xref: /freebsd/sys/kern/uipc_socket.c (revision 65990b68a2cd89a08f0350e187df1968b16f4255)
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_MPSAFE, 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_MPSAFE, 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 && newmaxsockets != maxsockets) {
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_MPSAFE, &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_locked(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_locked(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_locked(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  * Release a reference on a socket while holding the socket lock.
1218  * Unlocks the socket lock before returning.
1219  */
1220 void
1221 sorele_locked(struct socket *so)
1222 {
1223 	SOCK_LOCK_ASSERT(so);
1224 	if (refcount_release(&so->so_count))
1225 		sofree(so);
1226 	else
1227 		SOCK_UNLOCK(so);
1228 }
1229 
1230 /*
1231  * Close a socket on last file table reference removal.  Initiate disconnect
1232  * if connected.  Free socket when disconnect complete.
1233  *
1234  * This function will sorele() the socket.  Note that soclose() may be called
1235  * prior to the ref count reaching zero.  The actual socket structure will
1236  * not be freed until the ref count reaches zero.
1237  */
1238 int
1239 soclose(struct socket *so)
1240 {
1241 	struct accept_queue lqueue;
1242 	int error = 0;
1243 	bool listening, last __diagused;
1244 
1245 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
1246 
1247 	CURVNET_SET(so->so_vnet);
1248 	funsetown(&so->so_sigio);
1249 	if (so->so_state & SS_ISCONNECTED) {
1250 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
1251 			error = sodisconnect(so);
1252 			if (error) {
1253 				if (error == ENOTCONN)
1254 					error = 0;
1255 				goto drop;
1256 			}
1257 		}
1258 
1259 		if ((so->so_options & SO_LINGER) != 0 && so->so_linger != 0) {
1260 			if ((so->so_state & SS_ISDISCONNECTING) &&
1261 			    (so->so_state & SS_NBIO))
1262 				goto drop;
1263 			while (so->so_state & SS_ISCONNECTED) {
1264 				error = tsleep(&so->so_timeo,
1265 				    PSOCK | PCATCH, "soclos",
1266 				    so->so_linger * hz);
1267 				if (error)
1268 					break;
1269 			}
1270 		}
1271 	}
1272 
1273 drop:
1274 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
1275 		(*so->so_proto->pr_usrreqs->pru_close)(so);
1276 
1277 	SOCK_LOCK(so);
1278 	if ((listening = SOLISTENING(so))) {
1279 		struct socket *sp;
1280 
1281 		TAILQ_INIT(&lqueue);
1282 		TAILQ_SWAP(&lqueue, &so->sol_incomp, socket, so_list);
1283 		TAILQ_CONCAT(&lqueue, &so->sol_comp, so_list);
1284 
1285 		so->sol_qlen = so->sol_incqlen = 0;
1286 
1287 		TAILQ_FOREACH(sp, &lqueue, so_list) {
1288 			SOCK_LOCK(sp);
1289 			sp->so_qstate = SQ_NONE;
1290 			sp->so_listen = NULL;
1291 			SOCK_UNLOCK(sp);
1292 			last = refcount_release(&so->so_count);
1293 			KASSERT(!last, ("%s: released last reference for %p",
1294 			    __func__, so));
1295 		}
1296 	}
1297 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
1298 	so->so_state |= SS_NOFDREF;
1299 	sorele_locked(so);
1300 	if (listening) {
1301 		struct socket *sp, *tsp;
1302 
1303 		TAILQ_FOREACH_SAFE(sp, &lqueue, so_list, tsp) {
1304 			SOCK_LOCK(sp);
1305 			if (refcount_load(&sp->so_count) == 0) {
1306 				SOCK_UNLOCK(sp);
1307 				soabort(sp);
1308 			} else {
1309 				/* See the handling of queued sockets
1310 				   in sofree(). */
1311 				SOCK_UNLOCK(sp);
1312 			}
1313 		}
1314 	}
1315 	CURVNET_RESTORE();
1316 	return (error);
1317 }
1318 
1319 /*
1320  * soabort() is used to abruptly tear down a connection, such as when a
1321  * resource limit is reached (listen queue depth exceeded), or if a listen
1322  * socket is closed while there are sockets waiting to be accepted.
1323  *
1324  * This interface is tricky, because it is called on an unreferenced socket,
1325  * and must be called only by a thread that has actually removed the socket
1326  * from the listen queue it was on, or races with other threads are risked.
1327  *
1328  * This interface will call into the protocol code, so must not be called
1329  * with any socket locks held.  Protocols do call it while holding their own
1330  * recursible protocol mutexes, but this is something that should be subject
1331  * to review in the future.
1332  */
1333 void
1334 soabort(struct socket *so)
1335 {
1336 
1337 	/*
1338 	 * In as much as is possible, assert that no references to this
1339 	 * socket are held.  This is not quite the same as asserting that the
1340 	 * current thread is responsible for arranging for no references, but
1341 	 * is as close as we can get for now.
1342 	 */
1343 	KASSERT(so->so_count == 0, ("soabort: so_count"));
1344 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
1345 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
1346 	VNET_SO_ASSERT(so);
1347 
1348 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
1349 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
1350 	SOCK_LOCK(so);
1351 	sofree(so);
1352 }
1353 
1354 int
1355 soaccept(struct socket *so, struct sockaddr **nam)
1356 {
1357 	int error;
1358 
1359 	SOCK_LOCK(so);
1360 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
1361 	so->so_state &= ~SS_NOFDREF;
1362 	SOCK_UNLOCK(so);
1363 
1364 	CURVNET_SET(so->so_vnet);
1365 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
1366 	CURVNET_RESTORE();
1367 	return (error);
1368 }
1369 
1370 int
1371 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
1372 {
1373 
1374 	return (soconnectat(AT_FDCWD, so, nam, td));
1375 }
1376 
1377 int
1378 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
1379 {
1380 	int error;
1381 
1382 	CURVNET_SET(so->so_vnet);
1383 	/*
1384 	 * If protocol is connection-based, can only connect once.
1385 	 * Otherwise, if connected, try to disconnect first.  This allows
1386 	 * user to disconnect by connecting to, e.g., a null address.
1387 	 */
1388 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
1389 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
1390 	    (error = sodisconnect(so)))) {
1391 		error = EISCONN;
1392 	} else {
1393 		/*
1394 		 * Prevent accumulated error from previous connection from
1395 		 * biting us.
1396 		 */
1397 		so->so_error = 0;
1398 		if (fd == AT_FDCWD) {
1399 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1400 			    nam, td);
1401 		} else {
1402 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1403 			    so, nam, td);
1404 		}
1405 	}
1406 	CURVNET_RESTORE();
1407 
1408 	return (error);
1409 }
1410 
1411 int
1412 soconnect2(struct socket *so1, struct socket *so2)
1413 {
1414 	int error;
1415 
1416 	CURVNET_SET(so1->so_vnet);
1417 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1418 	CURVNET_RESTORE();
1419 	return (error);
1420 }
1421 
1422 int
1423 sodisconnect(struct socket *so)
1424 {
1425 	int error;
1426 
1427 	if ((so->so_state & SS_ISCONNECTED) == 0)
1428 		return (ENOTCONN);
1429 	if (so->so_state & SS_ISDISCONNECTING)
1430 		return (EALREADY);
1431 	VNET_SO_ASSERT(so);
1432 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1433 	return (error);
1434 }
1435 
1436 int
1437 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1438     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1439 {
1440 	long space;
1441 	ssize_t resid;
1442 	int clen = 0, error, dontroute;
1443 
1444 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1445 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1446 	    ("sosend_dgram: !PR_ATOMIC"));
1447 
1448 	if (uio != NULL)
1449 		resid = uio->uio_resid;
1450 	else
1451 		resid = top->m_pkthdr.len;
1452 	/*
1453 	 * In theory resid should be unsigned.  However, space must be
1454 	 * signed, as it might be less than 0 if we over-committed, and we
1455 	 * must use a signed comparison of space and resid.  On the other
1456 	 * hand, a negative resid causes us to loop sending 0-length
1457 	 * segments to the protocol.
1458 	 */
1459 	if (resid < 0) {
1460 		error = EINVAL;
1461 		goto out;
1462 	}
1463 
1464 	dontroute =
1465 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1466 	if (td != NULL)
1467 		td->td_ru.ru_msgsnd++;
1468 	if (control != NULL)
1469 		clen = control->m_len;
1470 
1471 	SOCKBUF_LOCK(&so->so_snd);
1472 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1473 		SOCKBUF_UNLOCK(&so->so_snd);
1474 		error = EPIPE;
1475 		goto out;
1476 	}
1477 	if (so->so_error) {
1478 		error = so->so_error;
1479 		so->so_error = 0;
1480 		SOCKBUF_UNLOCK(&so->so_snd);
1481 		goto out;
1482 	}
1483 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1484 		/*
1485 		 * `sendto' and `sendmsg' is allowed on a connection-based
1486 		 * socket if it supports implied connect.  Return ENOTCONN if
1487 		 * not connected and no address is supplied.
1488 		 */
1489 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1490 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1491 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1492 			    !(resid == 0 && clen != 0)) {
1493 				SOCKBUF_UNLOCK(&so->so_snd);
1494 				error = ENOTCONN;
1495 				goto out;
1496 			}
1497 		} else if (addr == NULL) {
1498 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1499 				error = ENOTCONN;
1500 			else
1501 				error = EDESTADDRREQ;
1502 			SOCKBUF_UNLOCK(&so->so_snd);
1503 			goto out;
1504 		}
1505 	}
1506 
1507 	/*
1508 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1509 	 * problem and need fixing.
1510 	 */
1511 	space = sbspace(&so->so_snd);
1512 	if (flags & MSG_OOB)
1513 		space += 1024;
1514 	space -= clen;
1515 	SOCKBUF_UNLOCK(&so->so_snd);
1516 	if (resid > space) {
1517 		error = EMSGSIZE;
1518 		goto out;
1519 	}
1520 	if (uio == NULL) {
1521 		resid = 0;
1522 		if (flags & MSG_EOR)
1523 			top->m_flags |= M_EOR;
1524 	} else {
1525 		/*
1526 		 * Copy the data from userland into a mbuf chain.
1527 		 * If no data is to be copied in, a single empty mbuf
1528 		 * is returned.
1529 		 */
1530 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1531 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1532 		if (top == NULL) {
1533 			error = EFAULT;	/* only possible error */
1534 			goto out;
1535 		}
1536 		space -= resid - uio->uio_resid;
1537 		resid = uio->uio_resid;
1538 	}
1539 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1540 	/*
1541 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1542 	 * than with.
1543 	 */
1544 	if (dontroute) {
1545 		SOCK_LOCK(so);
1546 		so->so_options |= SO_DONTROUTE;
1547 		SOCK_UNLOCK(so);
1548 	}
1549 	/*
1550 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1551 	 * of date.  We could have received a reset packet in an interrupt or
1552 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1553 	 * probably recheck again inside the locking protection here, but
1554 	 * there are probably other places that this also happens.  We must
1555 	 * rethink this.
1556 	 */
1557 	VNET_SO_ASSERT(so);
1558 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1559 	    (flags & MSG_OOB) ? PRUS_OOB :
1560 	/*
1561 	 * If the user set MSG_EOF, the protocol understands this flag and
1562 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1563 	 */
1564 	    ((flags & MSG_EOF) &&
1565 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1566 	     (resid <= 0)) ?
1567 		PRUS_EOF :
1568 		/* If there is more to send set PRUS_MORETOCOME */
1569 		(flags & MSG_MORETOCOME) ||
1570 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1571 		top, addr, control, td);
1572 	if (dontroute) {
1573 		SOCK_LOCK(so);
1574 		so->so_options &= ~SO_DONTROUTE;
1575 		SOCK_UNLOCK(so);
1576 	}
1577 	clen = 0;
1578 	control = NULL;
1579 	top = NULL;
1580 out:
1581 	if (top != NULL)
1582 		m_freem(top);
1583 	if (control != NULL)
1584 		m_freem(control);
1585 	return (error);
1586 }
1587 
1588 /*
1589  * Send on a socket.  If send must go all at once and message is larger than
1590  * send buffering, then hard error.  Lock against other senders.  If must go
1591  * all at once and not enough room now, then inform user that this would
1592  * block and do nothing.  Otherwise, if nonblocking, send as much as
1593  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1594  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1595  * in mbuf chain must be small enough to send all at once.
1596  *
1597  * Returns nonzero on error, timeout or signal; callers must check for short
1598  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1599  * on return.
1600  */
1601 int
1602 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1603     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1604 {
1605 	long space;
1606 	ssize_t resid;
1607 	int clen = 0, error, dontroute;
1608 	int atomic = sosendallatonce(so) || top;
1609 	int pru_flag;
1610 #ifdef KERN_TLS
1611 	struct ktls_session *tls;
1612 	int tls_enq_cnt, tls_pruflag;
1613 	uint8_t tls_rtype;
1614 
1615 	tls = NULL;
1616 	tls_rtype = TLS_RLTYPE_APP;
1617 #endif
1618 	if (uio != NULL)
1619 		resid = uio->uio_resid;
1620 	else if ((top->m_flags & M_PKTHDR) != 0)
1621 		resid = top->m_pkthdr.len;
1622 	else
1623 		resid = m_length(top, NULL);
1624 	/*
1625 	 * In theory resid should be unsigned.  However, space must be
1626 	 * signed, as it might be less than 0 if we over-committed, and we
1627 	 * must use a signed comparison of space and resid.  On the other
1628 	 * hand, a negative resid causes us to loop sending 0-length
1629 	 * segments to the protocol.
1630 	 *
1631 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1632 	 * type sockets since that's an error.
1633 	 */
1634 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1635 		error = EINVAL;
1636 		goto out;
1637 	}
1638 
1639 	dontroute =
1640 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1641 	    (so->so_proto->pr_flags & PR_ATOMIC);
1642 	if (td != NULL)
1643 		td->td_ru.ru_msgsnd++;
1644 	if (control != NULL)
1645 		clen = control->m_len;
1646 
1647 	error = SOCK_IO_SEND_LOCK(so, SBLOCKWAIT(flags));
1648 	if (error)
1649 		goto out;
1650 
1651 #ifdef KERN_TLS
1652 	tls_pruflag = 0;
1653 	tls = ktls_hold(so->so_snd.sb_tls_info);
1654 	if (tls != NULL) {
1655 		if (tls->mode == TCP_TLS_MODE_SW)
1656 			tls_pruflag = PRUS_NOTREADY;
1657 
1658 		if (control != NULL) {
1659 			struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1660 
1661 			if (clen >= sizeof(*cm) &&
1662 			    cm->cmsg_type == TLS_SET_RECORD_TYPE) {
1663 				tls_rtype = *((uint8_t *)CMSG_DATA(cm));
1664 				clen = 0;
1665 				m_freem(control);
1666 				control = NULL;
1667 				atomic = 1;
1668 			}
1669 		}
1670 	}
1671 #endif
1672 
1673 restart:
1674 	do {
1675 		SOCKBUF_LOCK(&so->so_snd);
1676 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1677 			SOCKBUF_UNLOCK(&so->so_snd);
1678 			error = EPIPE;
1679 			goto release;
1680 		}
1681 		if (so->so_error) {
1682 			error = so->so_error;
1683 			so->so_error = 0;
1684 			SOCKBUF_UNLOCK(&so->so_snd);
1685 			goto release;
1686 		}
1687 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1688 			/*
1689 			 * `sendto' and `sendmsg' is allowed on a connection-
1690 			 * based socket if it supports implied connect.
1691 			 * Return ENOTCONN if not connected and no address is
1692 			 * supplied.
1693 			 */
1694 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1695 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1696 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1697 				    !(resid == 0 && clen != 0)) {
1698 					SOCKBUF_UNLOCK(&so->so_snd);
1699 					error = ENOTCONN;
1700 					goto release;
1701 				}
1702 			} else if (addr == NULL) {
1703 				SOCKBUF_UNLOCK(&so->so_snd);
1704 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1705 					error = ENOTCONN;
1706 				else
1707 					error = EDESTADDRREQ;
1708 				goto release;
1709 			}
1710 		}
1711 		space = sbspace(&so->so_snd);
1712 		if (flags & MSG_OOB)
1713 			space += 1024;
1714 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1715 		    clen > so->so_snd.sb_hiwat) {
1716 			SOCKBUF_UNLOCK(&so->so_snd);
1717 			error = EMSGSIZE;
1718 			goto release;
1719 		}
1720 		if (space < resid + clen &&
1721 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1722 			if ((so->so_state & SS_NBIO) ||
1723 			    (flags & (MSG_NBIO | MSG_DONTWAIT)) != 0) {
1724 				SOCKBUF_UNLOCK(&so->so_snd);
1725 				error = EWOULDBLOCK;
1726 				goto release;
1727 			}
1728 			error = sbwait(&so->so_snd);
1729 			SOCKBUF_UNLOCK(&so->so_snd);
1730 			if (error)
1731 				goto release;
1732 			goto restart;
1733 		}
1734 		SOCKBUF_UNLOCK(&so->so_snd);
1735 		space -= clen;
1736 		do {
1737 			if (uio == NULL) {
1738 				resid = 0;
1739 				if (flags & MSG_EOR)
1740 					top->m_flags |= M_EOR;
1741 #ifdef KERN_TLS
1742 				if (tls != NULL) {
1743 					ktls_frame(top, tls, &tls_enq_cnt,
1744 					    tls_rtype);
1745 					tls_rtype = TLS_RLTYPE_APP;
1746 				}
1747 #endif
1748 			} else {
1749 				/*
1750 				 * Copy the data from userland into a mbuf
1751 				 * chain.  If resid is 0, which can happen
1752 				 * only if we have control to send, then
1753 				 * a single empty mbuf is returned.  This
1754 				 * is a workaround to prevent protocol send
1755 				 * methods to panic.
1756 				 */
1757 #ifdef KERN_TLS
1758 				if (tls != NULL) {
1759 					top = m_uiotombuf(uio, M_WAITOK, space,
1760 					    tls->params.max_frame_len,
1761 					    M_EXTPG |
1762 					    ((flags & MSG_EOR) ? M_EOR : 0));
1763 					if (top != NULL) {
1764 						ktls_frame(top, tls,
1765 						    &tls_enq_cnt, tls_rtype);
1766 					}
1767 					tls_rtype = TLS_RLTYPE_APP;
1768 				} else
1769 #endif
1770 					top = m_uiotombuf(uio, M_WAITOK, space,
1771 					    (atomic ? max_hdr : 0),
1772 					    (atomic ? M_PKTHDR : 0) |
1773 					    ((flags & MSG_EOR) ? M_EOR : 0));
1774 				if (top == NULL) {
1775 					error = EFAULT; /* only possible error */
1776 					goto release;
1777 				}
1778 				space -= resid - uio->uio_resid;
1779 				resid = uio->uio_resid;
1780 			}
1781 			if (dontroute) {
1782 				SOCK_LOCK(so);
1783 				so->so_options |= SO_DONTROUTE;
1784 				SOCK_UNLOCK(so);
1785 			}
1786 			/*
1787 			 * XXX all the SBS_CANTSENDMORE checks previously
1788 			 * done could be out of date.  We could have received
1789 			 * a reset packet in an interrupt or maybe we slept
1790 			 * while doing page faults in uiomove() etc.  We
1791 			 * could probably recheck again inside the locking
1792 			 * protection here, but there are probably other
1793 			 * places that this also happens.  We must rethink
1794 			 * this.
1795 			 */
1796 			VNET_SO_ASSERT(so);
1797 
1798 			pru_flag = (flags & MSG_OOB) ? PRUS_OOB :
1799 			/*
1800 			 * If the user set MSG_EOF, the protocol understands
1801 			 * this flag and nothing left to send then use
1802 			 * PRU_SEND_EOF instead of PRU_SEND.
1803 			 */
1804 			    ((flags & MSG_EOF) &&
1805 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1806 			     (resid <= 0)) ?
1807 				PRUS_EOF :
1808 			/* If there is more to send set PRUS_MORETOCOME. */
1809 			    (flags & MSG_MORETOCOME) ||
1810 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0;
1811 
1812 #ifdef KERN_TLS
1813 			pru_flag |= tls_pruflag;
1814 #endif
1815 
1816 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1817 			    pru_flag, top, addr, control, td);
1818 
1819 			if (dontroute) {
1820 				SOCK_LOCK(so);
1821 				so->so_options &= ~SO_DONTROUTE;
1822 				SOCK_UNLOCK(so);
1823 			}
1824 
1825 #ifdef KERN_TLS
1826 			if (tls != NULL && tls->mode == TCP_TLS_MODE_SW) {
1827 				if (error != 0) {
1828 					m_freem(top);
1829 					top = NULL;
1830 				} else {
1831 					soref(so);
1832 					ktls_enqueue(top, so, tls_enq_cnt);
1833 				}
1834 			}
1835 #endif
1836 			clen = 0;
1837 			control = NULL;
1838 			top = NULL;
1839 			if (error)
1840 				goto release;
1841 		} while (resid && space > 0);
1842 	} while (resid);
1843 
1844 release:
1845 	SOCK_IO_SEND_UNLOCK(so);
1846 out:
1847 #ifdef KERN_TLS
1848 	if (tls != NULL)
1849 		ktls_free(tls);
1850 #endif
1851 	if (top != NULL)
1852 		m_freem(top);
1853 	if (control != NULL)
1854 		m_freem(control);
1855 	return (error);
1856 }
1857 
1858 int
1859 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1860     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1861 {
1862 	int error;
1863 
1864 	CURVNET_SET(so->so_vnet);
1865 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio,
1866 	    top, control, flags, td);
1867 	CURVNET_RESTORE();
1868 	return (error);
1869 }
1870 
1871 /*
1872  * The part of soreceive() that implements reading non-inline out-of-band
1873  * data from a socket.  For more complete comments, see soreceive(), from
1874  * which this code originated.
1875  *
1876  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1877  * unable to return an mbuf chain to the caller.
1878  */
1879 static int
1880 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1881 {
1882 	struct protosw *pr = so->so_proto;
1883 	struct mbuf *m;
1884 	int error;
1885 
1886 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1887 	VNET_SO_ASSERT(so);
1888 
1889 	m = m_get(M_WAITOK, MT_DATA);
1890 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1891 	if (error)
1892 		goto bad;
1893 	do {
1894 		error = uiomove(mtod(m, void *),
1895 		    (int) min(uio->uio_resid, m->m_len), uio);
1896 		m = m_free(m);
1897 	} while (uio->uio_resid && error == 0 && m);
1898 bad:
1899 	if (m != NULL)
1900 		m_freem(m);
1901 	return (error);
1902 }
1903 
1904 /*
1905  * Following replacement or removal of the first mbuf on the first mbuf chain
1906  * of a socket buffer, push necessary state changes back into the socket
1907  * buffer so that other consumers see the values consistently.  'nextrecord'
1908  * is the callers locally stored value of the original value of
1909  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1910  * NOTE: 'nextrecord' may be NULL.
1911  */
1912 static __inline void
1913 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1914 {
1915 
1916 	SOCKBUF_LOCK_ASSERT(sb);
1917 	/*
1918 	 * First, update for the new value of nextrecord.  If necessary, make
1919 	 * it the first record.
1920 	 */
1921 	if (sb->sb_mb != NULL)
1922 		sb->sb_mb->m_nextpkt = nextrecord;
1923 	else
1924 		sb->sb_mb = nextrecord;
1925 
1926 	/*
1927 	 * Now update any dependent socket buffer fields to reflect the new
1928 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1929 	 * addition of a second clause that takes care of the case where
1930 	 * sb_mb has been updated, but remains the last record.
1931 	 */
1932 	if (sb->sb_mb == NULL) {
1933 		sb->sb_mbtail = NULL;
1934 		sb->sb_lastrecord = NULL;
1935 	} else if (sb->sb_mb->m_nextpkt == NULL)
1936 		sb->sb_lastrecord = sb->sb_mb;
1937 }
1938 
1939 /*
1940  * Implement receive operations on a socket.  We depend on the way that
1941  * records are added to the sockbuf by sbappend.  In particular, each record
1942  * (mbufs linked through m_next) must begin with an address if the protocol
1943  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1944  * data, and then zero or more mbufs of data.  In order to allow parallelism
1945  * between network receive and copying to user space, as well as avoid
1946  * sleeping with a mutex held, we release the socket buffer mutex during the
1947  * user space copy.  Although the sockbuf is locked, new data may still be
1948  * appended, and thus we must maintain consistency of the sockbuf during that
1949  * time.
1950  *
1951  * The caller may receive the data as a single mbuf chain by supplying an
1952  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1953  * the count in uio_resid.
1954  */
1955 int
1956 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1957     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1958 {
1959 	struct mbuf *m, **mp;
1960 	int flags, error, offset;
1961 	ssize_t len;
1962 	struct protosw *pr = so->so_proto;
1963 	struct mbuf *nextrecord;
1964 	int moff, type = 0;
1965 	ssize_t orig_resid = uio->uio_resid;
1966 
1967 	mp = mp0;
1968 	if (psa != NULL)
1969 		*psa = NULL;
1970 	if (controlp != NULL)
1971 		*controlp = NULL;
1972 	if (flagsp != NULL)
1973 		flags = *flagsp &~ MSG_EOR;
1974 	else
1975 		flags = 0;
1976 	if (flags & MSG_OOB)
1977 		return (soreceive_rcvoob(so, uio, flags));
1978 	if (mp != NULL)
1979 		*mp = NULL;
1980 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1981 	    && uio->uio_resid) {
1982 		VNET_SO_ASSERT(so);
1983 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1984 	}
1985 
1986 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
1987 	if (error)
1988 		return (error);
1989 
1990 restart:
1991 	SOCKBUF_LOCK(&so->so_rcv);
1992 	m = so->so_rcv.sb_mb;
1993 	/*
1994 	 * If we have less data than requested, block awaiting more (subject
1995 	 * to any timeout) if:
1996 	 *   1. the current count is less than the low water mark, or
1997 	 *   2. MSG_DONTWAIT is not set
1998 	 */
1999 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
2000 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
2001 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
2002 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
2003 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
2004 		    ("receive: m == %p sbavail == %u",
2005 		    m, sbavail(&so->so_rcv)));
2006 		if (so->so_error || so->so_rerror) {
2007 			if (m != NULL)
2008 				goto dontblock;
2009 			if (so->so_error)
2010 				error = so->so_error;
2011 			else
2012 				error = so->so_rerror;
2013 			if ((flags & MSG_PEEK) == 0) {
2014 				if (so->so_error)
2015 					so->so_error = 0;
2016 				else
2017 					so->so_rerror = 0;
2018 			}
2019 			SOCKBUF_UNLOCK(&so->so_rcv);
2020 			goto release;
2021 		}
2022 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2023 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2024 			if (m != NULL)
2025 				goto dontblock;
2026 #ifdef KERN_TLS
2027 			else if (so->so_rcv.sb_tlsdcc == 0 &&
2028 			    so->so_rcv.sb_tlscc == 0) {
2029 #else
2030 			else {
2031 #endif
2032 				SOCKBUF_UNLOCK(&so->so_rcv);
2033 				goto release;
2034 			}
2035 		}
2036 		for (; m != NULL; m = m->m_next)
2037 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
2038 				m = so->so_rcv.sb_mb;
2039 				goto dontblock;
2040 			}
2041 		if ((so->so_state & (SS_ISCONNECTING | SS_ISCONNECTED |
2042 		    SS_ISDISCONNECTING | SS_ISDISCONNECTED)) == 0 &&
2043 		    (so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
2044 			SOCKBUF_UNLOCK(&so->so_rcv);
2045 			error = ENOTCONN;
2046 			goto release;
2047 		}
2048 		if (uio->uio_resid == 0) {
2049 			SOCKBUF_UNLOCK(&so->so_rcv);
2050 			goto release;
2051 		}
2052 		if ((so->so_state & SS_NBIO) ||
2053 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2054 			SOCKBUF_UNLOCK(&so->so_rcv);
2055 			error = EWOULDBLOCK;
2056 			goto release;
2057 		}
2058 		SBLASTRECORDCHK(&so->so_rcv);
2059 		SBLASTMBUFCHK(&so->so_rcv);
2060 		error = sbwait(&so->so_rcv);
2061 		SOCKBUF_UNLOCK(&so->so_rcv);
2062 		if (error)
2063 			goto release;
2064 		goto restart;
2065 	}
2066 dontblock:
2067 	/*
2068 	 * From this point onward, we maintain 'nextrecord' as a cache of the
2069 	 * pointer to the next record in the socket buffer.  We must keep the
2070 	 * various socket buffer pointers and local stack versions of the
2071 	 * pointers in sync, pushing out modifications before dropping the
2072 	 * socket buffer mutex, and re-reading them when picking it up.
2073 	 *
2074 	 * Otherwise, we will race with the network stack appending new data
2075 	 * or records onto the socket buffer by using inconsistent/stale
2076 	 * versions of the field, possibly resulting in socket buffer
2077 	 * corruption.
2078 	 *
2079 	 * By holding the high-level sblock(), we prevent simultaneous
2080 	 * readers from pulling off the front of the socket buffer.
2081 	 */
2082 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2083 	if (uio->uio_td)
2084 		uio->uio_td->td_ru.ru_msgrcv++;
2085 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
2086 	SBLASTRECORDCHK(&so->so_rcv);
2087 	SBLASTMBUFCHK(&so->so_rcv);
2088 	nextrecord = m->m_nextpkt;
2089 	if (pr->pr_flags & PR_ADDR) {
2090 		KASSERT(m->m_type == MT_SONAME,
2091 		    ("m->m_type == %d", m->m_type));
2092 		orig_resid = 0;
2093 		if (psa != NULL)
2094 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2095 			    M_NOWAIT);
2096 		if (flags & MSG_PEEK) {
2097 			m = m->m_next;
2098 		} else {
2099 			sbfree(&so->so_rcv, m);
2100 			so->so_rcv.sb_mb = m_free(m);
2101 			m = so->so_rcv.sb_mb;
2102 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2103 		}
2104 	}
2105 
2106 	/*
2107 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2108 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
2109 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
2110 	 * perform externalization (or freeing if controlp == NULL).
2111 	 */
2112 	if (m != NULL && m->m_type == MT_CONTROL) {
2113 		struct mbuf *cm = NULL, *cmn;
2114 		struct mbuf **cme = &cm;
2115 #ifdef KERN_TLS
2116 		struct cmsghdr *cmsg;
2117 		struct tls_get_record tgr;
2118 
2119 		/*
2120 		 * For MSG_TLSAPPDATA, check for a non-application data
2121 		 * record.  If found, return ENXIO without removing
2122 		 * it from the receive queue.  This allows a subsequent
2123 		 * call without MSG_TLSAPPDATA to receive it.
2124 		 * Note that, for TLS, there should only be a single
2125 		 * control mbuf with the TLS_GET_RECORD message in it.
2126 		 */
2127 		if (flags & MSG_TLSAPPDATA) {
2128 			cmsg = mtod(m, struct cmsghdr *);
2129 			if (cmsg->cmsg_type == TLS_GET_RECORD &&
2130 			    cmsg->cmsg_len == CMSG_LEN(sizeof(tgr))) {
2131 				memcpy(&tgr, CMSG_DATA(cmsg), sizeof(tgr));
2132 				/* This will need to change for TLS 1.3. */
2133 				if (tgr.tls_type != TLS_RLTYPE_APP) {
2134 					SOCKBUF_UNLOCK(&so->so_rcv);
2135 					error = ENXIO;
2136 					goto release;
2137 				}
2138 			}
2139 		}
2140 #endif
2141 
2142 		do {
2143 			if (flags & MSG_PEEK) {
2144 				if (controlp != NULL) {
2145 					*controlp = m_copym(m, 0, m->m_len,
2146 					    M_NOWAIT);
2147 					controlp = &(*controlp)->m_next;
2148 				}
2149 				m = m->m_next;
2150 			} else {
2151 				sbfree(&so->so_rcv, m);
2152 				so->so_rcv.sb_mb = m->m_next;
2153 				m->m_next = NULL;
2154 				*cme = m;
2155 				cme = &(*cme)->m_next;
2156 				m = so->so_rcv.sb_mb;
2157 			}
2158 		} while (m != NULL && m->m_type == MT_CONTROL);
2159 		if ((flags & MSG_PEEK) == 0)
2160 			sockbuf_pushsync(&so->so_rcv, nextrecord);
2161 		while (cm != NULL) {
2162 			cmn = cm->m_next;
2163 			cm->m_next = NULL;
2164 			if (pr->pr_domain->dom_externalize != NULL) {
2165 				SOCKBUF_UNLOCK(&so->so_rcv);
2166 				VNET_SO_ASSERT(so);
2167 				error = (*pr->pr_domain->dom_externalize)
2168 				    (cm, controlp, flags);
2169 				SOCKBUF_LOCK(&so->so_rcv);
2170 			} else if (controlp != NULL)
2171 				*controlp = cm;
2172 			else
2173 				m_freem(cm);
2174 			if (controlp != NULL) {
2175 				while (*controlp != NULL)
2176 					controlp = &(*controlp)->m_next;
2177 			}
2178 			cm = cmn;
2179 		}
2180 		if (m != NULL)
2181 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
2182 		else
2183 			nextrecord = so->so_rcv.sb_mb;
2184 		orig_resid = 0;
2185 	}
2186 	if (m != NULL) {
2187 		if ((flags & MSG_PEEK) == 0) {
2188 			KASSERT(m->m_nextpkt == nextrecord,
2189 			    ("soreceive: post-control, nextrecord !sync"));
2190 			if (nextrecord == NULL) {
2191 				KASSERT(so->so_rcv.sb_mb == m,
2192 				    ("soreceive: post-control, sb_mb!=m"));
2193 				KASSERT(so->so_rcv.sb_lastrecord == m,
2194 				    ("soreceive: post-control, lastrecord!=m"));
2195 			}
2196 		}
2197 		type = m->m_type;
2198 		if (type == MT_OOBDATA)
2199 			flags |= MSG_OOB;
2200 	} else {
2201 		if ((flags & MSG_PEEK) == 0) {
2202 			KASSERT(so->so_rcv.sb_mb == nextrecord,
2203 			    ("soreceive: sb_mb != nextrecord"));
2204 			if (so->so_rcv.sb_mb == NULL) {
2205 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
2206 				    ("soreceive: sb_lastercord != NULL"));
2207 			}
2208 		}
2209 	}
2210 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2211 	SBLASTRECORDCHK(&so->so_rcv);
2212 	SBLASTMBUFCHK(&so->so_rcv);
2213 
2214 	/*
2215 	 * Now continue to read any data mbufs off of the head of the socket
2216 	 * buffer until the read request is satisfied.  Note that 'type' is
2217 	 * used to store the type of any mbuf reads that have happened so far
2218 	 * such that soreceive() can stop reading if the type changes, which
2219 	 * causes soreceive() to return only one of regular data and inline
2220 	 * out-of-band data in a single socket receive operation.
2221 	 */
2222 	moff = 0;
2223 	offset = 0;
2224 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
2225 	    && error == 0) {
2226 		/*
2227 		 * If the type of mbuf has changed since the last mbuf
2228 		 * examined ('type'), end the receive operation.
2229 		 */
2230 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2231 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
2232 			if (type != m->m_type)
2233 				break;
2234 		} else if (type == MT_OOBDATA)
2235 			break;
2236 		else
2237 		    KASSERT(m->m_type == MT_DATA,
2238 			("m->m_type == %d", m->m_type));
2239 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
2240 		len = uio->uio_resid;
2241 		if (so->so_oobmark && len > so->so_oobmark - offset)
2242 			len = so->so_oobmark - offset;
2243 		if (len > m->m_len - moff)
2244 			len = m->m_len - moff;
2245 		/*
2246 		 * If mp is set, just pass back the mbufs.  Otherwise copy
2247 		 * them out via the uio, then free.  Sockbuf must be
2248 		 * consistent here (points to current mbuf, it points to next
2249 		 * record) when we drop priority; we must note any additions
2250 		 * to the sockbuf when we block interrupts again.
2251 		 */
2252 		if (mp == NULL) {
2253 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2254 			SBLASTRECORDCHK(&so->so_rcv);
2255 			SBLASTMBUFCHK(&so->so_rcv);
2256 			SOCKBUF_UNLOCK(&so->so_rcv);
2257 			if ((m->m_flags & M_EXTPG) != 0)
2258 				error = m_unmapped_uiomove(m, moff, uio,
2259 				    (int)len);
2260 			else
2261 				error = uiomove(mtod(m, char *) + moff,
2262 				    (int)len, uio);
2263 			SOCKBUF_LOCK(&so->so_rcv);
2264 			if (error) {
2265 				/*
2266 				 * The MT_SONAME mbuf has already been removed
2267 				 * from the record, so it is necessary to
2268 				 * remove the data mbufs, if any, to preserve
2269 				 * the invariant in the case of PR_ADDR that
2270 				 * requires MT_SONAME mbufs at the head of
2271 				 * each record.
2272 				 */
2273 				if (pr->pr_flags & PR_ATOMIC &&
2274 				    ((flags & MSG_PEEK) == 0))
2275 					(void)sbdroprecord_locked(&so->so_rcv);
2276 				SOCKBUF_UNLOCK(&so->so_rcv);
2277 				goto release;
2278 			}
2279 		} else
2280 			uio->uio_resid -= len;
2281 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2282 		if (len == m->m_len - moff) {
2283 			if (m->m_flags & M_EOR)
2284 				flags |= MSG_EOR;
2285 			if (flags & MSG_PEEK) {
2286 				m = m->m_next;
2287 				moff = 0;
2288 			} else {
2289 				nextrecord = m->m_nextpkt;
2290 				sbfree(&so->so_rcv, m);
2291 				if (mp != NULL) {
2292 					m->m_nextpkt = NULL;
2293 					*mp = m;
2294 					mp = &m->m_next;
2295 					so->so_rcv.sb_mb = m = m->m_next;
2296 					*mp = NULL;
2297 				} else {
2298 					so->so_rcv.sb_mb = m_free(m);
2299 					m = so->so_rcv.sb_mb;
2300 				}
2301 				sockbuf_pushsync(&so->so_rcv, nextrecord);
2302 				SBLASTRECORDCHK(&so->so_rcv);
2303 				SBLASTMBUFCHK(&so->so_rcv);
2304 			}
2305 		} else {
2306 			if (flags & MSG_PEEK)
2307 				moff += len;
2308 			else {
2309 				if (mp != NULL) {
2310 					if (flags & MSG_DONTWAIT) {
2311 						*mp = m_copym(m, 0, len,
2312 						    M_NOWAIT);
2313 						if (*mp == NULL) {
2314 							/*
2315 							 * m_copym() couldn't
2316 							 * allocate an mbuf.
2317 							 * Adjust uio_resid back
2318 							 * (it was adjusted
2319 							 * down by len bytes,
2320 							 * which we didn't end
2321 							 * up "copying" over).
2322 							 */
2323 							uio->uio_resid += len;
2324 							break;
2325 						}
2326 					} else {
2327 						SOCKBUF_UNLOCK(&so->so_rcv);
2328 						*mp = m_copym(m, 0, len,
2329 						    M_WAITOK);
2330 						SOCKBUF_LOCK(&so->so_rcv);
2331 					}
2332 				}
2333 				sbcut_locked(&so->so_rcv, len);
2334 			}
2335 		}
2336 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2337 		if (so->so_oobmark) {
2338 			if ((flags & MSG_PEEK) == 0) {
2339 				so->so_oobmark -= len;
2340 				if (so->so_oobmark == 0) {
2341 					so->so_rcv.sb_state |= SBS_RCVATMARK;
2342 					break;
2343 				}
2344 			} else {
2345 				offset += len;
2346 				if (offset == so->so_oobmark)
2347 					break;
2348 			}
2349 		}
2350 		if (flags & MSG_EOR)
2351 			break;
2352 		/*
2353 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
2354 		 * must not quit until "uio->uio_resid == 0" or an error
2355 		 * termination.  If a signal/timeout occurs, return with a
2356 		 * short count but without error.  Keep sockbuf locked
2357 		 * against other readers.
2358 		 */
2359 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
2360 		    !sosendallatonce(so) && nextrecord == NULL) {
2361 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2362 			if (so->so_error || so->so_rerror ||
2363 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
2364 				break;
2365 			/*
2366 			 * Notify the protocol that some data has been
2367 			 * drained before blocking.
2368 			 */
2369 			if (pr->pr_flags & PR_WANTRCVD) {
2370 				SOCKBUF_UNLOCK(&so->so_rcv);
2371 				VNET_SO_ASSERT(so);
2372 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2373 				SOCKBUF_LOCK(&so->so_rcv);
2374 			}
2375 			SBLASTRECORDCHK(&so->so_rcv);
2376 			SBLASTMBUFCHK(&so->so_rcv);
2377 			/*
2378 			 * We could receive some data while was notifying
2379 			 * the protocol. Skip blocking in this case.
2380 			 */
2381 			if (so->so_rcv.sb_mb == NULL) {
2382 				error = sbwait(&so->so_rcv);
2383 				if (error) {
2384 					SOCKBUF_UNLOCK(&so->so_rcv);
2385 					goto release;
2386 				}
2387 			}
2388 			m = so->so_rcv.sb_mb;
2389 			if (m != NULL)
2390 				nextrecord = m->m_nextpkt;
2391 		}
2392 	}
2393 
2394 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2395 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
2396 		flags |= MSG_TRUNC;
2397 		if ((flags & MSG_PEEK) == 0)
2398 			(void) sbdroprecord_locked(&so->so_rcv);
2399 	}
2400 	if ((flags & MSG_PEEK) == 0) {
2401 		if (m == NULL) {
2402 			/*
2403 			 * First part is an inline SB_EMPTY_FIXUP().  Second
2404 			 * part makes sure sb_lastrecord is up-to-date if
2405 			 * there is still data in the socket buffer.
2406 			 */
2407 			so->so_rcv.sb_mb = nextrecord;
2408 			if (so->so_rcv.sb_mb == NULL) {
2409 				so->so_rcv.sb_mbtail = NULL;
2410 				so->so_rcv.sb_lastrecord = NULL;
2411 			} else if (nextrecord->m_nextpkt == NULL)
2412 				so->so_rcv.sb_lastrecord = nextrecord;
2413 		}
2414 		SBLASTRECORDCHK(&so->so_rcv);
2415 		SBLASTMBUFCHK(&so->so_rcv);
2416 		/*
2417 		 * If soreceive() is being done from the socket callback,
2418 		 * then don't need to generate ACK to peer to update window,
2419 		 * since ACK will be generated on return to TCP.
2420 		 */
2421 		if (!(flags & MSG_SOCALLBCK) &&
2422 		    (pr->pr_flags & PR_WANTRCVD)) {
2423 			SOCKBUF_UNLOCK(&so->so_rcv);
2424 			VNET_SO_ASSERT(so);
2425 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
2426 			SOCKBUF_LOCK(&so->so_rcv);
2427 		}
2428 	}
2429 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2430 	if (orig_resid == uio->uio_resid && orig_resid &&
2431 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
2432 		SOCKBUF_UNLOCK(&so->so_rcv);
2433 		goto restart;
2434 	}
2435 	SOCKBUF_UNLOCK(&so->so_rcv);
2436 
2437 	if (flagsp != NULL)
2438 		*flagsp |= flags;
2439 release:
2440 	SOCK_IO_RECV_UNLOCK(so);
2441 	return (error);
2442 }
2443 
2444 /*
2445  * Optimized version of soreceive() for stream (TCP) sockets.
2446  */
2447 int
2448 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
2449     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2450 {
2451 	int len = 0, error = 0, flags, oresid;
2452 	struct sockbuf *sb;
2453 	struct mbuf *m, *n = NULL;
2454 
2455 	/* We only do stream sockets. */
2456 	if (so->so_type != SOCK_STREAM)
2457 		return (EINVAL);
2458 	if (psa != NULL)
2459 		*psa = NULL;
2460 	if (flagsp != NULL)
2461 		flags = *flagsp &~ MSG_EOR;
2462 	else
2463 		flags = 0;
2464 	if (controlp != NULL)
2465 		*controlp = NULL;
2466 	if (flags & MSG_OOB)
2467 		return (soreceive_rcvoob(so, uio, flags));
2468 	if (mp0 != NULL)
2469 		*mp0 = NULL;
2470 
2471 	sb = &so->so_rcv;
2472 
2473 #ifdef KERN_TLS
2474 	/*
2475 	 * KTLS store TLS records as records with a control message to
2476 	 * describe the framing.
2477 	 *
2478 	 * We check once here before acquiring locks to optimize the
2479 	 * common case.
2480 	 */
2481 	if (sb->sb_tls_info != NULL)
2482 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2483 		    flagsp));
2484 #endif
2485 
2486 	/* Prevent other readers from entering the socket. */
2487 	error = SOCK_IO_RECV_LOCK(so, SBLOCKWAIT(flags));
2488 	if (error)
2489 		return (error);
2490 	SOCKBUF_LOCK(sb);
2491 
2492 #ifdef KERN_TLS
2493 	if (sb->sb_tls_info != NULL) {
2494 		SOCKBUF_UNLOCK(sb);
2495 		SOCK_IO_RECV_UNLOCK(so);
2496 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2497 		    flagsp));
2498 	}
2499 #endif
2500 
2501 	/* Easy one, no space to copyout anything. */
2502 	if (uio->uio_resid == 0) {
2503 		error = EINVAL;
2504 		goto out;
2505 	}
2506 	oresid = uio->uio_resid;
2507 
2508 	/* We will never ever get anything unless we are or were connected. */
2509 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
2510 		error = ENOTCONN;
2511 		goto out;
2512 	}
2513 
2514 restart:
2515 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2516 
2517 	/* Abort if socket has reported problems. */
2518 	if (so->so_error) {
2519 		if (sbavail(sb) > 0)
2520 			goto deliver;
2521 		if (oresid > uio->uio_resid)
2522 			goto out;
2523 		error = so->so_error;
2524 		if (!(flags & MSG_PEEK))
2525 			so->so_error = 0;
2526 		goto out;
2527 	}
2528 
2529 	/* Door is closed.  Deliver what is left, if any. */
2530 	if (sb->sb_state & SBS_CANTRCVMORE) {
2531 		if (sbavail(sb) > 0)
2532 			goto deliver;
2533 		else
2534 			goto out;
2535 	}
2536 
2537 	/* Socket buffer is empty and we shall not block. */
2538 	if (sbavail(sb) == 0 &&
2539 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
2540 		error = EAGAIN;
2541 		goto out;
2542 	}
2543 
2544 	/* Socket buffer got some data that we shall deliver now. */
2545 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2546 	    ((so->so_state & SS_NBIO) ||
2547 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2548 	     sbavail(sb) >= sb->sb_lowat ||
2549 	     sbavail(sb) >= uio->uio_resid ||
2550 	     sbavail(sb) >= sb->sb_hiwat) ) {
2551 		goto deliver;
2552 	}
2553 
2554 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2555 	if ((flags & MSG_WAITALL) &&
2556 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2557 		goto deliver;
2558 
2559 	/*
2560 	 * Wait and block until (more) data comes in.
2561 	 * NB: Drops the sockbuf lock during wait.
2562 	 */
2563 	error = sbwait(sb);
2564 	if (error)
2565 		goto out;
2566 	goto restart;
2567 
2568 deliver:
2569 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2570 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2571 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2572 
2573 	/* Statistics. */
2574 	if (uio->uio_td)
2575 		uio->uio_td->td_ru.ru_msgrcv++;
2576 
2577 	/* Fill uio until full or current end of socket buffer is reached. */
2578 	len = min(uio->uio_resid, sbavail(sb));
2579 	if (mp0 != NULL) {
2580 		/* Dequeue as many mbufs as possible. */
2581 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2582 			if (*mp0 == NULL)
2583 				*mp0 = sb->sb_mb;
2584 			else
2585 				m_cat(*mp0, sb->sb_mb);
2586 			for (m = sb->sb_mb;
2587 			     m != NULL && m->m_len <= len;
2588 			     m = m->m_next) {
2589 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2590 				    ("%s: m %p not available", __func__, m));
2591 				len -= m->m_len;
2592 				uio->uio_resid -= m->m_len;
2593 				sbfree(sb, m);
2594 				n = m;
2595 			}
2596 			n->m_next = NULL;
2597 			sb->sb_mb = m;
2598 			sb->sb_lastrecord = sb->sb_mb;
2599 			if (sb->sb_mb == NULL)
2600 				SB_EMPTY_FIXUP(sb);
2601 		}
2602 		/* Copy the remainder. */
2603 		if (len > 0) {
2604 			KASSERT(sb->sb_mb != NULL,
2605 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2606 
2607 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2608 			if (m == NULL)
2609 				len = 0;	/* Don't flush data from sockbuf. */
2610 			else
2611 				uio->uio_resid -= len;
2612 			if (*mp0 != NULL)
2613 				m_cat(*mp0, m);
2614 			else
2615 				*mp0 = m;
2616 			if (*mp0 == NULL) {
2617 				error = ENOBUFS;
2618 				goto out;
2619 			}
2620 		}
2621 	} else {
2622 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2623 		SOCKBUF_UNLOCK(sb);
2624 		error = m_mbuftouio(uio, sb->sb_mb, len);
2625 		SOCKBUF_LOCK(sb);
2626 		if (error)
2627 			goto out;
2628 	}
2629 	SBLASTRECORDCHK(sb);
2630 	SBLASTMBUFCHK(sb);
2631 
2632 	/*
2633 	 * Remove the delivered data from the socket buffer unless we
2634 	 * were only peeking.
2635 	 */
2636 	if (!(flags & MSG_PEEK)) {
2637 		if (len > 0)
2638 			sbdrop_locked(sb, len);
2639 
2640 		/* Notify protocol that we drained some data. */
2641 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2642 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2643 		     !(flags & MSG_SOCALLBCK))) {
2644 			SOCKBUF_UNLOCK(sb);
2645 			VNET_SO_ASSERT(so);
2646 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2647 			SOCKBUF_LOCK(sb);
2648 		}
2649 	}
2650 
2651 	/*
2652 	 * For MSG_WAITALL we may have to loop again and wait for
2653 	 * more data to come in.
2654 	 */
2655 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2656 		goto restart;
2657 out:
2658 	SBLASTRECORDCHK(sb);
2659 	SBLASTMBUFCHK(sb);
2660 	SOCKBUF_UNLOCK(sb);
2661 	SOCK_IO_RECV_UNLOCK(so);
2662 	return (error);
2663 }
2664 
2665 /*
2666  * Optimized version of soreceive() for simple datagram cases from userspace.
2667  * Unlike in the stream case, we're able to drop a datagram if copyout()
2668  * fails, and because we handle datagrams atomically, we don't need to use a
2669  * sleep lock to prevent I/O interlacing.
2670  */
2671 int
2672 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2673     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2674 {
2675 	struct mbuf *m, *m2;
2676 	int flags, error;
2677 	ssize_t len;
2678 	struct protosw *pr = so->so_proto;
2679 	struct mbuf *nextrecord;
2680 
2681 	if (psa != NULL)
2682 		*psa = NULL;
2683 	if (controlp != NULL)
2684 		*controlp = NULL;
2685 	if (flagsp != NULL)
2686 		flags = *flagsp &~ MSG_EOR;
2687 	else
2688 		flags = 0;
2689 
2690 	/*
2691 	 * For any complicated cases, fall back to the full
2692 	 * soreceive_generic().
2693 	 */
2694 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2695 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2696 		    flagsp));
2697 
2698 	/*
2699 	 * Enforce restrictions on use.
2700 	 */
2701 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2702 	    ("soreceive_dgram: wantrcvd"));
2703 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2704 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2705 	    ("soreceive_dgram: SBS_RCVATMARK"));
2706 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2707 	    ("soreceive_dgram: P_CONNREQUIRED"));
2708 
2709 	/*
2710 	 * Loop blocking while waiting for a datagram.
2711 	 */
2712 	SOCKBUF_LOCK(&so->so_rcv);
2713 	while ((m = so->so_rcv.sb_mb) == NULL) {
2714 		KASSERT(sbavail(&so->so_rcv) == 0,
2715 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2716 		    sbavail(&so->so_rcv)));
2717 		if (so->so_error) {
2718 			error = so->so_error;
2719 			so->so_error = 0;
2720 			SOCKBUF_UNLOCK(&so->so_rcv);
2721 			return (error);
2722 		}
2723 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2724 		    uio->uio_resid == 0) {
2725 			SOCKBUF_UNLOCK(&so->so_rcv);
2726 			return (0);
2727 		}
2728 		if ((so->so_state & SS_NBIO) ||
2729 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2730 			SOCKBUF_UNLOCK(&so->so_rcv);
2731 			return (EWOULDBLOCK);
2732 		}
2733 		SBLASTRECORDCHK(&so->so_rcv);
2734 		SBLASTMBUFCHK(&so->so_rcv);
2735 		error = sbwait(&so->so_rcv);
2736 		if (error) {
2737 			SOCKBUF_UNLOCK(&so->so_rcv);
2738 			return (error);
2739 		}
2740 	}
2741 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2742 
2743 	if (uio->uio_td)
2744 		uio->uio_td->td_ru.ru_msgrcv++;
2745 	SBLASTRECORDCHK(&so->so_rcv);
2746 	SBLASTMBUFCHK(&so->so_rcv);
2747 	nextrecord = m->m_nextpkt;
2748 	if (nextrecord == NULL) {
2749 		KASSERT(so->so_rcv.sb_lastrecord == m,
2750 		    ("soreceive_dgram: lastrecord != m"));
2751 	}
2752 
2753 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2754 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2755 
2756 	/*
2757 	 * Pull 'm' and its chain off the front of the packet queue.
2758 	 */
2759 	so->so_rcv.sb_mb = NULL;
2760 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2761 
2762 	/*
2763 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2764 	 */
2765 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2766 		sbfree(&so->so_rcv, m2);
2767 
2768 	/*
2769 	 * Do a few last checks before we let go of the lock.
2770 	 */
2771 	SBLASTRECORDCHK(&so->so_rcv);
2772 	SBLASTMBUFCHK(&so->so_rcv);
2773 	SOCKBUF_UNLOCK(&so->so_rcv);
2774 
2775 	if (pr->pr_flags & PR_ADDR) {
2776 		KASSERT(m->m_type == MT_SONAME,
2777 		    ("m->m_type == %d", m->m_type));
2778 		if (psa != NULL)
2779 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2780 			    M_NOWAIT);
2781 		m = m_free(m);
2782 	}
2783 	if (m == NULL) {
2784 		/* XXXRW: Can this happen? */
2785 		return (0);
2786 	}
2787 
2788 	/*
2789 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2790 	 * queue.
2791 	 *
2792 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2793 	 * in the first mbuf chain on the socket buffer.  We call into the
2794 	 * protocol to perform externalization (or freeing if controlp ==
2795 	 * NULL). In some cases there can be only MT_CONTROL mbufs without
2796 	 * MT_DATA mbufs.
2797 	 */
2798 	if (m->m_type == MT_CONTROL) {
2799 		struct mbuf *cm = NULL, *cmn;
2800 		struct mbuf **cme = &cm;
2801 
2802 		do {
2803 			m2 = m->m_next;
2804 			m->m_next = NULL;
2805 			*cme = m;
2806 			cme = &(*cme)->m_next;
2807 			m = m2;
2808 		} while (m != NULL && m->m_type == MT_CONTROL);
2809 		while (cm != NULL) {
2810 			cmn = cm->m_next;
2811 			cm->m_next = NULL;
2812 			if (pr->pr_domain->dom_externalize != NULL) {
2813 				error = (*pr->pr_domain->dom_externalize)
2814 				    (cm, controlp, flags);
2815 			} else if (controlp != NULL)
2816 				*controlp = cm;
2817 			else
2818 				m_freem(cm);
2819 			if (controlp != NULL) {
2820 				while (*controlp != NULL)
2821 					controlp = &(*controlp)->m_next;
2822 			}
2823 			cm = cmn;
2824 		}
2825 	}
2826 	KASSERT(m == NULL || m->m_type == MT_DATA,
2827 	    ("soreceive_dgram: !data"));
2828 	while (m != NULL && uio->uio_resid > 0) {
2829 		len = uio->uio_resid;
2830 		if (len > m->m_len)
2831 			len = m->m_len;
2832 		error = uiomove(mtod(m, char *), (int)len, uio);
2833 		if (error) {
2834 			m_freem(m);
2835 			return (error);
2836 		}
2837 		if (len == m->m_len)
2838 			m = m_free(m);
2839 		else {
2840 			m->m_data += len;
2841 			m->m_len -= len;
2842 		}
2843 	}
2844 	if (m != NULL) {
2845 		flags |= MSG_TRUNC;
2846 		m_freem(m);
2847 	}
2848 	if (flagsp != NULL)
2849 		*flagsp |= flags;
2850 	return (0);
2851 }
2852 
2853 int
2854 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2855     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2856 {
2857 	int error;
2858 
2859 	CURVNET_SET(so->so_vnet);
2860 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio,
2861 	    mp0, controlp, flagsp));
2862 	CURVNET_RESTORE();
2863 	return (error);
2864 }
2865 
2866 int
2867 soshutdown(struct socket *so, int how)
2868 {
2869 	struct protosw *pr;
2870 	int error, soerror_enotconn;
2871 
2872 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2873 		return (EINVAL);
2874 
2875 	soerror_enotconn = 0;
2876 	SOCK_LOCK(so);
2877 	if ((so->so_state &
2878 	    (SS_ISCONNECTED | SS_ISCONNECTING | SS_ISDISCONNECTING)) == 0) {
2879 		/*
2880 		 * POSIX mandates us to return ENOTCONN when shutdown(2) is
2881 		 * invoked on a datagram sockets, however historically we would
2882 		 * actually tear socket down. This is known to be leveraged by
2883 		 * some applications to unblock process waiting in recvXXX(2)
2884 		 * by other process that it shares that socket with. Try to meet
2885 		 * both backward-compatibility and POSIX requirements by forcing
2886 		 * ENOTCONN but still asking protocol to perform pru_shutdown().
2887 		 */
2888 		if (so->so_type != SOCK_DGRAM && !SOLISTENING(so)) {
2889 			SOCK_UNLOCK(so);
2890 			return (ENOTCONN);
2891 		}
2892 		soerror_enotconn = 1;
2893 	}
2894 
2895 	if (SOLISTENING(so)) {
2896 		if (how != SHUT_WR) {
2897 			so->so_error = ECONNABORTED;
2898 			solisten_wakeup(so);	/* unlocks so */
2899 		} else {
2900 			SOCK_UNLOCK(so);
2901 		}
2902 		goto done;
2903 	}
2904 	SOCK_UNLOCK(so);
2905 
2906 	CURVNET_SET(so->so_vnet);
2907 	pr = so->so_proto;
2908 	if (pr->pr_usrreqs->pru_flush != NULL)
2909 		(*pr->pr_usrreqs->pru_flush)(so, how);
2910 	if (how != SHUT_WR)
2911 		sorflush(so);
2912 	if (how != SHUT_RD) {
2913 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2914 		wakeup(&so->so_timeo);
2915 		CURVNET_RESTORE();
2916 		return ((error == 0 && soerror_enotconn) ? ENOTCONN : error);
2917 	}
2918 	wakeup(&so->so_timeo);
2919 	CURVNET_RESTORE();
2920 
2921 done:
2922 	return (soerror_enotconn ? ENOTCONN : 0);
2923 }
2924 
2925 void
2926 sorflush(struct socket *so)
2927 {
2928 	struct socket aso;
2929 	struct protosw *pr;
2930 	int error;
2931 
2932 	VNET_SO_ASSERT(so);
2933 
2934 	/*
2935 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2936 	 * held, we make a partial copy of the socket buffer and clear the
2937 	 * original.  The new socket buffer copy won't have initialized locks so
2938 	 * we can only call routines that won't use or assert those locks.
2939 	 * Ideally calling socantrcvmore() would prevent data from being added
2940 	 * to the buffer, but currently it merely prevents buffered data from
2941 	 * being read by userspace.  We make this effort to free buffered data
2942 	 * nonetheless.
2943 	 *
2944 	 * Dislodge threads currently blocked in receive and wait to acquire
2945 	 * a lock against other simultaneous readers before clearing the
2946 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2947 	 * despite any existing socket disposition on interruptable waiting.
2948 	 */
2949 	socantrcvmore(so);
2950 
2951 	error = SOCK_IO_RECV_LOCK(so, SBL_WAIT | SBL_NOINTR);
2952 	if (error != 0) {
2953 		KASSERT(SOLISTENING(so),
2954 		    ("%s: soiolock(%p) failed", __func__, so));
2955 		return;
2956 	}
2957 
2958 	SOCK_RECVBUF_LOCK(so);
2959 	bzero(&aso, sizeof(aso));
2960 	aso.so_pcb = so->so_pcb;
2961 	bcopy(&so->so_rcv.sb_startzero, &aso.so_rcv.sb_startzero,
2962 	    offsetof(struct sockbuf, sb_endzero) -
2963 	    offsetof(struct sockbuf, sb_startzero));
2964 	bzero(&so->so_rcv.sb_startzero,
2965 	    offsetof(struct sockbuf, sb_endzero) -
2966 	    offsetof(struct sockbuf, sb_startzero));
2967 	SOCK_RECVBUF_UNLOCK(so);
2968 	SOCK_IO_RECV_UNLOCK(so);
2969 
2970 	/*
2971 	 * Dispose of special rights and flush the copied socket.  Don't call
2972 	 * any unsafe routines (that rely on locks being initialized) on aso.
2973 	 */
2974 	pr = so->so_proto;
2975 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2976 		(*pr->pr_domain->dom_dispose)(&aso);
2977 	sbrelease_internal(&aso.so_rcv, so);
2978 }
2979 
2980 /*
2981  * Wrapper for Socket established helper hook.
2982  * Parameters: socket, context of the hook point, hook id.
2983  */
2984 static int inline
2985 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2986 {
2987 	struct socket_hhook_data hhook_data = {
2988 		.so = so,
2989 		.hctx = hctx,
2990 		.m = NULL,
2991 		.status = 0
2992 	};
2993 
2994 	CURVNET_SET(so->so_vnet);
2995 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2996 	CURVNET_RESTORE();
2997 
2998 	/* Ugly but needed, since hhooks return void for now */
2999 	return (hhook_data.status);
3000 }
3001 
3002 /*
3003  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
3004  * additional variant to handle the case where the option value needs to be
3005  * some kind of integer, but not a specific size.  In addition to their use
3006  * here, these functions are also called by the protocol-level pr_ctloutput()
3007  * routines.
3008  */
3009 int
3010 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
3011 {
3012 	size_t	valsize;
3013 
3014 	/*
3015 	 * If the user gives us more than we wanted, we ignore it, but if we
3016 	 * don't get the minimum length the caller wants, we return EINVAL.
3017 	 * On success, sopt->sopt_valsize is set to however much we actually
3018 	 * retrieved.
3019 	 */
3020 	if ((valsize = sopt->sopt_valsize) < minlen)
3021 		return EINVAL;
3022 	if (valsize > len)
3023 		sopt->sopt_valsize = valsize = len;
3024 
3025 	if (sopt->sopt_td != NULL)
3026 		return (copyin(sopt->sopt_val, buf, valsize));
3027 
3028 	bcopy(sopt->sopt_val, buf, valsize);
3029 	return (0);
3030 }
3031 
3032 /*
3033  * Kernel version of setsockopt(2).
3034  *
3035  * XXX: optlen is size_t, not socklen_t
3036  */
3037 int
3038 so_setsockopt(struct socket *so, int level, int optname, void *optval,
3039     size_t optlen)
3040 {
3041 	struct sockopt sopt;
3042 
3043 	sopt.sopt_level = level;
3044 	sopt.sopt_name = optname;
3045 	sopt.sopt_dir = SOPT_SET;
3046 	sopt.sopt_val = optval;
3047 	sopt.sopt_valsize = optlen;
3048 	sopt.sopt_td = NULL;
3049 	return (sosetopt(so, &sopt));
3050 }
3051 
3052 int
3053 sosetopt(struct socket *so, struct sockopt *sopt)
3054 {
3055 	int	error, optval;
3056 	struct	linger l;
3057 	struct	timeval tv;
3058 	sbintime_t val;
3059 	uint32_t val32;
3060 #ifdef MAC
3061 	struct mac extmac;
3062 #endif
3063 
3064 	CURVNET_SET(so->so_vnet);
3065 	error = 0;
3066 	if (sopt->sopt_level != SOL_SOCKET) {
3067 		if (so->so_proto->pr_ctloutput != NULL)
3068 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3069 		else
3070 			error = ENOPROTOOPT;
3071 	} else {
3072 		switch (sopt->sopt_name) {
3073 		case SO_ACCEPTFILTER:
3074 			error = accept_filt_setopt(so, sopt);
3075 			if (error)
3076 				goto bad;
3077 			break;
3078 
3079 		case SO_LINGER:
3080 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
3081 			if (error)
3082 				goto bad;
3083 			if (l.l_linger < 0 ||
3084 			    l.l_linger > USHRT_MAX ||
3085 			    l.l_linger > (INT_MAX / hz)) {
3086 				error = EDOM;
3087 				goto bad;
3088 			}
3089 			SOCK_LOCK(so);
3090 			so->so_linger = l.l_linger;
3091 			if (l.l_onoff)
3092 				so->so_options |= SO_LINGER;
3093 			else
3094 				so->so_options &= ~SO_LINGER;
3095 			SOCK_UNLOCK(so);
3096 			break;
3097 
3098 		case SO_DEBUG:
3099 		case SO_KEEPALIVE:
3100 		case SO_DONTROUTE:
3101 		case SO_USELOOPBACK:
3102 		case SO_BROADCAST:
3103 		case SO_REUSEADDR:
3104 		case SO_REUSEPORT:
3105 		case SO_REUSEPORT_LB:
3106 		case SO_OOBINLINE:
3107 		case SO_TIMESTAMP:
3108 		case SO_BINTIME:
3109 		case SO_NOSIGPIPE:
3110 		case SO_NO_DDP:
3111 		case SO_NO_OFFLOAD:
3112 		case SO_RERROR:
3113 			error = sooptcopyin(sopt, &optval, sizeof optval,
3114 			    sizeof optval);
3115 			if (error)
3116 				goto bad;
3117 			SOCK_LOCK(so);
3118 			if (optval)
3119 				so->so_options |= sopt->sopt_name;
3120 			else
3121 				so->so_options &= ~sopt->sopt_name;
3122 			SOCK_UNLOCK(so);
3123 			break;
3124 
3125 		case SO_SETFIB:
3126 			error = sooptcopyin(sopt, &optval, sizeof optval,
3127 			    sizeof optval);
3128 			if (error)
3129 				goto bad;
3130 
3131 			if (optval < 0 || optval >= rt_numfibs) {
3132 				error = EINVAL;
3133 				goto bad;
3134 			}
3135 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
3136 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
3137 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
3138 				so->so_fibnum = optval;
3139 			else
3140 				so->so_fibnum = 0;
3141 			break;
3142 
3143 		case SO_USER_COOKIE:
3144 			error = sooptcopyin(sopt, &val32, sizeof val32,
3145 			    sizeof val32);
3146 			if (error)
3147 				goto bad;
3148 			so->so_user_cookie = val32;
3149 			break;
3150 
3151 		case SO_SNDBUF:
3152 		case SO_RCVBUF:
3153 		case SO_SNDLOWAT:
3154 		case SO_RCVLOWAT:
3155 			error = sooptcopyin(sopt, &optval, sizeof optval,
3156 			    sizeof optval);
3157 			if (error)
3158 				goto bad;
3159 
3160 			/*
3161 			 * Values < 1 make no sense for any of these options,
3162 			 * so disallow them.
3163 			 */
3164 			if (optval < 1) {
3165 				error = EINVAL;
3166 				goto bad;
3167 			}
3168 
3169 			error = sbsetopt(so, sopt->sopt_name, optval);
3170 			break;
3171 
3172 		case SO_SNDTIMEO:
3173 		case SO_RCVTIMEO:
3174 #ifdef COMPAT_FREEBSD32
3175 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3176 				struct timeval32 tv32;
3177 
3178 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
3179 				    sizeof tv32);
3180 				CP(tv32, tv, tv_sec);
3181 				CP(tv32, tv, tv_usec);
3182 			} else
3183 #endif
3184 				error = sooptcopyin(sopt, &tv, sizeof tv,
3185 				    sizeof tv);
3186 			if (error)
3187 				goto bad;
3188 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
3189 			    tv.tv_usec >= 1000000) {
3190 				error = EDOM;
3191 				goto bad;
3192 			}
3193 			if (tv.tv_sec > INT32_MAX)
3194 				val = SBT_MAX;
3195 			else
3196 				val = tvtosbt(tv);
3197 			switch (sopt->sopt_name) {
3198 			case SO_SNDTIMEO:
3199 				so->so_snd.sb_timeo = val;
3200 				break;
3201 			case SO_RCVTIMEO:
3202 				so->so_rcv.sb_timeo = val;
3203 				break;
3204 			}
3205 			break;
3206 
3207 		case SO_LABEL:
3208 #ifdef MAC
3209 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
3210 			    sizeof extmac);
3211 			if (error)
3212 				goto bad;
3213 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
3214 			    so, &extmac);
3215 #else
3216 			error = EOPNOTSUPP;
3217 #endif
3218 			break;
3219 
3220 		case SO_TS_CLOCK:
3221 			error = sooptcopyin(sopt, &optval, sizeof optval,
3222 			    sizeof optval);
3223 			if (error)
3224 				goto bad;
3225 			if (optval < 0 || optval > SO_TS_CLOCK_MAX) {
3226 				error = EINVAL;
3227 				goto bad;
3228 			}
3229 			so->so_ts_clock = optval;
3230 			break;
3231 
3232 		case SO_MAX_PACING_RATE:
3233 			error = sooptcopyin(sopt, &val32, sizeof(val32),
3234 			    sizeof(val32));
3235 			if (error)
3236 				goto bad;
3237 			so->so_max_pacing_rate = val32;
3238 			break;
3239 
3240 		default:
3241 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3242 				error = hhook_run_socket(so, sopt,
3243 				    HHOOK_SOCKET_OPT);
3244 			else
3245 				error = ENOPROTOOPT;
3246 			break;
3247 		}
3248 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
3249 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
3250 	}
3251 bad:
3252 	CURVNET_RESTORE();
3253 	return (error);
3254 }
3255 
3256 /*
3257  * Helper routine for getsockopt.
3258  */
3259 int
3260 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
3261 {
3262 	int	error;
3263 	size_t	valsize;
3264 
3265 	error = 0;
3266 
3267 	/*
3268 	 * Documented get behavior is that we always return a value, possibly
3269 	 * truncated to fit in the user's buffer.  Traditional behavior is
3270 	 * that we always tell the user precisely how much we copied, rather
3271 	 * than something useful like the total amount we had available for
3272 	 * her.  Note that this interface is not idempotent; the entire
3273 	 * answer must be generated ahead of time.
3274 	 */
3275 	valsize = min(len, sopt->sopt_valsize);
3276 	sopt->sopt_valsize = valsize;
3277 	if (sopt->sopt_val != NULL) {
3278 		if (sopt->sopt_td != NULL)
3279 			error = copyout(buf, sopt->sopt_val, valsize);
3280 		else
3281 			bcopy(buf, sopt->sopt_val, valsize);
3282 	}
3283 	return (error);
3284 }
3285 
3286 int
3287 sogetopt(struct socket *so, struct sockopt *sopt)
3288 {
3289 	int	error, optval;
3290 	struct	linger l;
3291 	struct	timeval tv;
3292 #ifdef MAC
3293 	struct mac extmac;
3294 #endif
3295 
3296 	CURVNET_SET(so->so_vnet);
3297 	error = 0;
3298 	if (sopt->sopt_level != SOL_SOCKET) {
3299 		if (so->so_proto->pr_ctloutput != NULL)
3300 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
3301 		else
3302 			error = ENOPROTOOPT;
3303 		CURVNET_RESTORE();
3304 		return (error);
3305 	} else {
3306 		switch (sopt->sopt_name) {
3307 		case SO_ACCEPTFILTER:
3308 			error = accept_filt_getopt(so, sopt);
3309 			break;
3310 
3311 		case SO_LINGER:
3312 			SOCK_LOCK(so);
3313 			l.l_onoff = so->so_options & SO_LINGER;
3314 			l.l_linger = so->so_linger;
3315 			SOCK_UNLOCK(so);
3316 			error = sooptcopyout(sopt, &l, sizeof l);
3317 			break;
3318 
3319 		case SO_USELOOPBACK:
3320 		case SO_DONTROUTE:
3321 		case SO_DEBUG:
3322 		case SO_KEEPALIVE:
3323 		case SO_REUSEADDR:
3324 		case SO_REUSEPORT:
3325 		case SO_REUSEPORT_LB:
3326 		case SO_BROADCAST:
3327 		case SO_OOBINLINE:
3328 		case SO_ACCEPTCONN:
3329 		case SO_TIMESTAMP:
3330 		case SO_BINTIME:
3331 		case SO_NOSIGPIPE:
3332 		case SO_NO_DDP:
3333 		case SO_NO_OFFLOAD:
3334 		case SO_RERROR:
3335 			optval = so->so_options & sopt->sopt_name;
3336 integer:
3337 			error = sooptcopyout(sopt, &optval, sizeof optval);
3338 			break;
3339 
3340 		case SO_DOMAIN:
3341 			optval = so->so_proto->pr_domain->dom_family;
3342 			goto integer;
3343 
3344 		case SO_TYPE:
3345 			optval = so->so_type;
3346 			goto integer;
3347 
3348 		case SO_PROTOCOL:
3349 			optval = so->so_proto->pr_protocol;
3350 			goto integer;
3351 
3352 		case SO_ERROR:
3353 			SOCK_LOCK(so);
3354 			if (so->so_error) {
3355 				optval = so->so_error;
3356 				so->so_error = 0;
3357 			} else {
3358 				optval = so->so_rerror;
3359 				so->so_rerror = 0;
3360 			}
3361 			SOCK_UNLOCK(so);
3362 			goto integer;
3363 
3364 		case SO_SNDBUF:
3365 			optval = SOLISTENING(so) ? so->sol_sbsnd_hiwat :
3366 			    so->so_snd.sb_hiwat;
3367 			goto integer;
3368 
3369 		case SO_RCVBUF:
3370 			optval = SOLISTENING(so) ? so->sol_sbrcv_hiwat :
3371 			    so->so_rcv.sb_hiwat;
3372 			goto integer;
3373 
3374 		case SO_SNDLOWAT:
3375 			optval = SOLISTENING(so) ? so->sol_sbsnd_lowat :
3376 			    so->so_snd.sb_lowat;
3377 			goto integer;
3378 
3379 		case SO_RCVLOWAT:
3380 			optval = SOLISTENING(so) ? so->sol_sbrcv_lowat :
3381 			    so->so_rcv.sb_lowat;
3382 			goto integer;
3383 
3384 		case SO_SNDTIMEO:
3385 		case SO_RCVTIMEO:
3386 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
3387 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
3388 #ifdef COMPAT_FREEBSD32
3389 			if (SV_CURPROC_FLAG(SV_ILP32)) {
3390 				struct timeval32 tv32;
3391 
3392 				CP(tv, tv32, tv_sec);
3393 				CP(tv, tv32, tv_usec);
3394 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
3395 			} else
3396 #endif
3397 				error = sooptcopyout(sopt, &tv, sizeof tv);
3398 			break;
3399 
3400 		case SO_LABEL:
3401 #ifdef MAC
3402 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3403 			    sizeof(extmac));
3404 			if (error)
3405 				goto bad;
3406 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
3407 			    so, &extmac);
3408 			if (error)
3409 				goto bad;
3410 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3411 #else
3412 			error = EOPNOTSUPP;
3413 #endif
3414 			break;
3415 
3416 		case SO_PEERLABEL:
3417 #ifdef MAC
3418 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
3419 			    sizeof(extmac));
3420 			if (error)
3421 				goto bad;
3422 			error = mac_getsockopt_peerlabel(
3423 			    sopt->sopt_td->td_ucred, so, &extmac);
3424 			if (error)
3425 				goto bad;
3426 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
3427 #else
3428 			error = EOPNOTSUPP;
3429 #endif
3430 			break;
3431 
3432 		case SO_LISTENQLIMIT:
3433 			optval = SOLISTENING(so) ? so->sol_qlimit : 0;
3434 			goto integer;
3435 
3436 		case SO_LISTENQLEN:
3437 			optval = SOLISTENING(so) ? so->sol_qlen : 0;
3438 			goto integer;
3439 
3440 		case SO_LISTENINCQLEN:
3441 			optval = SOLISTENING(so) ? so->sol_incqlen : 0;
3442 			goto integer;
3443 
3444 		case SO_TS_CLOCK:
3445 			optval = so->so_ts_clock;
3446 			goto integer;
3447 
3448 		case SO_MAX_PACING_RATE:
3449 			optval = so->so_max_pacing_rate;
3450 			goto integer;
3451 
3452 		default:
3453 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
3454 				error = hhook_run_socket(so, sopt,
3455 				    HHOOK_SOCKET_OPT);
3456 			else
3457 				error = ENOPROTOOPT;
3458 			break;
3459 		}
3460 	}
3461 #ifdef MAC
3462 bad:
3463 #endif
3464 	CURVNET_RESTORE();
3465 	return (error);
3466 }
3467 
3468 int
3469 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
3470 {
3471 	struct mbuf *m, *m_prev;
3472 	int sopt_size = sopt->sopt_valsize;
3473 
3474 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3475 	if (m == NULL)
3476 		return ENOBUFS;
3477 	if (sopt_size > MLEN) {
3478 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
3479 		if ((m->m_flags & M_EXT) == 0) {
3480 			m_free(m);
3481 			return ENOBUFS;
3482 		}
3483 		m->m_len = min(MCLBYTES, sopt_size);
3484 	} else {
3485 		m->m_len = min(MLEN, sopt_size);
3486 	}
3487 	sopt_size -= m->m_len;
3488 	*mp = m;
3489 	m_prev = m;
3490 
3491 	while (sopt_size) {
3492 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
3493 		if (m == NULL) {
3494 			m_freem(*mp);
3495 			return ENOBUFS;
3496 		}
3497 		if (sopt_size > MLEN) {
3498 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
3499 			    M_NOWAIT);
3500 			if ((m->m_flags & M_EXT) == 0) {
3501 				m_freem(m);
3502 				m_freem(*mp);
3503 				return ENOBUFS;
3504 			}
3505 			m->m_len = min(MCLBYTES, sopt_size);
3506 		} else {
3507 			m->m_len = min(MLEN, sopt_size);
3508 		}
3509 		sopt_size -= m->m_len;
3510 		m_prev->m_next = m;
3511 		m_prev = m;
3512 	}
3513 	return (0);
3514 }
3515 
3516 int
3517 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
3518 {
3519 	struct mbuf *m0 = m;
3520 
3521 	if (sopt->sopt_val == NULL)
3522 		return (0);
3523 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3524 		if (sopt->sopt_td != NULL) {
3525 			int error;
3526 
3527 			error = copyin(sopt->sopt_val, mtod(m, char *),
3528 			    m->m_len);
3529 			if (error != 0) {
3530 				m_freem(m0);
3531 				return(error);
3532 			}
3533 		} else
3534 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
3535 		sopt->sopt_valsize -= m->m_len;
3536 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3537 		m = m->m_next;
3538 	}
3539 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
3540 		panic("ip6_sooptmcopyin");
3541 	return (0);
3542 }
3543 
3544 int
3545 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
3546 {
3547 	struct mbuf *m0 = m;
3548 	size_t valsize = 0;
3549 
3550 	if (sopt->sopt_val == NULL)
3551 		return (0);
3552 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
3553 		if (sopt->sopt_td != NULL) {
3554 			int error;
3555 
3556 			error = copyout(mtod(m, char *), sopt->sopt_val,
3557 			    m->m_len);
3558 			if (error != 0) {
3559 				m_freem(m0);
3560 				return(error);
3561 			}
3562 		} else
3563 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
3564 		sopt->sopt_valsize -= m->m_len;
3565 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
3566 		valsize += m->m_len;
3567 		m = m->m_next;
3568 	}
3569 	if (m != NULL) {
3570 		/* enough soopt buffer should be given from user-land */
3571 		m_freem(m0);
3572 		return(EINVAL);
3573 	}
3574 	sopt->sopt_valsize = valsize;
3575 	return (0);
3576 }
3577 
3578 /*
3579  * sohasoutofband(): protocol notifies socket layer of the arrival of new
3580  * out-of-band data, which will then notify socket consumers.
3581  */
3582 void
3583 sohasoutofband(struct socket *so)
3584 {
3585 
3586 	if (so->so_sigio != NULL)
3587 		pgsigio(&so->so_sigio, SIGURG, 0);
3588 	selwakeuppri(&so->so_rdsel, PSOCK);
3589 }
3590 
3591 int
3592 sopoll(struct socket *so, int events, struct ucred *active_cred,
3593     struct thread *td)
3594 {
3595 
3596 	/*
3597 	 * We do not need to set or assert curvnet as long as everyone uses
3598 	 * sopoll_generic().
3599 	 */
3600 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
3601 	    td));
3602 }
3603 
3604 int
3605 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3606     struct thread *td)
3607 {
3608 	int revents;
3609 
3610 	SOCK_LOCK(so);
3611 	if (SOLISTENING(so)) {
3612 		if (!(events & (POLLIN | POLLRDNORM)))
3613 			revents = 0;
3614 		else if (!TAILQ_EMPTY(&so->sol_comp))
3615 			revents = events & (POLLIN | POLLRDNORM);
3616 		else if ((events & POLLINIGNEOF) == 0 && so->so_error)
3617 			revents = (events & (POLLIN | POLLRDNORM)) | POLLHUP;
3618 		else {
3619 			selrecord(td, &so->so_rdsel);
3620 			revents = 0;
3621 		}
3622 	} else {
3623 		revents = 0;
3624 		SOCKBUF_LOCK(&so->so_snd);
3625 		SOCKBUF_LOCK(&so->so_rcv);
3626 		if (events & (POLLIN | POLLRDNORM))
3627 			if (soreadabledata(so))
3628 				revents |= events & (POLLIN | POLLRDNORM);
3629 		if (events & (POLLOUT | POLLWRNORM))
3630 			if (sowriteable(so))
3631 				revents |= events & (POLLOUT | POLLWRNORM);
3632 		if (events & (POLLPRI | POLLRDBAND))
3633 			if (so->so_oobmark ||
3634 			    (so->so_rcv.sb_state & SBS_RCVATMARK))
3635 				revents |= events & (POLLPRI | POLLRDBAND);
3636 		if ((events & POLLINIGNEOF) == 0) {
3637 			if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3638 				revents |= events & (POLLIN | POLLRDNORM);
3639 				if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3640 					revents |= POLLHUP;
3641 			}
3642 		}
3643 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE)
3644 			revents |= events & POLLRDHUP;
3645 		if (revents == 0) {
3646 			if (events &
3647 			    (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND | POLLRDHUP)) {
3648 				selrecord(td, &so->so_rdsel);
3649 				so->so_rcv.sb_flags |= SB_SEL;
3650 			}
3651 			if (events & (POLLOUT | POLLWRNORM)) {
3652 				selrecord(td, &so->so_wrsel);
3653 				so->so_snd.sb_flags |= SB_SEL;
3654 			}
3655 		}
3656 		SOCKBUF_UNLOCK(&so->so_rcv);
3657 		SOCKBUF_UNLOCK(&so->so_snd);
3658 	}
3659 	SOCK_UNLOCK(so);
3660 	return (revents);
3661 }
3662 
3663 int
3664 soo_kqfilter(struct file *fp, struct knote *kn)
3665 {
3666 	struct socket *so = kn->kn_fp->f_data;
3667 	struct sockbuf *sb;
3668 	struct knlist *knl;
3669 
3670 	switch (kn->kn_filter) {
3671 	case EVFILT_READ:
3672 		kn->kn_fop = &soread_filtops;
3673 		knl = &so->so_rdsel.si_note;
3674 		sb = &so->so_rcv;
3675 		break;
3676 	case EVFILT_WRITE:
3677 		kn->kn_fop = &sowrite_filtops;
3678 		knl = &so->so_wrsel.si_note;
3679 		sb = &so->so_snd;
3680 		break;
3681 	case EVFILT_EMPTY:
3682 		kn->kn_fop = &soempty_filtops;
3683 		knl = &so->so_wrsel.si_note;
3684 		sb = &so->so_snd;
3685 		break;
3686 	default:
3687 		return (EINVAL);
3688 	}
3689 
3690 	SOCK_LOCK(so);
3691 	if (SOLISTENING(so)) {
3692 		knlist_add(knl, kn, 1);
3693 	} else {
3694 		SOCKBUF_LOCK(sb);
3695 		knlist_add(knl, kn, 1);
3696 		sb->sb_flags |= SB_KNOTE;
3697 		SOCKBUF_UNLOCK(sb);
3698 	}
3699 	SOCK_UNLOCK(so);
3700 	return (0);
3701 }
3702 
3703 /*
3704  * Some routines that return EOPNOTSUPP for entry points that are not
3705  * supported by a protocol.  Fill in as needed.
3706  */
3707 int
3708 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3709 {
3710 
3711 	return EOPNOTSUPP;
3712 }
3713 
3714 int
3715 pru_aio_queue_notsupp(struct socket *so, struct kaiocb *job)
3716 {
3717 
3718 	return EOPNOTSUPP;
3719 }
3720 
3721 int
3722 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3723 {
3724 
3725 	return EOPNOTSUPP;
3726 }
3727 
3728 int
3729 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3730 {
3731 
3732 	return EOPNOTSUPP;
3733 }
3734 
3735 int
3736 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3737     struct thread *td)
3738 {
3739 
3740 	return EOPNOTSUPP;
3741 }
3742 
3743 int
3744 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3745 {
3746 
3747 	return EOPNOTSUPP;
3748 }
3749 
3750 int
3751 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3752     struct thread *td)
3753 {
3754 
3755 	return EOPNOTSUPP;
3756 }
3757 
3758 int
3759 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3760 {
3761 
3762 	return EOPNOTSUPP;
3763 }
3764 
3765 int
3766 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3767     struct ifnet *ifp, struct thread *td)
3768 {
3769 
3770 	return EOPNOTSUPP;
3771 }
3772 
3773 int
3774 pru_disconnect_notsupp(struct socket *so)
3775 {
3776 
3777 	return EOPNOTSUPP;
3778 }
3779 
3780 int
3781 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3782 {
3783 
3784 	return EOPNOTSUPP;
3785 }
3786 
3787 int
3788 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3789 {
3790 
3791 	return EOPNOTSUPP;
3792 }
3793 
3794 int
3795 pru_rcvd_notsupp(struct socket *so, int flags)
3796 {
3797 
3798 	return EOPNOTSUPP;
3799 }
3800 
3801 int
3802 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3803 {
3804 
3805 	return EOPNOTSUPP;
3806 }
3807 
3808 int
3809 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3810     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3811 {
3812 
3813 	if (control != NULL)
3814 		m_freem(control);
3815 	if ((flags & PRUS_NOTREADY) == 0)
3816 		m_freem(m);
3817 	return (EOPNOTSUPP);
3818 }
3819 
3820 int
3821 pru_ready_notsupp(struct socket *so, struct mbuf *m, int count)
3822 {
3823 
3824 	return (EOPNOTSUPP);
3825 }
3826 
3827 /*
3828  * This isn't really a ``null'' operation, but it's the default one and
3829  * doesn't do anything destructive.
3830  */
3831 int
3832 pru_sense_null(struct socket *so, struct stat *sb)
3833 {
3834 
3835 	sb->st_blksize = so->so_snd.sb_hiwat;
3836 	return 0;
3837 }
3838 
3839 int
3840 pru_shutdown_notsupp(struct socket *so)
3841 {
3842 
3843 	return EOPNOTSUPP;
3844 }
3845 
3846 int
3847 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3848 {
3849 
3850 	return EOPNOTSUPP;
3851 }
3852 
3853 int
3854 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3855     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3856 {
3857 
3858 	return EOPNOTSUPP;
3859 }
3860 
3861 int
3862 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3863     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3864 {
3865 
3866 	return EOPNOTSUPP;
3867 }
3868 
3869 int
3870 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3871     struct thread *td)
3872 {
3873 
3874 	return EOPNOTSUPP;
3875 }
3876 
3877 static void
3878 filt_sordetach(struct knote *kn)
3879 {
3880 	struct socket *so = kn->kn_fp->f_data;
3881 
3882 	so_rdknl_lock(so);
3883 	knlist_remove(&so->so_rdsel.si_note, kn, 1);
3884 	if (!SOLISTENING(so) && knlist_empty(&so->so_rdsel.si_note))
3885 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3886 	so_rdknl_unlock(so);
3887 }
3888 
3889 /*ARGSUSED*/
3890 static int
3891 filt_soread(struct knote *kn, long hint)
3892 {
3893 	struct socket *so;
3894 
3895 	so = kn->kn_fp->f_data;
3896 
3897 	if (SOLISTENING(so)) {
3898 		SOCK_LOCK_ASSERT(so);
3899 		kn->kn_data = so->sol_qlen;
3900 		if (so->so_error) {
3901 			kn->kn_flags |= EV_EOF;
3902 			kn->kn_fflags = so->so_error;
3903 			return (1);
3904 		}
3905 		return (!TAILQ_EMPTY(&so->sol_comp));
3906 	}
3907 
3908 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3909 
3910 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3911 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3912 		kn->kn_flags |= EV_EOF;
3913 		kn->kn_fflags = so->so_error;
3914 		return (1);
3915 	} else if (so->so_error || so->so_rerror)
3916 		return (1);
3917 
3918 	if (kn->kn_sfflags & NOTE_LOWAT) {
3919 		if (kn->kn_data >= kn->kn_sdata)
3920 			return (1);
3921 	} else if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3922 		return (1);
3923 
3924 	/* This hook returning non-zero indicates an event, not error */
3925 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3926 }
3927 
3928 static void
3929 filt_sowdetach(struct knote *kn)
3930 {
3931 	struct socket *so = kn->kn_fp->f_data;
3932 
3933 	so_wrknl_lock(so);
3934 	knlist_remove(&so->so_wrsel.si_note, kn, 1);
3935 	if (!SOLISTENING(so) && knlist_empty(&so->so_wrsel.si_note))
3936 		so->so_snd.sb_flags &= ~SB_KNOTE;
3937 	so_wrknl_unlock(so);
3938 }
3939 
3940 /*ARGSUSED*/
3941 static int
3942 filt_sowrite(struct knote *kn, long hint)
3943 {
3944 	struct socket *so;
3945 
3946 	so = kn->kn_fp->f_data;
3947 
3948 	if (SOLISTENING(so))
3949 		return (0);
3950 
3951 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3952 	kn->kn_data = sbspace(&so->so_snd);
3953 
3954 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3955 
3956 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3957 		kn->kn_flags |= EV_EOF;
3958 		kn->kn_fflags = so->so_error;
3959 		return (1);
3960 	} else if (so->so_error)	/* temporary udp error */
3961 		return (1);
3962 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3963 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3964 		return (0);
3965 	else if (kn->kn_sfflags & NOTE_LOWAT)
3966 		return (kn->kn_data >= kn->kn_sdata);
3967 	else
3968 		return (kn->kn_data >= so->so_snd.sb_lowat);
3969 }
3970 
3971 static int
3972 filt_soempty(struct knote *kn, long hint)
3973 {
3974 	struct socket *so;
3975 
3976 	so = kn->kn_fp->f_data;
3977 
3978 	if (SOLISTENING(so))
3979 		return (1);
3980 
3981 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3982 	kn->kn_data = sbused(&so->so_snd);
3983 
3984 	if (kn->kn_data == 0)
3985 		return (1);
3986 	else
3987 		return (0);
3988 }
3989 
3990 int
3991 socheckuid(struct socket *so, uid_t uid)
3992 {
3993 
3994 	if (so == NULL)
3995 		return (EPERM);
3996 	if (so->so_cred->cr_uid != uid)
3997 		return (EPERM);
3998 	return (0);
3999 }
4000 
4001 /*
4002  * These functions are used by protocols to notify the socket layer (and its
4003  * consumers) of state changes in the sockets driven by protocol-side events.
4004  */
4005 
4006 /*
4007  * Procedures to manipulate state flags of socket and do appropriate wakeups.
4008  *
4009  * Normal sequence from the active (originating) side is that
4010  * soisconnecting() is called during processing of connect() call, resulting
4011  * in an eventual call to soisconnected() if/when the connection is
4012  * established.  When the connection is torn down soisdisconnecting() is
4013  * called during processing of disconnect() call, and soisdisconnected() is
4014  * called when the connection to the peer is totally severed.  The semantics
4015  * of these routines are such that connectionless protocols can call
4016  * soisconnected() and soisdisconnected() only, bypassing the in-progress
4017  * calls when setting up a ``connection'' takes no time.
4018  *
4019  * From the passive side, a socket is created with two queues of sockets:
4020  * so_incomp for connections in progress and so_comp for connections already
4021  * made and awaiting user acceptance.  As a protocol is preparing incoming
4022  * connections, it creates a socket structure queued on so_incomp by calling
4023  * sonewconn().  When the connection is established, soisconnected() is
4024  * called, and transfers the socket structure to so_comp, making it available
4025  * to accept().
4026  *
4027  * If a socket is closed with sockets on either so_incomp or so_comp, these
4028  * sockets are dropped.
4029  *
4030  * If higher-level protocols are implemented in the kernel, the wakeups done
4031  * here will sometimes cause software-interrupt process scheduling.
4032  */
4033 void
4034 soisconnecting(struct socket *so)
4035 {
4036 
4037 	SOCK_LOCK(so);
4038 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
4039 	so->so_state |= SS_ISCONNECTING;
4040 	SOCK_UNLOCK(so);
4041 }
4042 
4043 void
4044 soisconnected(struct socket *so)
4045 {
4046 	bool last __diagused;
4047 
4048 	SOCK_LOCK(so);
4049 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
4050 	so->so_state |= SS_ISCONNECTED;
4051 
4052 	if (so->so_qstate == SQ_INCOMP) {
4053 		struct socket *head = so->so_listen;
4054 		int ret;
4055 
4056 		KASSERT(head, ("%s: so %p on incomp of NULL", __func__, so));
4057 		/*
4058 		 * Promoting a socket from incomplete queue to complete, we
4059 		 * need to go through reverse order of locking.  We first do
4060 		 * trylock, and if that doesn't succeed, we go the hard way
4061 		 * leaving a reference and rechecking consistency after proper
4062 		 * locking.
4063 		 */
4064 		if (__predict_false(SOLISTEN_TRYLOCK(head) == 0)) {
4065 			soref(head);
4066 			SOCK_UNLOCK(so);
4067 			SOLISTEN_LOCK(head);
4068 			SOCK_LOCK(so);
4069 			if (__predict_false(head != so->so_listen)) {
4070 				/*
4071 				 * The socket went off the listen queue,
4072 				 * should be lost race to close(2) of sol.
4073 				 * The socket is about to soabort().
4074 				 */
4075 				SOCK_UNLOCK(so);
4076 				sorele_locked(head);
4077 				return;
4078 			}
4079 			last = refcount_release(&head->so_count);
4080 			KASSERT(!last, ("%s: released last reference for %p",
4081 			    __func__, head));
4082 		}
4083 again:
4084 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
4085 			TAILQ_REMOVE(&head->sol_incomp, so, so_list);
4086 			head->sol_incqlen--;
4087 			TAILQ_INSERT_TAIL(&head->sol_comp, so, so_list);
4088 			head->sol_qlen++;
4089 			so->so_qstate = SQ_COMP;
4090 			SOCK_UNLOCK(so);
4091 			solisten_wakeup(head);	/* unlocks */
4092 		} else {
4093 			SOCKBUF_LOCK(&so->so_rcv);
4094 			soupcall_set(so, SO_RCV,
4095 			    head->sol_accept_filter->accf_callback,
4096 			    head->sol_accept_filter_arg);
4097 			so->so_options &= ~SO_ACCEPTFILTER;
4098 			ret = head->sol_accept_filter->accf_callback(so,
4099 			    head->sol_accept_filter_arg, M_NOWAIT);
4100 			if (ret == SU_ISCONNECTED) {
4101 				soupcall_clear(so, SO_RCV);
4102 				SOCKBUF_UNLOCK(&so->so_rcv);
4103 				goto again;
4104 			}
4105 			SOCKBUF_UNLOCK(&so->so_rcv);
4106 			SOCK_UNLOCK(so);
4107 			SOLISTEN_UNLOCK(head);
4108 		}
4109 		return;
4110 	}
4111 	SOCK_UNLOCK(so);
4112 	wakeup(&so->so_timeo);
4113 	sorwakeup(so);
4114 	sowwakeup(so);
4115 }
4116 
4117 void
4118 soisdisconnecting(struct socket *so)
4119 {
4120 
4121 	SOCK_LOCK(so);
4122 	so->so_state &= ~SS_ISCONNECTING;
4123 	so->so_state |= SS_ISDISCONNECTING;
4124 
4125 	if (!SOLISTENING(so)) {
4126 		SOCKBUF_LOCK(&so->so_rcv);
4127 		socantrcvmore_locked(so);
4128 		SOCKBUF_LOCK(&so->so_snd);
4129 		socantsendmore_locked(so);
4130 	}
4131 	SOCK_UNLOCK(so);
4132 	wakeup(&so->so_timeo);
4133 }
4134 
4135 void
4136 soisdisconnected(struct socket *so)
4137 {
4138 
4139 	SOCK_LOCK(so);
4140 
4141 	/*
4142 	 * There is at least one reader of so_state that does not
4143 	 * acquire socket lock, namely soreceive_generic().  Ensure
4144 	 * that it never sees all flags that track connection status
4145 	 * cleared, by ordering the update with a barrier semantic of
4146 	 * our release thread fence.
4147 	 */
4148 	so->so_state |= SS_ISDISCONNECTED;
4149 	atomic_thread_fence_rel();
4150 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
4151 
4152 	if (!SOLISTENING(so)) {
4153 		SOCK_UNLOCK(so);
4154 		SOCKBUF_LOCK(&so->so_rcv);
4155 		socantrcvmore_locked(so);
4156 		SOCKBUF_LOCK(&so->so_snd);
4157 		sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
4158 		socantsendmore_locked(so);
4159 	} else
4160 		SOCK_UNLOCK(so);
4161 	wakeup(&so->so_timeo);
4162 }
4163 
4164 int
4165 soiolock(struct socket *so, struct sx *sx, int flags)
4166 {
4167 	int error;
4168 
4169 	KASSERT((flags & SBL_VALID) == flags,
4170 	    ("soiolock: invalid flags %#x", flags));
4171 
4172 	if ((flags & SBL_WAIT) != 0) {
4173 		if ((flags & SBL_NOINTR) != 0) {
4174 			sx_xlock(sx);
4175 		} else {
4176 			error = sx_xlock_sig(sx);
4177 			if (error != 0)
4178 				return (error);
4179 		}
4180 	} else if (!sx_try_xlock(sx)) {
4181 		return (EWOULDBLOCK);
4182 	}
4183 
4184 	if (__predict_false(SOLISTENING(so))) {
4185 		sx_xunlock(sx);
4186 		return (ENOTCONN);
4187 	}
4188 	return (0);
4189 }
4190 
4191 void
4192 soiounlock(struct sx *sx)
4193 {
4194 	sx_xunlock(sx);
4195 }
4196 
4197 /*
4198  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
4199  */
4200 struct sockaddr *
4201 sodupsockaddr(const struct sockaddr *sa, int mflags)
4202 {
4203 	struct sockaddr *sa2;
4204 
4205 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
4206 	if (sa2)
4207 		bcopy(sa, sa2, sa->sa_len);
4208 	return sa2;
4209 }
4210 
4211 /*
4212  * Register per-socket destructor.
4213  */
4214 void
4215 sodtor_set(struct socket *so, so_dtor_t *func)
4216 {
4217 
4218 	SOCK_LOCK_ASSERT(so);
4219 	so->so_dtor = func;
4220 }
4221 
4222 /*
4223  * Register per-socket buffer upcalls.
4224  */
4225 void
4226 soupcall_set(struct socket *so, int which, so_upcall_t func, void *arg)
4227 {
4228 	struct sockbuf *sb;
4229 
4230 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4231 
4232 	switch (which) {
4233 	case SO_RCV:
4234 		sb = &so->so_rcv;
4235 		break;
4236 	case SO_SND:
4237 		sb = &so->so_snd;
4238 		break;
4239 	default:
4240 		panic("soupcall_set: bad which");
4241 	}
4242 	SOCKBUF_LOCK_ASSERT(sb);
4243 	sb->sb_upcall = func;
4244 	sb->sb_upcallarg = arg;
4245 	sb->sb_flags |= SB_UPCALL;
4246 }
4247 
4248 void
4249 soupcall_clear(struct socket *so, int which)
4250 {
4251 	struct sockbuf *sb;
4252 
4253 	KASSERT(!SOLISTENING(so), ("%s: so %p listening", __func__, so));
4254 
4255 	switch (which) {
4256 	case SO_RCV:
4257 		sb = &so->so_rcv;
4258 		break;
4259 	case SO_SND:
4260 		sb = &so->so_snd;
4261 		break;
4262 	default:
4263 		panic("soupcall_clear: bad which");
4264 	}
4265 	SOCKBUF_LOCK_ASSERT(sb);
4266 	KASSERT(sb->sb_upcall != NULL,
4267 	    ("%s: so %p no upcall to clear", __func__, so));
4268 	sb->sb_upcall = NULL;
4269 	sb->sb_upcallarg = NULL;
4270 	sb->sb_flags &= ~SB_UPCALL;
4271 }
4272 
4273 void
4274 solisten_upcall_set(struct socket *so, so_upcall_t func, void *arg)
4275 {
4276 
4277 	SOLISTEN_LOCK_ASSERT(so);
4278 	so->sol_upcall = func;
4279 	so->sol_upcallarg = arg;
4280 }
4281 
4282 static void
4283 so_rdknl_lock(void *arg)
4284 {
4285 	struct socket *so = arg;
4286 
4287 	if (SOLISTENING(so))
4288 		SOCK_LOCK(so);
4289 	else
4290 		SOCKBUF_LOCK(&so->so_rcv);
4291 }
4292 
4293 static void
4294 so_rdknl_unlock(void *arg)
4295 {
4296 	struct socket *so = arg;
4297 
4298 	if (SOLISTENING(so))
4299 		SOCK_UNLOCK(so);
4300 	else
4301 		SOCKBUF_UNLOCK(&so->so_rcv);
4302 }
4303 
4304 static void
4305 so_rdknl_assert_lock(void *arg, int what)
4306 {
4307 	struct socket *so = arg;
4308 
4309 	if (what == LA_LOCKED) {
4310 		if (SOLISTENING(so))
4311 			SOCK_LOCK_ASSERT(so);
4312 		else
4313 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
4314 	} else {
4315 		if (SOLISTENING(so))
4316 			SOCK_UNLOCK_ASSERT(so);
4317 		else
4318 			SOCKBUF_UNLOCK_ASSERT(&so->so_rcv);
4319 	}
4320 }
4321 
4322 static void
4323 so_wrknl_lock(void *arg)
4324 {
4325 	struct socket *so = arg;
4326 
4327 	if (SOLISTENING(so))
4328 		SOCK_LOCK(so);
4329 	else
4330 		SOCKBUF_LOCK(&so->so_snd);
4331 }
4332 
4333 static void
4334 so_wrknl_unlock(void *arg)
4335 {
4336 	struct socket *so = arg;
4337 
4338 	if (SOLISTENING(so))
4339 		SOCK_UNLOCK(so);
4340 	else
4341 		SOCKBUF_UNLOCK(&so->so_snd);
4342 }
4343 
4344 static void
4345 so_wrknl_assert_lock(void *arg, int what)
4346 {
4347 	struct socket *so = arg;
4348 
4349 	if (what == LA_LOCKED) {
4350 		if (SOLISTENING(so))
4351 			SOCK_LOCK_ASSERT(so);
4352 		else
4353 			SOCKBUF_LOCK_ASSERT(&so->so_snd);
4354 	} else {
4355 		if (SOLISTENING(so))
4356 			SOCK_UNLOCK_ASSERT(so);
4357 		else
4358 			SOCKBUF_UNLOCK_ASSERT(&so->so_snd);
4359 	}
4360 }
4361 
4362 /*
4363  * Create an external-format (``xsocket'') structure using the information in
4364  * the kernel-format socket structure pointed to by so.  This is done to
4365  * reduce the spew of irrelevant information over this interface, to isolate
4366  * user code from changes in the kernel structure, and potentially to provide
4367  * information-hiding if we decide that some of this information should be
4368  * hidden from users.
4369  */
4370 void
4371 sotoxsocket(struct socket *so, struct xsocket *xso)
4372 {
4373 
4374 	bzero(xso, sizeof(*xso));
4375 	xso->xso_len = sizeof *xso;
4376 	xso->xso_so = (uintptr_t)so;
4377 	xso->so_type = so->so_type;
4378 	xso->so_options = so->so_options;
4379 	xso->so_linger = so->so_linger;
4380 	xso->so_state = so->so_state;
4381 	xso->so_pcb = (uintptr_t)so->so_pcb;
4382 	xso->xso_protocol = so->so_proto->pr_protocol;
4383 	xso->xso_family = so->so_proto->pr_domain->dom_family;
4384 	xso->so_timeo = so->so_timeo;
4385 	xso->so_error = so->so_error;
4386 	xso->so_uid = so->so_cred->cr_uid;
4387 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
4388 	if (SOLISTENING(so)) {
4389 		xso->so_qlen = so->sol_qlen;
4390 		xso->so_incqlen = so->sol_incqlen;
4391 		xso->so_qlimit = so->sol_qlimit;
4392 		xso->so_oobmark = 0;
4393 	} else {
4394 		xso->so_state |= so->so_qstate;
4395 		xso->so_qlen = xso->so_incqlen = xso->so_qlimit = 0;
4396 		xso->so_oobmark = so->so_oobmark;
4397 		sbtoxsockbuf(&so->so_snd, &xso->so_snd);
4398 		sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
4399 	}
4400 }
4401 
4402 struct sockbuf *
4403 so_sockbuf_rcv(struct socket *so)
4404 {
4405 
4406 	return (&so->so_rcv);
4407 }
4408 
4409 struct sockbuf *
4410 so_sockbuf_snd(struct socket *so)
4411 {
4412 
4413 	return (&so->so_snd);
4414 }
4415 
4416 int
4417 so_state_get(const struct socket *so)
4418 {
4419 
4420 	return (so->so_state);
4421 }
4422 
4423 void
4424 so_state_set(struct socket *so, int val)
4425 {
4426 
4427 	so->so_state = val;
4428 }
4429 
4430 int
4431 so_options_get(const struct socket *so)
4432 {
4433 
4434 	return (so->so_options);
4435 }
4436 
4437 void
4438 so_options_set(struct socket *so, int val)
4439 {
4440 
4441 	so->so_options = val;
4442 }
4443 
4444 int
4445 so_error_get(const struct socket *so)
4446 {
4447 
4448 	return (so->so_error);
4449 }
4450 
4451 void
4452 so_error_set(struct socket *so, int val)
4453 {
4454 
4455 	so->so_error = val;
4456 }
4457 
4458 int
4459 so_linger_get(const struct socket *so)
4460 {
4461 
4462 	return (so->so_linger);
4463 }
4464 
4465 void
4466 so_linger_set(struct socket *so, int val)
4467 {
4468 
4469 	KASSERT(val >= 0 && val <= USHRT_MAX && val <= (INT_MAX / hz),
4470 	    ("%s: val %d out of range", __func__, val));
4471 
4472 	so->so_linger = val;
4473 }
4474 
4475 struct protosw *
4476 so_protosw_get(const struct socket *so)
4477 {
4478 
4479 	return (so->so_proto);
4480 }
4481 
4482 void
4483 so_protosw_set(struct socket *so, struct protosw *val)
4484 {
4485 
4486 	so->so_proto = val;
4487 }
4488 
4489 void
4490 so_sorwakeup(struct socket *so)
4491 {
4492 
4493 	sorwakeup(so);
4494 }
4495 
4496 void
4497 so_sowwakeup(struct socket *so)
4498 {
4499 
4500 	sowwakeup(so);
4501 }
4502 
4503 void
4504 so_sorwakeup_locked(struct socket *so)
4505 {
4506 
4507 	sorwakeup_locked(so);
4508 }
4509 
4510 void
4511 so_sowwakeup_locked(struct socket *so)
4512 {
4513 
4514 	sowwakeup_locked(so);
4515 }
4516 
4517 void
4518 so_lock(struct socket *so)
4519 {
4520 
4521 	SOCK_LOCK(so);
4522 }
4523 
4524 void
4525 so_unlock(struct socket *so)
4526 {
4527 
4528 	SOCK_UNLOCK(so);
4529 }
4530