				so->so_options |= SO_DONTROUTE;
1764 				SOCK_UNLOCK(so);
1765 			}
1766 			/*
1767 			 * XXX all the SBS_CANTSENDMORE checks previously
1768 			 * done could be out of date.  We could have received
1769 			 * a reset packet in an interrupt or maybe we slept
1770 			 * while doing page faults in uiomove() etc.  We
1771 			 * could probably recheck again inside the locking
1772 			 * protection here, but there are probably other
1773 			 * places that this also happens.  We must rethink
1774 			 * this.
1775 			 */
1776 			VNET_SO_ASSERT(so);
1777 
1778 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1779 			/*
1780 			 * If the user set MSG_EOF, the protocol understands
1781 			 * this flag and nothing left to send then use
1782 			 * PRU_SEND_EOF instead of PRU_SEND.
1783 			 */
1784 			    ((flags & MSG_EOF) &&
1785 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1786 			     (resid <= 0)) ?
1787 				PRUS_EOF :
1788 			/* If there is more to send set PRUS_MORETOCOME. */
1789 			    (flags & MSG_MORETOCOME) ||
1790 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1791 
1792 #ifdef KERN_TLS
1793 			pru_flag |= tls_pruflag;
1794 #endif
1795 
1796 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1797 			    pru_flag, top, addr, control, td);
1798 
1799 			if (dontroute) {
1800 				SOCK_LOCK(so);
1801 				so->so_options &= ~SO_DONTROUTE;
1802 				SOCK_UNLOCK(so);
1803 			}
1804 
1805 #ifdef KERN_TLS
1806 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1807 				if (error != 0) {
1808 					m_freem(top);
1809 					top = NULL;
1810 				} else {
1811 					soref(so);
1812 					ktls_enqueue(top, so, tls_enq_cnt);
1813 				}
1814 			}
1815 #endif
1816 			clen = 0;
1817 			control = NULL;
1818 			top = NULL;
1819 			if (error)
1820 				goto release;
1821 		} while (resid && space > 0);
1822 	} while (resid);
1823 
1824 release:
1825 	SOCK_IO_SEND_UNLOCK(so);
1826 out:
1827 #ifdef KERN_TLS
1828 	if (tls != NULL)
1829 		ktls_free(tls);
1830 #endif
1831 	if (top != NULL)
1832 		m_freem(top);
1833 	if (control != NULL)
1834 		m_freem(control);
1835 	return (error);
1836 }
1837 
1838 int
1839 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1840     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1841 {
1842 	int error;
1843 
1844 	CURVNET_SET(so->so_vnet);
1845 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1846 	    top, control, flags, td);
1847 	CURVNET_RESTORE();
1848 	return (error);
1849 }
1850 
1851 /*
1852  * The part of soreceive() that implements reading non-inline out-of-band
1853  * data from a socket.  For more complete comments, see soreceive(), from
1854  * which this code originated.
1855  *
1856  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1857  * unable to return an mbuf chain to the caller.
1858  */
1859 static int
1860 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1861 {
1862 	struct protosw *pr = so->so_proto;
1863 	struct mbuf *m;
1864 	int error;
1865 
1866 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1867 	VNET_SO_ASSERT(so);
1868 
1869 	m = m_get(M_WAITOK, MT_DATA);
1870 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1871 	if (error)
1872 		goto bad;
1873 	do {
1874 		error = uiomove(mtod(m, void *),
1875 		    (int) min(uio->uio_resid, m->m_len), uio);
1876 		m = m_free(m);
1877 	} while (uio->uio_resid && error == 0 && m);
1878 bad:
1879 	if (m != NULL)
1880 		m_freem(m);
1881 	return (error);
1882 }
1883 
1884 /*
1885  * Following replacement or removal of the first mbuf on the first mbuf chain
1886  * of a socket buffer, push necessary state changes back into the socket
1887  * buffer so that other consumers see the values consistently.  'nextrecord'
1888  * is the callers locally stored value of the original value of
1889  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1890  * NOTE: 'nextrecord' may be NULL.
1891  */
1892 static __inline void
1893 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1894 {
1895 
1896 	SOCKBUF_LOCK_ASSERT(sb);
1897 	/*
1898 	 * First, update for the new value of nextrecord.  If necessary, make
1899 	 * it the first record.
1900 	 */
1901 	if (sb->sb_mb != NULL)
1902 		sb->sb_mb->m_nextpkt = nextrecord;
1903 	else
1904 		sb->sb_mb = nextrecord;
1905 
1906 	/*
1907 	 * Now update any dependent socket buffer fields to reflect the new
1908 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1909 	 * addition of a second clause that takes care of the case where
1910 	 * sb_mb has been updated, but remains the last record.
1911 	 */
1912 	if (sb->sb_mb == NULL) {
1913 		sb->sb_mbtail = NULL;
1914 		sb->sb_lastrecord = NULL;
1915 	} else if (sb->sb_mb->m_nextpkt == NULL)
1916 		sb->sb_lastrecord = sb->sb_mb;
1917 }
1918 
1919 /*
1920  * Implement receive operations on a socket.  We depend on the way that
1921  * records are added to the sockbuf by sbappend.  In particular, each record
1922  * (mbufs linked through m_next) must begin with an address if the protocol
1923  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1924  * data, and then zero or more mbufs of data.  In order to allow parallelism
1925  * between network receive and copying to user space, as well as avoid
1926  * sleeping with a mutex held, we release the socket buffer mutex during the
1927  * user space copy.  Although the sockbuf is locked, new data may still be
1928  * appended, and thus we must maintain consistency of the sockbuf during that
1929  * time.
1930  *
1931  * The caller may receive the data as a single mbuf chain by supplying an
1932  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1933  * the count in uio_resid.
1934  */
1935 int
1936 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1937     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1938 {
1939 	struct mbuf *m, **mp;
1940 	int flags, error, offset;
1941 	ssize_t len;
1942 	struct protosw *pr = so->so_proto;
1943 	struct mbuf *nextrecord;
1944 	int moff, type = 0;
1945 	ssize_t orig_resid = uio->uio_resid;
1946 
1947 	mp = mp0;
1948 	if (psa != NULL)
1949 		*psa = NULL;
1950 	if (controlp != NULL)
1951 		*controlp = NULL;
1952 	if (flagsp != NULL)
1953 		flags = *flagsp &~ MSG_EOR;
1954 	else
1955 		flags = 0;
1956 	if (flags & MSG_OOB)
1957 		return (soreceive_rcvoob(so, uio, flags));
1958 	if (mp != NULL)
1959 		*mp = NULL;
1960 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1961 	    && uio->uio_resid) {
1962 		VNET_SO_ASSERT(so);
1963 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1964 	}
1965 
1966 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1967 	if (error)
1968 		return (error);
1969 
1970 restart:
1971 	SOCKBUF_LOCK(&so->so_rcv);
1972 	m = so->so_rcv.sb_mb;
1973 	/*
1974 	 * If we have less data than requested, block awaiting more (subject
1975 	 * to any timeout) if:
1976 	 *   1. the current count is less than the low water mark, or
1977 	 *   2. MSG_DONTWAIT is not set
1978 	 */
1979 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1980 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1981 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1982 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1983 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1984 		    ("receive: m == %p sbavail == %u",
1985 		    m, sbavail(&so->so_rcv)));
1986 		if (so->so_error || so->so_rerror) {
1987 			if (m != NULL)
1988 				goto dontblock;
1989 			if (so->so_error)
1990 				error = so->so_error;
1991 			else
1992 				error = so->so_rerror;
1993 			if ((flags & MSG_PEEK) == 0) {
1994 				if (so->so_error)
1995 					so->so_error = 0;
1996 				else
1997 					so->so_rerror = 0;
1998 			}
1999 			SOCKBUF_UNLOCK(&so->so_rcv);
2000 			goto release;
2001 		}
2002 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2003 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2004 			if (m != NULL)
2005 				goto dontblock;
2006 #ifdef KERN_TLS
2007 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2008 			    so->so_rcv.sb_tlscc == 0) {
2009 #else
2010 			else {
2011 #endif
2012 				SOCKBUF_UNLOCK(&so->so_rcv);
2013 				goto release;
2014 			}
2015 		}
2016 		for (; m != NULL; m = m->m_next)
2017 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2018 				m = so->so_rcv.sb_mb;
2019 				goto dontblock;
2020 			}
2021 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2022 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2023 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2024 			SOCKBUF_UNLOCK(&so->so_rcv);
2025 			error = ENOTCONN;
2026 			goto release;
2027 		}
2028 		if (uio->uio_resid == 0) {
2029 			SOCKBUF_UNLOCK(&so->so_rcv);
2030 			goto release;
2031 		}
2032 		if ((so->so_state & SS_NBIO) ||
2033 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2034 			SOCKBUF_UNLOCK(&so->so_rcv);
2035 			error = EWOULDBLOCK;
2036 			goto release;
2037 		}
2038 		SBLASTRECORDCHK(&so->so_rcv);
2039 		SBLASTMBUFCHK(&so->so_rcv);
2040 		error = sbwait(&so->so_rcv);
2041 		SOCKBUF_UNLOCK(&so->so_rcv);
2042 		if (error)
2043 			goto release;
2044 		goto restart;
2045 	}
2046 dontblock:
2047 	/*
2048 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2049 	 * pointer to the next record in the socket buffer.  We must keep the
2050 	 * various socket buffer pointers and local stack versions of the
2051 	 * pointers in sync, pushing out modifications before dropping the
2052 	 * socket buffer mutex, and re-reading them when picking it up.
2053 	 *
2054 	 * Otherwise, we will race with the network stack appending new data
2055 	 * or records onto the socket buffer by using inconsistent/stale
2056 	 * versions of the field, possibly resulting in socket buffer
2057 	 * corruption.
2058 	 *
2059 	 * By holding the high-level sblock(), we prevent simultaneous
2060 	 * readers from pulling off the front of the socket buffer.
2061 	 */
2062 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2063 	if (uio->uio_td)
2064 		uio->uio_td->td_ru.ru_msgrcv++;
2065 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2066 	SBLASTRECORDCHK(&so->so_rcv);
2067 	SBLASTMBUFCHK(&so->so_rcv);
2068 	nextrecord = m->m_nextpkt;
2069 	if (pr->pr_flags & PR_ADDR) {
2070 		KASSERT(m->m_type == MT_SONAME,
2071 		    ("m->m_type == %d", m->m_type));
2072 		orig_resid = 0;
2073 		if (psa != NULL)
2074 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2075 			    M_NOWAIT);
2076 		if (flags & MSG_PEEK) {
2077 			m = m->m_next;
2078 		} else {
2079 			sbfree(&so->so_rcv, m);
2080 			so->so_rcv.sb_mb = m_free(m);
2081 			m = so->so_rcv.sb_mb;
2082 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2083 		}
2084 	}
2085 
2086 	/*
2087 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2088 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2089 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2090 	 * perform externalization (or freeing if controlp == NULL).
2091 	 */
2092 	if (m != NULL && m->m_type == MT_CONTROL) {
2093 		struct mbuf *cm = NULL, *cmn;
2094 		struct mbuf **cme = &cm;
2095 #ifdef KERN_TLS
2096 		struct cmsghdr *cmsg;
2097 		struct tls_get_record tgr;
2098 
2099 		/*
2100 		 * For MSG_TLSAPPDATA, check for a non-application data
2101 		 * record.  If found, return ENXIO without removing
2102 		 * it from the receive queue.  This allows a subsequent
2103 		 * call without MSG_TLSAPPDATA to receive it.
2104 		 * Note that, for TLS, there should only be a single
2105 		 * control mbuf with the TLS_GET_RECORD message in it.
2106 		 */
2107 		if (flags & MSG_TLSAPPDATA) {
2108 			cmsg = mtod(m, struct cmsghdr *);
2109 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2110 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2111 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2112 				/* This will need to change for TLS 1.3. */
2113 				if (tgr.tls_type != TLS_RLTYPE_APP) {
2114 					SOCKBUF_UNLOCK(&so->so_rcv);
2115 					error = ENXIO;
2116 					goto release;
2117 				}
2118 			}
2119 		}
2120 #endif
2121 
2122 		do {
2123 			if (flags & MSG_PEEK) {
2124 				if (controlp != NULL) {
2125 					*controlp = m_copym(m, 0, m->m_len,
2126 					    M_NOWAIT);
2127 					controlp = &(*controlp)->m_next;
2128 				}
2129 				m = m->m_next;
2130 			} else {
2131 				sbfree(&so->so_rcv, m);
2132 				so->so_rcv.sb_mb = m->m_next;
2133 				m->m_next = NULL;
2134 				*cme = m;
2135 				cme = &(*cme)->m_next;
2136 				m = so->so_rcv.sb_mb;
2137 			}
2138 		} while (m != NULL && m->m_type == MT_CONTROL);
2139 		if ((flags & MSG_PEEK) == 0)
2140 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2141 		while (cm != NULL) {
2142 			cmn = cm->m_next;
2143 			cm->m_next = NULL;
2144 			if (pr->pr_domain->dom_externalize != NULL) {
2145 				SOCKBUF_UNLOCK(&so->so_rcv);
2146 				VNET_SO_ASSERT(so);
2147 				error = (*pr->pr_domain->dom_externalize)
2148 				    (cm, controlp, flags);
2149 				SOCKBUF_LOCK(&so->so_rcv);
2150 			} else if (controlp != NULL)
2151 				*controlp = cm;
2152 			else
2153 				m_freem(cm);
2154 			if (controlp != NULL) {
2155 				orig_resid = 0;
2156 				while (*controlp != NULL)
2157 					controlp = &(*controlp)->m_next;
2158 			}
2159 			cm = cmn;
2160 		}
2161 		if (m != NULL)
2162 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2163 		else
2164 			nextrecord = so->so_rcv.sb_mb;
2165 		orig_resid = 0;
2166 	}
2167 	if (m != NULL) {
2168 		if ((flags & MSG_PEEK) == 0) {
2169 			KASSERT(m->m_nextpkt == nextrecord,
2170 			    ("soreceive: post-control, nextrecord !sync"));
2171 			if (nextrecord == NULL) {
2172 				KASSERT(so->so_rcv.sb_mb == m,
2173 				    ("soreceive: post-control, sb_mb!=m"));
2174 				KASSERT(so->so_rcv.sb_lastrecord == m,
2175 				    ("soreceive: post-control, lastrecord!=m"));
2176 			}
2177 		}
2178 		type = m->m_type;
2179 		if (type == MT_OOBDATA)
2180 			flags |= MSG_OOB;
2181 	} else {
2182 		if ((flags & MSG_PEEK) == 0) {
2183 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2184 			    ("soreceive: sb_mb != nextrecord"));
2185 			if (so->so_rcv.sb_mb == NULL) {
2186 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2187 				    ("soreceive: sb_lastercord != NULL"));
2188 			}
2189 		}
2190 	}
2191 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2192 	SBLASTRECORDCHK(&so->so_rcv);
2193 	SBLASTMBUFCHK(&so->so_rcv);
2194 
2195 	/*
2196 	 * Now continue to read any data mbufs off of the head of the socket
2197 	 * buffer until the read request is satisfied.  Note that 'type' is
2198 	 * used to store the type of any mbuf reads that have happened so far
2199 	 * such that soreceive() can stop reading if the type changes, which
2200 	 * causes soreceive() to return only one of regular data and inline
2201 	 * out-of-band data in a single socket receive operation.
2202 	 */
2203 	moff = 0;
2204 	offset = 0;
2205 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2206 	    && error == 0) {
2207 		/*
2208 		 * If the type of mbuf has changed since the last mbuf
2209 		 * examined ('type'), end the receive operation.
2210 		 */
2211 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2212 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2213 			if (type != m->m_type)
2214 				break;
2215 		} else if (type == MT_OOBDATA)
2216 			break;
2217 		else
2218 		    KASSERT(m->m_type == MT_DATA,
2219 			("m->m_type == %d", m->m_type));
2220 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2221 		len = uio->uio_resid;
2222 		if (so->so_oobmark && len > so->so_oobmark - offset)
2223 			len = so->so_oobmark - offset;
2224 		if (len > m->m_len - moff)
2225 			len = m->m_len - moff;
2226 		/*
2227 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2228 		 * them out via the uio, then free.  Sockbuf must be
2229 		 * consistent here (points to current mbuf, it points to next
2230 		 * record) when we drop priority; we must note any additions
2231 		 * to the sockbuf when we block interrupts again.
2232 		 */
2233 		if (mp == NULL) {
2234 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2235 			SBLASTRECORDCHK(&so->so_rcv);
2236 			SBLASTMBUFCHK(&so->so_rcv);
2237 			SOCKBUF_UNLOCK(&so->so_rcv);
2238 			if ((m->m_flags & M_EXTPG) != 0)
2239 				error = m_unmapped_uiomove(m, moff, uio,
2240 				    (int)len);
2241 			else
2242 				error = uiomove(mtod(m, char *) + moff,
2243 				    (int)len, uio);
2244 			SOCKBUF_LOCK(&so->so_rcv);
2245 			if (error) {
2246 				/*
2247 				 * The MT_SONAME mbuf has already been removed
2248 				 * from the record, so it is necessary to
2249 				 * remove the data mbufs, if any, to preserve
2250 				 * the invariant in the case of PR_ADDR that
2251 				 * requires MT_SONAME mbufs at the head of
2252 				 * each record.
2253 				 */
2254 				if (pr->pr_flags & PR_ATOMIC &&
2255 				    ((flags & MSG_PEEK) == 0))
2256 					(void)sbdroprecord_locked(&so->so_rcv);
2257 				SOCKBUF_UNLOCK(&so->so_rcv);
2258 				goto release;
2259 			}
2260 		} else
2261 			uio->uio_resid -= len;
2262 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2263 		if (len == m->m_len - moff) {
2264 			if (m->m_flags & M_EOR)
2265 				flags |= MSG_EOR;
2266 			if (flags & MSG_PEEK) {
2267 				m = m->m_next;
2268 				moff = 0;
2269 			} else {
2270 				nextrecord = m->m_nextpkt;
2271 				sbfree(&so->so_rcv, m);
2272 				if (mp != NULL) {
2273 					m->m_nextpkt = NULL;
2274 					*mp = m;
2275 					mp = &m->m_next;
2276 					so->so_rcv.sb_mb = m = m->m_next;
2277 					*mp = NULL;
2278 				} else {
2279 					so->so_rcv.sb_mb = m_free(m);
2280 					m = so->so_rcv.sb_mb;
2281 				}
2282 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2283 				SBLASTRECORDCHK(&so->so_rcv);
2284 				SBLASTMBUFCHK(&so->so_rcv);
2285 			}
2286 		} else {
2287 			if (flags & MSG_PEEK)
2288 				moff += len;
2289 			else {
2290 				if (mp != NULL) {
2291 					if (flags & MSG_DONTWAIT) {
2292 						*mp = m_copym(m, 0, len,
2293 						    M_NOWAIT);
2294 						if (*mp == NULL) {
2295 							/*
2296 							 * m_copym() couldn't
2297 							 * allocate an mbuf.
2298 							 * Adjust uio_resid back
2299 							 * (it was adjusted
2300 							 * down by len bytes,
2301 							 * which we didn't end
2302 							 * up "copying" over).
2303 							 */
2304 							uio->uio_resid += len;
2305 							break;
2306 						}
2307 					} else {
2308 						SOCKBUF_UNLOCK(&so->so_rcv);
2309 						*mp = m_copym(m, 0, len,
2310 						    M_WAITOK);
2311 						SOCKBUF_LOCK(&so->so_rcv);
2312 					}
2313 				}
2314 				sbcut_locked(&so->so_rcv, len);
2315 			}
2316 		}
2317 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2318 		if (so->so_oobmark) {
2319 			if ((flags & MSG_PEEK) == 0) {
2320 				so->so_oobmark -= len;
2321 				if (so->so_oobmark == 0) {
2322 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2323 					break;
2324 				}
2325 			} else {
2326 				offset += len;
2327 				if (offset == so->so_oobmark)
2328 					break;
2329 			}
2330 		}
2331 		if (flags & MSG_EOR)
2332 			break;
2333 		/*
2334 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2335 		 * must not quit until "uio->uio_resid == 0" or an error
2336 		 * termination.  If a signal/timeout occurs, return with a
2337 		 * short count but without error.  Keep sockbuf locked
2338 		 * against other readers.
2339 		 */
2340 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2341 		    !sosendallatonce(so) && nextrecord == NULL) {
2342 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2343 			if (so->so_error || so->so_rerror ||
2344 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2345 				break;
2346 			/*
2347 			 * Notify the protocol that some data has been
2348 			 * drained before blocking.
2349 			 */
2350 			if (pr->pr_flags & PR_WANTRCVD) {
2351 				SOCKBUF_UNLOCK(&so->so_rcv);
2352 				VNET_SO_ASSERT(so);
2353 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2354 				SOCKBUF_LOCK(&so->so_rcv);
2355 			}
2356 			SBLASTRECORDCHK(&so->so_rcv);
2357 			SBLASTMBUFCHK(&so->so_rcv);
2358 			/*
2359 			 * We could receive some data while was notifying
2360 			 * the protocol. Skip blocking in this case.
2361 			 */
2362 			if (so->so_rcv.sb_mb == NULL) {
2363 				error = sbwait(&so->so_rcv);
2364 				if (error) {
2365 					SOCKBUF_UNLOCK(&so->so_rcv);
2366 					goto release;
2367 				}
2368 			}
2369 			m = so->so_rcv.sb_mb;
2370 			if (m != NULL)
2371 				nextrecord = m->m_nextpkt;
2372 		}
2373 	}
2374 
2375 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2376 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2377 		flags |= MSG_TRUNC;
2378 		if ((flags & MSG_PEEK) == 0)
2379 			(void) sbdroprecord_locked(&so->so_rcv);
2380 	}
2381 	if ((flags & MSG_PEEK) == 0) {
2382 		if (m == NULL) {
2383 			/*
2384 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2385 			 * part makes sure sb_lastrecord is up-to-date if
2386 			 * there is still data in the socket buffer.
2387 			 */
2388 			so->so_rcv.sb_mb = nextrecord;
2389 			if (so->so_rcv.sb_mb == NULL) {
2390 				so->so_rcv.sb_mbtail = NULL;
2391 				so->so_rcv.sb_lastrecord = NULL;
2392 			} else if (nextrecord->m_nextpkt == NULL)
2393 				so->so_rcv.sb_lastrecord = nextrecord;
2394 		}
2395 		SBLASTRECORDCHK(&so->so_rcv);
2396 		SBLASTMBUFCHK(&so->so_rcv);
2397 		/*
2398 		 * If soreceive() is being done from the socket callback,
2399 		 * then don't need to generate ACK to peer to update window,
2400 		 * since ACK will be generated on return to TCP.
2401 		 */
2402 		if (!(flags & MSG_SOCALLBCK) &&
2403 		    (pr->pr_flags & PR_WANTRCVD)) {
2404 			SOCKBUF_UNLOCK(&so->so_rcv);
2405 			VNET_SO_ASSERT(so);
2406 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2407 			SOCKBUF_LOCK(&so->so_rcv);
2408 		}
2409 	}
2410 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2411 	if (orig_resid == uio->uio_resid && orig_resid &&
2412 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2413 		SOCKBUF_UNLOCK(&so->so_rcv);
2414 		goto restart;
2415 	}
2416 	SOCKBUF_UNLOCK(&so->so_rcv);
2417 
2418 	if (flagsp != NULL)
2419 		*flagsp |= flags;
2420 release:
2421 	SOCK_IO_RECV_UNLOCK(so);
2422 	return (error);
2423 }
2424 
2425 /*
2426  * Optimized version of soreceive() for stream (TCP) sockets.
2427  */
2428 int
2429 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2430     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2431 {
2432 	int len = 0, error = 0, flags, oresid;
2433 	struct sockbuf *sb;
2434 	struct mbuf *m, *n = NULL;
2435 
2436 	/* We only do stream sockets. */
2437 	if (so->so_type != SOCK_STREAM)
2438 		return (EINVAL);
2439 	if (psa != NULL)
2440 		*psa = NULL;
2441 	if (flagsp != NULL)
2442 		flags = *flagsp &~ MSG_EOR;
2443 	else
2444 		flags = 0;
2445 	if (controlp != NULL)
2446 		*controlp = NULL;
2447 	if (flags & MSG_OOB)
2448 		return (soreceive_rcvoob(so, uio, flags));
2449 	if (mp0 != NULL)
2450 		*mp0 = NULL;
2451 
2452 	sb = &so->so_rcv;
2453 
2454 #ifdef KERN_TLS
2455 	/*
2456 	 * KTLS store TLS records as records with a control message to
2457 	 * describe the framing.
2458 	 *
2459 	 * We check once here before acquiring locks to optimize the
2460 	 * common case.
2461 	 */
2462 	if (sb->sb_tls_info != NULL)
2463 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2464 		    flagsp));
2465 #endif
2466 
2467 	/* Prevent other readers from entering the socket. */
2468 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2469 	if (error)
2470 		return (error);
2471 	SOCKBUF_LOCK(sb);
2472 
2473 #ifdef KERN_TLS
2474 	if (sb->sb_tls_info != NULL) {
2475 		SOCKBUF_UNLOCK(sb);
2476 		SOCK_IO_RECV_UNLOCK(so);
2477 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2478 		    flagsp));
2479 	}
2480 #endif
2481 
2482 	/* Easy one, no space to copyout anything. */
2483 	if (uio->uio_resid == 0) {
2484 		error = EINVAL;
2485 		goto out;
2486 	}
2487 	oresid = uio->uio_resid;
2488 
2489 	/* We will never ever get anything unless we are or were connected. */
2490 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2491 		error = ENOTCONN;
2492 		goto out;
2493 	}
2494 
2495 restart:
2496 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2497 
2498 	/* Abort if socket has reported problems. */
2499 	if (so->so_error) {
2500 		if (sbavail(sb) > 0)
2501 			goto deliver;
2502 		if (oresid > uio->uio_resid)
2503 			goto out;
2504 		error = so->so_error;
2505 		if (!(flags & MSG_PEEK))
2506 			so->so_error = 0;
2507 		goto out;
2508 	}
2509 
2510 	/* Door is closed.  Deliver what is left, if any. */
2511 	if (sb->sb_state & SBS_CANTRCVMORE) {
2512 		if (sbavail(sb) > 0)
2513 			goto deliver;
2514 		else
2515 			goto out;
2516 	}
2517 
2518 	/* Socket buffer is empty and we shall not block. */
2519 	if (sbavail(sb) == 0 &&
2520 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2521 		error = EAGAIN;
2522 		goto out;
2523 	}
2524 
2525 	/* Socket buffer got some data that we shall deliver now. */
2526 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2527 	    ((so->so_state & SS_NBIO) ||
2528 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2529 	     sbavail(sb) >= sb->sb_lowat ||
2530 	     sbavail(sb) >= uio->uio_resid ||
2531 	     sbavail(sb) >= sb->sb_hiwat) ) {
2532 		goto deliver;
2533 	}
2534 
2535 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2536 	if ((flags & MSG_WAITALL) &&
2537 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2538 		goto deliver;
2539 
2540 	/*
2541 	 * Wait and block until (more) data comes in.
2542 	 * NB: Drops the sockbuf lock during wait.
2543 	 */
2544 	error = sbwait(sb);
2545 	if (error)
2546 		goto out;
2547 	goto restart;
2548 
2549 deliver:
2550 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2551 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2552 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2553 
2554 	/* Statistics. */
2555 	if (uio->uio_td)
2556 		uio->uio_td->td_ru.ru_msgrcv++;
2557 
2558 	/* Fill uio until full or current end of socket buffer is reached. */
2559 	len = min(uio->uio_resid, sbavail(sb));
2560 	if (mp0 != NULL) {
2561 		/* Dequeue as many mbufs as possible. */
2562 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2563 			if (*mp0 == NULL)
2564 				*mp0 = sb->sb_mb;
2565 			else
2566 				m_cat(*mp0, sb->sb_mb);
2567 			for (m = sb->sb_mb;
2568 			     m != NULL && m->m_len <= len;
2569 			     m = m->m_next) {
2570 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2571 				    ("%s: m %p not available", __func__, m));
2572 				len -= m->m_len;
2573 				uio->uio_resid -= m->m_len;
2574 				sbfree(sb, m);
2575 				n = m;
2576 			}
2577 			n->m_next = NULL;
2578 			sb->sb_mb = m;
2579 			sb->sb_lastrecord = sb->sb_mb;
2580 			if (sb->sb_mb == NULL)
2581 				SB_EMPTY_FIXUP(sb);
2582 		}
2583 		/* Copy the remainder. */
2584 		if (len > 0) {
2585 			KASSERT(sb->sb_mb != NULL,
2586 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2587 
2588 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2589 			if (m == NULL)
2590 				len = 0;	/* Don't flush data from sockbuf. */
2591 			else
2592 				uio->uio_resid -= len;
2593 			if (*mp0 != NULL)
2594 				m_cat(*mp0, m);
2595 			else
2596 				*mp0 = m;
2597 			if (*mp0 == NULL) {
2598 				error = ENOBUFS;
2599 				goto out;
2600 			}
2601 		}
2602 	} else {
2603 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2604 		SOCKBUF_UNLOCK(sb);
2605 		error = m_mbuftouio(uio, sb->sb_mb, len);
2606 		SOCKBUF_LOCK(sb);
2607 		if (error)
2608 			goto out;
2609 	}
2610 	SBLASTRECORDCHK(sb);
2611 	SBLASTMBUFCHK(sb);
2612 
2613 	/*
2614 	 * Remove the delivered data from the socket buffer unless we
2615 	 * were only peeking.
2616 	 */
2617 	if (!(flags & MSG_PEEK)) {
2618 		if (len > 0)
2619 			sbdrop_locked(sb, len);
2620 
2621 		/* Notify protocol that we drained some data. */
2622 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2623 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2624 		     !(flags & MSG_SOCALLBCK))) {
2625 			SOCKBUF_UNLOCK(sb);
2626 			VNET_SO_ASSERT(so);
2627 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2628 			SOCKBUF_LOCK(sb);
2629 		}
2630 	}
2631 
2632 	/*
2633 	 * For MSG_WAITALL we may have to loop again and wait for
2634 	 * more data to come in.
2635 	 */
2636 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2637 		goto restart;
2638 out:
2639 	SBLASTRECORDCHK(sb);
2640 	SBLASTMBUFCHK(sb);
2641 	SOCKBUF_UNLOCK(sb);
2642 	SOCK_IO_RECV_UNLOCK(so);
2643 	return (error);
2644 }
2645 
2646 /*
2647  * Optimized version of soreceive() for simple datagram cases from userspace.
2648  * Unlike in the stream case, we're able to drop a datagram if copyout()
2649  * fails, and because we handle datagrams atomically, we don't need to use a
2650  * sleep lock to prevent I/O interlacing.
2651  */
2652 int
2653 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2654     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2655 {
2656 	struct mbuf *m, *m2;
2657 	int flags, error;
2658 	ssize_t len;
2659 	struct protosw *pr = so->so_proto;
2660 	struct mbuf *nextrecord;
2661 
2662 	if (psa != NULL)
2663 		*psa = NULL;
2664 	if (controlp != NULL)
2665 		*controlp = NULL;
2666 	if (flagsp != NULL)
2667 		flags = *flagsp &~ MSG_EOR;
2668 	else
2669 		flags = 0;
2670 
2671 	/*
2672 	 * For any complicated cases, fall back to the full
2673 	 * soreceive_generic().
2674 	 */
2675 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2676 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2677 		    flagsp));
2678 
2679 	/*
2680 	 * Enforce restrictions on use.
2681 	 */
2682 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2683 	    ("soreceive_dgram: wantrcvd"));
2684 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2685 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2686 	    ("soreceive_dgram: SBS_RCVATMARK"));
2687 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2688 	    ("soreceive_dgram: P_CONNREQUIRED"));
2689 
2690 	/*
2691 	 * Loop blocking while waiting for a datagram.
2692 	 */
2693 	SOCKBUF_LOCK(&so->so_rcv);
2694 	while ((m = so->so_rcv.sb_mb) == NULL) {
2695 		KASSERT(sbavail(&so->so_rcv) == 0,
2696 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2697 		    sbavail(&so->so_rcv)));
2698 		if (so->so_error) {
2699 			error = so->so_error;
2700 			so->so_error = 0;
2701 			SOCKBUF_UNLOCK(&so->so_rcv);
2702 			return (error);
2703 		}
2704 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2705 		    uio->uio_resid == 0) {
2706 			SOCKBUF_UNLOCK(&so->so_rcv);
2707 			return (0);
2708 		}
2709 		if ((so->so_state & SS_NBIO) ||
2710 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2711 			SOCKBUF_UNLOCK(&so->so_rcv);
2712 			return (EWOULDBLOCK);
2713 		}
2714 		SBLASTRECORDCHK(&so->so_rcv);
2715 		SBLASTMBUFCHK(&so->so_rcv);
2716 		error = sbwait(&so->so_rcv);
2717 		if (error) {
2718 			SOCKBUF_UNLOCK(&so->so_rcv);
2719 			return (error);
2720 		}
2721 	}
2722 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2723 
2724 	if (uio->uio_td)
2725 		uio->uio_td->td_ru.ru_msgrcv++;
2726 	SBLASTRECORDCHK(&so->so_rcv);
2727 	SBLASTMBUFCHK(&so->so_rcv);
2728 	nextrecord = m->m_nextpkt;
2729 	if (nextrecord == NULL) {
2730 		KASSERT(so->so_rcv.sb_lastrecord == m,
2731 		    ("soreceive_dgram: lastrecord != m"));
2732 	}
2733 
2734 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2735 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2736 
2737 	/*
2738 	 * Pull 'm' and its chain off the front of the packet queue.
2739 	 */
2740 	so->so_rcv.sb_mb = NULL;
2741 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2742 
2743 	/*
2744 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2745 	 */
2746 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2747 		sbfree(&so->so_rcv, m2);
2748 
2749 	/*
2750 	 * Do a few last checks before we let go of the lock.
2751 	 */
2752 	SBLASTRECORDCHK(&so->so_rcv);
2753 	SBLASTMBUFCHK(&so->so_rcv);
2754 	SOCKBUF_UNLOCK(&so->so_rcv);
2755 
2756 	if (pr->pr_flags & PR_ADDR) {
2757 		KASSERT(m->m_type == MT_SONAME,
2758 		    ("m->m_type == %d", m->m_type));
2759 		if (psa != NULL)
2760 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2761 			    M_NOWAIT);
2762 		m = m_free(m);
2763 	}
2764 	if (m == NULL) {
2765 		/* XXXRW: Can this happen? */
2766 		return (0);
2767 	}
2768 
2769 	/*
2770 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2771 	 * queue.
2772 	 *
2773 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2774 	 * in the first mbuf chain on the socket buffer.  We call into the
2775 	 * protocol to perform externalization (or freeing if controlp ==
2776 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2777 	 * MT_DATA mbufs.
2778 	 */
2779 	if (m->m_type == MT_CONTROL) {
2780 		struct mbuf *cm = NULL, *cmn;
2781 		struct mbuf **cme = &cm;
2782 
2783 		do {
2784 			m2 = m->m_next;
2785 			m->m_next = NULL;
2786 			*cme = m;
2787 			cme = &(*cme)->m_next;
2788 			m = m2;
2789 		} while (m != NULL && m->m_type == MT_CONTROL);
2790 		while (cm != NULL) {
2791 			cmn = cm->m_next;
2792 			cm->m_next = NULL;
2793 			if (pr->pr_domain->dom_externalize != NULL) {
2794 				error = (*pr->pr_domain->dom_externalize)
2795 				    (cm, controlp, flags);
2796 			} else if (controlp != NULL)
2797 				*controlp = cm;
2798 			else
2799 				m_freem(cm);
2800 			if (controlp != NULL) {
2801 				while (*controlp != NULL)
2802 					controlp = &(*controlp)->m_next;
2803 			}
2804 			cm = cmn;
2805 		}
2806 	}
2807 	KASSERT(m == NULL || m->m_type == MT_DATA,
2808 	    ("soreceive_dgram: !data"));
2809 	while (m != NULL && uio->uio_resid > 0) {
2810 		len = uio->uio_resid;
2811 		if (len > m->m_len)
2812 			len = m->m_len;
2813 		error = uiomove(mtod(m, char *), (int)len, uio);
2814 		if (error) {
2815 			m_freem(m);
2816 			return (error);
2817 		}
2818 		if (len == m->m_len)
2819 			m = m_free(m);
2820 		else {
2821 			m->m_data += len;
2822 			m->m_len -= len;
2823 		}
2824 	}
2825 	if (m != NULL) {
2826 		flags |= MSG_TRUNC;
2827 		m_freem(m);
2828 	}
2829 	if (flagsp != NULL)
2830 		*flagsp |= flags;
2831 	return (0);
2832 }
2833 
2834 int
2835 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2836     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2837 {
2838 	int error;
2839 
2840 	CURVNET_SET(so->so_vnet);
2841 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2842 	    mp0, controlp, flagsp));
2843 	CURVNET_RESTORE();
2844 	return (error);
2845 }
2846 
2847 int
2848 soshutdown(struct socket *so, int how)
2849 {
2850 	struct protosw *pr;
2851 	int error, soerror_enotconn;
2852 
2853 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2854 		return (EINVAL);
2855 
2856 	soerror_enotconn = 0;
2857 	SOCK_LOCK(so);
2858 	if ((so->so_state &
2859 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2860 		/*
2861 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2862 		 * invoked on a datagram sockets, however historically we would
2863 		 * actually tear socket down. This is known to be leveraged by
2864 		 * some applications to unblock process waiting in recvXXX(2)
2865 		 * by other process that it shares that socket with. Try to meet
2866 		 * both backward-compatibility and POSIX requirements by forcing
2867 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2868 		 */
2869 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2870 			SOCK_UNLOCK(so);
2871 			return (ENOTCONN);
2872 		}
2873 		soerror_enotconn = 1;
2874 	}
2875 
2876 	if (SOLISTENING(so)) {
2877 		if (how != SHUT_WR) {
2878 			so->so_error = ECONNABORTED;
2879 			solisten_wakeup(so);	/* unlocks so */
2880 		} else {
2881 			SOCK_UNLOCK(so);
2882 		}
2883 		goto done;
2884 	}
2885 	SOCK_UNLOCK(so);
2886 
2887 	CURVNET_SET(so->so_vnet);
2888 	pr = so->so_proto;
2889 	if (pr->pr_usrreqs->pru_flush != NULL)
2890 		(*pr->pr_usrreqs->pru_flush)(so, how);
2891 	if (how != SHUT_WR)
2892 		sorflush(so);
2893 	if (how != SHUT_RD) {
2894 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2895 		wakeup(&so->so_timeo);
2896 		CURVNET_RESTORE();
2897 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2898 	}
2899 	wakeup(&so->so_timeo);
2900 	CURVNET_RESTORE();
2901 
2902 done:
2903 	return (soerror_enotconn ? ENOTCONN : 0);
2904 }
2905 
2906 void
2907 sorflush(struct socket *so)
2908 {
2909 	struct socket aso;
2910 	struct protosw *pr;
2911 	int error;
2912 
2913 	VNET_SO_ASSERT(so);
2914 
2915 	/*
2916 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2917 	 * held, we make a partial copy of the socket buffer and clear the
2918 	 * original.  The new socket buffer copy won't have initialized locks so
2919 	 * we can only call routines that won't use or assert those locks.
2920 	 * Ideally calling socantrcvmore() would prevent data from being added
2921 	 * to the buffer, but currently it merely prevents buffered data from
2922 	 * being read by userspace.  We make this effort to free buffered data
2923 	 * nonetheless.
2924 	 *
2925 	 * Dislodge threads currently blocked in receive and wait to acquire
2926 	 * a lock against other simultaneous readers before clearing the
2927 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2928 	 * despite any existing socket disposition on interruptable waiting.
2929 	 */
2930 	socantrcvmore(so);
2931 
2932 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2933 	if (error != 0) {
2934 		KASSERT(SOLISTENING(so),
2935 		    ("%s: soiolock(%p) failed", __func__, so));
2936 		return;
2937 	}
2938 
2939 	SOCK_RECVBUF_LOCK(so);
2940 	bzero(&aso, sizeof(aso));
2941 	aso.so_pcb = so->so_pcb;
2942 	bcopy(&so->so_rcv.sb_startzero, &aso.so_rcv.sb_startzero,
2943 	    offsetof(struct sockbuf, sb_endzero) -
2944 	    offsetof(struct sockbuf, sb_startzero));
2945 	bzero(&so->so_rcv.sb_startzero,
2946 	    offsetof(struct sockbuf, sb_endzero) -
2947 	    offsetof(struct sockbuf, sb_startzero));
2948 	SOCK_RECVBUF_UNLOCK(so);
2949 	SOCK_IO_RECV_UNLOCK(so);
2950 
2951 	/*
2952 	 * Dispose of special rights and flush the copied socket.  Don't call
2953 	 * any unsafe routines (that rely on locks being initialized) on aso.
2954 	 */
2955 	pr = so->so_proto;
2956 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2957 		(*pr->pr_domain->dom_dispose)(&aso);
2958 	sbrelease_internal(&aso.so_rcv, so);
2959 }
2960 
2961 /*
2962  * Wrapper for Socket established helper hook.
2963  * Parameters: socket, context of the hook point, hook id.
2964  */
2965 static int inline
2966 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2967 {
2968 	struct socket_hhook_data hhook_data = {
2969 		.so = so,
2970 		.hctx = hctx,
2971 		.m = NULL,
2972 		.status = 0
2973 	};
2974 
2975 	CURVNET_SET(so->so_vnet);
2976 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2977 	CURVNET_RESTORE();
2978 
2979 	/* Ugly but needed, since hhooks return void for now */
2980 	return (hhook_data.status);
2981 }
2982 
2983 /*
2984  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2985  * additional variant to handle the case where the option value needs to be
2986  * some kind of integer, but not a specific size.  In addition to their use
2987  * here, these functions are also called by the protocol-level pr_ctloutput()
2988  * routines.
2989  */
2990 int
2991 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2992 {
2993 	size_t	valsize;
2994 
2995 	/*
2996 	 * If the user gives us more than we wanted, we ignore it, but if we
2997 	 * don't get the minimum length the caller wants, we return EINVAL.
2998 	 * On success, sopt->sopt_valsize is set to however much we actually
2999 	 * retrieved.
3000 	 */
3001 	if ((valsize = sopt->sopt_valsize) < minlen)
3002 		return EINVAL;
3003 	if (valsize > len)
3004 		sopt->sopt_valsize = valsize = len;
3005 
3006 	if (sopt->sopt_td != NULL)
3007 		return (copyin(sopt->sopt_val, buf, valsize));
3008 
3009 	bcopy(sopt->sopt_val, buf, valsize);
3010 	return (0);
3011 }
3012 
3013 /*
3014  * Kernel version of setsockopt(2).
3015  *
3016  * XXX: optlen is size_t, not socklen_t
3017  */
3018 int
3019 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3020     size_t optlen)
3021 {
3022 	struct sockopt sopt;
3023 
3024 	sopt.sopt_level = level;
3025 	sopt.sopt_name = optname;
3026 	sopt.sopt_dir = SOPT_SET;
3027 	sopt.sopt_val = optval;
3028 	sopt.sopt_valsize = optlen;
3029 	sopt.sopt_td = NULL;
3030 	return (sosetopt(so, &sopt));
3031 }
3032 
3033 int
3034 sosetopt(struct socket *so, struct sockopt *sopt)
3035 {
3036 	int	error, optval;
3037 	struct	linger l;
3038 	struct	timeval tv;
3039 	sbintime_t val;
3040 	uint32_t val32;
3041 #ifdef MAC
3042 	struct mac extmac;
3043 #endif
3044 
3045 	CURVNET_SET(so->so_vnet);
3046 	error = 0;
3047 	if (sopt->sopt_level != SOL_SOCKET) {
3048 		if (so->so_proto->pr_ctloutput != NULL)
3049 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3050 		else
3051 			error = ENOPROTOOPT;
3052 	} else {
3053 		switch (sopt->sopt_name) {
3054 		case SO_ACCEPTFILTER:
3055 			error = accept_filt_setopt(so, sopt);
3056 			if (error)
3057 				goto bad;
3058 			break;
3059 
3060 		case SO_LINGER:
3061 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3062 			if (error)
3063 				goto bad;
3064 			if (l.l_linger < 0 ||
3065 			    l.l_linger > USHRT_MAX ||
3066 			    l.l_linger > (INT_MAX / hz)) {
3067 				error = EDOM;
3068 				goto bad;
3069 			}
3070 			SOCK_LOCK(so);
3071 			so->so_linger = l.l_linger;
3072 			if (l.l_onoff)
3073 				so->so_options |= SO_LINGER;
3074 			else
3075 				so->so_options &= ~SO_LINGER;
3076 			SOCK_UNLOCK(so);
3077 			break;
3078 
3079 		case SO_DEBUG:
3080 		case SO_KEEPALIVE:
3081 		case SO_DONTROUTE:
3082 		case SO_USELOOPBACK:
3083 		case SO_BROADCAST:
3084 		case SO_REUSEADDR:
3085 		case SO_REUSEPORT:
3086 		case SO_REUSEPORT_LB:
3087 		case SO_OOBINLINE:
3088 		case SO_TIMESTAMP:
3089 		case SO_BINTIME:
3090 		case SO_NOSIGPIPE:
3091 		case SO_NO_DDP:
3092 		case SO_NO_OFFLOAD:
3093 		case SO_RERROR:
3094 			error = sooptcopyin(sopt, &optval, sizeof optval,
3095 			    sizeof optval);
3096 			if (error)
3097 				goto bad;
3098 			SOCK_LOCK(so);
3099 			if (optval)
3100 				so->so_options |= sopt->sopt_name;
3101 			else
3102 				so->so_options &= ~sopt->sopt_name;
3103 			SOCK_UNLOCK(so);
3104 			break;
3105 
3106 		case SO_SETFIB:
3107 			error = sooptcopyin(sopt, &optval, sizeof optval,
3108 			    sizeof optval);
3109 			if (error)
3110 				goto bad;
3111 
3112 			if (optval < 0 || optval >= rt_numfibs) {
3113 				error = EINVAL;
3114 				goto bad;
3115 			}
3116 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3117 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3118 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3119 				so->so_fibnum = optval;
3120 			else
3121 				so->so_fibnum = 0;
3122 			break;
3123 
3124 		case SO_USER_COOKIE:
3125 			error = sooptcopyin(sopt, &val32, sizeof val32,
3126 			    sizeof val32);
3127 			if (error)
3128 				goto bad;
3129 			so->so_user_cookie = val32;
3130 			break;
3131 
3132 		case SO_SNDBUF:
3133 		case SO_RCVBUF:
3134 		case SO_SNDLOWAT:
3135 		case SO_RCVLOWAT:
3136 			error = sooptcopyin(sopt, &optval, sizeof optval,
3137 			    sizeof optval);
3138 			if (error)
3139 				goto bad;
3140 
3141 			/*
3142 			 * Values < 1 make no sense for any of these options,
3143 			 * so disallow them.
3144 			 */
3145 			if (optval < 1) {
3146 				error = EINVAL;
3147 				goto bad;
3148 			}
3149 
3150 			error = sbsetopt(so, sopt->sopt_name, optval);
3151 			break;
3152 
3153 		case SO_SNDTIMEO:
3154 		case SO_RCVTIMEO:
3155 #ifdef COMPAT_FREEBSD32
3156 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3157 				struct timeval32 tv32;
3158 
3159 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3160 				    sizeof tv32);
3161 				CP(tv32, tv, tv_sec);
3162 				CP(tv32, tv, tv_usec);
3163 			} else
3164 #endif
3165 				error = sooptcopyin(sopt, &tv, sizeof tv,
3166 				    sizeof tv);
3167 			if (error)
3168 				goto bad;
3169 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3170 			    tv.tv_usec >= 1000000) {
3171 				error = EDOM;
3172 				goto bad;
3173 			}
3174 			if (tv.tv_sec > INT32_MAX)
3175 				val = SBT_MAX;
3176 			else
3177 				val = tvtosbt(tv);
3178 			switch (sopt->sopt_name) {
3179 			case SO_SNDTIMEO:
3180 				so->so_snd.sb_timeo = val;
3181 				break;
3182 			case SO_RCVTIMEO:
3183 				so->so_rcv.sb_timeo = val;
3184 				break;
3185 			}
3186 			break;
3187 
3188 		case SO_LABEL:
3189 #ifdef MAC
3190 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3191 			    sizeof extmac);
3192 			if (error)
3193 				goto bad;
3194 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3195 			    so, &extmac);
3196 #else
3197 			error = EOPNOTSUPP;
3198 #endif
3199 			break;
3200 
3201 		case SO_TS_CLOCK:
3202 			error = sooptcopyin(sopt, &optval, sizeof optval,
3203 			    sizeof optval);
3204 			if (error)
3205 				goto bad;
3206 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3207 				error = EINVAL;
3208 				goto bad;
3209 			}
3210 			so->so_ts_clock = optval;
3211 			break;
3212 
3213 		case SO_MAX_PACING_RATE:
3214 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3215 			    sizeof(val32));
3216 			if (error)
3217 				goto bad;
3218 			so->so_max_pacing_rate = val32;
3219 			break;
3220 
3221 		default:
3222 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3223 				error = hhook_run_socket(so, sopt,
3224 				    HHOOK_SOCKET_OPT);
3225 			else
3226 				error = ENOPROTOOPT;
3227 			break;
3228 		}
3229 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3230 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3231 	}
3232 bad:
3233 	CURVNET_RESTORE();
3234 	return (error);
3235 }
3236 
3237 /*
3238  * Helper routine for getsockopt.
3239  */
3240 int
3241 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3242 {
3243 	int	error;
3244 	size_t	valsize;
3245 
3246 	error = 0;
3247 
3248 	/*
3249 	 * Documented get behavior is that we always return a value, possibly
3250 	 * truncated to fit in the user's buffer.  Traditional behavior is
3251 	 * that we always tell the user precisely how much we copied, rather
3252 	 * than something useful like the total amount we had available for
3253 	 * her.  Note that this interface is not idempotent; the entire
3254 	 * answer must be generated ahead of time.
3255 	 */
3256 	valsize = min(len, sopt->sopt_valsize);
3257 	sopt->sopt_valsize = valsize;
3258 	if (sopt->sopt_val != NULL) {
3259 		if (sopt->sopt_td != NULL)
3260 			error = copyout(buf, sopt->sopt_val, valsize);
3261 		else
3262 			bcopy(buf, sopt->sopt_val, valsize);
3263 	}
3264 	return (error);
3265 }
3266 
3267 int
3268 sogetopt(struct socket *so, struct sockopt *sopt)
3269 {
3270 	int	error, optval;
3271 	struct	linger l;
3272 	struct	timeval tv;
3273 #ifdef MAC
3274 	struct mac extmac;
3275 #endif
3276 
3277 	CURVNET_SET(so->so_vnet);
3278 	error = 0;
3279 	if (sopt->sopt_level != SOL_SOCKET) {
3280 		if (so->so_proto->pr_ctloutput != NULL)
3281 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3282 		else
3283 			error = ENOPROTOOPT;
3284 		CURVNET_RESTORE();
3285 		return (error);
3286 	} else {
3287 		switch (sopt->sopt_name) {
3288 		case SO_ACCEPTFILTER:
3289 			error = accept_filt_getopt(so, sopt);
3290 			break;
3291 
3292 		case SO_LINGER:
3293 			SOCK_LOCK(so);
3294 			l.l_onoff = so->so_options & SO_LINGER;
3295 			l.l_linger = so->so_linger;
3296 			SOCK_UNLOCK(so);
3297 			error = sooptcopyout(sopt, &l, sizeof l);
3298 			break;
3299 
3300 		case SO_USELOOPBACK:
3301 		case SO_DONTROUTE:
3302 		case SO_DEBUG:
3303 		case SO_KEEPALIVE:
3304 		case SO_REUSEADDR:
3305 		case SO_REUSEPORT:
3306 		case SO_REUSEPORT_LB:
3307 		case SO_BROADCAST:
3308 		case SO_OOBINLINE:
3309 		case SO_ACCEPTCONN:
3310 		case SO_TIMESTAMP:
3311 		case SO_BINTIME:
3312 		case SO_NOSIGPIPE:
3313 		case SO_NO_DDP:
3314 		case SO_NO_OFFLOAD:
3315 		case SO_RERROR:
3316 			optval = so->so_options & sopt->sopt_name;
3317 integer:
3318 			error = sooptcopyout(sopt, &optval, sizeof optval);
3319 			break;
3320 
3321 		case SO_DOMAIN:
3322 			optval = so->so_proto->pr_domain->dom_family;
3323 			goto integer;
3324 
3325 		case SO_TYPE:
3326 			optval = so->so_type;
3327 			goto integer;
3328 
3329 		case SO_PROTOCOL:
3330 			optval = so->so_proto->pr_protocol;
3331 			goto integer;
3332 
3333 		case SO_ERROR:
3334 			SOCK_LOCK(so);
3335 			if (so->so_error) {
3336 				optval = so->so_error;
3337 				so->so_error = 0;
3338 			} else {
3339 				optval = so->so_rerror;
3340 				so->so_rerror = 0;
3341 			}
3342 			SOCK_UNLOCK(so);
3343 			goto integer;
3344 
3345 		case SO_SNDBUF:
3346 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3347 			    so->so_snd.sb_hiwat;
3348 			goto integer;
3349 
3350 		case SO_RCVBUF:
3351 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3352 			    so->so_rcv.sb_hiwat;
3353 			goto integer;
3354 
3355 		case SO_SNDLOWAT:
3356 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3357 			    so->so_snd.sb_lowat;
3358 			goto integer;
3359 
3360 		case SO_RCVLOWAT:
3361 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3362 			    so->so_rcv.sb_lowat;
3363 			goto integer;
3364 
3365 		case SO_SNDTIMEO:
3366 		case SO_RCVTIMEO:
3367 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3368 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3369 #ifdef COMPAT_FREEBSD32
3370 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3371 				struct timeval32 tv32;
3372 
3373 				CP(tv, tv32, tv_sec);
3374 				CP(tv, tv32, tv_usec);
3375 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3376 			} else
3377 #endif
3378 				error = sooptcopyout(sopt, &tv, sizeof tv);
3379 			break;
3380 
3381 		case SO_LABEL:
3382 #ifdef MAC
3383 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3384 			    sizeof(extmac));
3385 			if (error)
3386 				goto bad;
3387 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3388 			    so, &extmac);
3389 			if (error)
3390 				goto bad;
3391 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3392 #else
3393 			error = EOPNOTSUPP;
3394 #endif
3395 			break;
3396 
3397 		case SO_PEERLABEL:
3398 #ifdef MAC
3399 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3400 			    sizeof(extmac));
3401 			if (error)
3402 				goto bad;
3403 			error = mac_getsockopt_peerlabel(
3404 			    sopt->sopt_td->td_ucred, so, &extmac);
3405 			if (error)
3406 				goto bad;
3407 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3408 #else
3409 			error = EOPNOTSUPP;
3410 #endif
3411 			break;
3412 
3413 		case SO_LISTENQLIMIT:
3414 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3415 			goto integer;
3416 
3417 		case SO_LISTENQLEN:
3418 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3419 			goto integer;
3420 
3421 		case SO_LISTENINCQLEN:
3422 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3423 			goto integer;
3424 
3425 		case SO_TS_CLOCK:
3426 			optval = so->so_ts_clock;
3427 			goto integer;
3428 
3429 		case SO_MAX_PACING_RATE:
3430 			optval = so->so_max_pacing_rate;
3431 			goto integer;
3432 
3433 		default:
3434 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3435 				error = hhook_run_socket(so, sopt,
3436 				    HHOOK_SOCKET_OPT);
3437 			else
3438 				error = ENOPROTOOPT;
3439 			break;
3440 		}
3441 	}
3442 #ifdef MAC
3443 bad:
3444 #endif
3445 	CURVNET_RESTORE();
3446 	return (error);
3447 }
3448 
3449 int
3450 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3451 {
3452 	struct mbuf *m, *m_prev;
3453 	int sopt_size = sopt->sopt_valsize;
3454 
3455 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3456 	if (m == NULL)
3457 		return ENOBUFS;
3458 	if (sopt_size > MLEN) {
3459 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3460 		if ((m->m_flags & M_EXT) == 0) {
3461 			m_free(m);
3462 			return ENOBUFS;
3463 		}
3464 		m->m_len = min(MCLBYTES, sopt_size);
3465 	} else {
3466 		m->m_len = min(MLEN, sopt_size);
3467 	}
3468 	sopt_size -= m->m_len;
3469 	*mp = m;
3470 	m_prev = m;
3471 
3472 	while (sopt_size) {
3473 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3474 		if (m == NULL) {
3475 			m_freem(*mp);
3476 			return ENOBUFS;
3477 		}
3478 		if (sopt_size > MLEN) {
3479 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3480 			    M_NOWAIT);
3481 			if ((m->m_flags & M_EXT) == 0) {
3482 				m_freem(m);
3483 				m_freem(*mp);
3484 				return ENOBUFS;
3485 			}
3486 			m->m_len = min(MCLBYTES, sopt_size);
3487 		} else {
3488 			m->m_len = min(MLEN, sopt_size);
3489 		}
3490 		sopt_size -= m->m_len;
3491 		m_prev->m_next = m;
3492 		m_prev = m;
3493 	}
3494 	return (0);
3495 }
3496 
3497 int
3498 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3499 {
3500 	struct mbuf *m0 = m;
3501 
3502 	if (sopt->sopt_val == NULL)
3503 		return (0);
3504 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3505 		if (sopt->sopt_td != NULL) {
3506 			int error;
3507 
3508 			error = copyin(sopt->sopt_val, mtod(m, char *),
3509 			    m->m_len);
3510 			if (error != 0) {
3511 				m_freem(m0);
3512 				return(error);
3513 			}
3514 		} else
3515 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3516 		sopt->sopt_valsize -= m->m_len;
3517 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3518 		m = m->m_next;
3519 	}
3520 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3521 		panic("ip6_sooptmcopyin");
3522 	return (0);
3523 }
3524 
3525 int
3526 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3527 {
3528 	struct mbuf *m0 = m;
3529 	size_t valsize = 0;
3530 
3531 	if (sopt->sopt_val == NULL)
3532 		return (0);
3533 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3534 		if (sopt->sopt_td != NULL) {
3535 			int error;
3536 
3537 			error = copyout(mtod(m, char *), sopt->sopt_val,
3538 			    m->m_len);
3539 			if (error != 0) {
3540 				m_freem(m0);
3541 				return(error);
3542 			}
3543 		} else
3544 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3545 		sopt->sopt_valsize -= m->m_len;
3546 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3547 		valsize += m->m_len;
3548 		m = m->m_next;
3549 	}
3550 	if (m != NULL) {
3551 		/* enough soopt buffer should be given from user-land */
3552 		m_freem(m0);
3553 		return(EINVAL);
3554 	}
3555 	sopt->sopt_valsize = valsize;
3556 	return (0);
3557 }
3558 
3559 /*
3560  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3561  * out-of-band data, which will then notify socket consumers.
3562  */
3563 void
3564 sohasoutofband(struct socket *so)
3565 {
3566 
3567 	if (so->so_sigio != NULL)
3568 		pgsigio(&so->so_sigio, SIGURG, 0);
3569 	selwakeuppri(&so->so_rdsel, PSOCK);
3570 }
3571 
3572 int
3573 sopoll(struct socket *so, int events, struct ucred *active_cred,
3574     struct thread *td)
3575 {
3576 
3577 	/*
3578 	 * We do not need to set or assert curvnet as long as everyone uses
3579 	 * sopoll_generic().
3580 	 */
3581 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3582 	    td));
3583 }
3584 
3585 int
3586 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3587     struct thread *td)
3588 {
3589 	int revents;
3590 
3591 	SOCK_LOCK(so);
3592 	if (SOLISTENING(so)) {
3593 		if (!(events & (POLLIN | POLLRDNORM)))
3594 			revents = 0;
3595 		else if (!TAILQ_EMPTY(&so->sol_comp))
3596 			revents = events & (POLLIN | POLLRDNORM);
3597 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3598 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3599 		else {
3600 			selrecord(td, &so->so_rdsel);
3601 			revents = 0;
3602 		}
3603 	} else {
3604 		revents = 0;
3605 		SOCKBUF_LOCK(&so->so_snd);
3606 		SOCKBUF_LOCK(&so->so_rcv);
3607 		if (events & (POLLIN | POLLRDNORM))
3608 			if (soreadabledata(so))
3609 				revents |= events & (POLLIN | POLLRDNORM);
3610 		if (events & (POLLOUT | POLLWRNORM))
3611 			if (sowriteable(so))
3612 				revents |= events & (POLLOUT | POLLWRNORM);
3613 		if (events & (POLLPRI | POLLRDBAND))
3614 			if (so->so_oobmark ||
3615 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3616 				revents |= events & (POLLPRI | POLLRDBAND);
3617 		if ((events & POLLINIGNEOF) == 0) {
3618 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3619 				revents |= events & (POLLIN | POLLRDNORM);
3620 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3621 					revents |= POLLHUP;
3622 			}
3623 		}
3624 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3625 			revents |= events & POLLRDHUP;
3626 		if (revents == 0) {
3627 			if (events &
3628 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3629 				selrecord(td, &so->so_rdsel);
3630 				so->so_rcv.sb_flags |= SB_SEL;
3631 			}
3632 			if (events & (POLLOUT | POLLWRNORM)) {
3633 				selrecord(td, &so->so_wrsel);
3634 				so->so_snd.sb_flags |= SB_SEL;
3635 			}
3636 		}
3637 		SOCKBUF_UNLOCK(&so->so_rcv);
3638 		SOCKBUF_UNLOCK(&so->so_snd);
3639 	}
3640 	SOCK_UNLOCK(so);
3641 	return (revents);
3642 }
3643 
3644 int
3645 soo_kqfilter(struct file *fp, struct knote *kn)
3646 {
3647 	struct socket *so = kn->kn_fp->f_data;
3648 	struct sockbuf *sb;
3649 	struct knlist *knl;
3650 
3651 	switch (kn->kn_filter) {
3652 	case EVFILT_READ:
3653 		kn->kn_fop = &soread_filtops;
3654 		knl = &so->so_rdsel.si_note;
3655 		sb = &so->so_rcv;
3656 		break;
3657 	case EVFILT_WRITE:
3658 		kn->kn_fop = &sowrite_filtops;
3659 		knl = &so->so_wrsel.si_note;
3660 		sb = &so->so_snd;
3661 		break;
3662 	case EVFILT_EMPTY:
3663 		kn->kn_fop = &soempty_filtops;
3664 		knl = &so->so_wrsel.si_note;
3665 		sb = &so->so_snd;
3666 		break;
3667 	default:
3668 		return (EINVAL);
3669 	}
3670 
3671 	SOCK_LOCK(so);
3672 	if (SOLISTENING(so)) {
3673 		knlist_add(knl, kn, 1);
3674 	} else {
3675 		SOCKBUF_LOCK(sb);
3676 		knlist_add(knl, kn, 1);
3677 		sb->sb_flags |= SB_KNOTE;
3678 		SOCKBUF_UNLOCK(sb);
3679 	}
3680 	SOCK_UNLOCK(so);
3681 	return (0);
3682 }
3683 
3684 /*
3685  * Some routines that return EOPNOTSUPP for entry points that are not
3686  * supported by a protocol.  Fill in as needed.
3687  */
3688 int
3689 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3690 {
3691 
3692 	return EOPNOTSUPP;
3693 }
3694 
3695 int
3696 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3697 {
3698 
3699 	return EOPNOTSUPP;
3700 }
3701 
3702 int
3703 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3704 {
3705 
3706 	return EOPNOTSUPP;
3707 }
3708 
3709 int
3710 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3711 {
3712 
3713 	return EOPNOTSUPP;
3714 }
3715 
3716 int
3717 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3718     struct thread *td)
3719 {
3720 
3721 	return EOPNOTSUPP;
3722 }
3723 
3724 int
3725 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3726 {
3727 
3728 	return EOPNOTSUPP;
3729 }
3730 
3731 int
3732 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3733     struct thread *td)
3734 {
3735 
3736 	return EOPNOTSUPP;
3737 }
3738 
3739 int
3740 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3741 {
3742 
3743 	return EOPNOTSUPP;
3744 }
3745 
3746 int
3747 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3748     struct ifnet *ifp, struct thread *td)
3749 {
3750 
3751 	return EOPNOTSUPP;
3752 }
3753 
3754 int
3755 pru_disconnect_notsupp(struct socket *so)
3756 {
3757 
3758 	return EOPNOTSUPP;
3759 }
3760 
3761 int
3762 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3763 {
3764 
3765 	return EOPNOTSUPP;
3766 }
3767 
3768 int
3769 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3770 {
3771 
3772 	return EOPNOTSUPP;
3773 }
3774 
3775 int
3776 pru_rcvd_notsupp(struct socket *so, int flags)
3777 {
3778 
3779 	return EOPNOTSUPP;
3780 }
3781 
3782 int
3783 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3784 {
3785 
3786 	return EOPNOTSUPP;
3787 }
3788 
3789 int
3790 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3791     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3792 {
3793 
3794 	if (control != NULL)
3795 		m_freem(control);
3796 	if ((flags & PRUS_NOTREADY) == 0)
3797 		m_freem(m);
3798 	return (EOPNOTSUPP);
3799 }
3800 
3801 int
3802 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3803 {
3804 
3805 	return (EOPNOTSUPP);
3806 }
3807 
3808 /*
3809  * This isn't really a ``null'' operation, but it's the default one and
3810  * doesn't do anything destructive.
3811  */
3812 int
3813 pru_sense_null(struct socket *so, struct stat *sb)
3814 {
3815 
3816 	sb->st_blksize = so->so_snd.sb_hiwat;
3817 	return 0;
3818 }
3819 
3820 int
3821 pru_shutdown_notsupp(struct socket *so)
3822 {
3823 
3824 	return EOPNOTSUPP;
3825 }
3826 
3827 int
3828 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3829 {
3830 
3831 	return EOPNOTSUPP;
3832 }
3833 
3834 int
3835 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3836     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3837 {
3838 
3839 	return EOPNOTSUPP;
3840 }
3841 
3842 int
3843 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3844     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3845 {
3846 
3847 	return EOPNOTSUPP;
3848 }
3849 
3850 int
3851 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3852     struct thread *td)
3853 {
3854 
3855 	return EOPNOTSUPP;
3856 }
3857 
3858 static void
3859 filt_sordetach(struct knote *kn)
3860 {
3861 	struct socket *so = kn->kn_fp->f_data;
3862 
3863 	so_rdknl_lock(so);
3864 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3865 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3866 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3867 	so_rdknl_unlock(so);
3868 }
3869 
3870 /*ARGSUSED*/
3871 static int
3872 filt_soread(struct knote *kn, long hint)
3873 {
3874 	struct socket *so;
3875 
3876 	so = kn->kn_fp->f_data;
3877 
3878 	if (SOLISTENING(so)) {
3879 		SOCK_LOCK_ASSERT(so);
3880 		kn->kn_data = so->sol_qlen;
3881 		if (so->so_error) {
3882 			kn->kn_flags |= EV_EOF;
3883 			kn->kn_fflags = so->so_error;
3884 			return (1);
3885 		}
3886 		return (!TAILQ_EMPTY(&so->sol_comp));
3887 	}
3888 
3889 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3890 
3891 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3892 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3893 		kn->kn_flags |= EV_EOF;
3894 		kn->kn_fflags = so->so_error;
3895 		return (1);
3896 	} else if (so->so_error || so->so_rerror)
3897 		return (1);
3898 
3899 	if (kn->kn_sfflags & NOTE_LOWAT) {
3900 		if (kn->kn_data >= kn->kn_sdata)
3901 			return (1);
3902 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3903 		return (1);
3904 
3905 	/* This hook returning non-zero indicates an event, not error */
3906 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3907 }
3908 
3909 static void
3910 filt_sowdetach(struct knote *kn)
3911 {
3912 	struct socket *so = kn->kn_fp->f_data;
3913 
3914 	so_wrknl_lock(so);
3915 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3916 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3917 		so->so_snd.sb_flags &= ~SB_KNOTE;
3918 	so_wrknl_unlock(so);
3919 }
3920 
3921 /*ARGSUSED*/
3922 static int
3923 filt_sowrite(struct knote *kn, long hint)
3924 {
3925 	struct socket *so;
3926 
3927 	so = kn->kn_fp->f_data;
3928 
3929 	if (SOLISTENING(so))
3930 		return (0);
3931 
3932 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3933 	kn->kn_data = sbspace(&so->so_snd);
3934 
3935 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3936 
3937 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3938 		kn->kn_flags |= EV_EOF;
3939 		kn->kn_fflags = so->so_error;
3940 		return (1);
3941 	} else if (so->so_error)	/* temporary udp error */
3942 		return (1);
3943 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3944 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3945 		return (0);
3946 	else if (kn->kn_sfflags & NOTE_LOWAT)
3947 		return (kn->kn_data >= kn->kn_sdata);
3948 	else
3949 		return (kn->kn_data >= so->so_snd.sb_lowat);
3950 }
3951 
3952 static int
3953 filt_soempty(struct knote *kn, long hint)
3954 {
3955 	struct socket *so;
3956 
3957 	so = kn->kn_fp->f_data;
3958 
3959 	if (SOLISTENING(so))
3960 		return (1);
3961 
3962 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3963 	kn->kn_data = sbused(&so->so_snd);
3964 
3965 	if (kn->kn_data == 0)
3966 		return (1);
3967 	else
3968 		return (0);
3969 }
3970 
3971 int
3972 socheckuid(struct socket *so, uid_t uid)
3973 {
3974 
3975 	if (so == NULL)
3976 		return (EPERM);
3977 	if (so->so_cred->cr_uid != uid)
3978 		return (EPERM);
3979 	return (0);
3980 }
3981 
3982 /*
3983  * These functions are used by protocols to notify the socket layer (and its
3984  * consumers) of state changes in the sockets driven by protocol-side events.
3985  */
3986 
3987 /*
3988  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3989  *
3990  * Normal sequence from the active (originating) side is that
3991  * soisconnecting() is called during processing of connect() call, resulting
3992  * in an eventual call to soisconnected() if/when the connection is
3993  * established.  When the connection is torn down soisdisconnecting() is
3994  * called during processing of disconnect() call, and soisdisconnected() is
3995  * called when the connection to the peer is totally severed.  The semantics
3996  * of these routines are such that connectionless protocols can call
3997  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3998  * calls when setting up a ``connection'' takes no time.
3999  *
4000  * From the passive side, a socket is created with two queues of sockets:
4001  * so_incomp for connections in progress and so_comp for connections already
4002  * made and awaiting user acceptance.  As a protocol is preparing incoming
4003  * connections, it creates a socket structure queued on so_incomp by calling
4004  * sonewconn().  When the connection is established, soisconnected() is
4005  * called, and transfers the socket structure to so_comp, making it available
4006  * to accept().
4007  *
4008  * If a socket is closed with sockets on either so_incomp or so_comp, these
4009  * sockets are dropped.
4010  *
4011  * If higher-level protocols are implemented in the kernel, the wakeups done
4012  * here will sometimes cause software-interrupt process scheduling.
4013  */
4014 void
4015 soisconnecting(struct socket *so)
4016 {
4017 
4018 	SOCK_LOCK(so);
4019 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4020 	so->so_state |= SS_ISCONNECTING;
4021 	SOCK_UNLOCK(so);
4022 }
4023 
4024 void
4025 soisconnected(struct socket *so)
4026 {
4027 	bool last __diagused;
4028 
4029 	SOCK_LOCK(so);
4030 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
4031 	so->so_state |= SS_ISCONNECTED;
4032 
4033 	if (so->so_qstate == SQ_INCOMP) {
4034 		struct socket *head = so->so_listen;
4035 		int ret;
4036 
4037 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4038 		/*
4039 		 * Promoting a socket from incomplete queue to complete, we
4040 		 * need to go through reverse order of locking.  We first do
4041 		 * trylock, and if that doesn't succeed, we go the hard way
4042 		 * leaving a reference and rechecking consistency after proper
4043 		 * locking.
4044 		 */
4045 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4046 			soref(head);
4047 			SOCK_UNLOCK(so);
4048 			SOLISTEN_LOCK(head);
4049 			SOCK_LOCK(so);
4050 			if (__predict_false(head != so->so_listen)) {
4051 				/*
4052 				 * The socket went off the listen queue,
4053 				 * should be lost race to close(2) of sol.
4054 				 * The socket is about to soabort().
4055 				 */
4056 				SOCK_UNLOCK(so);
4057 				sorele(head);
4058 				return;
4059 			}
4060 			last = refcount_release(&head->so_count);
4061 			KASSERT(!last, ("%s: released last reference for %p",
4062 			    __func__, head));
4063 		}
4064 again:
4065 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4066 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4067 			head->sol_incqlen--;
4068 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4069 			head->sol_qlen++;
4070 			so->so_qstate = SQ_COMP;
4071 			SOCK_UNLOCK(so);
4072 			solisten_wakeup(head);	/* unlocks */
4073 		} else {
4074 			SOCKBUF_LOCK(&so->so_rcv);
4075 			soupcall_set(so, SO_RCV,
4076 			    head->sol_accept_filter->accf_callback,
4077 			    head->sol_accept_filter_arg);
4078 			so->so_options &= ~SO_ACCEPTFILTER;
4079 			ret = head->sol_accept_filter->accf_callback(so,
4080 			    head->sol_accept_filter_arg, M_NOWAIT);
4081 			if (ret == SU_ISCONNECTED) {
4082 				soupcall_clear(so, SO_RCV);
4083 				SOCKBUF_UNLOCK(&so->so_rcv);
4084 				goto again;
4085 			}
4086 			SOCKBUF_UNLOCK(&so->so_rcv);
4087 			SOCK_UNLOCK(so);
4088 			SOLISTEN_UNLOCK(head);
4089 		}
4090 		return;
4091 	}
4092 	SOCK_UNLOCK(so);
4093 	wakeup(&so->so_timeo);
4094 	sorwakeup(so);
4095 	sowwakeup(so);
4096 }
4097 
4098 void
4099 soisdisconnecting(struct socket *so)
4100 {
4101 
4102 	SOCK_LOCK(so);
4103 	so->so_state &= ~SS_ISCONNECTING;
4104 	so->so_state |= SS_ISDISCONNECTING;
4105 
4106 	if (!SOLISTENING(so)) {
4107 		SOCKBUF_LOCK(&so->so_rcv);
4108 		socantrcvmore_locked(so);
4109 		SOCKBUF_LOCK(&so->so_snd);
4110 		socantsendmore_locked(so);
4111 	}
4112 	SOCK_UNLOCK(so);
4113 	wakeup(&so->so_timeo);
4114 }
4115 
4116 void
4117 soisdisconnected(struct socket *so)
4118 {
4119 
4120 	SOCK_LOCK(so);
4121 
4122 	/*
4123 	 * There is at least one reader of so_state that does not
4124 	 * acquire socket lock, namely soreceive_generic().  Ensure
4125 	 * that it never sees all flags that track connection status
4126 	 * cleared, by ordering the update with a barrier semantic of
4127 	 * our release thread fence.
4128 	 */
4129 	so->so_state |= SS_ISDISCONNECTED;
4130 	atomic_thread_fence_rel();
4131 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4132 
4133 	if (!SOLISTENING(so)) {
4134 		SOCK_UNLOCK(so);
4135 		SOCKBUF_LOCK(&so->so_rcv);
4136 		socantrcvmore_locked(so);
4137 		SOCKBUF_LOCK(&so->so_snd);
4138 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4139 		socantsendmore_locked(so);
4140 	} else
4141 		SOCK_UNLOCK(so);
4142 	wakeup(&so->so_timeo);
4143 }
4144 
4145 int
4146 soiolock(struct socket *so, struct sx *sx, int flags)
4147 {
4148 	int error;
4149 
4150 	KASSERT((flags & SBL_VALID) == flags,
4151 	    ("soiolock: invalid flags %#x", flags));
4152 
4153 	if ((flags & SBL_WAIT) != 0) {
4154 		if ((flags & SBL_NOINTR) != 0) {
4155 			sx_xlock(sx);
4156 		} else {
4157 			error = sx_xlock_sig(sx);
4158 			if (error != 0)
4159 				return (error);
4160 		}
4161 	} else if (!sx_try_xlock(sx)) {
4162 		return (EWOULDBLOCK);
4163 	}
4164 
4165 	if (__predict_false(SOLISTENING(so))) {
4166 		sx_xunlock(sx);
4167 		return (ENOTCONN);
4168 	}
4169 	return (0);
4170 }
4171 
4172 void
4173 soiounlock(struct sx *sx)
4174 {
4175 	sx_xunlock(sx);
4176 }
4177 
4178 /*
4179  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4180  */
4181 struct sockaddr *
4182 sodupsockaddr(const struct sockaddr *sa, int mflags)
4183 {
4184 	struct sockaddr *sa2;
4185 
4186 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4187 	if (sa2)
4188 		bcopy(sa, sa2, sa->sa_len);
4189 	return sa2;
4190 }
4191 
4192 /*
4193  * Register per-socket destructor.
4194  */
4195 void
4196 sodtor_set(struct socket *so, so_dtor_t *func)
4197 {
4198 
4199 	SOCK_LOCK_ASSERT(so);
4200 	so->so_dtor = func;
4201 }
4202 
4203 /*
4204  * Register per-socket buffer upcalls.
4205  */
4206 void
4207 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4208 {
4209 	struct sockbuf *sb;
4210 
4211 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4212 
4213 	switch (which) {
4214 	case SO_RCV:
4215 		sb = &so->so_rcv;
4216 		break;
4217 	case SO_SND:
4218 		sb = &so->so_snd;
4219 		break;
4220 	default:
4221 		panic("soupcall_set: bad which");
4222 	}
4223 	SOCKBUF_LOCK_ASSERT(sb);
4224 	sb->sb_upcall = func;
4225 	sb->sb_upcallarg = arg;
4226 	sb->sb_flags |= SB_UPCALL;
4227 }
4228 
4229 void
4230 soupcall_clear(struct socket *so, int which)
4231 {
4232 	struct sockbuf *sb;
4233 
4234 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4235 
4236 	switch (which) {
4237 	case SO_RCV:
4238 		sb = &so->so_rcv;
4239 		break;
4240 	case SO_SND:
4241 		sb = &so->so_snd;
4242 		break;
4243 	default:
4244 		panic("soupcall_clear: bad which");
4245 	}
4246 	SOCKBUF_LOCK_ASSERT(sb);
4247 	KASSERT(sb->sb_upcall != NULL,
4248 	    ("%s: so %p no upcall to clear", __func__, so));
4249 	sb->sb_upcall = NULL;
4250 	sb->sb_upcallarg = NULL;
4251 	sb->sb_flags &= ~SB_UPCALL;
4252 }
4253 
4254 void
4255 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4256 {
4257 
4258 	SOLISTEN_LOCK_ASSERT(so);
4259 	so->sol_upcall = func;
4260 	so->sol_upcallarg = arg;
4261 }
4262 
4263 static void
4264 so_rdknl_lock(void *arg)
4265 {
4266 	struct socket *so = arg;
4267 
4268 	if (SOLISTENING(so))
4269 		SOCK_LOCK(so);
4270 	else
4271 		SOCKBUF_LOCK(&so->so_rcv);
4272 }
4273 
4274 static void
4275 so_rdknl_unlock(void *arg)
4276 {
4277 	struct socket *so = arg;
4278 
4279 	if (SOLISTENING(so))
4280 		SOCK_UNLOCK(so);
4281 	else
4282 		SOCKBUF_UNLOCK(&so->so_rcv);
4283 }
4284 
4285 static void
4286 so_rdknl_assert_lock(void *arg, int what)
4287 {
4288 	struct socket *so = arg;
4289 
4290 	if (what == LA_LOCKED) {
4291 		if (SOLISTENING(so))
4292 			SOCK_LOCK_ASSERT(so);
4293 		else
4294 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4295 	} else {
4296 		if (SOLISTENING(so))
4297 			SOCK_UNLOCK_ASSERT(so);
4298 		else
4299 			SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4300 	}
4301 }
4302 
4303 static void
4304 so_wrknl_lock(void *arg)
4305 {
4306 	struct socket *so = arg;
4307 
4308 	if (SOLISTENING(so))
4309 		SOCK_LOCK(so);
4310 	else
4311 		SOCKBUF_LOCK(&so->so_snd);
4312 }
4313 
4314 static void
4315 so_wrknl_unlock(void *arg)
4316 {
4317 	struct socket *so = arg;
4318 
4319 	if (SOLISTENING(so))
4320 		SOCK_UNLOCK(so);
4321 	else
4322 		SOCKBUF_UNLOCK(&so->so_snd);
4323 }
4324 
4325 static void
4326 so_wrknl_assert_lock(void *arg, int what)
4327 {
4328 	struct socket *so = arg;
4329 
4330 	if (what == LA_LOCKED) {
4331 		if (SOLISTENING(so))
4332 			SOCK_LOCK_ASSERT(so);
4333 		else
4334 			SOCKBUF_LOCK_ASSERT(&so->so_snd);
4335 	} else {
4336 		if (SOLISTENING(so))
4337 			SOCK_UNLOCK_ASSERT(so);
4338 		else
4339 			SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4340 	}
4341 }
4342 
4343 /*
4344  * Create an external-format (``xsocket'') structure using the information in
4345  * the kernel-format socket structure pointed to by so.  This is done to
4346  * reduce the spew of irrelevant information over this interface, to isolate
4347  * user code from changes in the kernel structure, and potentially to provide
4348  * information-hiding if we decide that some of this information should be
4349  * hidden from users.
4350  */
4351 void
4352 sotoxsocket(struct socket *so, struct xsocket *xso)
4353 {
4354 
4355 	bzero(xso, sizeof(*xso));
4356 	xso->xso_len = sizeof *xso;
4357 	xso->xso_so = (uintptr_t)so;
4358 	xso->so_type = so->so_type;
4359 	xso->so_options = so->so_options;
4360 	xso->so_linger = so->so_linger;
4361 	xso->so_state = so->so_state;
4362 	xso->so_pcb = (uintptr_t)so->so_pcb;
4363 	xso->xso_protocol = so->so_proto->pr_protocol;
4364 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4365 	xso->so_timeo = so->so_timeo;
4366 	xso->so_error = so->so_error;
4367 	xso->so_uid = so->so_cred->cr_uid;
4368 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4369 	if (SOLISTENING(so)) {
4370 		xso->so_qlen = so->sol_qlen;
4371 		xso->so_incqlen = so->sol_incqlen;
4372 		xso->so_qlimit = so->sol_qlimit;
4373 		xso->so_oobmark = 0;
4374 	} else {
4375 		xso->so_state |= so->so_qstate;
4376 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4377 		xso->so_oobmark = so->so_oobmark;
4378 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4379 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4380 	}
4381 }
4382 
4383 struct sockbuf *
4384 so_sockbuf_rcv(struct socket *so)
4385 {
4386 
4387 	return (&so->so_rcv);
4388 }
4389 
4390 struct sockbuf *
4391 so_sockbuf_snd(struct socket *so)
4392 {
4393 
4394 	return (&so->so_snd);
4395 }
4396 
4397 int
4398 so_state_get(const struct socket *so)
4399 {
4400 
4401 	return (so->so_state);
4402 }
4403 
4404 void
4405 so_state_set(struct socket *so, int val)
4406 {
4407 
4408 	so->so_state = val;
4409 }
4410 
4411 int
4412 so_options_get(const struct socket *so)
4413 {
4414 
4415 	return (so->so_options);
4416 }
4417 
4418 void
4419 so_options_set(struct socket *so, int val)
4420 {
4421 
4422 	so->so_options = val;
4423 }
4424 
4425 int
4426 so_error_get(const struct socket *so)
4427 {
4428 
4429 	return (so->so_error);
4430 }
4431 
4432 void
4433 so_error_set(struct socket *so, int val)
4434 {
4435 
4436 	so->so_error = val;
4437 }
4438 
4439 int
4440 so_linger_get(const struct socket *so)
4441 {
4442 
4443 	return (so->so_linger);
4444 }
4445 
4446 void
4447 so_linger_set(struct socket *so, int val)
4448 {
4449 
4450 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4451 	    ("%s: val %d out of range", __func__, val));
4452 
4453 	so->so_linger = val;
4454 }
4455 
4456 struct protosw *
4457 so_protosw_get(const struct socket *so)
4458 {
4459 
4460 	return (so->so_proto);
4461 }
4462 
4463 void
4464 so_protosw_set(struct socket *so, struct protosw *val)
4465 {
4466 
4467 	so->so_proto = val;
4468 }
4469 
4470 void
4471 so_sorwakeup(struct socket *so)
4472 {
4473 
4474 	sorwakeup(so);
4475 }
4476 
4477 void
4478 so_sowwakeup(struct socket *so)
4479 {
4480 
4481 	sowwakeup(so);
4482 }
4483 
4484 void
4485 so_sorwakeup_locked(struct socket *so)
4486 {
4487 
4488 	sorwakeup_locked(so);
4489 }
4490 
4491 void
4492 so_sowwakeup_locked(struct socket *so)
4493 {
4494 
4495 	sowwakeup_locked(so);
4496 }
4497 
4498 void
4499 so_lock(struct socket *so)
4500 {
4501 
4502 	SOCK_LOCK(so);
4503 }
4504 
4505 void
4506 so_unlock(struct socket *so)
4507 {
4508 
4509 	SOCK_UNLOCK(so);
4510 }
4511