xref: /freebsd/sys/kern/uipc_socket.c (revision dcf58f92e2c19a32fc171f763698e711c719badc)
1 /*-
2  * Copyright (c) 1982, 1986, 1988, 1990, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004 The FreeBSD Foundation
5  * Copyright (c) 2004-2008 Robert N. M. Watson
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 4. Neither the name of the University nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  *
32  *	@(#)uipc_socket.c	8.3 (Berkeley) 4/15/94
33  */
34 
35 /*
36  * Comments on the socket life cycle:
37  *
38  * soalloc() sets of socket layer state for a socket, called only by
39  * socreate() and sonewconn().  Socket layer private.
40  *
41  * sodealloc() tears down socket layer state for a socket, called only by
42  * sofree() and sonewconn().  Socket layer private.
43  *
44  * pru_attach() associates protocol layer state with an allocated socket;
45  * called only once, may fail, aborting socket allocation.  This is called
46  * from socreate() and sonewconn().  Socket layer private.
47  *
48  * pru_detach() disassociates protocol layer state from an attached socket,
49  * and will be called exactly once for sockets in which pru_attach() has
50  * been successfully called.  If pru_attach() returned an error,
51  * pru_detach() will not be called.  Socket layer private.
52  *
53  * pru_abort() and pru_close() notify the protocol layer that the last
54  * consumer of a socket is starting to tear down the socket, and that the
55  * protocol should terminate the connection.  Historically, pru_abort() also
56  * detached protocol state from the socket state, but this is no longer the
57  * case.
58  *
59  * socreate() creates a socket and attaches protocol state.  This is a public
60  * interface that may be used by socket layer consumers to create new
61  * sockets.
62  *
63  * sonewconn() creates a socket and attaches protocol state.  This is a
64  * public interface  that may be used by protocols to create new sockets when
65  * a new connection is received and will be available for accept() on a
66  * listen socket.
67  *
68  * soclose() destroys a socket after possibly waiting for it to disconnect.
69  * This is a public interface that socket consumers should use to close and
70  * release a socket when done with it.
71  *
72  * soabort() destroys a socket without waiting for it to disconnect (used
73  * only for incoming connections that are already partially or fully
74  * connected).  This is used internally by the socket layer when clearing
75  * listen socket queues (due to overflow or close on the listen socket), but
76  * is also a public interface protocols may use to abort connections in
77  * their incomplete listen queues should they no longer be required.  Sockets
78  * placed in completed connection listen queues should not be aborted for
79  * reasons described in the comment above the soclose() implementation.  This
80  * is not a general purpose close routine, and except in the specific
81  * circumstances described here, should not be used.
82  *
83  * sofree() will free a socket and its protocol state if all references on
84  * the socket have been released, and is the public interface to attempt to
85  * free a socket when a reference is removed.  This is a socket layer private
86  * interface.
87  *
88  * NOTE: In addition to socreate() and soclose(), which provide a single
89  * socket reference to the consumer to be managed as required, there are two
90  * calls to explicitly manage socket references, soref(), and sorele().
91  * Currently, these are generally required only when transitioning a socket
92  * from a listen queue to a file descriptor, in order to prevent garbage
93  * collection of the socket at an untimely moment.  For a number of reasons,
94  * these interfaces are not preferred, and should be avoided.
95  *
96  * NOTE: With regard to VNETs the general rule is that callers do not set
97  * curvnet. Exceptions to this rule include soabort(), sodisconnect(),
98  * sofree() (and with that sorele(), sotryfree()), as well as sonewconn()
99  * and sorflush(), which are usually called from a pre-set VNET context.
100  * sopoll() currently does not need a VNET context to be set.
101  */
102 
103 #include <sys/cdefs.h>
104 __FBSDID("$FreeBSD$");
105 
106 #include "opt_inet.h"
107 #include "opt_inet6.h"
108 #include "opt_compat.h"
109 
110 #include <sys/param.h>
111 #include <sys/systm.h>
112 #include <sys/fcntl.h>
113 #include <sys/limits.h>
114 #include <sys/lock.h>
115 #include <sys/mac.h>
116 #include <sys/malloc.h>
117 #include <sys/mbuf.h>
118 #include <sys/mutex.h>
119 #include <sys/domain.h>
120 #include <sys/file.h>			/* for struct knote */
121 #include <sys/hhook.h>
122 #include <sys/kernel.h>
123 #include <sys/khelp.h>
124 #include <sys/event.h>
125 #include <sys/eventhandler.h>
126 #include <sys/poll.h>
127 #include <sys/proc.h>
128 #include <sys/protosw.h>
129 #include <sys/socket.h>
130 #include <sys/socketvar.h>
131 #include <sys/resourcevar.h>
132 #include <net/route.h>
133 #include <sys/signalvar.h>
134 #include <sys/stat.h>
135 #include <sys/sx.h>
136 #include <sys/sysctl.h>
137 #include <sys/uio.h>
138 #include <sys/jail.h>
139 #include <sys/syslog.h>
140 #include <netinet/in.h>
141 
142 #include <net/vnet.h>
143 
144 #include <security/mac/mac_framework.h>
145 
146 #include <vm/uma.h>
147 
148 #ifdef COMPAT_FREEBSD32
149 #include <sys/mount.h>
150 #include <sys/sysent.h>
151 #include <compat/freebsd32/freebsd32.h>
152 #endif
153 
154 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
155 		    int flags);
156 
157 static void	filt_sordetach(struct knote *kn);
158 static int	filt_soread(struct knote *kn, long hint);
159 static void	filt_sowdetach(struct knote *kn);
160 static int	filt_sowrite(struct knote *kn, long hint);
161 static int	filt_solisten(struct knote *kn, long hint);
162 static int inline hhook_run_socket(struct socket *so, void *hctx, int32_t h_id);
163 fo_kqfilter_t	soo_kqfilter;
164 
165 static struct filterops solisten_filtops = {
166 	.f_isfd = 1,
167 	.f_detach = filt_sordetach,
168 	.f_event = filt_solisten,
169 };
170 static struct filterops soread_filtops = {
171 	.f_isfd = 1,
172 	.f_detach = filt_sordetach,
173 	.f_event = filt_soread,
174 };
175 static struct filterops sowrite_filtops = {
176 	.f_isfd = 1,
177 	.f_detach = filt_sowdetach,
178 	.f_event = filt_sowrite,
179 };
180 
181 so_gen_t	so_gencnt;	/* generation count for sockets */
182 
183 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
184 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
185 
186 #define	VNET_SO_ASSERT(so)						\
187 	VNET_ASSERT(curvnet != NULL,					\
188 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
189 
190 VNET_DEFINE(struct hhook_head *, socket_hhh[HHOOK_SOCKET_LAST + 1]);
191 #define	V_socket_hhh		VNET(socket_hhh)
192 
193 /*
194  * Limit on the number of connections in the listen queue waiting
195  * for accept(2).
196  * NB: The orginal sysctl somaxconn is still available but hidden
197  * to prevent confusion about the actual purpose of this number.
198  */
199 static int somaxconn = SOMAXCONN;
200 
201 static int
202 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
203 {
204 	int error;
205 	int val;
206 
207 	val = somaxconn;
208 	error = sysctl_handle_int(oidp, &val, 0, req);
209 	if (error || !req->newptr )
210 		return (error);
211 
212 	if (val < 1 || val > USHRT_MAX)
213 		return (EINVAL);
214 
215 	somaxconn = val;
216 	return (0);
217 }
218 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
219     0, sizeof(int), sysctl_somaxconn, "I",
220     "Maximum listen socket pending connection accept queue size");
221 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
222     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
223     0, sizeof(int), sysctl_somaxconn, "I",
224     "Maximum listen socket pending connection accept queue size (compat)");
225 
226 static int numopensockets;
227 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
228     &numopensockets, 0, "Number of open sockets");
229 
230 /*
231  * accept_mtx locks down per-socket fields relating to accept queues.  See
232  * socketvar.h for an annotation of the protected fields of struct socket.
233  */
234 struct mtx accept_mtx;
235 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
236 
237 /*
238  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
239  * so_gencnt field.
240  */
241 static struct mtx so_global_mtx;
242 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
243 
244 /*
245  * General IPC sysctl name space, used by sockets and a variety of other IPC
246  * types.
247  */
248 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
249 
250 /*
251  * Initialize the socket subsystem and set up the socket
252  * memory allocator.
253  */
254 static uma_zone_t socket_zone;
255 int	maxsockets;
256 
257 static void
258 socket_zone_change(void *tag)
259 {
260 
261 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
262 }
263 
264 static void
265 socket_hhook_register(int subtype)
266 {
267 
268 	if (hhook_head_register(HHOOK_TYPE_SOCKET, subtype,
269 	    &V_socket_hhh[subtype],
270 	    HHOOK_NOWAIT|HHOOK_HEADISINVNET) != 0)
271 		printf("%s: WARNING: unable to register hook\n", __func__);
272 }
273 
274 static void
275 socket_hhook_deregister(int subtype)
276 {
277 
278 	if (hhook_head_deregister(V_socket_hhh[subtype]) != 0)
279 		printf("%s: WARNING: unable to deregister hook\n", __func__);
280 }
281 
282 static void
283 socket_init(void *tag)
284 {
285 
286 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
287 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
288 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
289 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
290 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
291 	    EVENTHANDLER_PRI_FIRST);
292 }
293 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
294 
295 static void
296 socket_vnet_init(const void *unused __unused)
297 {
298 	int i;
299 
300 	/* We expect a contiguous range */
301 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
302 		socket_hhook_register(i);
303 }
304 VNET_SYSINIT(socket_vnet_init, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
305     socket_vnet_init, NULL);
306 
307 static void
308 socket_vnet_uninit(const void *unused __unused)
309 {
310 	int i;
311 
312 	for (i = 0; i <= HHOOK_SOCKET_LAST; i++)
313 		socket_hhook_deregister(i);
314 }
315 VNET_SYSUNINIT(socket_vnet_uninit, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY,
316     socket_vnet_uninit, NULL);
317 
318 /*
319  * Initialise maxsockets.  This SYSINIT must be run after
320  * tunable_mbinit().
321  */
322 static void
323 init_maxsockets(void *ignored)
324 {
325 
326 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
327 	maxsockets = imax(maxsockets, maxfiles);
328 }
329 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
330 
331 /*
332  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
333  * of the change so that they can update their dependent limits as required.
334  */
335 static int
336 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
337 {
338 	int error, newmaxsockets;
339 
340 	newmaxsockets = maxsockets;
341 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
342 	if (error == 0 && req->newptr) {
343 		if (newmaxsockets > maxsockets &&
344 		    newmaxsockets <= maxfiles) {
345 			maxsockets = newmaxsockets;
346 			EVENTHANDLER_INVOKE(maxsockets_change);
347 		} else
348 			error = EINVAL;
349 	}
350 	return (error);
351 }
352 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
353     &maxsockets, 0, sysctl_maxsockets, "IU",
354     "Maximum number of sockets avaliable");
355 
356 /*
357  * Socket operation routines.  These routines are called by the routines in
358  * sys_socket.c or from a system process, and implement the semantics of
359  * socket operations by switching out to the protocol specific routines.
360  */
361 
362 /*
363  * Get a socket structure from our zone, and initialize it.  Note that it
364  * would probably be better to allocate socket and PCB at the same time, but
365  * I'm not convinced that all the protocols can be easily modified to do
366  * this.
367  *
368  * soalloc() returns a socket with a ref count of 0.
369  */
370 static struct socket *
371 soalloc(struct vnet *vnet)
372 {
373 	struct socket *so;
374 
375 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
376 	if (so == NULL)
377 		return (NULL);
378 #ifdef MAC
379 	if (mac_socket_init(so, M_NOWAIT) != 0) {
380 		uma_zfree(socket_zone, so);
381 		return (NULL);
382 	}
383 #endif
384 	if (khelp_init_osd(HELPER_CLASS_SOCKET, &so->osd)) {
385 		uma_zfree(socket_zone, so);
386 		return (NULL);
387 	}
388 
389 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
390 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
391 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
392 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
393 	TAILQ_INIT(&so->so_aiojobq);
394 #ifdef VIMAGE
395 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
396 	    __func__, __LINE__, so));
397 	so->so_vnet = vnet;
398 #endif
399 	/* We shouldn't need the so_global_mtx */
400 	if (hhook_run_socket(so, NULL, HHOOK_SOCKET_CREATE)) {
401 		/* Do we need more comprehensive error returns? */
402 		uma_zfree(socket_zone, so);
403 		return (NULL);
404 	}
405 	mtx_lock(&so_global_mtx);
406 	so->so_gencnt = ++so_gencnt;
407 	++numopensockets;
408 #ifdef VIMAGE
409 	vnet->vnet_sockcnt++;
410 #endif
411 	mtx_unlock(&so_global_mtx);
412 
413 	return (so);
414 }
415 
416 /*
417  * Free the storage associated with a socket at the socket layer, tear down
418  * locks, labels, etc.  All protocol state is assumed already to have been
419  * torn down (and possibly never set up) by the caller.
420  */
421 static void
422 sodealloc(struct socket *so)
423 {
424 
425 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
426 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
427 
428 	mtx_lock(&so_global_mtx);
429 	so->so_gencnt = ++so_gencnt;
430 	--numopensockets;	/* Could be below, but faster here. */
431 #ifdef VIMAGE
432 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
433 	    __func__, __LINE__, so));
434 	so->so_vnet->vnet_sockcnt--;
435 #endif
436 	mtx_unlock(&so_global_mtx);
437 	if (so->so_rcv.sb_hiwat)
438 		(void)chgsbsize(so->so_cred->cr_uidinfo,
439 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
440 	if (so->so_snd.sb_hiwat)
441 		(void)chgsbsize(so->so_cred->cr_uidinfo,
442 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
443 	/* remove acccept filter if one is present. */
444 	if (so->so_accf != NULL)
445 		do_setopt_accept_filter(so, NULL);
446 #ifdef MAC
447 	mac_socket_destroy(so);
448 #endif
449 	hhook_run_socket(so, NULL, HHOOK_SOCKET_CLOSE);
450 
451 	crfree(so->so_cred);
452 	khelp_destroy_osd(&so->osd);
453 	sx_destroy(&so->so_snd.sb_sx);
454 	sx_destroy(&so->so_rcv.sb_sx);
455 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
456 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
457 	uma_zfree(socket_zone, so);
458 }
459 
460 /*
461  * socreate returns a socket with a ref count of 1.  The socket should be
462  * closed with soclose().
463  */
464 int
465 socreate(int dom, struct socket **aso, int type, int proto,
466     struct ucred *cred, struct thread *td)
467 {
468 	struct protosw *prp;
469 	struct socket *so;
470 	int error;
471 
472 	if (proto)
473 		prp = pffindproto(dom, proto, type);
474 	else
475 		prp = pffindtype(dom, type);
476 
477 	if (prp == NULL) {
478 		/* No support for domain. */
479 		if (pffinddomain(dom) == NULL)
480 			return (EAFNOSUPPORT);
481 		/* No support for socket type. */
482 		if (proto == 0 && type != 0)
483 			return (EPROTOTYPE);
484 		return (EPROTONOSUPPORT);
485 	}
486 	if (prp->pr_usrreqs->pru_attach == NULL ||
487 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
488 		return (EPROTONOSUPPORT);
489 
490 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
491 		return (EPROTONOSUPPORT);
492 
493 	if (prp->pr_type != type)
494 		return (EPROTOTYPE);
495 	so = soalloc(CRED_TO_VNET(cred));
496 	if (so == NULL)
497 		return (ENOBUFS);
498 
499 	TAILQ_INIT(&so->so_incomp);
500 	TAILQ_INIT(&so->so_comp);
501 	so->so_type = type;
502 	so->so_cred = crhold(cred);
503 	if ((prp->pr_domain->dom_family == PF_INET) ||
504 	    (prp->pr_domain->dom_family == PF_INET6) ||
505 	    (prp->pr_domain->dom_family == PF_ROUTE))
506 		so->so_fibnum = td->td_proc->p_fibnum;
507 	else
508 		so->so_fibnum = 0;
509 	so->so_proto = prp;
510 #ifdef MAC
511 	mac_socket_create(cred, so);
512 #endif
513 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
514 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
515 	so->so_count = 1;
516 	/*
517 	 * Auto-sizing of socket buffers is managed by the protocols and
518 	 * the appropriate flags must be set in the pru_attach function.
519 	 */
520 	CURVNET_SET(so->so_vnet);
521 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
522 	CURVNET_RESTORE();
523 	if (error) {
524 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
525 		    so->so_count));
526 		so->so_count = 0;
527 		sodealloc(so);
528 		return (error);
529 	}
530 	*aso = so;
531 	return (0);
532 }
533 
534 #ifdef REGRESSION
535 static int regression_sonewconn_earlytest = 1;
536 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
537     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
538 #endif
539 
540 /*
541  * When an attempt at a new connection is noted on a socket which accepts
542  * connections, sonewconn is called.  If the connection is possible (subject
543  * to space constraints, etc.) then we allocate a new structure, propoerly
544  * linked into the data structure of the original socket, and return this.
545  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
546  *
547  * Note: the ref count on the socket is 0 on return.
548  */
549 struct socket *
550 sonewconn(struct socket *head, int connstatus)
551 {
552 	static struct timeval lastover;
553 	static struct timeval overinterval = { 60, 0 };
554 	static int overcount;
555 
556 	struct socket *so;
557 	int over;
558 
559 	ACCEPT_LOCK();
560 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
561 	ACCEPT_UNLOCK();
562 #ifdef REGRESSION
563 	if (regression_sonewconn_earlytest && over) {
564 #else
565 	if (over) {
566 #endif
567 		overcount++;
568 
569 		if (ratecheck(&lastover, &overinterval)) {
570 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
571 			    "%i already in queue awaiting acceptance "
572 			    "(%d occurrences)\n",
573 			    __func__, head->so_pcb, head->so_qlen, overcount);
574 
575 			overcount = 0;
576 		}
577 
578 		return (NULL);
579 	}
580 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
581 	    __func__, __LINE__, head));
582 	so = soalloc(head->so_vnet);
583 	if (so == NULL) {
584 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
585 		    "limit reached or out of memory\n",
586 		    __func__, head->so_pcb);
587 		return (NULL);
588 	}
589 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
590 		connstatus = 0;
591 	so->so_head = head;
592 	so->so_type = head->so_type;
593 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
594 	so->so_linger = head->so_linger;
595 	so->so_state = head->so_state | SS_NOFDREF;
596 	so->so_fibnum = head->so_fibnum;
597 	so->so_proto = head->so_proto;
598 	so->so_cred = crhold(head->so_cred);
599 #ifdef MAC
600 	mac_socket_newconn(head, so);
601 #endif
602 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
603 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
604 	VNET_SO_ASSERT(head);
605 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
606 		sodealloc(so);
607 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
608 		    __func__, head->so_pcb);
609 		return (NULL);
610 	}
611 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
612 		sodealloc(so);
613 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
614 		    __func__, head->so_pcb);
615 		return (NULL);
616 	}
617 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
618 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
619 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
620 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
621 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
622 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
623 	so->so_state |= connstatus;
624 	ACCEPT_LOCK();
625 	/*
626 	 * The accept socket may be tearing down but we just
627 	 * won a race on the ACCEPT_LOCK.
628 	 * However, if sctp_peeloff() is called on a 1-to-many
629 	 * style socket, the SO_ACCEPTCONN doesn't need to be set.
630 	 */
631 	if (!(head->so_options & SO_ACCEPTCONN) &&
632 	    ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
633 	     (head->so_type != SOCK_SEQPACKET))) {
634 		SOCK_LOCK(so);
635 		so->so_head = NULL;
636 		sofree(so);		/* NB: returns ACCEPT_UNLOCK'ed. */
637 		return (NULL);
638 	}
639 	if (connstatus) {
640 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
641 		so->so_qstate |= SQ_COMP;
642 		head->so_qlen++;
643 	} else {
644 		/*
645 		 * Keep removing sockets from the head until there's room for
646 		 * us to insert on the tail.  In pre-locking revisions, this
647 		 * was a simple if(), but as we could be racing with other
648 		 * threads and soabort() requires dropping locks, we must
649 		 * loop waiting for the condition to be true.
650 		 */
651 		while (head->so_incqlen > head->so_qlimit) {
652 			struct socket *sp;
653 			sp = TAILQ_FIRST(&head->so_incomp);
654 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
655 			head->so_incqlen--;
656 			sp->so_qstate &= ~SQ_INCOMP;
657 			sp->so_head = NULL;
658 			ACCEPT_UNLOCK();
659 			soabort(sp);
660 			ACCEPT_LOCK();
661 		}
662 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
663 		so->so_qstate |= SQ_INCOMP;
664 		head->so_incqlen++;
665 	}
666 	ACCEPT_UNLOCK();
667 	if (connstatus) {
668 		sorwakeup(head);
669 		wakeup_one(&head->so_timeo);
670 	}
671 	return (so);
672 }
673 
674 int
675 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
676 {
677 	int error;
678 
679 	CURVNET_SET(so->so_vnet);
680 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
681 	CURVNET_RESTORE();
682 	return (error);
683 }
684 
685 int
686 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
687 {
688 	int error;
689 
690 	CURVNET_SET(so->so_vnet);
691 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
692 	CURVNET_RESTORE();
693 	return (error);
694 }
695 
696 /*
697  * solisten() transitions a socket from a non-listening state to a listening
698  * state, but can also be used to update the listen queue depth on an
699  * existing listen socket.  The protocol will call back into the sockets
700  * layer using solisten_proto_check() and solisten_proto() to check and set
701  * socket-layer listen state.  Call backs are used so that the protocol can
702  * acquire both protocol and socket layer locks in whatever order is required
703  * by the protocol.
704  *
705  * Protocol implementors are advised to hold the socket lock across the
706  * socket-layer test and set to avoid races at the socket layer.
707  */
708 int
709 solisten(struct socket *so, int backlog, struct thread *td)
710 {
711 	int error;
712 
713 	CURVNET_SET(so->so_vnet);
714 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
715 	CURVNET_RESTORE();
716 	return (error);
717 }
718 
719 int
720 solisten_proto_check(struct socket *so)
721 {
722 
723 	SOCK_LOCK_ASSERT(so);
724 
725 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
726 	    SS_ISDISCONNECTING))
727 		return (EINVAL);
728 	return (0);
729 }
730 
731 void
732 solisten_proto(struct socket *so, int backlog)
733 {
734 
735 	SOCK_LOCK_ASSERT(so);
736 
737 	if (backlog < 0 || backlog > somaxconn)
738 		backlog = somaxconn;
739 	so->so_qlimit = backlog;
740 	so->so_options |= SO_ACCEPTCONN;
741 }
742 
743 /*
744  * Evaluate the reference count and named references on a socket; if no
745  * references remain, free it.  This should be called whenever a reference is
746  * released, such as in sorele(), but also when named reference flags are
747  * cleared in socket or protocol code.
748  *
749  * sofree() will free the socket if:
750  *
751  * - There are no outstanding file descriptor references or related consumers
752  *   (so_count == 0).
753  *
754  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
755  *
756  * - The protocol does not have an outstanding strong reference on the socket
757  *   (SS_PROTOREF).
758  *
759  * - The socket is not in a completed connection queue, so a process has been
760  *   notified that it is present.  If it is removed, the user process may
761  *   block in accept() despite select() saying the socket was ready.
762  */
763 void
764 sofree(struct socket *so)
765 {
766 	struct protosw *pr = so->so_proto;
767 	struct socket *head;
768 
769 	ACCEPT_LOCK_ASSERT();
770 	SOCK_LOCK_ASSERT(so);
771 
772 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
773 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
774 		SOCK_UNLOCK(so);
775 		ACCEPT_UNLOCK();
776 		return;
777 	}
778 
779 	head = so->so_head;
780 	if (head != NULL) {
781 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
782 		    (so->so_qstate & SQ_INCOMP) != 0,
783 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
784 		    "SQ_INCOMP"));
785 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
786 		    (so->so_qstate & SQ_INCOMP) == 0,
787 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
788 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
789 		head->so_incqlen--;
790 		so->so_qstate &= ~SQ_INCOMP;
791 		so->so_head = NULL;
792 	}
793 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
794 	    (so->so_qstate & SQ_INCOMP) == 0,
795 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
796 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
797 	if (so->so_options & SO_ACCEPTCONN) {
798 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
799 		    ("sofree: so_comp populated"));
800 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
801 		    ("sofree: so_incomp populated"));
802 	}
803 	SOCK_UNLOCK(so);
804 	ACCEPT_UNLOCK();
805 
806 	VNET_SO_ASSERT(so);
807 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
808 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
809 	if (pr->pr_usrreqs->pru_detach != NULL)
810 		(*pr->pr_usrreqs->pru_detach)(so);
811 
812 	/*
813 	 * From this point on, we assume that no other references to this
814 	 * socket exist anywhere else in the stack.  Therefore, no locks need
815 	 * to be acquired or held.
816 	 *
817 	 * We used to do a lot of socket buffer and socket locking here, as
818 	 * well as invoke sorflush() and perform wakeups.  The direct call to
819 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
820 	 * necessary from sorflush().
821 	 *
822 	 * Notice that the socket buffer and kqueue state are torn down
823 	 * before calling pru_detach.  This means that protocols shold not
824 	 * assume they can perform socket wakeups, etc, in their detach code.
825 	 */
826 	sbdestroy(&so->so_snd, so);
827 	sbdestroy(&so->so_rcv, so);
828 	seldrain(&so->so_snd.sb_sel);
829 	seldrain(&so->so_rcv.sb_sel);
830 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
831 	knlist_destroy(&so->so_snd.sb_sel.si_note);
832 	sodealloc(so);
833 }
834 
835 /*
836  * Close a socket on last file table reference removal.  Initiate disconnect
837  * if connected.  Free socket when disconnect complete.
838  *
839  * This function will sorele() the socket.  Note that soclose() may be called
840  * prior to the ref count reaching zero.  The actual socket structure will
841  * not be freed until the ref count reaches zero.
842  */
843 int
844 soclose(struct socket *so)
845 {
846 	int error = 0;
847 
848 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
849 
850 	CURVNET_SET(so->so_vnet);
851 	funsetown(&so->so_sigio);
852 	if (so->so_state & SS_ISCONNECTED) {
853 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
854 			error = sodisconnect(so);
855 			if (error) {
856 				if (error == ENOTCONN)
857 					error = 0;
858 				goto drop;
859 			}
860 		}
861 		if (so->so_options & SO_LINGER) {
862 			if ((so->so_state & SS_ISDISCONNECTING) &&
863 			    (so->so_state & SS_NBIO))
864 				goto drop;
865 			while (so->so_state & SS_ISCONNECTED) {
866 				error = tsleep(&so->so_timeo,
867 				    PSOCK | PCATCH, "soclos",
868 				    so->so_linger * hz);
869 				if (error)
870 					break;
871 			}
872 		}
873 	}
874 
875 drop:
876 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
877 		(*so->so_proto->pr_usrreqs->pru_close)(so);
878 	ACCEPT_LOCK();
879 	if (so->so_options & SO_ACCEPTCONN) {
880 		struct socket *sp;
881 		/*
882 		 * Prevent new additions to the accept queues due
883 		 * to ACCEPT_LOCK races while we are draining them.
884 		 */
885 		so->so_options &= ~SO_ACCEPTCONN;
886 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
887 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
888 			so->so_incqlen--;
889 			sp->so_qstate &= ~SQ_INCOMP;
890 			sp->so_head = NULL;
891 			ACCEPT_UNLOCK();
892 			soabort(sp);
893 			ACCEPT_LOCK();
894 		}
895 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
896 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
897 			so->so_qlen--;
898 			sp->so_qstate &= ~SQ_COMP;
899 			sp->so_head = NULL;
900 			ACCEPT_UNLOCK();
901 			soabort(sp);
902 			ACCEPT_LOCK();
903 		}
904 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
905 		    ("%s: so_comp populated", __func__));
906 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
907 		    ("%s: so_incomp populated", __func__));
908 	}
909 	SOCK_LOCK(so);
910 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
911 	so->so_state |= SS_NOFDREF;
912 	sorele(so);			/* NB: Returns with ACCEPT_UNLOCK(). */
913 	CURVNET_RESTORE();
914 	return (error);
915 }
916 
917 /*
918  * soabort() is used to abruptly tear down a connection, such as when a
919  * resource limit is reached (listen queue depth exceeded), or if a listen
920  * socket is closed while there are sockets waiting to be accepted.
921  *
922  * This interface is tricky, because it is called on an unreferenced socket,
923  * and must be called only by a thread that has actually removed the socket
924  * from the listen queue it was on, or races with other threads are risked.
925  *
926  * This interface will call into the protocol code, so must not be called
927  * with any socket locks held.  Protocols do call it while holding their own
928  * recursible protocol mutexes, but this is something that should be subject
929  * to review in the future.
930  */
931 void
932 soabort(struct socket *so)
933 {
934 
935 	/*
936 	 * In as much as is possible, assert that no references to this
937 	 * socket are held.  This is not quite the same as asserting that the
938 	 * current thread is responsible for arranging for no references, but
939 	 * is as close as we can get for now.
940 	 */
941 	KASSERT(so->so_count == 0, ("soabort: so_count"));
942 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
943 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
944 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
945 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
946 	VNET_SO_ASSERT(so);
947 
948 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
949 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
950 	ACCEPT_LOCK();
951 	SOCK_LOCK(so);
952 	sofree(so);
953 }
954 
955 int
956 soaccept(struct socket *so, struct sockaddr **nam)
957 {
958 	int error;
959 
960 	SOCK_LOCK(so);
961 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
962 	so->so_state &= ~SS_NOFDREF;
963 	SOCK_UNLOCK(so);
964 
965 	CURVNET_SET(so->so_vnet);
966 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
967 	CURVNET_RESTORE();
968 	return (error);
969 }
970 
971 int
972 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
973 {
974 
975 	return (soconnectat(AT_FDCWD, so, nam, td));
976 }
977 
978 int
979 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
980 {
981 	int error;
982 
983 	if (so->so_options & SO_ACCEPTCONN)
984 		return (EOPNOTSUPP);
985 
986 	CURVNET_SET(so->so_vnet);
987 	/*
988 	 * If protocol is connection-based, can only connect once.
989 	 * Otherwise, if connected, try to disconnect first.  This allows
990 	 * user to disconnect by connecting to, e.g., a null address.
991 	 */
992 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
993 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
994 	    (error = sodisconnect(so)))) {
995 		error = EISCONN;
996 	} else {
997 		/*
998 		 * Prevent accumulated error from previous connection from
999 		 * biting us.
1000 		 */
1001 		so->so_error = 0;
1002 		if (fd == AT_FDCWD) {
1003 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
1004 			    nam, td);
1005 		} else {
1006 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
1007 			    so, nam, td);
1008 		}
1009 	}
1010 	CURVNET_RESTORE();
1011 
1012 	return (error);
1013 }
1014 
1015 int
1016 soconnect2(struct socket *so1, struct socket *so2)
1017 {
1018 	int error;
1019 
1020 	CURVNET_SET(so1->so_vnet);
1021 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
1022 	CURVNET_RESTORE();
1023 	return (error);
1024 }
1025 
1026 int
1027 sodisconnect(struct socket *so)
1028 {
1029 	int error;
1030 
1031 	if ((so->so_state & SS_ISCONNECTED) == 0)
1032 		return (ENOTCONN);
1033 	if (so->so_state & SS_ISDISCONNECTING)
1034 		return (EALREADY);
1035 	VNET_SO_ASSERT(so);
1036 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
1037 	return (error);
1038 }
1039 
1040 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
1041 
1042 int
1043 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
1044     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1045 {
1046 	long space;
1047 	ssize_t resid;
1048 	int clen = 0, error, dontroute;
1049 
1050 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
1051 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
1052 	    ("sosend_dgram: !PR_ATOMIC"));
1053 
1054 	if (uio != NULL)
1055 		resid = uio->uio_resid;
1056 	else
1057 		resid = top->m_pkthdr.len;
1058 	/*
1059 	 * In theory resid should be unsigned.  However, space must be
1060 	 * signed, as it might be less than 0 if we over-committed, and we
1061 	 * must use a signed comparison of space and resid.  On the other
1062 	 * hand, a negative resid causes us to loop sending 0-length
1063 	 * segments to the protocol.
1064 	 */
1065 	if (resid < 0) {
1066 		error = EINVAL;
1067 		goto out;
1068 	}
1069 
1070 	dontroute =
1071 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1072 	if (td != NULL)
1073 		td->td_ru.ru_msgsnd++;
1074 	if (control != NULL)
1075 		clen = control->m_len;
1076 
1077 	SOCKBUF_LOCK(&so->so_snd);
1078 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1079 		SOCKBUF_UNLOCK(&so->so_snd);
1080 		error = EPIPE;
1081 		goto out;
1082 	}
1083 	if (so->so_error) {
1084 		error = so->so_error;
1085 		so->so_error = 0;
1086 		SOCKBUF_UNLOCK(&so->so_snd);
1087 		goto out;
1088 	}
1089 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1090 		/*
1091 		 * `sendto' and `sendmsg' is allowed on a connection-based
1092 		 * socket if it supports implied connect.  Return ENOTCONN if
1093 		 * not connected and no address is supplied.
1094 		 */
1095 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1096 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1097 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1098 			    !(resid == 0 && clen != 0)) {
1099 				SOCKBUF_UNLOCK(&so->so_snd);
1100 				error = ENOTCONN;
1101 				goto out;
1102 			}
1103 		} else if (addr == NULL) {
1104 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1105 				error = ENOTCONN;
1106 			else
1107 				error = EDESTADDRREQ;
1108 			SOCKBUF_UNLOCK(&so->so_snd);
1109 			goto out;
1110 		}
1111 	}
1112 
1113 	/*
1114 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1115 	 * problem and need fixing.
1116 	 */
1117 	space = sbspace(&so->so_snd);
1118 	if (flags & MSG_OOB)
1119 		space += 1024;
1120 	space -= clen;
1121 	SOCKBUF_UNLOCK(&so->so_snd);
1122 	if (resid > space) {
1123 		error = EMSGSIZE;
1124 		goto out;
1125 	}
1126 	if (uio == NULL) {
1127 		resid = 0;
1128 		if (flags & MSG_EOR)
1129 			top->m_flags |= M_EOR;
1130 	} else {
1131 		/*
1132 		 * Copy the data from userland into a mbuf chain.
1133 		 * If no data is to be copied in, a single empty mbuf
1134 		 * is returned.
1135 		 */
1136 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1137 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1138 		if (top == NULL) {
1139 			error = EFAULT;	/* only possible error */
1140 			goto out;
1141 		}
1142 		space -= resid - uio->uio_resid;
1143 		resid = uio->uio_resid;
1144 	}
1145 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1146 	/*
1147 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1148 	 * than with.
1149 	 */
1150 	if (dontroute) {
1151 		SOCK_LOCK(so);
1152 		so->so_options |= SO_DONTROUTE;
1153 		SOCK_UNLOCK(so);
1154 	}
1155 	/*
1156 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1157 	 * of date.  We could have recieved a reset packet in an interrupt or
1158 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1159 	 * probably recheck again inside the locking protection here, but
1160 	 * there are probably other places that this also happens.  We must
1161 	 * rethink this.
1162 	 */
1163 	VNET_SO_ASSERT(so);
1164 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1165 	    (flags & MSG_OOB) ? PRUS_OOB :
1166 	/*
1167 	 * If the user set MSG_EOF, the protocol understands this flag and
1168 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1169 	 */
1170 	    ((flags & MSG_EOF) &&
1171 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1172 	     (resid <= 0)) ?
1173 		PRUS_EOF :
1174 		/* If there is more to send set PRUS_MORETOCOME */
1175 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1176 		top, addr, control, td);
1177 	if (dontroute) {
1178 		SOCK_LOCK(so);
1179 		so->so_options &= ~SO_DONTROUTE;
1180 		SOCK_UNLOCK(so);
1181 	}
1182 	clen = 0;
1183 	control = NULL;
1184 	top = NULL;
1185 out:
1186 	if (top != NULL)
1187 		m_freem(top);
1188 	if (control != NULL)
1189 		m_freem(control);
1190 	return (error);
1191 }
1192 
1193 /*
1194  * Send on a socket.  If send must go all at once and message is larger than
1195  * send buffering, then hard error.  Lock against other senders.  If must go
1196  * all at once and not enough room now, then inform user that this would
1197  * block and do nothing.  Otherwise, if nonblocking, send as much as
1198  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1199  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1200  * in mbuf chain must be small enough to send all at once.
1201  *
1202  * Returns nonzero on error, timeout or signal; callers must check for short
1203  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1204  * on return.
1205  */
1206 int
1207 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1208     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1209 {
1210 	long space;
1211 	ssize_t resid;
1212 	int clen = 0, error, dontroute;
1213 	int atomic = sosendallatonce(so) || top;
1214 
1215 	if (uio != NULL)
1216 		resid = uio->uio_resid;
1217 	else
1218 		resid = top->m_pkthdr.len;
1219 	/*
1220 	 * In theory resid should be unsigned.  However, space must be
1221 	 * signed, as it might be less than 0 if we over-committed, and we
1222 	 * must use a signed comparison of space and resid.  On the other
1223 	 * hand, a negative resid causes us to loop sending 0-length
1224 	 * segments to the protocol.
1225 	 *
1226 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1227 	 * type sockets since that's an error.
1228 	 */
1229 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1230 		error = EINVAL;
1231 		goto out;
1232 	}
1233 
1234 	dontroute =
1235 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1236 	    (so->so_proto->pr_flags & PR_ATOMIC);
1237 	if (td != NULL)
1238 		td->td_ru.ru_msgsnd++;
1239 	if (control != NULL)
1240 		clen = control->m_len;
1241 
1242 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1243 	if (error)
1244 		goto out;
1245 
1246 restart:
1247 	do {
1248 		SOCKBUF_LOCK(&so->so_snd);
1249 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1250 			SOCKBUF_UNLOCK(&so->so_snd);
1251 			error = EPIPE;
1252 			goto release;
1253 		}
1254 		if (so->so_error) {
1255 			error = so->so_error;
1256 			so->so_error = 0;
1257 			SOCKBUF_UNLOCK(&so->so_snd);
1258 			goto release;
1259 		}
1260 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1261 			/*
1262 			 * `sendto' and `sendmsg' is allowed on a connection-
1263 			 * based socket if it supports implied connect.
1264 			 * Return ENOTCONN if not connected and no address is
1265 			 * supplied.
1266 			 */
1267 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1268 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1269 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1270 				    !(resid == 0 && clen != 0)) {
1271 					SOCKBUF_UNLOCK(&so->so_snd);
1272 					error = ENOTCONN;
1273 					goto release;
1274 				}
1275 			} else if (addr == NULL) {
1276 				SOCKBUF_UNLOCK(&so->so_snd);
1277 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1278 					error = ENOTCONN;
1279 				else
1280 					error = EDESTADDRREQ;
1281 				goto release;
1282 			}
1283 		}
1284 		space = sbspace(&so->so_snd);
1285 		if (flags & MSG_OOB)
1286 			space += 1024;
1287 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1288 		    clen > so->so_snd.sb_hiwat) {
1289 			SOCKBUF_UNLOCK(&so->so_snd);
1290 			error = EMSGSIZE;
1291 			goto release;
1292 		}
1293 		if (space < resid + clen &&
1294 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1295 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1296 				SOCKBUF_UNLOCK(&so->so_snd);
1297 				error = EWOULDBLOCK;
1298 				goto release;
1299 			}
1300 			error = sbwait(&so->so_snd);
1301 			SOCKBUF_UNLOCK(&so->so_snd);
1302 			if (error)
1303 				goto release;
1304 			goto restart;
1305 		}
1306 		SOCKBUF_UNLOCK(&so->so_snd);
1307 		space -= clen;
1308 		do {
1309 			if (uio == NULL) {
1310 				resid = 0;
1311 				if (flags & MSG_EOR)
1312 					top->m_flags |= M_EOR;
1313 			} else if (resid > 0) {
1314 				/*
1315 				 * Copy the data from userland into a mbuf
1316 				 * chain.  If no data is to be copied in,
1317 				 * a single empty mbuf is returned.
1318 				 */
1319 				top = m_uiotombuf(uio, M_WAITOK, space,
1320 				    (atomic ? max_hdr : 0),
1321 				    (atomic ? M_PKTHDR : 0) |
1322 				    ((flags & MSG_EOR) ? M_EOR : 0));
1323 				if (top == NULL) {
1324 					error = EFAULT; /* only possible error */
1325 					goto release;
1326 				}
1327 				space -= resid - uio->uio_resid;
1328 				resid = uio->uio_resid;
1329 			}
1330 			if (dontroute) {
1331 				SOCK_LOCK(so);
1332 				so->so_options |= SO_DONTROUTE;
1333 				SOCK_UNLOCK(so);
1334 			}
1335 			/*
1336 			 * XXX all the SBS_CANTSENDMORE checks previously
1337 			 * done could be out of date.  We could have recieved
1338 			 * a reset packet in an interrupt or maybe we slept
1339 			 * while doing page faults in uiomove() etc.  We
1340 			 * could probably recheck again inside the locking
1341 			 * protection here, but there are probably other
1342 			 * places that this also happens.  We must rethink
1343 			 * this.
1344 			 */
1345 			VNET_SO_ASSERT(so);
1346 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1347 			    (flags & MSG_OOB) ? PRUS_OOB :
1348 			/*
1349 			 * If the user set MSG_EOF, the protocol understands
1350 			 * this flag and nothing left to send then use
1351 			 * PRU_SEND_EOF instead of PRU_SEND.
1352 			 */
1353 			    ((flags & MSG_EOF) &&
1354 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1355 			     (resid <= 0)) ?
1356 				PRUS_EOF :
1357 			/* If there is more to send set PRUS_MORETOCOME. */
1358 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1359 			    top, addr, control, td);
1360 			if (dontroute) {
1361 				SOCK_LOCK(so);
1362 				so->so_options &= ~SO_DONTROUTE;
1363 				SOCK_UNLOCK(so);
1364 			}
1365 			clen = 0;
1366 			control = NULL;
1367 			top = NULL;
1368 			if (error)
1369 				goto release;
1370 		} while (resid && space > 0);
1371 	} while (resid);
1372 
1373 release:
1374 	sbunlock(&so->so_snd);
1375 out:
1376 	if (top != NULL)
1377 		m_freem(top);
1378 	if (control != NULL)
1379 		m_freem(control);
1380 	return (error);
1381 }
1382 
1383 int
1384 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1385     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1386 {
1387 	int error;
1388 
1389 	CURVNET_SET(so->so_vnet);
1390 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1391 	    control, flags, td);
1392 	CURVNET_RESTORE();
1393 	return (error);
1394 }
1395 
1396 /*
1397  * The part of soreceive() that implements reading non-inline out-of-band
1398  * data from a socket.  For more complete comments, see soreceive(), from
1399  * which this code originated.
1400  *
1401  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1402  * unable to return an mbuf chain to the caller.
1403  */
1404 static int
1405 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1406 {
1407 	struct protosw *pr = so->so_proto;
1408 	struct mbuf *m;
1409 	int error;
1410 
1411 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1412 	VNET_SO_ASSERT(so);
1413 
1414 	m = m_get(M_WAITOK, MT_DATA);
1415 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1416 	if (error)
1417 		goto bad;
1418 	do {
1419 		error = uiomove(mtod(m, void *),
1420 		    (int) min(uio->uio_resid, m->m_len), uio);
1421 		m = m_free(m);
1422 	} while (uio->uio_resid && error == 0 && m);
1423 bad:
1424 	if (m != NULL)
1425 		m_freem(m);
1426 	return (error);
1427 }
1428 
1429 /*
1430  * Following replacement or removal of the first mbuf on the first mbuf chain
1431  * of a socket buffer, push necessary state changes back into the socket
1432  * buffer so that other consumers see the values consistently.  'nextrecord'
1433  * is the callers locally stored value of the original value of
1434  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1435  * NOTE: 'nextrecord' may be NULL.
1436  */
1437 static __inline void
1438 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1439 {
1440 
1441 	SOCKBUF_LOCK_ASSERT(sb);
1442 	/*
1443 	 * First, update for the new value of nextrecord.  If necessary, make
1444 	 * it the first record.
1445 	 */
1446 	if (sb->sb_mb != NULL)
1447 		sb->sb_mb->m_nextpkt = nextrecord;
1448 	else
1449 		sb->sb_mb = nextrecord;
1450 
1451 	/*
1452 	 * Now update any dependent socket buffer fields to reflect the new
1453 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1454 	 * addition of a second clause that takes care of the case where
1455 	 * sb_mb has been updated, but remains the last record.
1456 	 */
1457 	if (sb->sb_mb == NULL) {
1458 		sb->sb_mbtail = NULL;
1459 		sb->sb_lastrecord = NULL;
1460 	} else if (sb->sb_mb->m_nextpkt == NULL)
1461 		sb->sb_lastrecord = sb->sb_mb;
1462 }
1463 
1464 /*
1465  * Implement receive operations on a socket.  We depend on the way that
1466  * records are added to the sockbuf by sbappend.  In particular, each record
1467  * (mbufs linked through m_next) must begin with an address if the protocol
1468  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1469  * data, and then zero or more mbufs of data.  In order to allow parallelism
1470  * between network receive and copying to user space, as well as avoid
1471  * sleeping with a mutex held, we release the socket buffer mutex during the
1472  * user space copy.  Although the sockbuf is locked, new data may still be
1473  * appended, and thus we must maintain consistency of the sockbuf during that
1474  * time.
1475  *
1476  * The caller may receive the data as a single mbuf chain by supplying an
1477  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1478  * the count in uio_resid.
1479  */
1480 int
1481 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1482     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1483 {
1484 	struct mbuf *m, **mp;
1485 	int flags, error, offset;
1486 	ssize_t len;
1487 	struct protosw *pr = so->so_proto;
1488 	struct mbuf *nextrecord;
1489 	int moff, type = 0;
1490 	ssize_t orig_resid = uio->uio_resid;
1491 
1492 	mp = mp0;
1493 	if (psa != NULL)
1494 		*psa = NULL;
1495 	if (controlp != NULL)
1496 		*controlp = NULL;
1497 	if (flagsp != NULL)
1498 		flags = *flagsp &~ MSG_EOR;
1499 	else
1500 		flags = 0;
1501 	if (flags & MSG_OOB)
1502 		return (soreceive_rcvoob(so, uio, flags));
1503 	if (mp != NULL)
1504 		*mp = NULL;
1505 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1506 	    && uio->uio_resid) {
1507 		VNET_SO_ASSERT(so);
1508 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1509 	}
1510 
1511 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1512 	if (error)
1513 		return (error);
1514 
1515 restart:
1516 	SOCKBUF_LOCK(&so->so_rcv);
1517 	m = so->so_rcv.sb_mb;
1518 	/*
1519 	 * If we have less data than requested, block awaiting more (subject
1520 	 * to any timeout) if:
1521 	 *   1. the current count is less than the low water mark, or
1522 	 *   2. MSG_DONTWAIT is not set
1523 	 */
1524 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1525 	    sbavail(&so->so_rcv) < uio->uio_resid) &&
1526 	    sbavail(&so->so_rcv) < so->so_rcv.sb_lowat &&
1527 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1528 		KASSERT(m != NULL || !sbavail(&so->so_rcv),
1529 		    ("receive: m == %p sbavail == %u",
1530 		    m, sbavail(&so->so_rcv)));
1531 		if (so->so_error) {
1532 			if (m != NULL)
1533 				goto dontblock;
1534 			error = so->so_error;
1535 			if ((flags & MSG_PEEK) == 0)
1536 				so->so_error = 0;
1537 			SOCKBUF_UNLOCK(&so->so_rcv);
1538 			goto release;
1539 		}
1540 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1541 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1542 			if (m == NULL) {
1543 				SOCKBUF_UNLOCK(&so->so_rcv);
1544 				goto release;
1545 			} else
1546 				goto dontblock;
1547 		}
1548 		for (; m != NULL; m = m->m_next)
1549 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1550 				m = so->so_rcv.sb_mb;
1551 				goto dontblock;
1552 			}
1553 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1554 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1555 			SOCKBUF_UNLOCK(&so->so_rcv);
1556 			error = ENOTCONN;
1557 			goto release;
1558 		}
1559 		if (uio->uio_resid == 0) {
1560 			SOCKBUF_UNLOCK(&so->so_rcv);
1561 			goto release;
1562 		}
1563 		if ((so->so_state & SS_NBIO) ||
1564 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1565 			SOCKBUF_UNLOCK(&so->so_rcv);
1566 			error = EWOULDBLOCK;
1567 			goto release;
1568 		}
1569 		SBLASTRECORDCHK(&so->so_rcv);
1570 		SBLASTMBUFCHK(&so->so_rcv);
1571 		error = sbwait(&so->so_rcv);
1572 		SOCKBUF_UNLOCK(&so->so_rcv);
1573 		if (error)
1574 			goto release;
1575 		goto restart;
1576 	}
1577 dontblock:
1578 	/*
1579 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1580 	 * pointer to the next record in the socket buffer.  We must keep the
1581 	 * various socket buffer pointers and local stack versions of the
1582 	 * pointers in sync, pushing out modifications before dropping the
1583 	 * socket buffer mutex, and re-reading them when picking it up.
1584 	 *
1585 	 * Otherwise, we will race with the network stack appending new data
1586 	 * or records onto the socket buffer by using inconsistent/stale
1587 	 * versions of the field, possibly resulting in socket buffer
1588 	 * corruption.
1589 	 *
1590 	 * By holding the high-level sblock(), we prevent simultaneous
1591 	 * readers from pulling off the front of the socket buffer.
1592 	 */
1593 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1594 	if (uio->uio_td)
1595 		uio->uio_td->td_ru.ru_msgrcv++;
1596 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1597 	SBLASTRECORDCHK(&so->so_rcv);
1598 	SBLASTMBUFCHK(&so->so_rcv);
1599 	nextrecord = m->m_nextpkt;
1600 	if (pr->pr_flags & PR_ADDR) {
1601 		KASSERT(m->m_type == MT_SONAME,
1602 		    ("m->m_type == %d", m->m_type));
1603 		orig_resid = 0;
1604 		if (psa != NULL)
1605 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1606 			    M_NOWAIT);
1607 		if (flags & MSG_PEEK) {
1608 			m = m->m_next;
1609 		} else {
1610 			sbfree(&so->so_rcv, m);
1611 			so->so_rcv.sb_mb = m_free(m);
1612 			m = so->so_rcv.sb_mb;
1613 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1614 		}
1615 	}
1616 
1617 	/*
1618 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1619 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1620 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1621 	 * perform externalization (or freeing if controlp == NULL).
1622 	 */
1623 	if (m != NULL && m->m_type == MT_CONTROL) {
1624 		struct mbuf *cm = NULL, *cmn;
1625 		struct mbuf **cme = &cm;
1626 
1627 		do {
1628 			if (flags & MSG_PEEK) {
1629 				if (controlp != NULL) {
1630 					*controlp = m_copy(m, 0, m->m_len);
1631 					controlp = &(*controlp)->m_next;
1632 				}
1633 				m = m->m_next;
1634 			} else {
1635 				sbfree(&so->so_rcv, m);
1636 				so->so_rcv.sb_mb = m->m_next;
1637 				m->m_next = NULL;
1638 				*cme = m;
1639 				cme = &(*cme)->m_next;
1640 				m = so->so_rcv.sb_mb;
1641 			}
1642 		} while (m != NULL && m->m_type == MT_CONTROL);
1643 		if ((flags & MSG_PEEK) == 0)
1644 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1645 		while (cm != NULL) {
1646 			cmn = cm->m_next;
1647 			cm->m_next = NULL;
1648 			if (pr->pr_domain->dom_externalize != NULL) {
1649 				SOCKBUF_UNLOCK(&so->so_rcv);
1650 				VNET_SO_ASSERT(so);
1651 				error = (*pr->pr_domain->dom_externalize)
1652 				    (cm, controlp, flags);
1653 				SOCKBUF_LOCK(&so->so_rcv);
1654 			} else if (controlp != NULL)
1655 				*controlp = cm;
1656 			else
1657 				m_freem(cm);
1658 			if (controlp != NULL) {
1659 				orig_resid = 0;
1660 				while (*controlp != NULL)
1661 					controlp = &(*controlp)->m_next;
1662 			}
1663 			cm = cmn;
1664 		}
1665 		if (m != NULL)
1666 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1667 		else
1668 			nextrecord = so->so_rcv.sb_mb;
1669 		orig_resid = 0;
1670 	}
1671 	if (m != NULL) {
1672 		if ((flags & MSG_PEEK) == 0) {
1673 			KASSERT(m->m_nextpkt == nextrecord,
1674 			    ("soreceive: post-control, nextrecord !sync"));
1675 			if (nextrecord == NULL) {
1676 				KASSERT(so->so_rcv.sb_mb == m,
1677 				    ("soreceive: post-control, sb_mb!=m"));
1678 				KASSERT(so->so_rcv.sb_lastrecord == m,
1679 				    ("soreceive: post-control, lastrecord!=m"));
1680 			}
1681 		}
1682 		type = m->m_type;
1683 		if (type == MT_OOBDATA)
1684 			flags |= MSG_OOB;
1685 	} else {
1686 		if ((flags & MSG_PEEK) == 0) {
1687 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1688 			    ("soreceive: sb_mb != nextrecord"));
1689 			if (so->so_rcv.sb_mb == NULL) {
1690 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1691 				    ("soreceive: sb_lastercord != NULL"));
1692 			}
1693 		}
1694 	}
1695 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1696 	SBLASTRECORDCHK(&so->so_rcv);
1697 	SBLASTMBUFCHK(&so->so_rcv);
1698 
1699 	/*
1700 	 * Now continue to read any data mbufs off of the head of the socket
1701 	 * buffer until the read request is satisfied.  Note that 'type' is
1702 	 * used to store the type of any mbuf reads that have happened so far
1703 	 * such that soreceive() can stop reading if the type changes, which
1704 	 * causes soreceive() to return only one of regular data and inline
1705 	 * out-of-band data in a single socket receive operation.
1706 	 */
1707 	moff = 0;
1708 	offset = 0;
1709 	while (m != NULL && !(m->m_flags & M_NOTAVAIL) && uio->uio_resid > 0
1710 	    && error == 0) {
1711 		/*
1712 		 * If the type of mbuf has changed since the last mbuf
1713 		 * examined ('type'), end the receive operation.
1714 		 */
1715 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1716 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1717 			if (type != m->m_type)
1718 				break;
1719 		} else if (type == MT_OOBDATA)
1720 			break;
1721 		else
1722 		    KASSERT(m->m_type == MT_DATA,
1723 			("m->m_type == %d", m->m_type));
1724 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1725 		len = uio->uio_resid;
1726 		if (so->so_oobmark && len > so->so_oobmark - offset)
1727 			len = so->so_oobmark - offset;
1728 		if (len > m->m_len - moff)
1729 			len = m->m_len - moff;
1730 		/*
1731 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1732 		 * them out via the uio, then free.  Sockbuf must be
1733 		 * consistent here (points to current mbuf, it points to next
1734 		 * record) when we drop priority; we must note any additions
1735 		 * to the sockbuf when we block interrupts again.
1736 		 */
1737 		if (mp == NULL) {
1738 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1739 			SBLASTRECORDCHK(&so->so_rcv);
1740 			SBLASTMBUFCHK(&so->so_rcv);
1741 			SOCKBUF_UNLOCK(&so->so_rcv);
1742 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1743 			SOCKBUF_LOCK(&so->so_rcv);
1744 			if (error) {
1745 				/*
1746 				 * The MT_SONAME mbuf has already been removed
1747 				 * from the record, so it is necessary to
1748 				 * remove the data mbufs, if any, to preserve
1749 				 * the invariant in the case of PR_ADDR that
1750 				 * requires MT_SONAME mbufs at the head of
1751 				 * each record.
1752 				 */
1753 				if (m && pr->pr_flags & PR_ATOMIC &&
1754 				    ((flags & MSG_PEEK) == 0))
1755 					(void)sbdroprecord_locked(&so->so_rcv);
1756 				SOCKBUF_UNLOCK(&so->so_rcv);
1757 				goto release;
1758 			}
1759 		} else
1760 			uio->uio_resid -= len;
1761 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1762 		if (len == m->m_len - moff) {
1763 			if (m->m_flags & M_EOR)
1764 				flags |= MSG_EOR;
1765 			if (flags & MSG_PEEK) {
1766 				m = m->m_next;
1767 				moff = 0;
1768 			} else {
1769 				nextrecord = m->m_nextpkt;
1770 				sbfree(&so->so_rcv, m);
1771 				if (mp != NULL) {
1772 					m->m_nextpkt = NULL;
1773 					*mp = m;
1774 					mp = &m->m_next;
1775 					so->so_rcv.sb_mb = m = m->m_next;
1776 					*mp = NULL;
1777 				} else {
1778 					so->so_rcv.sb_mb = m_free(m);
1779 					m = so->so_rcv.sb_mb;
1780 				}
1781 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1782 				SBLASTRECORDCHK(&so->so_rcv);
1783 				SBLASTMBUFCHK(&so->so_rcv);
1784 			}
1785 		} else {
1786 			if (flags & MSG_PEEK)
1787 				moff += len;
1788 			else {
1789 				if (mp != NULL) {
1790 					if (flags & MSG_DONTWAIT) {
1791 						*mp = m_copym(m, 0, len,
1792 						    M_NOWAIT);
1793 						if (*mp == NULL) {
1794 							/*
1795 							 * m_copym() couldn't
1796 							 * allocate an mbuf.
1797 							 * Adjust uio_resid back
1798 							 * (it was adjusted
1799 							 * down by len bytes,
1800 							 * which we didn't end
1801 							 * up "copying" over).
1802 							 */
1803 							uio->uio_resid += len;
1804 							break;
1805 						}
1806 					} else {
1807 						SOCKBUF_UNLOCK(&so->so_rcv);
1808 						*mp = m_copym(m, 0, len,
1809 						    M_WAITOK);
1810 						SOCKBUF_LOCK(&so->so_rcv);
1811 					}
1812 				}
1813 				sbcut_locked(&so->so_rcv, len);
1814 			}
1815 		}
1816 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1817 		if (so->so_oobmark) {
1818 			if ((flags & MSG_PEEK) == 0) {
1819 				so->so_oobmark -= len;
1820 				if (so->so_oobmark == 0) {
1821 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1822 					break;
1823 				}
1824 			} else {
1825 				offset += len;
1826 				if (offset == so->so_oobmark)
1827 					break;
1828 			}
1829 		}
1830 		if (flags & MSG_EOR)
1831 			break;
1832 		/*
1833 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1834 		 * must not quit until "uio->uio_resid == 0" or an error
1835 		 * termination.  If a signal/timeout occurs, return with a
1836 		 * short count but without error.  Keep sockbuf locked
1837 		 * against other readers.
1838 		 */
1839 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1840 		    !sosendallatonce(so) && nextrecord == NULL) {
1841 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1842 			if (so->so_error ||
1843 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
1844 				break;
1845 			/*
1846 			 * Notify the protocol that some data has been
1847 			 * drained before blocking.
1848 			 */
1849 			if (pr->pr_flags & PR_WANTRCVD) {
1850 				SOCKBUF_UNLOCK(&so->so_rcv);
1851 				VNET_SO_ASSERT(so);
1852 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1853 				SOCKBUF_LOCK(&so->so_rcv);
1854 			}
1855 			SBLASTRECORDCHK(&so->so_rcv);
1856 			SBLASTMBUFCHK(&so->so_rcv);
1857 			/*
1858 			 * We could receive some data while was notifying
1859 			 * the protocol. Skip blocking in this case.
1860 			 */
1861 			if (so->so_rcv.sb_mb == NULL) {
1862 				error = sbwait(&so->so_rcv);
1863 				if (error) {
1864 					SOCKBUF_UNLOCK(&so->so_rcv);
1865 					goto release;
1866 				}
1867 			}
1868 			m = so->so_rcv.sb_mb;
1869 			if (m != NULL)
1870 				nextrecord = m->m_nextpkt;
1871 		}
1872 	}
1873 
1874 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1875 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1876 		flags |= MSG_TRUNC;
1877 		if ((flags & MSG_PEEK) == 0)
1878 			(void) sbdroprecord_locked(&so->so_rcv);
1879 	}
1880 	if ((flags & MSG_PEEK) == 0) {
1881 		if (m == NULL) {
1882 			/*
1883 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1884 			 * part makes sure sb_lastrecord is up-to-date if
1885 			 * there is still data in the socket buffer.
1886 			 */
1887 			so->so_rcv.sb_mb = nextrecord;
1888 			if (so->so_rcv.sb_mb == NULL) {
1889 				so->so_rcv.sb_mbtail = NULL;
1890 				so->so_rcv.sb_lastrecord = NULL;
1891 			} else if (nextrecord->m_nextpkt == NULL)
1892 				so->so_rcv.sb_lastrecord = nextrecord;
1893 		}
1894 		SBLASTRECORDCHK(&so->so_rcv);
1895 		SBLASTMBUFCHK(&so->so_rcv);
1896 		/*
1897 		 * If soreceive() is being done from the socket callback,
1898 		 * then don't need to generate ACK to peer to update window,
1899 		 * since ACK will be generated on return to TCP.
1900 		 */
1901 		if (!(flags & MSG_SOCALLBCK) &&
1902 		    (pr->pr_flags & PR_WANTRCVD)) {
1903 			SOCKBUF_UNLOCK(&so->so_rcv);
1904 			VNET_SO_ASSERT(so);
1905 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1906 			SOCKBUF_LOCK(&so->so_rcv);
1907 		}
1908 	}
1909 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1910 	if (orig_resid == uio->uio_resid && orig_resid &&
1911 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1912 		SOCKBUF_UNLOCK(&so->so_rcv);
1913 		goto restart;
1914 	}
1915 	SOCKBUF_UNLOCK(&so->so_rcv);
1916 
1917 	if (flagsp != NULL)
1918 		*flagsp |= flags;
1919 release:
1920 	sbunlock(&so->so_rcv);
1921 	return (error);
1922 }
1923 
1924 /*
1925  * Optimized version of soreceive() for stream (TCP) sockets.
1926  * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1927  */
1928 int
1929 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1930     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1931 {
1932 	int len = 0, error = 0, flags, oresid;
1933 	struct sockbuf *sb;
1934 	struct mbuf *m, *n = NULL;
1935 
1936 	/* We only do stream sockets. */
1937 	if (so->so_type != SOCK_STREAM)
1938 		return (EINVAL);
1939 	if (psa != NULL)
1940 		*psa = NULL;
1941 	if (controlp != NULL)
1942 		return (EINVAL);
1943 	if (flagsp != NULL)
1944 		flags = *flagsp &~ MSG_EOR;
1945 	else
1946 		flags = 0;
1947 	if (flags & MSG_OOB)
1948 		return (soreceive_rcvoob(so, uio, flags));
1949 	if (mp0 != NULL)
1950 		*mp0 = NULL;
1951 
1952 	sb = &so->so_rcv;
1953 
1954 	/* Prevent other readers from entering the socket. */
1955 	error = sblock(sb, SBLOCKWAIT(flags));
1956 	if (error)
1957 		goto out;
1958 	SOCKBUF_LOCK(sb);
1959 
1960 	/* Easy one, no space to copyout anything. */
1961 	if (uio->uio_resid == 0) {
1962 		error = EINVAL;
1963 		goto out;
1964 	}
1965 	oresid = uio->uio_resid;
1966 
1967 	/* We will never ever get anything unless we are or were connected. */
1968 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1969 		error = ENOTCONN;
1970 		goto out;
1971 	}
1972 
1973 restart:
1974 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1975 
1976 	/* Abort if socket has reported problems. */
1977 	if (so->so_error) {
1978 		if (sbavail(sb) > 0)
1979 			goto deliver;
1980 		if (oresid > uio->uio_resid)
1981 			goto out;
1982 		error = so->so_error;
1983 		if (!(flags & MSG_PEEK))
1984 			so->so_error = 0;
1985 		goto out;
1986 	}
1987 
1988 	/* Door is closed.  Deliver what is left, if any. */
1989 	if (sb->sb_state & SBS_CANTRCVMORE) {
1990 		if (sbavail(sb) > 0)
1991 			goto deliver;
1992 		else
1993 			goto out;
1994 	}
1995 
1996 	/* Socket buffer is empty and we shall not block. */
1997 	if (sbavail(sb) == 0 &&
1998 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1999 		error = EAGAIN;
2000 		goto out;
2001 	}
2002 
2003 	/* Socket buffer got some data that we shall deliver now. */
2004 	if (sbavail(sb) > 0 && !(flags & MSG_WAITALL) &&
2005 	    ((sb->sb_flags & SS_NBIO) ||
2006 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
2007 	     sbavail(sb) >= sb->sb_lowat ||
2008 	     sbavail(sb) >= uio->uio_resid ||
2009 	     sbavail(sb) >= sb->sb_hiwat) ) {
2010 		goto deliver;
2011 	}
2012 
2013 	/* On MSG_WAITALL we must wait until all data or error arrives. */
2014 	if ((flags & MSG_WAITALL) &&
2015 	    (sbavail(sb) >= uio->uio_resid || sbavail(sb) >= sb->sb_hiwat))
2016 		goto deliver;
2017 
2018 	/*
2019 	 * Wait and block until (more) data comes in.
2020 	 * NB: Drops the sockbuf lock during wait.
2021 	 */
2022 	error = sbwait(sb);
2023 	if (error)
2024 		goto out;
2025 	goto restart;
2026 
2027 deliver:
2028 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2029 	KASSERT(sbavail(sb) > 0, ("%s: sockbuf empty", __func__));
2030 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
2031 
2032 	/* Statistics. */
2033 	if (uio->uio_td)
2034 		uio->uio_td->td_ru.ru_msgrcv++;
2035 
2036 	/* Fill uio until full or current end of socket buffer is reached. */
2037 	len = min(uio->uio_resid, sbavail(sb));
2038 	if (mp0 != NULL) {
2039 		/* Dequeue as many mbufs as possible. */
2040 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
2041 			if (*mp0 == NULL)
2042 				*mp0 = sb->sb_mb;
2043 			else
2044 				m_cat(*mp0, sb->sb_mb);
2045 			for (m = sb->sb_mb;
2046 			     m != NULL && m->m_len <= len;
2047 			     m = m->m_next) {
2048 				KASSERT(!(m->m_flags & M_NOTAVAIL),
2049 				    ("%s: m %p not available", __func__, m));
2050 				len -= m->m_len;
2051 				uio->uio_resid -= m->m_len;
2052 				sbfree(sb, m);
2053 				n = m;
2054 			}
2055 			n->m_next = NULL;
2056 			sb->sb_mb = m;
2057 			sb->sb_lastrecord = sb->sb_mb;
2058 			if (sb->sb_mb == NULL)
2059 				SB_EMPTY_FIXUP(sb);
2060 		}
2061 		/* Copy the remainder. */
2062 		if (len > 0) {
2063 			KASSERT(sb->sb_mb != NULL,
2064 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2065 
2066 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2067 			if (m == NULL)
2068 				len = 0;	/* Don't flush data from sockbuf. */
2069 			else
2070 				uio->uio_resid -= len;
2071 			if (*mp0 != NULL)
2072 				m_cat(*mp0, m);
2073 			else
2074 				*mp0 = m;
2075 			if (*mp0 == NULL) {
2076 				error = ENOBUFS;
2077 				goto out;
2078 			}
2079 		}
2080 	} else {
2081 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2082 		SOCKBUF_UNLOCK(sb);
2083 		error = m_mbuftouio(uio, sb->sb_mb, len);
2084 		SOCKBUF_LOCK(sb);
2085 		if (error)
2086 			goto out;
2087 	}
2088 	SBLASTRECORDCHK(sb);
2089 	SBLASTMBUFCHK(sb);
2090 
2091 	/*
2092 	 * Remove the delivered data from the socket buffer unless we
2093 	 * were only peeking.
2094 	 */
2095 	if (!(flags & MSG_PEEK)) {
2096 		if (len > 0)
2097 			sbdrop_locked(sb, len);
2098 
2099 		/* Notify protocol that we drained some data. */
2100 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2101 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2102 		     !(flags & MSG_SOCALLBCK))) {
2103 			SOCKBUF_UNLOCK(sb);
2104 			VNET_SO_ASSERT(so);
2105 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2106 			SOCKBUF_LOCK(sb);
2107 		}
2108 	}
2109 
2110 	/*
2111 	 * For MSG_WAITALL we may have to loop again and wait for
2112 	 * more data to come in.
2113 	 */
2114 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2115 		goto restart;
2116 out:
2117 	SOCKBUF_LOCK_ASSERT(sb);
2118 	SBLASTRECORDCHK(sb);
2119 	SBLASTMBUFCHK(sb);
2120 	SOCKBUF_UNLOCK(sb);
2121 	sbunlock(sb);
2122 	return (error);
2123 }
2124 
2125 /*
2126  * Optimized version of soreceive() for simple datagram cases from userspace.
2127  * Unlike in the stream case, we're able to drop a datagram if copyout()
2128  * fails, and because we handle datagrams atomically, we don't need to use a
2129  * sleep lock to prevent I/O interlacing.
2130  */
2131 int
2132 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2133     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2134 {
2135 	struct mbuf *m, *m2;
2136 	int flags, error;
2137 	ssize_t len;
2138 	struct protosw *pr = so->so_proto;
2139 	struct mbuf *nextrecord;
2140 
2141 	if (psa != NULL)
2142 		*psa = NULL;
2143 	if (controlp != NULL)
2144 		*controlp = NULL;
2145 	if (flagsp != NULL)
2146 		flags = *flagsp &~ MSG_EOR;
2147 	else
2148 		flags = 0;
2149 
2150 	/*
2151 	 * For any complicated cases, fall back to the full
2152 	 * soreceive_generic().
2153 	 */
2154 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2155 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2156 		    flagsp));
2157 
2158 	/*
2159 	 * Enforce restrictions on use.
2160 	 */
2161 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2162 	    ("soreceive_dgram: wantrcvd"));
2163 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2164 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2165 	    ("soreceive_dgram: SBS_RCVATMARK"));
2166 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2167 	    ("soreceive_dgram: P_CONNREQUIRED"));
2168 
2169 	/*
2170 	 * Loop blocking while waiting for a datagram.
2171 	 */
2172 	SOCKBUF_LOCK(&so->so_rcv);
2173 	while ((m = so->so_rcv.sb_mb) == NULL) {
2174 		KASSERT(sbavail(&so->so_rcv) == 0,
2175 		    ("soreceive_dgram: sb_mb NULL but sbavail %u",
2176 		    sbavail(&so->so_rcv)));
2177 		if (so->so_error) {
2178 			error = so->so_error;
2179 			so->so_error = 0;
2180 			SOCKBUF_UNLOCK(&so->so_rcv);
2181 			return (error);
2182 		}
2183 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2184 		    uio->uio_resid == 0) {
2185 			SOCKBUF_UNLOCK(&so->so_rcv);
2186 			return (0);
2187 		}
2188 		if ((so->so_state & SS_NBIO) ||
2189 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2190 			SOCKBUF_UNLOCK(&so->so_rcv);
2191 			return (EWOULDBLOCK);
2192 		}
2193 		SBLASTRECORDCHK(&so->so_rcv);
2194 		SBLASTMBUFCHK(&so->so_rcv);
2195 		error = sbwait(&so->so_rcv);
2196 		if (error) {
2197 			SOCKBUF_UNLOCK(&so->so_rcv);
2198 			return (error);
2199 		}
2200 	}
2201 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2202 
2203 	if (uio->uio_td)
2204 		uio->uio_td->td_ru.ru_msgrcv++;
2205 	SBLASTRECORDCHK(&so->so_rcv);
2206 	SBLASTMBUFCHK(&so->so_rcv);
2207 	nextrecord = m->m_nextpkt;
2208 	if (nextrecord == NULL) {
2209 		KASSERT(so->so_rcv.sb_lastrecord == m,
2210 		    ("soreceive_dgram: lastrecord != m"));
2211 	}
2212 
2213 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2214 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2215 
2216 	/*
2217 	 * Pull 'm' and its chain off the front of the packet queue.
2218 	 */
2219 	so->so_rcv.sb_mb = NULL;
2220 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2221 
2222 	/*
2223 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2224 	 */
2225 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2226 		sbfree(&so->so_rcv, m2);
2227 
2228 	/*
2229 	 * Do a few last checks before we let go of the lock.
2230 	 */
2231 	SBLASTRECORDCHK(&so->so_rcv);
2232 	SBLASTMBUFCHK(&so->so_rcv);
2233 	SOCKBUF_UNLOCK(&so->so_rcv);
2234 
2235 	if (pr->pr_flags & PR_ADDR) {
2236 		KASSERT(m->m_type == MT_SONAME,
2237 		    ("m->m_type == %d", m->m_type));
2238 		if (psa != NULL)
2239 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2240 			    M_NOWAIT);
2241 		m = m_free(m);
2242 	}
2243 	if (m == NULL) {
2244 		/* XXXRW: Can this happen? */
2245 		return (0);
2246 	}
2247 
2248 	/*
2249 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2250 	 * queue.
2251 	 *
2252 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2253 	 * in the first mbuf chain on the socket buffer.  We call into the
2254 	 * protocol to perform externalization (or freeing if controlp ==
2255 	 * NULL).
2256 	 */
2257 	if (m->m_type == MT_CONTROL) {
2258 		struct mbuf *cm = NULL, *cmn;
2259 		struct mbuf **cme = &cm;
2260 
2261 		do {
2262 			m2 = m->m_next;
2263 			m->m_next = NULL;
2264 			*cme = m;
2265 			cme = &(*cme)->m_next;
2266 			m = m2;
2267 		} while (m != NULL && m->m_type == MT_CONTROL);
2268 		while (cm != NULL) {
2269 			cmn = cm->m_next;
2270 			cm->m_next = NULL;
2271 			if (pr->pr_domain->dom_externalize != NULL) {
2272 				error = (*pr->pr_domain->dom_externalize)
2273 				    (cm, controlp, flags);
2274 			} else if (controlp != NULL)
2275 				*controlp = cm;
2276 			else
2277 				m_freem(cm);
2278 			if (controlp != NULL) {
2279 				while (*controlp != NULL)
2280 					controlp = &(*controlp)->m_next;
2281 			}
2282 			cm = cmn;
2283 		}
2284 	}
2285 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2286 
2287 	while (m != NULL && uio->uio_resid > 0) {
2288 		len = uio->uio_resid;
2289 		if (len > m->m_len)
2290 			len = m->m_len;
2291 		error = uiomove(mtod(m, char *), (int)len, uio);
2292 		if (error) {
2293 			m_freem(m);
2294 			return (error);
2295 		}
2296 		if (len == m->m_len)
2297 			m = m_free(m);
2298 		else {
2299 			m->m_data += len;
2300 			m->m_len -= len;
2301 		}
2302 	}
2303 	if (m != NULL)
2304 		flags |= MSG_TRUNC;
2305 	m_freem(m);
2306 	if (flagsp != NULL)
2307 		*flagsp |= flags;
2308 	return (0);
2309 }
2310 
2311 int
2312 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2313     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2314 {
2315 	int error;
2316 
2317 	CURVNET_SET(so->so_vnet);
2318 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2319 	    controlp, flagsp));
2320 	CURVNET_RESTORE();
2321 	return (error);
2322 }
2323 
2324 int
2325 soshutdown(struct socket *so, int how)
2326 {
2327 	struct protosw *pr = so->so_proto;
2328 	int error;
2329 
2330 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2331 		return (EINVAL);
2332 
2333 	CURVNET_SET(so->so_vnet);
2334 	if (pr->pr_usrreqs->pru_flush != NULL)
2335 		(*pr->pr_usrreqs->pru_flush)(so, how);
2336 	if (how != SHUT_WR)
2337 		sorflush(so);
2338 	if (how != SHUT_RD) {
2339 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2340 		wakeup(&so->so_timeo);
2341 		CURVNET_RESTORE();
2342 		return (error);
2343 	}
2344 	wakeup(&so->so_timeo);
2345 	CURVNET_RESTORE();
2346 	return (0);
2347 }
2348 
2349 void
2350 sorflush(struct socket *so)
2351 {
2352 	struct sockbuf *sb = &so->so_rcv;
2353 	struct protosw *pr = so->so_proto;
2354 	struct sockbuf asb;
2355 
2356 	VNET_SO_ASSERT(so);
2357 
2358 	/*
2359 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2360 	 * held, and in order to generally avoid holding the lock for a long
2361 	 * time, we make a copy of the socket buffer and clear the original
2362 	 * (except locks, state).  The new socket buffer copy won't have
2363 	 * initialized locks so we can only call routines that won't use or
2364 	 * assert those locks.
2365 	 *
2366 	 * Dislodge threads currently blocked in receive and wait to acquire
2367 	 * a lock against other simultaneous readers before clearing the
2368 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2369 	 * despite any existing socket disposition on interruptable waiting.
2370 	 */
2371 	socantrcvmore(so);
2372 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2373 
2374 	/*
2375 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2376 	 * and mutex data unchanged.
2377 	 */
2378 	SOCKBUF_LOCK(sb);
2379 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2380 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2381 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2382 	bzero(&sb->sb_startzero,
2383 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2384 	SOCKBUF_UNLOCK(sb);
2385 	sbunlock(sb);
2386 
2387 	/*
2388 	 * Dispose of special rights and flush the socket buffer.  Don't call
2389 	 * any unsafe routines (that rely on locks being initialized) on asb.
2390 	 */
2391 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2392 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2393 	sbrelease_internal(&asb, so);
2394 }
2395 
2396 /*
2397  * Wrapper for Socket established helper hook.
2398  * Parameters: socket, context of the hook point, hook id.
2399  */
2400 static int inline
2401 hhook_run_socket(struct socket *so, void *hctx, int32_t h_id)
2402 {
2403 	struct socket_hhook_data hhook_data = {
2404 		.so = so,
2405 		.hctx = hctx,
2406 		.m = NULL,
2407 		.status = 0
2408 	};
2409 
2410 	CURVNET_SET(so->so_vnet);
2411 	HHOOKS_RUN_IF(V_socket_hhh[h_id], &hhook_data, &so->osd);
2412 	CURVNET_RESTORE();
2413 
2414 	/* Ugly but needed, since hhooks return void for now */
2415 	return (hhook_data.status);
2416 }
2417 
2418 /*
2419  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2420  * additional variant to handle the case where the option value needs to be
2421  * some kind of integer, but not a specific size.  In addition to their use
2422  * here, these functions are also called by the protocol-level pr_ctloutput()
2423  * routines.
2424  */
2425 int
2426 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2427 {
2428 	size_t	valsize;
2429 
2430 	/*
2431 	 * If the user gives us more than we wanted, we ignore it, but if we
2432 	 * don't get the minimum length the caller wants, we return EINVAL.
2433 	 * On success, sopt->sopt_valsize is set to however much we actually
2434 	 * retrieved.
2435 	 */
2436 	if ((valsize = sopt->sopt_valsize) < minlen)
2437 		return EINVAL;
2438 	if (valsize > len)
2439 		sopt->sopt_valsize = valsize = len;
2440 
2441 	if (sopt->sopt_td != NULL)
2442 		return (copyin(sopt->sopt_val, buf, valsize));
2443 
2444 	bcopy(sopt->sopt_val, buf, valsize);
2445 	return (0);
2446 }
2447 
2448 /*
2449  * Kernel version of setsockopt(2).
2450  *
2451  * XXX: optlen is size_t, not socklen_t
2452  */
2453 int
2454 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2455     size_t optlen)
2456 {
2457 	struct sockopt sopt;
2458 
2459 	sopt.sopt_level = level;
2460 	sopt.sopt_name = optname;
2461 	sopt.sopt_dir = SOPT_SET;
2462 	sopt.sopt_val = optval;
2463 	sopt.sopt_valsize = optlen;
2464 	sopt.sopt_td = NULL;
2465 	return (sosetopt(so, &sopt));
2466 }
2467 
2468 int
2469 sosetopt(struct socket *so, struct sockopt *sopt)
2470 {
2471 	int	error, optval;
2472 	struct	linger l;
2473 	struct	timeval tv;
2474 	sbintime_t val;
2475 	uint32_t val32;
2476 #ifdef MAC
2477 	struct mac extmac;
2478 #endif
2479 
2480 	CURVNET_SET(so->so_vnet);
2481 	error = 0;
2482 	if (sopt->sopt_level != SOL_SOCKET) {
2483 		if (so->so_proto->pr_ctloutput != NULL) {
2484 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2485 			CURVNET_RESTORE();
2486 			return (error);
2487 		}
2488 		error = ENOPROTOOPT;
2489 	} else {
2490 		switch (sopt->sopt_name) {
2491 		case SO_ACCEPTFILTER:
2492 			error = do_setopt_accept_filter(so, sopt);
2493 			if (error)
2494 				goto bad;
2495 			break;
2496 
2497 		case SO_LINGER:
2498 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2499 			if (error)
2500 				goto bad;
2501 
2502 			SOCK_LOCK(so);
2503 			so->so_linger = l.l_linger;
2504 			if (l.l_onoff)
2505 				so->so_options |= SO_LINGER;
2506 			else
2507 				so->so_options &= ~SO_LINGER;
2508 			SOCK_UNLOCK(so);
2509 			break;
2510 
2511 		case SO_DEBUG:
2512 		case SO_KEEPALIVE:
2513 		case SO_DONTROUTE:
2514 		case SO_USELOOPBACK:
2515 		case SO_BROADCAST:
2516 		case SO_REUSEADDR:
2517 		case SO_REUSEPORT:
2518 		case SO_OOBINLINE:
2519 		case SO_TIMESTAMP:
2520 		case SO_BINTIME:
2521 		case SO_NOSIGPIPE:
2522 		case SO_NO_DDP:
2523 		case SO_NO_OFFLOAD:
2524 			error = sooptcopyin(sopt, &optval, sizeof optval,
2525 			    sizeof optval);
2526 			if (error)
2527 				goto bad;
2528 			SOCK_LOCK(so);
2529 			if (optval)
2530 				so->so_options |= sopt->sopt_name;
2531 			else
2532 				so->so_options &= ~sopt->sopt_name;
2533 			SOCK_UNLOCK(so);
2534 			break;
2535 
2536 		case SO_SETFIB:
2537 			error = sooptcopyin(sopt, &optval, sizeof optval,
2538 			    sizeof optval);
2539 			if (error)
2540 				goto bad;
2541 
2542 			if (optval < 0 || optval >= rt_numfibs) {
2543 				error = EINVAL;
2544 				goto bad;
2545 			}
2546 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2547 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2548 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2549 				so->so_fibnum = optval;
2550 			else
2551 				so->so_fibnum = 0;
2552 			break;
2553 
2554 		case SO_USER_COOKIE:
2555 			error = sooptcopyin(sopt, &val32, sizeof val32,
2556 			    sizeof val32);
2557 			if (error)
2558 				goto bad;
2559 			so->so_user_cookie = val32;
2560 			break;
2561 
2562 		case SO_SNDBUF:
2563 		case SO_RCVBUF:
2564 		case SO_SNDLOWAT:
2565 		case SO_RCVLOWAT:
2566 			error = sooptcopyin(sopt, &optval, sizeof optval,
2567 			    sizeof optval);
2568 			if (error)
2569 				goto bad;
2570 
2571 			/*
2572 			 * Values < 1 make no sense for any of these options,
2573 			 * so disallow them.
2574 			 */
2575 			if (optval < 1) {
2576 				error = EINVAL;
2577 				goto bad;
2578 			}
2579 
2580 			switch (sopt->sopt_name) {
2581 			case SO_SNDBUF:
2582 			case SO_RCVBUF:
2583 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2584 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2585 				    so, curthread) == 0) {
2586 					error = ENOBUFS;
2587 					goto bad;
2588 				}
2589 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2590 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2591 				break;
2592 
2593 			/*
2594 			 * Make sure the low-water is never greater than the
2595 			 * high-water.
2596 			 */
2597 			case SO_SNDLOWAT:
2598 				SOCKBUF_LOCK(&so->so_snd);
2599 				so->so_snd.sb_lowat =
2600 				    (optval > so->so_snd.sb_hiwat) ?
2601 				    so->so_snd.sb_hiwat : optval;
2602 				SOCKBUF_UNLOCK(&so->so_snd);
2603 				break;
2604 			case SO_RCVLOWAT:
2605 				SOCKBUF_LOCK(&so->so_rcv);
2606 				so->so_rcv.sb_lowat =
2607 				    (optval > so->so_rcv.sb_hiwat) ?
2608 				    so->so_rcv.sb_hiwat : optval;
2609 				SOCKBUF_UNLOCK(&so->so_rcv);
2610 				break;
2611 			}
2612 			break;
2613 
2614 		case SO_SNDTIMEO:
2615 		case SO_RCVTIMEO:
2616 #ifdef COMPAT_FREEBSD32
2617 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2618 				struct timeval32 tv32;
2619 
2620 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2621 				    sizeof tv32);
2622 				CP(tv32, tv, tv_sec);
2623 				CP(tv32, tv, tv_usec);
2624 			} else
2625 #endif
2626 				error = sooptcopyin(sopt, &tv, sizeof tv,
2627 				    sizeof tv);
2628 			if (error)
2629 				goto bad;
2630 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2631 			    tv.tv_usec >= 1000000) {
2632 				error = EDOM;
2633 				goto bad;
2634 			}
2635 			if (tv.tv_sec > INT32_MAX)
2636 				val = SBT_MAX;
2637 			else
2638 				val = tvtosbt(tv);
2639 			switch (sopt->sopt_name) {
2640 			case SO_SNDTIMEO:
2641 				so->so_snd.sb_timeo = val;
2642 				break;
2643 			case SO_RCVTIMEO:
2644 				so->so_rcv.sb_timeo = val;
2645 				break;
2646 			}
2647 			break;
2648 
2649 		case SO_LABEL:
2650 #ifdef MAC
2651 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2652 			    sizeof extmac);
2653 			if (error)
2654 				goto bad;
2655 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2656 			    so, &extmac);
2657 #else
2658 			error = EOPNOTSUPP;
2659 #endif
2660 			break;
2661 
2662 		default:
2663 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2664 				error = hhook_run_socket(so, sopt,
2665 				    HHOOK_SOCKET_OPT);
2666 			else
2667 				error = ENOPROTOOPT;
2668 			break;
2669 		}
2670 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2671 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2672 	}
2673 bad:
2674 	CURVNET_RESTORE();
2675 	return (error);
2676 }
2677 
2678 /*
2679  * Helper routine for getsockopt.
2680  */
2681 int
2682 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2683 {
2684 	int	error;
2685 	size_t	valsize;
2686 
2687 	error = 0;
2688 
2689 	/*
2690 	 * Documented get behavior is that we always return a value, possibly
2691 	 * truncated to fit in the user's buffer.  Traditional behavior is
2692 	 * that we always tell the user precisely how much we copied, rather
2693 	 * than something useful like the total amount we had available for
2694 	 * her.  Note that this interface is not idempotent; the entire
2695 	 * answer must generated ahead of time.
2696 	 */
2697 	valsize = min(len, sopt->sopt_valsize);
2698 	sopt->sopt_valsize = valsize;
2699 	if (sopt->sopt_val != NULL) {
2700 		if (sopt->sopt_td != NULL)
2701 			error = copyout(buf, sopt->sopt_val, valsize);
2702 		else
2703 			bcopy(buf, sopt->sopt_val, valsize);
2704 	}
2705 	return (error);
2706 }
2707 
2708 int
2709 sogetopt(struct socket *so, struct sockopt *sopt)
2710 {
2711 	int	error, optval;
2712 	struct	linger l;
2713 	struct	timeval tv;
2714 #ifdef MAC
2715 	struct mac extmac;
2716 #endif
2717 
2718 	CURVNET_SET(so->so_vnet);
2719 	error = 0;
2720 	if (sopt->sopt_level != SOL_SOCKET) {
2721 		if (so->so_proto->pr_ctloutput != NULL)
2722 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2723 		else
2724 			error = ENOPROTOOPT;
2725 		CURVNET_RESTORE();
2726 		return (error);
2727 	} else {
2728 		switch (sopt->sopt_name) {
2729 		case SO_ACCEPTFILTER:
2730 			error = do_getopt_accept_filter(so, sopt);
2731 			break;
2732 
2733 		case SO_LINGER:
2734 			SOCK_LOCK(so);
2735 			l.l_onoff = so->so_options & SO_LINGER;
2736 			l.l_linger = so->so_linger;
2737 			SOCK_UNLOCK(so);
2738 			error = sooptcopyout(sopt, &l, sizeof l);
2739 			break;
2740 
2741 		case SO_USELOOPBACK:
2742 		case SO_DONTROUTE:
2743 		case SO_DEBUG:
2744 		case SO_KEEPALIVE:
2745 		case SO_REUSEADDR:
2746 		case SO_REUSEPORT:
2747 		case SO_BROADCAST:
2748 		case SO_OOBINLINE:
2749 		case SO_ACCEPTCONN:
2750 		case SO_TIMESTAMP:
2751 		case SO_BINTIME:
2752 		case SO_NOSIGPIPE:
2753 			optval = so->so_options & sopt->sopt_name;
2754 integer:
2755 			error = sooptcopyout(sopt, &optval, sizeof optval);
2756 			break;
2757 
2758 		case SO_TYPE:
2759 			optval = so->so_type;
2760 			goto integer;
2761 
2762 		case SO_PROTOCOL:
2763 			optval = so->so_proto->pr_protocol;
2764 			goto integer;
2765 
2766 		case SO_ERROR:
2767 			SOCK_LOCK(so);
2768 			optval = so->so_error;
2769 			so->so_error = 0;
2770 			SOCK_UNLOCK(so);
2771 			goto integer;
2772 
2773 		case SO_SNDBUF:
2774 			optval = so->so_snd.sb_hiwat;
2775 			goto integer;
2776 
2777 		case SO_RCVBUF:
2778 			optval = so->so_rcv.sb_hiwat;
2779 			goto integer;
2780 
2781 		case SO_SNDLOWAT:
2782 			optval = so->so_snd.sb_lowat;
2783 			goto integer;
2784 
2785 		case SO_RCVLOWAT:
2786 			optval = so->so_rcv.sb_lowat;
2787 			goto integer;
2788 
2789 		case SO_SNDTIMEO:
2790 		case SO_RCVTIMEO:
2791 			tv = sbttotv(sopt->sopt_name == SO_SNDTIMEO ?
2792 			    so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2793 #ifdef COMPAT_FREEBSD32
2794 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2795 				struct timeval32 tv32;
2796 
2797 				CP(tv, tv32, tv_sec);
2798 				CP(tv, tv32, tv_usec);
2799 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2800 			} else
2801 #endif
2802 				error = sooptcopyout(sopt, &tv, sizeof tv);
2803 			break;
2804 
2805 		case SO_LABEL:
2806 #ifdef MAC
2807 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2808 			    sizeof(extmac));
2809 			if (error)
2810 				goto bad;
2811 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2812 			    so, &extmac);
2813 			if (error)
2814 				goto bad;
2815 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2816 #else
2817 			error = EOPNOTSUPP;
2818 #endif
2819 			break;
2820 
2821 		case SO_PEERLABEL:
2822 #ifdef MAC
2823 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2824 			    sizeof(extmac));
2825 			if (error)
2826 				goto bad;
2827 			error = mac_getsockopt_peerlabel(
2828 			    sopt->sopt_td->td_ucred, so, &extmac);
2829 			if (error)
2830 				goto bad;
2831 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2832 #else
2833 			error = EOPNOTSUPP;
2834 #endif
2835 			break;
2836 
2837 		case SO_LISTENQLIMIT:
2838 			optval = so->so_qlimit;
2839 			goto integer;
2840 
2841 		case SO_LISTENQLEN:
2842 			optval = so->so_qlen;
2843 			goto integer;
2844 
2845 		case SO_LISTENINCQLEN:
2846 			optval = so->so_incqlen;
2847 			goto integer;
2848 
2849 		default:
2850 			if (V_socket_hhh[HHOOK_SOCKET_OPT]->hhh_nhooks > 0)
2851 				error = hhook_run_socket(so, sopt,
2852 				    HHOOK_SOCKET_OPT);
2853 			else
2854 				error = ENOPROTOOPT;
2855 			break;
2856 		}
2857 	}
2858 #ifdef MAC
2859 bad:
2860 #endif
2861 	CURVNET_RESTORE();
2862 	return (error);
2863 }
2864 
2865 int
2866 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2867 {
2868 	struct mbuf *m, *m_prev;
2869 	int sopt_size = sopt->sopt_valsize;
2870 
2871 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2872 	if (m == NULL)
2873 		return ENOBUFS;
2874 	if (sopt_size > MLEN) {
2875 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2876 		if ((m->m_flags & M_EXT) == 0) {
2877 			m_free(m);
2878 			return ENOBUFS;
2879 		}
2880 		m->m_len = min(MCLBYTES, sopt_size);
2881 	} else {
2882 		m->m_len = min(MLEN, sopt_size);
2883 	}
2884 	sopt_size -= m->m_len;
2885 	*mp = m;
2886 	m_prev = m;
2887 
2888 	while (sopt_size) {
2889 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2890 		if (m == NULL) {
2891 			m_freem(*mp);
2892 			return ENOBUFS;
2893 		}
2894 		if (sopt_size > MLEN) {
2895 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2896 			    M_NOWAIT);
2897 			if ((m->m_flags & M_EXT) == 0) {
2898 				m_freem(m);
2899 				m_freem(*mp);
2900 				return ENOBUFS;
2901 			}
2902 			m->m_len = min(MCLBYTES, sopt_size);
2903 		} else {
2904 			m->m_len = min(MLEN, sopt_size);
2905 		}
2906 		sopt_size -= m->m_len;
2907 		m_prev->m_next = m;
2908 		m_prev = m;
2909 	}
2910 	return (0);
2911 }
2912 
2913 int
2914 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2915 {
2916 	struct mbuf *m0 = m;
2917 
2918 	if (sopt->sopt_val == NULL)
2919 		return (0);
2920 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2921 		if (sopt->sopt_td != NULL) {
2922 			int error;
2923 
2924 			error = copyin(sopt->sopt_val, mtod(m, char *),
2925 			    m->m_len);
2926 			if (error != 0) {
2927 				m_freem(m0);
2928 				return(error);
2929 			}
2930 		} else
2931 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2932 		sopt->sopt_valsize -= m->m_len;
2933 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2934 		m = m->m_next;
2935 	}
2936 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2937 		panic("ip6_sooptmcopyin");
2938 	return (0);
2939 }
2940 
2941 int
2942 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2943 {
2944 	struct mbuf *m0 = m;
2945 	size_t valsize = 0;
2946 
2947 	if (sopt->sopt_val == NULL)
2948 		return (0);
2949 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2950 		if (sopt->sopt_td != NULL) {
2951 			int error;
2952 
2953 			error = copyout(mtod(m, char *), sopt->sopt_val,
2954 			    m->m_len);
2955 			if (error != 0) {
2956 				m_freem(m0);
2957 				return(error);
2958 			}
2959 		} else
2960 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2961 		sopt->sopt_valsize -= m->m_len;
2962 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2963 		valsize += m->m_len;
2964 		m = m->m_next;
2965 	}
2966 	if (m != NULL) {
2967 		/* enough soopt buffer should be given from user-land */
2968 		m_freem(m0);
2969 		return(EINVAL);
2970 	}
2971 	sopt->sopt_valsize = valsize;
2972 	return (0);
2973 }
2974 
2975 /*
2976  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2977  * out-of-band data, which will then notify socket consumers.
2978  */
2979 void
2980 sohasoutofband(struct socket *so)
2981 {
2982 
2983 	if (so->so_sigio != NULL)
2984 		pgsigio(&so->so_sigio, SIGURG, 0);
2985 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2986 }
2987 
2988 int
2989 sopoll(struct socket *so, int events, struct ucred *active_cred,
2990     struct thread *td)
2991 {
2992 
2993 	/*
2994 	 * We do not need to set or assert curvnet as long as everyone uses
2995 	 * sopoll_generic().
2996 	 */
2997 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2998 	    td));
2999 }
3000 
3001 int
3002 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
3003     struct thread *td)
3004 {
3005 	int revents = 0;
3006 
3007 	SOCKBUF_LOCK(&so->so_snd);
3008 	SOCKBUF_LOCK(&so->so_rcv);
3009 	if (events & (POLLIN | POLLRDNORM))
3010 		if (soreadabledata(so))
3011 			revents |= events & (POLLIN | POLLRDNORM);
3012 
3013 	if (events & (POLLOUT | POLLWRNORM))
3014 		if (sowriteable(so))
3015 			revents |= events & (POLLOUT | POLLWRNORM);
3016 
3017 	if (events & (POLLPRI | POLLRDBAND))
3018 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
3019 			revents |= events & (POLLPRI | POLLRDBAND);
3020 
3021 	if ((events & POLLINIGNEOF) == 0) {
3022 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3023 			revents |= events & (POLLIN | POLLRDNORM);
3024 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
3025 				revents |= POLLHUP;
3026 		}
3027 	}
3028 
3029 	if (revents == 0) {
3030 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
3031 			selrecord(td, &so->so_rcv.sb_sel);
3032 			so->so_rcv.sb_flags |= SB_SEL;
3033 		}
3034 
3035 		if (events & (POLLOUT | POLLWRNORM)) {
3036 			selrecord(td, &so->so_snd.sb_sel);
3037 			so->so_snd.sb_flags |= SB_SEL;
3038 		}
3039 	}
3040 
3041 	SOCKBUF_UNLOCK(&so->so_rcv);
3042 	SOCKBUF_UNLOCK(&so->so_snd);
3043 	return (revents);
3044 }
3045 
3046 int
3047 soo_kqfilter(struct file *fp, struct knote *kn)
3048 {
3049 	struct socket *so = kn->kn_fp->f_data;
3050 	struct sockbuf *sb;
3051 
3052 	switch (kn->kn_filter) {
3053 	case EVFILT_READ:
3054 		if (so->so_options & SO_ACCEPTCONN)
3055 			kn->kn_fop = &solisten_filtops;
3056 		else
3057 			kn->kn_fop = &soread_filtops;
3058 		sb = &so->so_rcv;
3059 		break;
3060 	case EVFILT_WRITE:
3061 		kn->kn_fop = &sowrite_filtops;
3062 		sb = &so->so_snd;
3063 		break;
3064 	default:
3065 		return (EINVAL);
3066 	}
3067 
3068 	SOCKBUF_LOCK(sb);
3069 	knlist_add(&sb->sb_sel.si_note, kn, 1);
3070 	sb->sb_flags |= SB_KNOTE;
3071 	SOCKBUF_UNLOCK(sb);
3072 	return (0);
3073 }
3074 
3075 /*
3076  * Some routines that return EOPNOTSUPP for entry points that are not
3077  * supported by a protocol.  Fill in as needed.
3078  */
3079 int
3080 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
3081 {
3082 
3083 	return EOPNOTSUPP;
3084 }
3085 
3086 int
3087 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
3088 {
3089 
3090 	return EOPNOTSUPP;
3091 }
3092 
3093 int
3094 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3095 {
3096 
3097 	return EOPNOTSUPP;
3098 }
3099 
3100 int
3101 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3102     struct thread *td)
3103 {
3104 
3105 	return EOPNOTSUPP;
3106 }
3107 
3108 int
3109 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3110 {
3111 
3112 	return EOPNOTSUPP;
3113 }
3114 
3115 int
3116 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3117     struct thread *td)
3118 {
3119 
3120 	return EOPNOTSUPP;
3121 }
3122 
3123 int
3124 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3125 {
3126 
3127 	return EOPNOTSUPP;
3128 }
3129 
3130 int
3131 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3132     struct ifnet *ifp, struct thread *td)
3133 {
3134 
3135 	return EOPNOTSUPP;
3136 }
3137 
3138 int
3139 pru_disconnect_notsupp(struct socket *so)
3140 {
3141 
3142 	return EOPNOTSUPP;
3143 }
3144 
3145 int
3146 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3147 {
3148 
3149 	return EOPNOTSUPP;
3150 }
3151 
3152 int
3153 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3154 {
3155 
3156 	return EOPNOTSUPP;
3157 }
3158 
3159 int
3160 pru_rcvd_notsupp(struct socket *so, int flags)
3161 {
3162 
3163 	return EOPNOTSUPP;
3164 }
3165 
3166 int
3167 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3168 {
3169 
3170 	return EOPNOTSUPP;
3171 }
3172 
3173 int
3174 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3175     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3176 {
3177 
3178 	return EOPNOTSUPP;
3179 }
3180 
3181 /*
3182  * This isn't really a ``null'' operation, but it's the default one and
3183  * doesn't do anything destructive.
3184  */
3185 int
3186 pru_sense_null(struct socket *so, struct stat *sb)
3187 {
3188 
3189 	sb->st_blksize = so->so_snd.sb_hiwat;
3190 	return 0;
3191 }
3192 
3193 int
3194 pru_shutdown_notsupp(struct socket *so)
3195 {
3196 
3197 	return EOPNOTSUPP;
3198 }
3199 
3200 int
3201 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3202 {
3203 
3204 	return EOPNOTSUPP;
3205 }
3206 
3207 int
3208 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3209     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3210 {
3211 
3212 	return EOPNOTSUPP;
3213 }
3214 
3215 int
3216 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3217     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3218 {
3219 
3220 	return EOPNOTSUPP;
3221 }
3222 
3223 int
3224 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3225     struct thread *td)
3226 {
3227 
3228 	return EOPNOTSUPP;
3229 }
3230 
3231 static void
3232 filt_sordetach(struct knote *kn)
3233 {
3234 	struct socket *so = kn->kn_fp->f_data;
3235 
3236 	SOCKBUF_LOCK(&so->so_rcv);
3237 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3238 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3239 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3240 	SOCKBUF_UNLOCK(&so->so_rcv);
3241 }
3242 
3243 /*ARGSUSED*/
3244 static int
3245 filt_soread(struct knote *kn, long hint)
3246 {
3247 	struct socket *so;
3248 
3249 	so = kn->kn_fp->f_data;
3250 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3251 
3252 	kn->kn_data = sbavail(&so->so_rcv) - so->so_rcv.sb_ctl;
3253 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3254 		kn->kn_flags |= EV_EOF;
3255 		kn->kn_fflags = so->so_error;
3256 		return (1);
3257 	} else if (so->so_error)	/* temporary udp error */
3258 		return (1);
3259 
3260 	if (kn->kn_sfflags & NOTE_LOWAT) {
3261 		if (kn->kn_data >= kn->kn_sdata)
3262 			return 1;
3263 	} else {
3264 		if (sbavail(&so->so_rcv) >= so->so_rcv.sb_lowat)
3265 			return 1;
3266 	}
3267 
3268 	/* This hook returning non-zero indicates an event, not error */
3269 	return (hhook_run_socket(so, NULL, HHOOK_FILT_SOREAD));
3270 }
3271 
3272 static void
3273 filt_sowdetach(struct knote *kn)
3274 {
3275 	struct socket *so = kn->kn_fp->f_data;
3276 
3277 	SOCKBUF_LOCK(&so->so_snd);
3278 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3279 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3280 		so->so_snd.sb_flags &= ~SB_KNOTE;
3281 	SOCKBUF_UNLOCK(&so->so_snd);
3282 }
3283 
3284 /*ARGSUSED*/
3285 static int
3286 filt_sowrite(struct knote *kn, long hint)
3287 {
3288 	struct socket *so;
3289 
3290 	so = kn->kn_fp->f_data;
3291 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3292 	kn->kn_data = sbspace(&so->so_snd);
3293 
3294 	hhook_run_socket(so, kn, HHOOK_FILT_SOWRITE);
3295 
3296 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3297 		kn->kn_flags |= EV_EOF;
3298 		kn->kn_fflags = so->so_error;
3299 		return (1);
3300 	} else if (so->so_error)	/* temporary udp error */
3301 		return (1);
3302 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3303 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3304 		return (0);
3305 	else if (kn->kn_sfflags & NOTE_LOWAT)
3306 		return (kn->kn_data >= kn->kn_sdata);
3307 	else
3308 		return (kn->kn_data >= so->so_snd.sb_lowat);
3309 }
3310 
3311 /*ARGSUSED*/
3312 static int
3313 filt_solisten(struct knote *kn, long hint)
3314 {
3315 	struct socket *so = kn->kn_fp->f_data;
3316 
3317 	kn->kn_data = so->so_qlen;
3318 	return (!TAILQ_EMPTY(&so->so_comp));
3319 }
3320 
3321 int
3322 socheckuid(struct socket *so, uid_t uid)
3323 {
3324 
3325 	if (so == NULL)
3326 		return (EPERM);
3327 	if (so->so_cred->cr_uid != uid)
3328 		return (EPERM);
3329 	return (0);
3330 }
3331 
3332 /*
3333  * These functions are used by protocols to notify the socket layer (and its
3334  * consumers) of state changes in the sockets driven by protocol-side events.
3335  */
3336 
3337 /*
3338  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3339  *
3340  * Normal sequence from the active (originating) side is that
3341  * soisconnecting() is called during processing of connect() call, resulting
3342  * in an eventual call to soisconnected() if/when the connection is
3343  * established.  When the connection is torn down soisdisconnecting() is
3344  * called during processing of disconnect() call, and soisdisconnected() is
3345  * called when the connection to the peer is totally severed.  The semantics
3346  * of these routines are such that connectionless protocols can call
3347  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3348  * calls when setting up a ``connection'' takes no time.
3349  *
3350  * From the passive side, a socket is created with two queues of sockets:
3351  * so_incomp for connections in progress and so_comp for connections already
3352  * made and awaiting user acceptance.  As a protocol is preparing incoming
3353  * connections, it creates a socket structure queued on so_incomp by calling
3354  * sonewconn().  When the connection is established, soisconnected() is
3355  * called, and transfers the socket structure to so_comp, making it available
3356  * to accept().
3357  *
3358  * If a socket is closed with sockets on either so_incomp or so_comp, these
3359  * sockets are dropped.
3360  *
3361  * If higher-level protocols are implemented in the kernel, the wakeups done
3362  * here will sometimes cause software-interrupt process scheduling.
3363  */
3364 void
3365 soisconnecting(struct socket *so)
3366 {
3367 
3368 	SOCK_LOCK(so);
3369 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3370 	so->so_state |= SS_ISCONNECTING;
3371 	SOCK_UNLOCK(so);
3372 }
3373 
3374 void
3375 soisconnected(struct socket *so)
3376 {
3377 	struct socket *head;
3378 	int ret;
3379 
3380 restart:
3381 	ACCEPT_LOCK();
3382 	SOCK_LOCK(so);
3383 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3384 	so->so_state |= SS_ISCONNECTED;
3385 	head = so->so_head;
3386 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3387 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3388 			SOCK_UNLOCK(so);
3389 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3390 			head->so_incqlen--;
3391 			so->so_qstate &= ~SQ_INCOMP;
3392 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3393 			head->so_qlen++;
3394 			so->so_qstate |= SQ_COMP;
3395 			ACCEPT_UNLOCK();
3396 			sorwakeup(head);
3397 			wakeup_one(&head->so_timeo);
3398 		} else {
3399 			ACCEPT_UNLOCK();
3400 			soupcall_set(so, SO_RCV,
3401 			    head->so_accf->so_accept_filter->accf_callback,
3402 			    head->so_accf->so_accept_filter_arg);
3403 			so->so_options &= ~SO_ACCEPTFILTER;
3404 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3405 			    head->so_accf->so_accept_filter_arg, M_NOWAIT);
3406 			if (ret == SU_ISCONNECTED)
3407 				soupcall_clear(so, SO_RCV);
3408 			SOCK_UNLOCK(so);
3409 			if (ret == SU_ISCONNECTED)
3410 				goto restart;
3411 		}
3412 		return;
3413 	}
3414 	SOCK_UNLOCK(so);
3415 	ACCEPT_UNLOCK();
3416 	wakeup(&so->so_timeo);
3417 	sorwakeup(so);
3418 	sowwakeup(so);
3419 }
3420 
3421 void
3422 soisdisconnecting(struct socket *so)
3423 {
3424 
3425 	/*
3426 	 * Note: This code assumes that SOCK_LOCK(so) and
3427 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3428 	 */
3429 	SOCKBUF_LOCK(&so->so_rcv);
3430 	so->so_state &= ~SS_ISCONNECTING;
3431 	so->so_state |= SS_ISDISCONNECTING;
3432 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3433 	sorwakeup_locked(so);
3434 	SOCKBUF_LOCK(&so->so_snd);
3435 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3436 	sowwakeup_locked(so);
3437 	wakeup(&so->so_timeo);
3438 }
3439 
3440 void
3441 soisdisconnected(struct socket *so)
3442 {
3443 
3444 	/*
3445 	 * Note: This code assumes that SOCK_LOCK(so) and
3446 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3447 	 */
3448 	SOCKBUF_LOCK(&so->so_rcv);
3449 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3450 	so->so_state |= SS_ISDISCONNECTED;
3451 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3452 	sorwakeup_locked(so);
3453 	SOCKBUF_LOCK(&so->so_snd);
3454 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3455 	sbdrop_locked(&so->so_snd, sbused(&so->so_snd));
3456 	sowwakeup_locked(so);
3457 	wakeup(&so->so_timeo);
3458 }
3459 
3460 /*
3461  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3462  */
3463 struct sockaddr *
3464 sodupsockaddr(const struct sockaddr *sa, int mflags)
3465 {
3466 	struct sockaddr *sa2;
3467 
3468 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3469 	if (sa2)
3470 		bcopy(sa, sa2, sa->sa_len);
3471 	return sa2;
3472 }
3473 
3474 /*
3475  * Register per-socket buffer upcalls.
3476  */
3477 void
3478 soupcall_set(struct socket *so, int which,
3479     int (*func)(struct socket *, void *, int), void *arg)
3480 {
3481 	struct sockbuf *sb;
3482 
3483 	switch (which) {
3484 	case SO_RCV:
3485 		sb = &so->so_rcv;
3486 		break;
3487 	case SO_SND:
3488 		sb = &so->so_snd;
3489 		break;
3490 	default:
3491 		panic("soupcall_set: bad which");
3492 	}
3493 	SOCKBUF_LOCK_ASSERT(sb);
3494 #if 0
3495 	/* XXX: accf_http actually wants to do this on purpose. */
3496 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3497 #endif
3498 	sb->sb_upcall = func;
3499 	sb->sb_upcallarg = arg;
3500 	sb->sb_flags |= SB_UPCALL;
3501 }
3502 
3503 void
3504 soupcall_clear(struct socket *so, int which)
3505 {
3506 	struct sockbuf *sb;
3507 
3508 	switch (which) {
3509 	case SO_RCV:
3510 		sb = &so->so_rcv;
3511 		break;
3512 	case SO_SND:
3513 		sb = &so->so_snd;
3514 		break;
3515 	default:
3516 		panic("soupcall_clear: bad which");
3517 	}
3518 	SOCKBUF_LOCK_ASSERT(sb);
3519 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3520 	sb->sb_upcall = NULL;
3521 	sb->sb_upcallarg = NULL;
3522 	sb->sb_flags &= ~SB_UPCALL;
3523 }
3524 
3525 /*
3526  * Create an external-format (``xsocket'') structure using the information in
3527  * the kernel-format socket structure pointed to by so.  This is done to
3528  * reduce the spew of irrelevant information over this interface, to isolate
3529  * user code from changes in the kernel structure, and potentially to provide
3530  * information-hiding if we decide that some of this information should be
3531  * hidden from users.
3532  */
3533 void
3534 sotoxsocket(struct socket *so, struct xsocket *xso)
3535 {
3536 
3537 	xso->xso_len = sizeof *xso;
3538 	xso->xso_so = so;
3539 	xso->so_type = so->so_type;
3540 	xso->so_options = so->so_options;
3541 	xso->so_linger = so->so_linger;
3542 	xso->so_state = so->so_state;
3543 	xso->so_pcb = so->so_pcb;
3544 	xso->xso_protocol = so->so_proto->pr_protocol;
3545 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3546 	xso->so_qlen = so->so_qlen;
3547 	xso->so_incqlen = so->so_incqlen;
3548 	xso->so_qlimit = so->so_qlimit;
3549 	xso->so_timeo = so->so_timeo;
3550 	xso->so_error = so->so_error;
3551 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3552 	xso->so_oobmark = so->so_oobmark;
3553 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3554 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3555 	xso->so_uid = so->so_cred->cr_uid;
3556 }
3557 
3558 
3559 /*
3560  * Socket accessor functions to provide external consumers with
3561  * a safe interface to socket state
3562  *
3563  */
3564 
3565 void
3566 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3567     void *arg)
3568 {
3569 
3570 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3571 		func(so, arg);
3572 }
3573 
3574 struct sockbuf *
3575 so_sockbuf_rcv(struct socket *so)
3576 {
3577 
3578 	return (&so->so_rcv);
3579 }
3580 
3581 struct sockbuf *
3582 so_sockbuf_snd(struct socket *so)
3583 {
3584 
3585 	return (&so->so_snd);
3586 }
3587 
3588 int
3589 so_state_get(const struct socket *so)
3590 {
3591 
3592 	return (so->so_state);
3593 }
3594 
3595 void
3596 so_state_set(struct socket *so, int val)
3597 {
3598 
3599 	so->so_state = val;
3600 }
3601 
3602 int
3603 so_options_get(const struct socket *so)
3604 {
3605 
3606 	return (so->so_options);
3607 }
3608 
3609 void
3610 so_options_set(struct socket *so, int val)
3611 {
3612 
3613 	so->so_options = val;
3614 }
3615 
3616 int
3617 so_error_get(const struct socket *so)
3618 {
3619 
3620 	return (so->so_error);
3621 }
3622 
3623 void
3624 so_error_set(struct socket *so, int val)
3625 {
3626 
3627 	so->so_error = val;
3628 }
3629 
3630 int
3631 so_linger_get(const struct socket *so)
3632 {
3633 
3634 	return (so->so_linger);
3635 }
3636 
3637 void
3638 so_linger_set(struct socket *so, int val)
3639 {
3640 
3641 	so->so_linger = val;
3642 }
3643 
3644 struct protosw *
3645 so_protosw_get(const struct socket *so)
3646 {
3647 
3648 	return (so->so_proto);
3649 }
3650 
3651 void
3652 so_protosw_set(struct socket *so, struct protosw *val)
3653 {
3654 
3655 	so->so_proto = val;
3656 }
3657 
3658 void
3659 so_sorwakeup(struct socket *so)
3660 {
3661 
3662 	sorwakeup(so);
3663 }
3664 
3665 void
3666 so_sowwakeup(struct socket *so)
3667 {
3668 
3669 	sowwakeup(so);
3670 }
3671 
3672 void
3673 so_sorwakeup_locked(struct socket *so)
3674 {
3675 
3676 	sorwakeup_locked(so);
3677 }
3678 
3679 void
3680 so_sowwakeup_locked(struct socket *so)
3681 {
3682 
3683 	sowwakeup_locked(so);
3684 }
3685 
3686 void
3687 so_lock(struct socket *so)
3688 {
3689 
3690 	SOCK_LOCK(so);
3691 }
3692 
3693 void
3694 so_unlock(struct socket *so)
3695 {
3696 
3697 	SOCK_UNLOCK(so);
3698 }
3699