xref: /freebsd/sys/kern/uipc_socket.c (revision 13de33a5dc2304b13d595d75d48c51793958474f)
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/kernel.h>
122 #include <sys/event.h>
123 #include <sys/eventhandler.h>
124 #include <sys/poll.h>
125 #include <sys/proc.h>
126 #include <sys/protosw.h>
127 #include <sys/socket.h>
128 #include <sys/socketvar.h>
129 #include <sys/resourcevar.h>
130 #include <net/route.h>
131 #include <sys/signalvar.h>
132 #include <sys/stat.h>
133 #include <sys/sx.h>
134 #include <sys/sysctl.h>
135 #include <sys/uio.h>
136 #include <sys/jail.h>
137 #include <sys/syslog.h>
138 #include <netinet/in.h>
139 
140 #include <net/vnet.h>
141 
142 #include <security/mac/mac_framework.h>
143 
144 #include <vm/uma.h>
145 
146 #ifdef COMPAT_FREEBSD32
147 #include <sys/mount.h>
148 #include <sys/sysent.h>
149 #include <compat/freebsd32/freebsd32.h>
150 #endif
151 
152 static int	soreceive_rcvoob(struct socket *so, struct uio *uio,
153 		    int flags);
154 
155 static void	filt_sordetach(struct knote *kn);
156 static int	filt_soread(struct knote *kn, long hint);
157 static void	filt_sowdetach(struct knote *kn);
158 static int	filt_sowrite(struct knote *kn, long hint);
159 static int	filt_solisten(struct knote *kn, long hint);
160 
161 static struct filterops solisten_filtops = {
162 	.f_isfd = 1,
163 	.f_detach = filt_sordetach,
164 	.f_event = filt_solisten,
165 };
166 static struct filterops soread_filtops = {
167 	.f_isfd = 1,
168 	.f_detach = filt_sordetach,
169 	.f_event = filt_soread,
170 };
171 static struct filterops sowrite_filtops = {
172 	.f_isfd = 1,
173 	.f_detach = filt_sowdetach,
174 	.f_event = filt_sowrite,
175 };
176 
177 so_gen_t	so_gencnt;	/* generation count for sockets */
178 
179 MALLOC_DEFINE(M_SONAME, "soname", "socket name");
180 MALLOC_DEFINE(M_PCB, "pcb", "protocol control block");
181 
182 #define	VNET_SO_ASSERT(so)						\
183 	VNET_ASSERT(curvnet != NULL,					\
184 	    ("%s:%d curvnet is NULL, so=%p", __func__, __LINE__, (so)));
185 
186 /*
187  * Limit on the number of connections in the listen queue waiting
188  * for accept(2).
189  * NB: The orginal sysctl somaxconn is still available but hidden
190  * to prevent confusion about the actual purpose of this number.
191  */
192 static int somaxconn = SOMAXCONN;
193 
194 static int
195 sysctl_somaxconn(SYSCTL_HANDLER_ARGS)
196 {
197 	int error;
198 	int val;
199 
200 	val = somaxconn;
201 	error = sysctl_handle_int(oidp, &val, 0, req);
202 	if (error || !req->newptr )
203 		return (error);
204 
205 	if (val < 1 || val > USHRT_MAX)
206 		return (EINVAL);
207 
208 	somaxconn = val;
209 	return (0);
210 }
211 SYSCTL_PROC(_kern_ipc, OID_AUTO, soacceptqueue, CTLTYPE_UINT | CTLFLAG_RW,
212     0, sizeof(int), sysctl_somaxconn, "I",
213     "Maximum listen socket pending connection accept queue size");
214 SYSCTL_PROC(_kern_ipc, KIPC_SOMAXCONN, somaxconn,
215     CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_SKIP,
216     0, sizeof(int), sysctl_somaxconn, "I",
217     "Maximum listen socket pending connection accept queue size (compat)");
218 
219 static int numopensockets;
220 SYSCTL_INT(_kern_ipc, OID_AUTO, numopensockets, CTLFLAG_RD,
221     &numopensockets, 0, "Number of open sockets");
222 
223 /*
224  * accept_mtx locks down per-socket fields relating to accept queues.  See
225  * socketvar.h for an annotation of the protected fields of struct socket.
226  */
227 struct mtx accept_mtx;
228 MTX_SYSINIT(accept_mtx, &accept_mtx, "accept", MTX_DEF);
229 
230 /*
231  * so_global_mtx protects so_gencnt, numopensockets, and the per-socket
232  * so_gencnt field.
233  */
234 static struct mtx so_global_mtx;
235 MTX_SYSINIT(so_global_mtx, &so_global_mtx, "so_glabel", MTX_DEF);
236 
237 /*
238  * General IPC sysctl name space, used by sockets and a variety of other IPC
239  * types.
240  */
241 SYSCTL_NODE(_kern, KERN_IPC, ipc, CTLFLAG_RW, 0, "IPC");
242 
243 /*
244  * Initialize the socket subsystem and set up the socket
245  * memory allocator.
246  */
247 static uma_zone_t socket_zone;
248 int	maxsockets;
249 
250 static void
251 socket_zone_change(void *tag)
252 {
253 
254 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
255 }
256 
257 static void
258 socket_init(void *tag)
259 {
260 
261 	socket_zone = uma_zcreate("socket", sizeof(struct socket), NULL, NULL,
262 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
263 	maxsockets = uma_zone_set_max(socket_zone, maxsockets);
264 	uma_zone_set_warning(socket_zone, "kern.ipc.maxsockets limit reached");
265 	EVENTHANDLER_REGISTER(maxsockets_change, socket_zone_change, NULL,
266 	    EVENTHANDLER_PRI_FIRST);
267 }
268 SYSINIT(socket, SI_SUB_PROTO_DOMAININIT, SI_ORDER_ANY, socket_init, NULL);
269 
270 /*
271  * Initialise maxsockets.  This SYSINIT must be run after
272  * tunable_mbinit().
273  */
274 static void
275 init_maxsockets(void *ignored)
276 {
277 
278 	TUNABLE_INT_FETCH("kern.ipc.maxsockets", &maxsockets);
279 	maxsockets = imax(maxsockets, maxfiles);
280 }
281 SYSINIT(param, SI_SUB_TUNABLES, SI_ORDER_ANY, init_maxsockets, NULL);
282 
283 /*
284  * Sysctl to get and set the maximum global sockets limit.  Notify protocols
285  * of the change so that they can update their dependent limits as required.
286  */
287 static int
288 sysctl_maxsockets(SYSCTL_HANDLER_ARGS)
289 {
290 	int error, newmaxsockets;
291 
292 	newmaxsockets = maxsockets;
293 	error = sysctl_handle_int(oidp, &newmaxsockets, 0, req);
294 	if (error == 0 && req->newptr) {
295 		if (newmaxsockets > maxsockets &&
296 		    newmaxsockets <= maxfiles) {
297 			maxsockets = newmaxsockets;
298 			EVENTHANDLER_INVOKE(maxsockets_change);
299 		} else
300 			error = EINVAL;
301 	}
302 	return (error);
303 }
304 SYSCTL_PROC(_kern_ipc, OID_AUTO, maxsockets, CTLTYPE_INT|CTLFLAG_RW,
305     &maxsockets, 0, sysctl_maxsockets, "IU",
306     "Maximum number of sockets avaliable");
307 
308 /*
309  * Socket operation routines.  These routines are called by the routines in
310  * sys_socket.c or from a system process, and implement the semantics of
311  * socket operations by switching out to the protocol specific routines.
312  */
313 
314 /*
315  * Get a socket structure from our zone, and initialize it.  Note that it
316  * would probably be better to allocate socket and PCB at the same time, but
317  * I'm not convinced that all the protocols can be easily modified to do
318  * this.
319  *
320  * soalloc() returns a socket with a ref count of 0.
321  */
322 static struct socket *
323 soalloc(struct vnet *vnet)
324 {
325 	struct socket *so;
326 
327 	so = uma_zalloc(socket_zone, M_NOWAIT | M_ZERO);
328 	if (so == NULL)
329 		return (NULL);
330 #ifdef MAC
331 	if (mac_socket_init(so, M_NOWAIT) != 0) {
332 		uma_zfree(socket_zone, so);
333 		return (NULL);
334 	}
335 #endif
336 	SOCKBUF_LOCK_INIT(&so->so_snd, "so_snd");
337 	SOCKBUF_LOCK_INIT(&so->so_rcv, "so_rcv");
338 	sx_init(&so->so_snd.sb_sx, "so_snd_sx");
339 	sx_init(&so->so_rcv.sb_sx, "so_rcv_sx");
340 	TAILQ_INIT(&so->so_aiojobq);
341 	mtx_lock(&so_global_mtx);
342 	so->so_gencnt = ++so_gencnt;
343 	++numopensockets;
344 #ifdef VIMAGE
345 	VNET_ASSERT(vnet != NULL, ("%s:%d vnet is NULL, so=%p",
346 	    __func__, __LINE__, so));
347 	vnet->vnet_sockcnt++;
348 	so->so_vnet = vnet;
349 #endif
350 	mtx_unlock(&so_global_mtx);
351 	return (so);
352 }
353 
354 /*
355  * Free the storage associated with a socket at the socket layer, tear down
356  * locks, labels, etc.  All protocol state is assumed already to have been
357  * torn down (and possibly never set up) by the caller.
358  */
359 static void
360 sodealloc(struct socket *so)
361 {
362 
363 	KASSERT(so->so_count == 0, ("sodealloc(): so_count %d", so->so_count));
364 	KASSERT(so->so_pcb == NULL, ("sodealloc(): so_pcb != NULL"));
365 
366 	mtx_lock(&so_global_mtx);
367 	so->so_gencnt = ++so_gencnt;
368 	--numopensockets;	/* Could be below, but faster here. */
369 #ifdef VIMAGE
370 	VNET_ASSERT(so->so_vnet != NULL, ("%s:%d so_vnet is NULL, so=%p",
371 	    __func__, __LINE__, so));
372 	so->so_vnet->vnet_sockcnt--;
373 #endif
374 	mtx_unlock(&so_global_mtx);
375 	if (so->so_rcv.sb_hiwat)
376 		(void)chgsbsize(so->so_cred->cr_uidinfo,
377 		    &so->so_rcv.sb_hiwat, 0, RLIM_INFINITY);
378 	if (so->so_snd.sb_hiwat)
379 		(void)chgsbsize(so->so_cred->cr_uidinfo,
380 		    &so->so_snd.sb_hiwat, 0, RLIM_INFINITY);
381 #ifdef INET
382 	/* remove acccept filter if one is present. */
383 	if (so->so_accf != NULL)
384 		do_setopt_accept_filter(so, NULL);
385 #endif
386 #ifdef MAC
387 	mac_socket_destroy(so);
388 #endif
389 	crfree(so->so_cred);
390 	sx_destroy(&so->so_snd.sb_sx);
391 	sx_destroy(&so->so_rcv.sb_sx);
392 	SOCKBUF_LOCK_DESTROY(&so->so_snd);
393 	SOCKBUF_LOCK_DESTROY(&so->so_rcv);
394 	uma_zfree(socket_zone, so);
395 }
396 
397 /*
398  * socreate returns a socket with a ref count of 1.  The socket should be
399  * closed with soclose().
400  */
401 int
402 socreate(int dom, struct socket **aso, int type, int proto,
403     struct ucred *cred, struct thread *td)
404 {
405 	struct protosw *prp;
406 	struct socket *so;
407 	int error;
408 
409 	if (proto)
410 		prp = pffindproto(dom, proto, type);
411 	else
412 		prp = pffindtype(dom, type);
413 
414 	if (prp == NULL) {
415 		/* No support for domain. */
416 		if (pffinddomain(dom) == NULL)
417 			return (EAFNOSUPPORT);
418 		/* No support for socket type. */
419 		if (proto == 0 && type != 0)
420 			return (EPROTOTYPE);
421 		return (EPROTONOSUPPORT);
422 	}
423 	if (prp->pr_usrreqs->pru_attach == NULL ||
424 	    prp->pr_usrreqs->pru_attach == pru_attach_notsupp)
425 		return (EPROTONOSUPPORT);
426 
427 	if (prison_check_af(cred, prp->pr_domain->dom_family) != 0)
428 		return (EPROTONOSUPPORT);
429 
430 	if (prp->pr_type != type)
431 		return (EPROTOTYPE);
432 	so = soalloc(CRED_TO_VNET(cred));
433 	if (so == NULL)
434 		return (ENOBUFS);
435 
436 	TAILQ_INIT(&so->so_incomp);
437 	TAILQ_INIT(&so->so_comp);
438 	so->so_type = type;
439 	so->so_cred = crhold(cred);
440 	if ((prp->pr_domain->dom_family == PF_INET) ||
441 	    (prp->pr_domain->dom_family == PF_INET6) ||
442 	    (prp->pr_domain->dom_family == PF_ROUTE))
443 		so->so_fibnum = td->td_proc->p_fibnum;
444 	else
445 		so->so_fibnum = 0;
446 	so->so_proto = prp;
447 #ifdef MAC
448 	mac_socket_create(cred, so);
449 #endif
450 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
451 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
452 	so->so_count = 1;
453 	/*
454 	 * Auto-sizing of socket buffers is managed by the protocols and
455 	 * the appropriate flags must be set in the pru_attach function.
456 	 */
457 	CURVNET_SET(so->so_vnet);
458 	error = (*prp->pr_usrreqs->pru_attach)(so, proto, td);
459 	CURVNET_RESTORE();
460 	if (error) {
461 		KASSERT(so->so_count == 1, ("socreate: so_count %d",
462 		    so->so_count));
463 		so->so_count = 0;
464 		sodealloc(so);
465 		return (error);
466 	}
467 	*aso = so;
468 	return (0);
469 }
470 
471 #ifdef REGRESSION
472 static int regression_sonewconn_earlytest = 1;
473 SYSCTL_INT(_regression, OID_AUTO, sonewconn_earlytest, CTLFLAG_RW,
474     &regression_sonewconn_earlytest, 0, "Perform early sonewconn limit test");
475 #endif
476 
477 /*
478  * When an attempt at a new connection is noted on a socket which accepts
479  * connections, sonewconn is called.  If the connection is possible (subject
480  * to space constraints, etc.) then we allocate a new structure, propoerly
481  * linked into the data structure of the original socket, and return this.
482  * Connstatus may be 0, or SS_ISCONFIRMING, or SS_ISCONNECTED.
483  *
484  * Note: the ref count on the socket is 0 on return.
485  */
486 struct socket *
487 sonewconn(struct socket *head, int connstatus)
488 {
489 	static struct timeval lastover;
490 	static struct timeval overinterval = { 60, 0 };
491 	static int overcount;
492 
493 	struct socket *so;
494 	int over;
495 
496 	ACCEPT_LOCK();
497 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
498 	ACCEPT_UNLOCK();
499 #ifdef REGRESSION
500 	if (regression_sonewconn_earlytest && over) {
501 #else
502 	if (over) {
503 #endif
504 		overcount++;
505 
506 		if (ratecheck(&lastover, &overinterval)) {
507 			log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
508 			    "%i already in queue awaiting acceptance "
509 			    "(%d occurrences)\n",
510 			    __func__, head->so_pcb, head->so_qlen, overcount);
511 
512 			overcount = 0;
513 		}
514 
515 		return (NULL);
516 	}
517 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
518 	    __func__, __LINE__, head));
519 	so = soalloc(head->so_vnet);
520 	if (so == NULL) {
521 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
522 		    "limit reached or out of memory\n",
523 		    __func__, head->so_pcb);
524 		return (NULL);
525 	}
526 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
527 		connstatus = 0;
528 	so->so_head = head;
529 	so->so_type = head->so_type;
530 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
531 	so->so_linger = head->so_linger;
532 	so->so_state = head->so_state | SS_NOFDREF;
533 	so->so_fibnum = head->so_fibnum;
534 	so->so_proto = head->so_proto;
535 	so->so_cred = crhold(head->so_cred);
536 #ifdef MAC
537 	mac_socket_newconn(head, so);
538 #endif
539 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
540 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
541 	VNET_SO_ASSERT(head);
542 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
543 		sodealloc(so);
544 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
545 		    __func__, head->so_pcb);
546 		return (NULL);
547 	}
548 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
549 		sodealloc(so);
550 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
551 		    __func__, head->so_pcb);
552 		return (NULL);
553 	}
554 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
555 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
556 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
557 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
558 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
559 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
560 	so->so_state |= connstatus;
561 	ACCEPT_LOCK();
562 	/*
563 	 * The accept socket may be tearing down but we just
564 	 * won a race on the ACCEPT_LOCK.
565 	 * However, if sctp_peeloff() is called on a 1-to-many
566 	 * style socket, the SO_ACCEPTCONN doesn't need to be set.
567 	 */
568 	if (!(head->so_options & SO_ACCEPTCONN) &&
569 	    ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
570 	     (head->so_type != SOCK_SEQPACKET))) {
571 		SOCK_LOCK(so);
572 		so->so_head = NULL;
573 		sofree(so);		/* NB: returns ACCEPT_UNLOCK'ed. */
574 		return (NULL);
575 	}
576 	if (connstatus) {
577 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
578 		so->so_qstate |= SQ_COMP;
579 		head->so_qlen++;
580 	} else {
581 		/*
582 		 * Keep removing sockets from the head until there's room for
583 		 * us to insert on the tail.  In pre-locking revisions, this
584 		 * was a simple if(), but as we could be racing with other
585 		 * threads and soabort() requires dropping locks, we must
586 		 * loop waiting for the condition to be true.
587 		 */
588 		while (head->so_incqlen > head->so_qlimit) {
589 			struct socket *sp;
590 			sp = TAILQ_FIRST(&head->so_incomp);
591 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
592 			head->so_incqlen--;
593 			sp->so_qstate &= ~SQ_INCOMP;
594 			sp->so_head = NULL;
595 			ACCEPT_UNLOCK();
596 			soabort(sp);
597 			ACCEPT_LOCK();
598 		}
599 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
600 		so->so_qstate |= SQ_INCOMP;
601 		head->so_incqlen++;
602 	}
603 	ACCEPT_UNLOCK();
604 	if (connstatus) {
605 		sorwakeup(head);
606 		wakeup_one(&head->so_timeo);
607 	}
608 	return (so);
609 }
610 
611 int
612 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
613 {
614 	int error;
615 
616 	CURVNET_SET(so->so_vnet);
617 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
618 	CURVNET_RESTORE();
619 	return (error);
620 }
621 
622 int
623 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
624 {
625 	int error;
626 
627 	CURVNET_SET(so->so_vnet);
628 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
629 	CURVNET_RESTORE();
630 	return (error);
631 }
632 
633 /*
634  * solisten() transitions a socket from a non-listening state to a listening
635  * state, but can also be used to update the listen queue depth on an
636  * existing listen socket.  The protocol will call back into the sockets
637  * layer using solisten_proto_check() and solisten_proto() to check and set
638  * socket-layer listen state.  Call backs are used so that the protocol can
639  * acquire both protocol and socket layer locks in whatever order is required
640  * by the protocol.
641  *
642  * Protocol implementors are advised to hold the socket lock across the
643  * socket-layer test and set to avoid races at the socket layer.
644  */
645 int
646 solisten(struct socket *so, int backlog, struct thread *td)
647 {
648 	int error;
649 
650 	CURVNET_SET(so->so_vnet);
651 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
652 	CURVNET_RESTORE();
653 	return (error);
654 }
655 
656 int
657 solisten_proto_check(struct socket *so)
658 {
659 
660 	SOCK_LOCK_ASSERT(so);
661 
662 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
663 	    SS_ISDISCONNECTING))
664 		return (EINVAL);
665 	return (0);
666 }
667 
668 void
669 solisten_proto(struct socket *so, int backlog)
670 {
671 
672 	SOCK_LOCK_ASSERT(so);
673 
674 	if (backlog < 0 || backlog > somaxconn)
675 		backlog = somaxconn;
676 	so->so_qlimit = backlog;
677 	so->so_options |= SO_ACCEPTCONN;
678 }
679 
680 /*
681  * Evaluate the reference count and named references on a socket; if no
682  * references remain, free it.  This should be called whenever a reference is
683  * released, such as in sorele(), but also when named reference flags are
684  * cleared in socket or protocol code.
685  *
686  * sofree() will free the socket if:
687  *
688  * - There are no outstanding file descriptor references or related consumers
689  *   (so_count == 0).
690  *
691  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
692  *
693  * - The protocol does not have an outstanding strong reference on the socket
694  *   (SS_PROTOREF).
695  *
696  * - The socket is not in a completed connection queue, so a process has been
697  *   notified that it is present.  If it is removed, the user process may
698  *   block in accept() despite select() saying the socket was ready.
699  */
700 void
701 sofree(struct socket *so)
702 {
703 	struct protosw *pr = so->so_proto;
704 	struct socket *head;
705 
706 	ACCEPT_LOCK_ASSERT();
707 	SOCK_LOCK_ASSERT(so);
708 
709 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
710 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
711 		SOCK_UNLOCK(so);
712 		ACCEPT_UNLOCK();
713 		return;
714 	}
715 
716 	head = so->so_head;
717 	if (head != NULL) {
718 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
719 		    (so->so_qstate & SQ_INCOMP) != 0,
720 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
721 		    "SQ_INCOMP"));
722 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
723 		    (so->so_qstate & SQ_INCOMP) == 0,
724 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
725 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
726 		head->so_incqlen--;
727 		so->so_qstate &= ~SQ_INCOMP;
728 		so->so_head = NULL;
729 	}
730 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
731 	    (so->so_qstate & SQ_INCOMP) == 0,
732 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
733 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
734 	if (so->so_options & SO_ACCEPTCONN) {
735 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
736 		    ("sofree: so_comp populated"));
737 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
738 		    ("sofree: so_incomp populated"));
739 	}
740 	SOCK_UNLOCK(so);
741 	ACCEPT_UNLOCK();
742 
743 	VNET_SO_ASSERT(so);
744 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
745 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
746 	if (pr->pr_usrreqs->pru_detach != NULL)
747 		(*pr->pr_usrreqs->pru_detach)(so);
748 
749 	/*
750 	 * From this point on, we assume that no other references to this
751 	 * socket exist anywhere else in the stack.  Therefore, no locks need
752 	 * to be acquired or held.
753 	 *
754 	 * We used to do a lot of socket buffer and socket locking here, as
755 	 * well as invoke sorflush() and perform wakeups.  The direct call to
756 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
757 	 * necessary from sorflush().
758 	 *
759 	 * Notice that the socket buffer and kqueue state are torn down
760 	 * before calling pru_detach.  This means that protocols shold not
761 	 * assume they can perform socket wakeups, etc, in their detach code.
762 	 */
763 	sbdestroy(&so->so_snd, so);
764 	sbdestroy(&so->so_rcv, so);
765 	seldrain(&so->so_snd.sb_sel);
766 	seldrain(&so->so_rcv.sb_sel);
767 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
768 	knlist_destroy(&so->so_snd.sb_sel.si_note);
769 	sodealloc(so);
770 }
771 
772 /*
773  * Close a socket on last file table reference removal.  Initiate disconnect
774  * if connected.  Free socket when disconnect complete.
775  *
776  * This function will sorele() the socket.  Note that soclose() may be called
777  * prior to the ref count reaching zero.  The actual socket structure will
778  * not be freed until the ref count reaches zero.
779  */
780 int
781 soclose(struct socket *so)
782 {
783 	int error = 0;
784 
785 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
786 
787 	CURVNET_SET(so->so_vnet);
788 	funsetown(&so->so_sigio);
789 	if (so->so_state & SS_ISCONNECTED) {
790 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
791 			error = sodisconnect(so);
792 			if (error) {
793 				if (error == ENOTCONN)
794 					error = 0;
795 				goto drop;
796 			}
797 		}
798 		if (so->so_options & SO_LINGER) {
799 			if ((so->so_state & SS_ISDISCONNECTING) &&
800 			    (so->so_state & SS_NBIO))
801 				goto drop;
802 			while (so->so_state & SS_ISCONNECTED) {
803 				error = tsleep(&so->so_timeo,
804 				    PSOCK | PCATCH, "soclos",
805 				    so->so_linger * hz);
806 				if (error)
807 					break;
808 			}
809 		}
810 	}
811 
812 drop:
813 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
814 		(*so->so_proto->pr_usrreqs->pru_close)(so);
815 	ACCEPT_LOCK();
816 	if (so->so_options & SO_ACCEPTCONN) {
817 		struct socket *sp;
818 		/*
819 		 * Prevent new additions to the accept queues due
820 		 * to ACCEPT_LOCK races while we are draining them.
821 		 */
822 		so->so_options &= ~SO_ACCEPTCONN;
823 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
824 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
825 			so->so_incqlen--;
826 			sp->so_qstate &= ~SQ_INCOMP;
827 			sp->so_head = NULL;
828 			ACCEPT_UNLOCK();
829 			soabort(sp);
830 			ACCEPT_LOCK();
831 		}
832 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
833 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
834 			so->so_qlen--;
835 			sp->so_qstate &= ~SQ_COMP;
836 			sp->so_head = NULL;
837 			ACCEPT_UNLOCK();
838 			soabort(sp);
839 			ACCEPT_LOCK();
840 		}
841 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
842 		    ("%s: so_comp populated", __func__));
843 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
844 		    ("%s: so_incomp populated", __func__));
845 	}
846 	SOCK_LOCK(so);
847 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
848 	so->so_state |= SS_NOFDREF;
849 	sorele(so);			/* NB: Returns with ACCEPT_UNLOCK(). */
850 	CURVNET_RESTORE();
851 	return (error);
852 }
853 
854 /*
855  * soabort() is used to abruptly tear down a connection, such as when a
856  * resource limit is reached (listen queue depth exceeded), or if a listen
857  * socket is closed while there are sockets waiting to be accepted.
858  *
859  * This interface is tricky, because it is called on an unreferenced socket,
860  * and must be called only by a thread that has actually removed the socket
861  * from the listen queue it was on, or races with other threads are risked.
862  *
863  * This interface will call into the protocol code, so must not be called
864  * with any socket locks held.  Protocols do call it while holding their own
865  * recursible protocol mutexes, but this is something that should be subject
866  * to review in the future.
867  */
868 void
869 soabort(struct socket *so)
870 {
871 
872 	/*
873 	 * In as much as is possible, assert that no references to this
874 	 * socket are held.  This is not quite the same as asserting that the
875 	 * current thread is responsible for arranging for no references, but
876 	 * is as close as we can get for now.
877 	 */
878 	KASSERT(so->so_count == 0, ("soabort: so_count"));
879 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
880 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
881 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
882 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
883 	VNET_SO_ASSERT(so);
884 
885 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
886 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
887 	ACCEPT_LOCK();
888 	SOCK_LOCK(so);
889 	sofree(so);
890 }
891 
892 int
893 soaccept(struct socket *so, struct sockaddr **nam)
894 {
895 	int error;
896 
897 	SOCK_LOCK(so);
898 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
899 	so->so_state &= ~SS_NOFDREF;
900 	SOCK_UNLOCK(so);
901 
902 	CURVNET_SET(so->so_vnet);
903 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
904 	CURVNET_RESTORE();
905 	return (error);
906 }
907 
908 int
909 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
910 {
911 
912 	return (soconnectat(AT_FDCWD, so, nam, td));
913 }
914 
915 int
916 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
917 {
918 	int error;
919 
920 	if (so->so_options & SO_ACCEPTCONN)
921 		return (EOPNOTSUPP);
922 
923 	CURVNET_SET(so->so_vnet);
924 	/*
925 	 * If protocol is connection-based, can only connect once.
926 	 * Otherwise, if connected, try to disconnect first.  This allows
927 	 * user to disconnect by connecting to, e.g., a null address.
928 	 */
929 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
930 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
931 	    (error = sodisconnect(so)))) {
932 		error = EISCONN;
933 	} else {
934 		/*
935 		 * Prevent accumulated error from previous connection from
936 		 * biting us.
937 		 */
938 		so->so_error = 0;
939 		if (fd == AT_FDCWD) {
940 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
941 			    nam, td);
942 		} else {
943 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
944 			    so, nam, td);
945 		}
946 	}
947 	CURVNET_RESTORE();
948 
949 	return (error);
950 }
951 
952 int
953 soconnect2(struct socket *so1, struct socket *so2)
954 {
955 	int error;
956 
957 	CURVNET_SET(so1->so_vnet);
958 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
959 	CURVNET_RESTORE();
960 	return (error);
961 }
962 
963 int
964 sodisconnect(struct socket *so)
965 {
966 	int error;
967 
968 	if ((so->so_state & SS_ISCONNECTED) == 0)
969 		return (ENOTCONN);
970 	if (so->so_state & SS_ISDISCONNECTING)
971 		return (EALREADY);
972 	VNET_SO_ASSERT(so);
973 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
974 	return (error);
975 }
976 
977 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
978 
979 int
980 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
981     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
982 {
983 	long space;
984 	ssize_t resid;
985 	int clen = 0, error, dontroute;
986 
987 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
988 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
989 	    ("sosend_dgram: !PR_ATOMIC"));
990 
991 	if (uio != NULL)
992 		resid = uio->uio_resid;
993 	else
994 		resid = top->m_pkthdr.len;
995 	/*
996 	 * In theory resid should be unsigned.  However, space must be
997 	 * signed, as it might be less than 0 if we over-committed, and we
998 	 * must use a signed comparison of space and resid.  On the other
999 	 * hand, a negative resid causes us to loop sending 0-length
1000 	 * segments to the protocol.
1001 	 */
1002 	if (resid < 0) {
1003 		error = EINVAL;
1004 		goto out;
1005 	}
1006 
1007 	dontroute =
1008 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
1009 	if (td != NULL)
1010 		td->td_ru.ru_msgsnd++;
1011 	if (control != NULL)
1012 		clen = control->m_len;
1013 
1014 	SOCKBUF_LOCK(&so->so_snd);
1015 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1016 		SOCKBUF_UNLOCK(&so->so_snd);
1017 		error = EPIPE;
1018 		goto out;
1019 	}
1020 	if (so->so_error) {
1021 		error = so->so_error;
1022 		so->so_error = 0;
1023 		SOCKBUF_UNLOCK(&so->so_snd);
1024 		goto out;
1025 	}
1026 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1027 		/*
1028 		 * `sendto' and `sendmsg' is allowed on a connection-based
1029 		 * socket if it supports implied connect.  Return ENOTCONN if
1030 		 * not connected and no address is supplied.
1031 		 */
1032 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1033 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1034 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1035 			    !(resid == 0 && clen != 0)) {
1036 				SOCKBUF_UNLOCK(&so->so_snd);
1037 				error = ENOTCONN;
1038 				goto out;
1039 			}
1040 		} else if (addr == NULL) {
1041 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1042 				error = ENOTCONN;
1043 			else
1044 				error = EDESTADDRREQ;
1045 			SOCKBUF_UNLOCK(&so->so_snd);
1046 			goto out;
1047 		}
1048 	}
1049 
1050 	/*
1051 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1052 	 * problem and need fixing.
1053 	 */
1054 	space = sbspace(&so->so_snd);
1055 	if (flags & MSG_OOB)
1056 		space += 1024;
1057 	space -= clen;
1058 	SOCKBUF_UNLOCK(&so->so_snd);
1059 	if (resid > space) {
1060 		error = EMSGSIZE;
1061 		goto out;
1062 	}
1063 	if (uio == NULL) {
1064 		resid = 0;
1065 		if (flags & MSG_EOR)
1066 			top->m_flags |= M_EOR;
1067 	} else {
1068 		/*
1069 		 * Copy the data from userland into a mbuf chain.
1070 		 * If no data is to be copied in, a single empty mbuf
1071 		 * is returned.
1072 		 */
1073 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1074 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1075 		if (top == NULL) {
1076 			error = EFAULT;	/* only possible error */
1077 			goto out;
1078 		}
1079 		space -= resid - uio->uio_resid;
1080 		resid = uio->uio_resid;
1081 	}
1082 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1083 	/*
1084 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1085 	 * than with.
1086 	 */
1087 	if (dontroute) {
1088 		SOCK_LOCK(so);
1089 		so->so_options |= SO_DONTROUTE;
1090 		SOCK_UNLOCK(so);
1091 	}
1092 	/*
1093 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1094 	 * of date.  We could have recieved a reset packet in an interrupt or
1095 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1096 	 * probably recheck again inside the locking protection here, but
1097 	 * there are probably other places that this also happens.  We must
1098 	 * rethink this.
1099 	 */
1100 	VNET_SO_ASSERT(so);
1101 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1102 	    (flags & MSG_OOB) ? PRUS_OOB :
1103 	/*
1104 	 * If the user set MSG_EOF, the protocol understands this flag and
1105 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1106 	 */
1107 	    ((flags & MSG_EOF) &&
1108 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1109 	     (resid <= 0)) ?
1110 		PRUS_EOF :
1111 		/* If there is more to send set PRUS_MORETOCOME */
1112 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1113 		top, addr, control, td);
1114 	if (dontroute) {
1115 		SOCK_LOCK(so);
1116 		so->so_options &= ~SO_DONTROUTE;
1117 		SOCK_UNLOCK(so);
1118 	}
1119 	clen = 0;
1120 	control = NULL;
1121 	top = NULL;
1122 out:
1123 	if (top != NULL)
1124 		m_freem(top);
1125 	if (control != NULL)
1126 		m_freem(control);
1127 	return (error);
1128 }
1129 
1130 /*
1131  * Send on a socket.  If send must go all at once and message is larger than
1132  * send buffering, then hard error.  Lock against other senders.  If must go
1133  * all at once and not enough room now, then inform user that this would
1134  * block and do nothing.  Otherwise, if nonblocking, send as much as
1135  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1136  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1137  * in mbuf chain must be small enough to send all at once.
1138  *
1139  * Returns nonzero on error, timeout or signal; callers must check for short
1140  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1141  * on return.
1142  */
1143 int
1144 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1145     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1146 {
1147 	long space;
1148 	ssize_t resid;
1149 	int clen = 0, error, dontroute;
1150 	int atomic = sosendallatonce(so) || top;
1151 
1152 	if (uio != NULL)
1153 		resid = uio->uio_resid;
1154 	else
1155 		resid = top->m_pkthdr.len;
1156 	/*
1157 	 * In theory resid should be unsigned.  However, space must be
1158 	 * signed, as it might be less than 0 if we over-committed, and we
1159 	 * must use a signed comparison of space and resid.  On the other
1160 	 * hand, a negative resid causes us to loop sending 0-length
1161 	 * segments to the protocol.
1162 	 *
1163 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1164 	 * type sockets since that's an error.
1165 	 */
1166 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1167 		error = EINVAL;
1168 		goto out;
1169 	}
1170 
1171 	dontroute =
1172 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1173 	    (so->so_proto->pr_flags & PR_ATOMIC);
1174 	if (td != NULL)
1175 		td->td_ru.ru_msgsnd++;
1176 	if (control != NULL)
1177 		clen = control->m_len;
1178 
1179 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1180 	if (error)
1181 		goto out;
1182 
1183 restart:
1184 	do {
1185 		SOCKBUF_LOCK(&so->so_snd);
1186 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1187 			SOCKBUF_UNLOCK(&so->so_snd);
1188 			error = EPIPE;
1189 			goto release;
1190 		}
1191 		if (so->so_error) {
1192 			error = so->so_error;
1193 			so->so_error = 0;
1194 			SOCKBUF_UNLOCK(&so->so_snd);
1195 			goto release;
1196 		}
1197 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1198 			/*
1199 			 * `sendto' and `sendmsg' is allowed on a connection-
1200 			 * based socket if it supports implied connect.
1201 			 * Return ENOTCONN if not connected and no address is
1202 			 * supplied.
1203 			 */
1204 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1205 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1206 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1207 				    !(resid == 0 && clen != 0)) {
1208 					SOCKBUF_UNLOCK(&so->so_snd);
1209 					error = ENOTCONN;
1210 					goto release;
1211 				}
1212 			} else if (addr == NULL) {
1213 				SOCKBUF_UNLOCK(&so->so_snd);
1214 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1215 					error = ENOTCONN;
1216 				else
1217 					error = EDESTADDRREQ;
1218 				goto release;
1219 			}
1220 		}
1221 		space = sbspace(&so->so_snd);
1222 		if (flags & MSG_OOB)
1223 			space += 1024;
1224 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1225 		    clen > so->so_snd.sb_hiwat) {
1226 			SOCKBUF_UNLOCK(&so->so_snd);
1227 			error = EMSGSIZE;
1228 			goto release;
1229 		}
1230 		if (space < resid + clen &&
1231 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1232 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1233 				SOCKBUF_UNLOCK(&so->so_snd);
1234 				error = EWOULDBLOCK;
1235 				goto release;
1236 			}
1237 			error = sbwait(&so->so_snd);
1238 			SOCKBUF_UNLOCK(&so->so_snd);
1239 			if (error)
1240 				goto release;
1241 			goto restart;
1242 		}
1243 		SOCKBUF_UNLOCK(&so->so_snd);
1244 		space -= clen;
1245 		do {
1246 			if (uio == NULL) {
1247 				resid = 0;
1248 				if (flags & MSG_EOR)
1249 					top->m_flags |= M_EOR;
1250 			} else {
1251 				/*
1252 				 * Copy the data from userland into a mbuf
1253 				 * chain.  If no data is to be copied in,
1254 				 * a single empty mbuf is returned.
1255 				 */
1256 				top = m_uiotombuf(uio, M_WAITOK, space,
1257 				    (atomic ? max_hdr : 0),
1258 				    (atomic ? M_PKTHDR : 0) |
1259 				    ((flags & MSG_EOR) ? M_EOR : 0));
1260 				if (top == NULL) {
1261 					error = EFAULT; /* only possible error */
1262 					goto release;
1263 				}
1264 				space -= resid - uio->uio_resid;
1265 				resid = uio->uio_resid;
1266 			}
1267 			if (dontroute) {
1268 				SOCK_LOCK(so);
1269 				so->so_options |= SO_DONTROUTE;
1270 				SOCK_UNLOCK(so);
1271 			}
1272 			/*
1273 			 * XXX all the SBS_CANTSENDMORE checks previously
1274 			 * done could be out of date.  We could have recieved
1275 			 * a reset packet in an interrupt or maybe we slept
1276 			 * while doing page faults in uiomove() etc.  We
1277 			 * could probably recheck again inside the locking
1278 			 * protection here, but there are probably other
1279 			 * places that this also happens.  We must rethink
1280 			 * this.
1281 			 */
1282 			VNET_SO_ASSERT(so);
1283 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1284 			    (flags & MSG_OOB) ? PRUS_OOB :
1285 			/*
1286 			 * If the user set MSG_EOF, the protocol understands
1287 			 * this flag and nothing left to send then use
1288 			 * PRU_SEND_EOF instead of PRU_SEND.
1289 			 */
1290 			    ((flags & MSG_EOF) &&
1291 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1292 			     (resid <= 0)) ?
1293 				PRUS_EOF :
1294 			/* If there is more to send set PRUS_MORETOCOME. */
1295 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1296 			    top, addr, control, td);
1297 			if (dontroute) {
1298 				SOCK_LOCK(so);
1299 				so->so_options &= ~SO_DONTROUTE;
1300 				SOCK_UNLOCK(so);
1301 			}
1302 			clen = 0;
1303 			control = NULL;
1304 			top = NULL;
1305 			if (error)
1306 				goto release;
1307 		} while (resid && space > 0);
1308 	} while (resid);
1309 
1310 release:
1311 	sbunlock(&so->so_snd);
1312 out:
1313 	if (top != NULL)
1314 		m_freem(top);
1315 	if (control != NULL)
1316 		m_freem(control);
1317 	return (error);
1318 }
1319 
1320 int
1321 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1322     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1323 {
1324 	int error;
1325 
1326 	CURVNET_SET(so->so_vnet);
1327 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1328 	    control, flags, td);
1329 	CURVNET_RESTORE();
1330 	return (error);
1331 }
1332 
1333 /*
1334  * The part of soreceive() that implements reading non-inline out-of-band
1335  * data from a socket.  For more complete comments, see soreceive(), from
1336  * which this code originated.
1337  *
1338  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1339  * unable to return an mbuf chain to the caller.
1340  */
1341 static int
1342 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1343 {
1344 	struct protosw *pr = so->so_proto;
1345 	struct mbuf *m;
1346 	int error;
1347 
1348 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1349 	VNET_SO_ASSERT(so);
1350 
1351 	m = m_get(M_WAITOK, MT_DATA);
1352 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1353 	if (error)
1354 		goto bad;
1355 	do {
1356 		error = uiomove(mtod(m, void *),
1357 		    (int) min(uio->uio_resid, m->m_len), uio);
1358 		m = m_free(m);
1359 	} while (uio->uio_resid && error == 0 && m);
1360 bad:
1361 	if (m != NULL)
1362 		m_freem(m);
1363 	return (error);
1364 }
1365 
1366 /*
1367  * Following replacement or removal of the first mbuf on the first mbuf chain
1368  * of a socket buffer, push necessary state changes back into the socket
1369  * buffer so that other consumers see the values consistently.  'nextrecord'
1370  * is the callers locally stored value of the original value of
1371  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1372  * NOTE: 'nextrecord' may be NULL.
1373  */
1374 static __inline void
1375 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1376 {
1377 
1378 	SOCKBUF_LOCK_ASSERT(sb);
1379 	/*
1380 	 * First, update for the new value of nextrecord.  If necessary, make
1381 	 * it the first record.
1382 	 */
1383 	if (sb->sb_mb != NULL)
1384 		sb->sb_mb->m_nextpkt = nextrecord;
1385 	else
1386 		sb->sb_mb = nextrecord;
1387 
1388 	/*
1389 	 * Now update any dependent socket buffer fields to reflect the new
1390 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1391 	 * addition of a second clause that takes care of the case where
1392 	 * sb_mb has been updated, but remains the last record.
1393 	 */
1394 	if (sb->sb_mb == NULL) {
1395 		sb->sb_mbtail = NULL;
1396 		sb->sb_lastrecord = NULL;
1397 	} else if (sb->sb_mb->m_nextpkt == NULL)
1398 		sb->sb_lastrecord = sb->sb_mb;
1399 }
1400 
1401 /*
1402  * Implement receive operations on a socket.  We depend on the way that
1403  * records are added to the sockbuf by sbappend.  In particular, each record
1404  * (mbufs linked through m_next) must begin with an address if the protocol
1405  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1406  * data, and then zero or more mbufs of data.  In order to allow parallelism
1407  * between network receive and copying to user space, as well as avoid
1408  * sleeping with a mutex held, we release the socket buffer mutex during the
1409  * user space copy.  Although the sockbuf is locked, new data may still be
1410  * appended, and thus we must maintain consistency of the sockbuf during that
1411  * time.
1412  *
1413  * The caller may receive the data as a single mbuf chain by supplying an
1414  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1415  * the count in uio_resid.
1416  */
1417 int
1418 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1419     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1420 {
1421 	struct mbuf *m, **mp;
1422 	int flags, error, offset;
1423 	ssize_t len;
1424 	struct protosw *pr = so->so_proto;
1425 	struct mbuf *nextrecord;
1426 	int moff, type = 0;
1427 	ssize_t orig_resid = uio->uio_resid;
1428 
1429 	mp = mp0;
1430 	if (psa != NULL)
1431 		*psa = NULL;
1432 	if (controlp != NULL)
1433 		*controlp = NULL;
1434 	if (flagsp != NULL)
1435 		flags = *flagsp &~ MSG_EOR;
1436 	else
1437 		flags = 0;
1438 	if (flags & MSG_OOB)
1439 		return (soreceive_rcvoob(so, uio, flags));
1440 	if (mp != NULL)
1441 		*mp = NULL;
1442 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1443 	    && uio->uio_resid) {
1444 		VNET_SO_ASSERT(so);
1445 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1446 	}
1447 
1448 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1449 	if (error)
1450 		return (error);
1451 
1452 restart:
1453 	SOCKBUF_LOCK(&so->so_rcv);
1454 	m = so->so_rcv.sb_mb;
1455 	/*
1456 	 * If we have less data than requested, block awaiting more (subject
1457 	 * to any timeout) if:
1458 	 *   1. the current count is less than the low water mark, or
1459 	 *   2. MSG_DONTWAIT is not set
1460 	 */
1461 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1462 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1463 	    so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1464 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1465 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1466 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1467 		    m, so->so_rcv.sb_cc));
1468 		if (so->so_error) {
1469 			if (m != NULL)
1470 				goto dontblock;
1471 			error = so->so_error;
1472 			if ((flags & MSG_PEEK) == 0)
1473 				so->so_error = 0;
1474 			SOCKBUF_UNLOCK(&so->so_rcv);
1475 			goto release;
1476 		}
1477 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1478 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1479 			if (m == NULL) {
1480 				SOCKBUF_UNLOCK(&so->so_rcv);
1481 				goto release;
1482 			} else
1483 				goto dontblock;
1484 		}
1485 		for (; m != NULL; m = m->m_next)
1486 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1487 				m = so->so_rcv.sb_mb;
1488 				goto dontblock;
1489 			}
1490 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1491 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1492 			SOCKBUF_UNLOCK(&so->so_rcv);
1493 			error = ENOTCONN;
1494 			goto release;
1495 		}
1496 		if (uio->uio_resid == 0) {
1497 			SOCKBUF_UNLOCK(&so->so_rcv);
1498 			goto release;
1499 		}
1500 		if ((so->so_state & SS_NBIO) ||
1501 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1502 			SOCKBUF_UNLOCK(&so->so_rcv);
1503 			error = EWOULDBLOCK;
1504 			goto release;
1505 		}
1506 		SBLASTRECORDCHK(&so->so_rcv);
1507 		SBLASTMBUFCHK(&so->so_rcv);
1508 		error = sbwait(&so->so_rcv);
1509 		SOCKBUF_UNLOCK(&so->so_rcv);
1510 		if (error)
1511 			goto release;
1512 		goto restart;
1513 	}
1514 dontblock:
1515 	/*
1516 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1517 	 * pointer to the next record in the socket buffer.  We must keep the
1518 	 * various socket buffer pointers and local stack versions of the
1519 	 * pointers in sync, pushing out modifications before dropping the
1520 	 * socket buffer mutex, and re-reading them when picking it up.
1521 	 *
1522 	 * Otherwise, we will race with the network stack appending new data
1523 	 * or records onto the socket buffer by using inconsistent/stale
1524 	 * versions of the field, possibly resulting in socket buffer
1525 	 * corruption.
1526 	 *
1527 	 * By holding the high-level sblock(), we prevent simultaneous
1528 	 * readers from pulling off the front of the socket buffer.
1529 	 */
1530 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1531 	if (uio->uio_td)
1532 		uio->uio_td->td_ru.ru_msgrcv++;
1533 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1534 	SBLASTRECORDCHK(&so->so_rcv);
1535 	SBLASTMBUFCHK(&so->so_rcv);
1536 	nextrecord = m->m_nextpkt;
1537 	if (pr->pr_flags & PR_ADDR) {
1538 		KASSERT(m->m_type == MT_SONAME,
1539 		    ("m->m_type == %d", m->m_type));
1540 		orig_resid = 0;
1541 		if (psa != NULL)
1542 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1543 			    M_NOWAIT);
1544 		if (flags & MSG_PEEK) {
1545 			m = m->m_next;
1546 		} else {
1547 			sbfree(&so->so_rcv, m);
1548 			so->so_rcv.sb_mb = m_free(m);
1549 			m = so->so_rcv.sb_mb;
1550 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1551 		}
1552 	}
1553 
1554 	/*
1555 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1556 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1557 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1558 	 * perform externalization (or freeing if controlp == NULL).
1559 	 */
1560 	if (m != NULL && m->m_type == MT_CONTROL) {
1561 		struct mbuf *cm = NULL, *cmn;
1562 		struct mbuf **cme = &cm;
1563 
1564 		do {
1565 			if (flags & MSG_PEEK) {
1566 				if (controlp != NULL) {
1567 					*controlp = m_copy(m, 0, m->m_len);
1568 					controlp = &(*controlp)->m_next;
1569 				}
1570 				m = m->m_next;
1571 			} else {
1572 				sbfree(&so->so_rcv, m);
1573 				so->so_rcv.sb_mb = m->m_next;
1574 				m->m_next = NULL;
1575 				*cme = m;
1576 				cme = &(*cme)->m_next;
1577 				m = so->so_rcv.sb_mb;
1578 			}
1579 		} while (m != NULL && m->m_type == MT_CONTROL);
1580 		if ((flags & MSG_PEEK) == 0)
1581 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1582 		while (cm != NULL) {
1583 			cmn = cm->m_next;
1584 			cm->m_next = NULL;
1585 			if (pr->pr_domain->dom_externalize != NULL) {
1586 				SOCKBUF_UNLOCK(&so->so_rcv);
1587 				VNET_SO_ASSERT(so);
1588 				error = (*pr->pr_domain->dom_externalize)
1589 				    (cm, controlp, flags);
1590 				SOCKBUF_LOCK(&so->so_rcv);
1591 			} else if (controlp != NULL)
1592 				*controlp = cm;
1593 			else
1594 				m_freem(cm);
1595 			if (controlp != NULL) {
1596 				orig_resid = 0;
1597 				while (*controlp != NULL)
1598 					controlp = &(*controlp)->m_next;
1599 			}
1600 			cm = cmn;
1601 		}
1602 		if (m != NULL)
1603 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1604 		else
1605 			nextrecord = so->so_rcv.sb_mb;
1606 		orig_resid = 0;
1607 	}
1608 	if (m != NULL) {
1609 		if ((flags & MSG_PEEK) == 0) {
1610 			KASSERT(m->m_nextpkt == nextrecord,
1611 			    ("soreceive: post-control, nextrecord !sync"));
1612 			if (nextrecord == NULL) {
1613 				KASSERT(so->so_rcv.sb_mb == m,
1614 				    ("soreceive: post-control, sb_mb!=m"));
1615 				KASSERT(so->so_rcv.sb_lastrecord == m,
1616 				    ("soreceive: post-control, lastrecord!=m"));
1617 			}
1618 		}
1619 		type = m->m_type;
1620 		if (type == MT_OOBDATA)
1621 			flags |= MSG_OOB;
1622 	} else {
1623 		if ((flags & MSG_PEEK) == 0) {
1624 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1625 			    ("soreceive: sb_mb != nextrecord"));
1626 			if (so->so_rcv.sb_mb == NULL) {
1627 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1628 				    ("soreceive: sb_lastercord != NULL"));
1629 			}
1630 		}
1631 	}
1632 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1633 	SBLASTRECORDCHK(&so->so_rcv);
1634 	SBLASTMBUFCHK(&so->so_rcv);
1635 
1636 	/*
1637 	 * Now continue to read any data mbufs off of the head of the socket
1638 	 * buffer until the read request is satisfied.  Note that 'type' is
1639 	 * used to store the type of any mbuf reads that have happened so far
1640 	 * such that soreceive() can stop reading if the type changes, which
1641 	 * causes soreceive() to return only one of regular data and inline
1642 	 * out-of-band data in a single socket receive operation.
1643 	 */
1644 	moff = 0;
1645 	offset = 0;
1646 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1647 		/*
1648 		 * If the type of mbuf has changed since the last mbuf
1649 		 * examined ('type'), end the receive operation.
1650 		 */
1651 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1652 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1653 			if (type != m->m_type)
1654 				break;
1655 		} else if (type == MT_OOBDATA)
1656 			break;
1657 		else
1658 		    KASSERT(m->m_type == MT_DATA,
1659 			("m->m_type == %d", m->m_type));
1660 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1661 		len = uio->uio_resid;
1662 		if (so->so_oobmark && len > so->so_oobmark - offset)
1663 			len = so->so_oobmark - offset;
1664 		if (len > m->m_len - moff)
1665 			len = m->m_len - moff;
1666 		/*
1667 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1668 		 * them out via the uio, then free.  Sockbuf must be
1669 		 * consistent here (points to current mbuf, it points to next
1670 		 * record) when we drop priority; we must note any additions
1671 		 * to the sockbuf when we block interrupts again.
1672 		 */
1673 		if (mp == NULL) {
1674 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1675 			SBLASTRECORDCHK(&so->so_rcv);
1676 			SBLASTMBUFCHK(&so->so_rcv);
1677 			SOCKBUF_UNLOCK(&so->so_rcv);
1678 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1679 			SOCKBUF_LOCK(&so->so_rcv);
1680 			if (error) {
1681 				/*
1682 				 * The MT_SONAME mbuf has already been removed
1683 				 * from the record, so it is necessary to
1684 				 * remove the data mbufs, if any, to preserve
1685 				 * the invariant in the case of PR_ADDR that
1686 				 * requires MT_SONAME mbufs at the head of
1687 				 * each record.
1688 				 */
1689 				if (m && pr->pr_flags & PR_ATOMIC &&
1690 				    ((flags & MSG_PEEK) == 0))
1691 					(void)sbdroprecord_locked(&so->so_rcv);
1692 				SOCKBUF_UNLOCK(&so->so_rcv);
1693 				goto release;
1694 			}
1695 		} else
1696 			uio->uio_resid -= len;
1697 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1698 		if (len == m->m_len - moff) {
1699 			if (m->m_flags & M_EOR)
1700 				flags |= MSG_EOR;
1701 			if (flags & MSG_PEEK) {
1702 				m = m->m_next;
1703 				moff = 0;
1704 			} else {
1705 				nextrecord = m->m_nextpkt;
1706 				sbfree(&so->so_rcv, m);
1707 				if (mp != NULL) {
1708 					m->m_nextpkt = NULL;
1709 					*mp = m;
1710 					mp = &m->m_next;
1711 					so->so_rcv.sb_mb = m = m->m_next;
1712 					*mp = NULL;
1713 				} else {
1714 					so->so_rcv.sb_mb = m_free(m);
1715 					m = so->so_rcv.sb_mb;
1716 				}
1717 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1718 				SBLASTRECORDCHK(&so->so_rcv);
1719 				SBLASTMBUFCHK(&so->so_rcv);
1720 			}
1721 		} else {
1722 			if (flags & MSG_PEEK)
1723 				moff += len;
1724 			else {
1725 				if (mp != NULL) {
1726 					int copy_flag;
1727 
1728 					if (flags & MSG_DONTWAIT)
1729 						copy_flag = M_NOWAIT;
1730 					else
1731 						copy_flag = M_WAIT;
1732 					if (copy_flag == M_WAITOK)
1733 						SOCKBUF_UNLOCK(&so->so_rcv);
1734 					*mp = m_copym(m, 0, len, copy_flag);
1735 					if (copy_flag == M_WAITOK)
1736 						SOCKBUF_LOCK(&so->so_rcv);
1737 					if (*mp == NULL) {
1738 						/*
1739 						 * m_copym() couldn't
1740 						 * allocate an mbuf.  Adjust
1741 						 * uio_resid back (it was
1742 						 * adjusted down by len
1743 						 * bytes, which we didn't end
1744 						 * up "copying" over).
1745 						 */
1746 						uio->uio_resid += len;
1747 						break;
1748 					}
1749 				}
1750 				m->m_data += len;
1751 				m->m_len -= len;
1752 				so->so_rcv.sb_cc -= len;
1753 			}
1754 		}
1755 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1756 		if (so->so_oobmark) {
1757 			if ((flags & MSG_PEEK) == 0) {
1758 				so->so_oobmark -= len;
1759 				if (so->so_oobmark == 0) {
1760 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1761 					break;
1762 				}
1763 			} else {
1764 				offset += len;
1765 				if (offset == so->so_oobmark)
1766 					break;
1767 			}
1768 		}
1769 		if (flags & MSG_EOR)
1770 			break;
1771 		/*
1772 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1773 		 * must not quit until "uio->uio_resid == 0" or an error
1774 		 * termination.  If a signal/timeout occurs, return with a
1775 		 * short count but without error.  Keep sockbuf locked
1776 		 * against other readers.
1777 		 */
1778 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1779 		    !sosendallatonce(so) && nextrecord == NULL) {
1780 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1781 			if (so->so_error ||
1782 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
1783 				break;
1784 			/*
1785 			 * Notify the protocol that some data has been
1786 			 * drained before blocking.
1787 			 */
1788 			if (pr->pr_flags & PR_WANTRCVD) {
1789 				SOCKBUF_UNLOCK(&so->so_rcv);
1790 				VNET_SO_ASSERT(so);
1791 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1792 				SOCKBUF_LOCK(&so->so_rcv);
1793 			}
1794 			SBLASTRECORDCHK(&so->so_rcv);
1795 			SBLASTMBUFCHK(&so->so_rcv);
1796 			/*
1797 			 * We could receive some data while was notifying
1798 			 * the protocol. Skip blocking in this case.
1799 			 */
1800 			if (so->so_rcv.sb_mb == NULL) {
1801 				error = sbwait(&so->so_rcv);
1802 				if (error) {
1803 					SOCKBUF_UNLOCK(&so->so_rcv);
1804 					goto release;
1805 				}
1806 			}
1807 			m = so->so_rcv.sb_mb;
1808 			if (m != NULL)
1809 				nextrecord = m->m_nextpkt;
1810 		}
1811 	}
1812 
1813 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1814 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1815 		flags |= MSG_TRUNC;
1816 		if ((flags & MSG_PEEK) == 0)
1817 			(void) sbdroprecord_locked(&so->so_rcv);
1818 	}
1819 	if ((flags & MSG_PEEK) == 0) {
1820 		if (m == NULL) {
1821 			/*
1822 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1823 			 * part makes sure sb_lastrecord is up-to-date if
1824 			 * there is still data in the socket buffer.
1825 			 */
1826 			so->so_rcv.sb_mb = nextrecord;
1827 			if (so->so_rcv.sb_mb == NULL) {
1828 				so->so_rcv.sb_mbtail = NULL;
1829 				so->so_rcv.sb_lastrecord = NULL;
1830 			} else if (nextrecord->m_nextpkt == NULL)
1831 				so->so_rcv.sb_lastrecord = nextrecord;
1832 		}
1833 		SBLASTRECORDCHK(&so->so_rcv);
1834 		SBLASTMBUFCHK(&so->so_rcv);
1835 		/*
1836 		 * If soreceive() is being done from the socket callback,
1837 		 * then don't need to generate ACK to peer to update window,
1838 		 * since ACK will be generated on return to TCP.
1839 		 */
1840 		if (!(flags & MSG_SOCALLBCK) &&
1841 		    (pr->pr_flags & PR_WANTRCVD)) {
1842 			SOCKBUF_UNLOCK(&so->so_rcv);
1843 			VNET_SO_ASSERT(so);
1844 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1845 			SOCKBUF_LOCK(&so->so_rcv);
1846 		}
1847 	}
1848 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1849 	if (orig_resid == uio->uio_resid && orig_resid &&
1850 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1851 		SOCKBUF_UNLOCK(&so->so_rcv);
1852 		goto restart;
1853 	}
1854 	SOCKBUF_UNLOCK(&so->so_rcv);
1855 
1856 	if (flagsp != NULL)
1857 		*flagsp |= flags;
1858 release:
1859 	sbunlock(&so->so_rcv);
1860 	return (error);
1861 }
1862 
1863 /*
1864  * Optimized version of soreceive() for stream (TCP) sockets.
1865  * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1866  */
1867 int
1868 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1869     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1870 {
1871 	int len = 0, error = 0, flags, oresid;
1872 	struct sockbuf *sb;
1873 	struct mbuf *m, *n = NULL;
1874 
1875 	/* We only do stream sockets. */
1876 	if (so->so_type != SOCK_STREAM)
1877 		return (EINVAL);
1878 	if (psa != NULL)
1879 		*psa = NULL;
1880 	if (controlp != NULL)
1881 		return (EINVAL);
1882 	if (flagsp != NULL)
1883 		flags = *flagsp &~ MSG_EOR;
1884 	else
1885 		flags = 0;
1886 	if (flags & MSG_OOB)
1887 		return (soreceive_rcvoob(so, uio, flags));
1888 	if (mp0 != NULL)
1889 		*mp0 = NULL;
1890 
1891 	sb = &so->so_rcv;
1892 
1893 	/* Prevent other readers from entering the socket. */
1894 	error = sblock(sb, SBLOCKWAIT(flags));
1895 	if (error)
1896 		goto out;
1897 	SOCKBUF_LOCK(sb);
1898 
1899 	/* Easy one, no space to copyout anything. */
1900 	if (uio->uio_resid == 0) {
1901 		error = EINVAL;
1902 		goto out;
1903 	}
1904 	oresid = uio->uio_resid;
1905 
1906 	/* We will never ever get anything unless we are or were connected. */
1907 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1908 		error = ENOTCONN;
1909 		goto out;
1910 	}
1911 
1912 restart:
1913 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1914 
1915 	/* Abort if socket has reported problems. */
1916 	if (so->so_error) {
1917 		if (sb->sb_cc > 0)
1918 			goto deliver;
1919 		if (oresid > uio->uio_resid)
1920 			goto out;
1921 		error = so->so_error;
1922 		if (!(flags & MSG_PEEK))
1923 			so->so_error = 0;
1924 		goto out;
1925 	}
1926 
1927 	/* Door is closed.  Deliver what is left, if any. */
1928 	if (sb->sb_state & SBS_CANTRCVMORE) {
1929 		if (sb->sb_cc > 0)
1930 			goto deliver;
1931 		else
1932 			goto out;
1933 	}
1934 
1935 	/* Socket buffer is empty and we shall not block. */
1936 	if (sb->sb_cc == 0 &&
1937 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1938 		error = EAGAIN;
1939 		goto out;
1940 	}
1941 
1942 	/* Socket buffer got some data that we shall deliver now. */
1943 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1944 	    ((sb->sb_flags & SS_NBIO) ||
1945 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1946 	     sb->sb_cc >= sb->sb_lowat ||
1947 	     sb->sb_cc >= uio->uio_resid ||
1948 	     sb->sb_cc >= sb->sb_hiwat) ) {
1949 		goto deliver;
1950 	}
1951 
1952 	/* On MSG_WAITALL we must wait until all data or error arrives. */
1953 	if ((flags & MSG_WAITALL) &&
1954 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
1955 		goto deliver;
1956 
1957 	/*
1958 	 * Wait and block until (more) data comes in.
1959 	 * NB: Drops the sockbuf lock during wait.
1960 	 */
1961 	error = sbwait(sb);
1962 	if (error)
1963 		goto out;
1964 	goto restart;
1965 
1966 deliver:
1967 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1968 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1969 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1970 
1971 	/* Statistics. */
1972 	if (uio->uio_td)
1973 		uio->uio_td->td_ru.ru_msgrcv++;
1974 
1975 	/* Fill uio until full or current end of socket buffer is reached. */
1976 	len = min(uio->uio_resid, sb->sb_cc);
1977 	if (mp0 != NULL) {
1978 		/* Dequeue as many mbufs as possible. */
1979 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1980 			if (*mp0 == NULL)
1981 				*mp0 = sb->sb_mb;
1982 			else
1983 				m_cat(*mp0, sb->sb_mb);
1984 			for (m = sb->sb_mb;
1985 			     m != NULL && m->m_len <= len;
1986 			     m = m->m_next) {
1987 				len -= m->m_len;
1988 				uio->uio_resid -= m->m_len;
1989 				sbfree(sb, m);
1990 				n = m;
1991 			}
1992 			n->m_next = NULL;
1993 			sb->sb_mb = m;
1994 			sb->sb_lastrecord = sb->sb_mb;
1995 			if (sb->sb_mb == NULL)
1996 				SB_EMPTY_FIXUP(sb);
1997 		}
1998 		/* Copy the remainder. */
1999 		if (len > 0) {
2000 			KASSERT(sb->sb_mb != NULL,
2001 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
2002 
2003 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
2004 			if (m == NULL)
2005 				len = 0;	/* Don't flush data from sockbuf. */
2006 			else
2007 				uio->uio_resid -= len;
2008 			if (*mp0 != NULL)
2009 				m_cat(*mp0, m);
2010 			else
2011 				*mp0 = m;
2012 			if (*mp0 == NULL) {
2013 				error = ENOBUFS;
2014 				goto out;
2015 			}
2016 		}
2017 	} else {
2018 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2019 		SOCKBUF_UNLOCK(sb);
2020 		error = m_mbuftouio(uio, sb->sb_mb, len);
2021 		SOCKBUF_LOCK(sb);
2022 		if (error)
2023 			goto out;
2024 	}
2025 	SBLASTRECORDCHK(sb);
2026 	SBLASTMBUFCHK(sb);
2027 
2028 	/*
2029 	 * Remove the delivered data from the socket buffer unless we
2030 	 * were only peeking.
2031 	 */
2032 	if (!(flags & MSG_PEEK)) {
2033 		if (len > 0)
2034 			sbdrop_locked(sb, len);
2035 
2036 		/* Notify protocol that we drained some data. */
2037 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2038 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2039 		     !(flags & MSG_SOCALLBCK))) {
2040 			SOCKBUF_UNLOCK(sb);
2041 			VNET_SO_ASSERT(so);
2042 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2043 			SOCKBUF_LOCK(sb);
2044 		}
2045 	}
2046 
2047 	/*
2048 	 * For MSG_WAITALL we may have to loop again and wait for
2049 	 * more data to come in.
2050 	 */
2051 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2052 		goto restart;
2053 out:
2054 	SOCKBUF_LOCK_ASSERT(sb);
2055 	SBLASTRECORDCHK(sb);
2056 	SBLASTMBUFCHK(sb);
2057 	SOCKBUF_UNLOCK(sb);
2058 	sbunlock(sb);
2059 	return (error);
2060 }
2061 
2062 /*
2063  * Optimized version of soreceive() for simple datagram cases from userspace.
2064  * Unlike in the stream case, we're able to drop a datagram if copyout()
2065  * fails, and because we handle datagrams atomically, we don't need to use a
2066  * sleep lock to prevent I/O interlacing.
2067  */
2068 int
2069 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2070     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2071 {
2072 	struct mbuf *m, *m2;
2073 	int flags, error;
2074 	ssize_t len;
2075 	struct protosw *pr = so->so_proto;
2076 	struct mbuf *nextrecord;
2077 
2078 	if (psa != NULL)
2079 		*psa = NULL;
2080 	if (controlp != NULL)
2081 		*controlp = NULL;
2082 	if (flagsp != NULL)
2083 		flags = *flagsp &~ MSG_EOR;
2084 	else
2085 		flags = 0;
2086 
2087 	/*
2088 	 * For any complicated cases, fall back to the full
2089 	 * soreceive_generic().
2090 	 */
2091 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2092 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2093 		    flagsp));
2094 
2095 	/*
2096 	 * Enforce restrictions on use.
2097 	 */
2098 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2099 	    ("soreceive_dgram: wantrcvd"));
2100 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2101 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2102 	    ("soreceive_dgram: SBS_RCVATMARK"));
2103 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2104 	    ("soreceive_dgram: P_CONNREQUIRED"));
2105 
2106 	/*
2107 	 * Loop blocking while waiting for a datagram.
2108 	 */
2109 	SOCKBUF_LOCK(&so->so_rcv);
2110 	while ((m = so->so_rcv.sb_mb) == NULL) {
2111 		KASSERT(so->so_rcv.sb_cc == 0,
2112 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2113 		    so->so_rcv.sb_cc));
2114 		if (so->so_error) {
2115 			error = so->so_error;
2116 			so->so_error = 0;
2117 			SOCKBUF_UNLOCK(&so->so_rcv);
2118 			return (error);
2119 		}
2120 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2121 		    uio->uio_resid == 0) {
2122 			SOCKBUF_UNLOCK(&so->so_rcv);
2123 			return (0);
2124 		}
2125 		if ((so->so_state & SS_NBIO) ||
2126 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2127 			SOCKBUF_UNLOCK(&so->so_rcv);
2128 			return (EWOULDBLOCK);
2129 		}
2130 		SBLASTRECORDCHK(&so->so_rcv);
2131 		SBLASTMBUFCHK(&so->so_rcv);
2132 		error = sbwait(&so->so_rcv);
2133 		if (error) {
2134 			SOCKBUF_UNLOCK(&so->so_rcv);
2135 			return (error);
2136 		}
2137 	}
2138 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2139 
2140 	if (uio->uio_td)
2141 		uio->uio_td->td_ru.ru_msgrcv++;
2142 	SBLASTRECORDCHK(&so->so_rcv);
2143 	SBLASTMBUFCHK(&so->so_rcv);
2144 	nextrecord = m->m_nextpkt;
2145 	if (nextrecord == NULL) {
2146 		KASSERT(so->so_rcv.sb_lastrecord == m,
2147 		    ("soreceive_dgram: lastrecord != m"));
2148 	}
2149 
2150 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2151 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2152 
2153 	/*
2154 	 * Pull 'm' and its chain off the front of the packet queue.
2155 	 */
2156 	so->so_rcv.sb_mb = NULL;
2157 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2158 
2159 	/*
2160 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2161 	 */
2162 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2163 		sbfree(&so->so_rcv, m2);
2164 
2165 	/*
2166 	 * Do a few last checks before we let go of the lock.
2167 	 */
2168 	SBLASTRECORDCHK(&so->so_rcv);
2169 	SBLASTMBUFCHK(&so->so_rcv);
2170 	SOCKBUF_UNLOCK(&so->so_rcv);
2171 
2172 	if (pr->pr_flags & PR_ADDR) {
2173 		KASSERT(m->m_type == MT_SONAME,
2174 		    ("m->m_type == %d", m->m_type));
2175 		if (psa != NULL)
2176 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2177 			    M_NOWAIT);
2178 		m = m_free(m);
2179 	}
2180 	if (m == NULL) {
2181 		/* XXXRW: Can this happen? */
2182 		return (0);
2183 	}
2184 
2185 	/*
2186 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2187 	 * queue.
2188 	 *
2189 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2190 	 * in the first mbuf chain on the socket buffer.  We call into the
2191 	 * protocol to perform externalization (or freeing if controlp ==
2192 	 * NULL).
2193 	 */
2194 	if (m->m_type == MT_CONTROL) {
2195 		struct mbuf *cm = NULL, *cmn;
2196 		struct mbuf **cme = &cm;
2197 
2198 		do {
2199 			m2 = m->m_next;
2200 			m->m_next = NULL;
2201 			*cme = m;
2202 			cme = &(*cme)->m_next;
2203 			m = m2;
2204 		} while (m != NULL && m->m_type == MT_CONTROL);
2205 		while (cm != NULL) {
2206 			cmn = cm->m_next;
2207 			cm->m_next = NULL;
2208 			if (pr->pr_domain->dom_externalize != NULL) {
2209 				error = (*pr->pr_domain->dom_externalize)
2210 				    (cm, controlp, flags);
2211 			} else if (controlp != NULL)
2212 				*controlp = cm;
2213 			else
2214 				m_freem(cm);
2215 			if (controlp != NULL) {
2216 				while (*controlp != NULL)
2217 					controlp = &(*controlp)->m_next;
2218 			}
2219 			cm = cmn;
2220 		}
2221 	}
2222 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2223 
2224 	while (m != NULL && uio->uio_resid > 0) {
2225 		len = uio->uio_resid;
2226 		if (len > m->m_len)
2227 			len = m->m_len;
2228 		error = uiomove(mtod(m, char *), (int)len, uio);
2229 		if (error) {
2230 			m_freem(m);
2231 			return (error);
2232 		}
2233 		if (len == m->m_len)
2234 			m = m_free(m);
2235 		else {
2236 			m->m_data += len;
2237 			m->m_len -= len;
2238 		}
2239 	}
2240 	if (m != NULL)
2241 		flags |= MSG_TRUNC;
2242 	m_freem(m);
2243 	if (flagsp != NULL)
2244 		*flagsp |= flags;
2245 	return (0);
2246 }
2247 
2248 int
2249 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2250     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2251 {
2252 	int error;
2253 
2254 	CURVNET_SET(so->so_vnet);
2255 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2256 	    controlp, flagsp));
2257 	CURVNET_RESTORE();
2258 	return (error);
2259 }
2260 
2261 int
2262 soshutdown(struct socket *so, int how)
2263 {
2264 	struct protosw *pr = so->so_proto;
2265 	int error;
2266 
2267 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2268 		return (EINVAL);
2269 
2270 	CURVNET_SET(so->so_vnet);
2271 	if (pr->pr_usrreqs->pru_flush != NULL)
2272 		(*pr->pr_usrreqs->pru_flush)(so, how);
2273 	if (how != SHUT_WR)
2274 		sorflush(so);
2275 	if (how != SHUT_RD) {
2276 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2277 		wakeup(&so->so_timeo);
2278 		CURVNET_RESTORE();
2279 		return (error);
2280 	}
2281 	wakeup(&so->so_timeo);
2282 	CURVNET_RESTORE();
2283 	return (0);
2284 }
2285 
2286 void
2287 sorflush(struct socket *so)
2288 {
2289 	struct sockbuf *sb = &so->so_rcv;
2290 	struct protosw *pr = so->so_proto;
2291 	struct sockbuf asb;
2292 
2293 	VNET_SO_ASSERT(so);
2294 
2295 	/*
2296 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2297 	 * held, and in order to generally avoid holding the lock for a long
2298 	 * time, we make a copy of the socket buffer and clear the original
2299 	 * (except locks, state).  The new socket buffer copy won't have
2300 	 * initialized locks so we can only call routines that won't use or
2301 	 * assert those locks.
2302 	 *
2303 	 * Dislodge threads currently blocked in receive and wait to acquire
2304 	 * a lock against other simultaneous readers before clearing the
2305 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2306 	 * despite any existing socket disposition on interruptable waiting.
2307 	 */
2308 	socantrcvmore(so);
2309 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2310 
2311 	/*
2312 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2313 	 * and mutex data unchanged.
2314 	 */
2315 	SOCKBUF_LOCK(sb);
2316 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2317 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2318 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2319 	bzero(&sb->sb_startzero,
2320 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2321 	SOCKBUF_UNLOCK(sb);
2322 	sbunlock(sb);
2323 
2324 	/*
2325 	 * Dispose of special rights and flush the socket buffer.  Don't call
2326 	 * any unsafe routines (that rely on locks being initialized) on asb.
2327 	 */
2328 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2329 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2330 	sbrelease_internal(&asb, so);
2331 }
2332 
2333 /*
2334  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2335  * additional variant to handle the case where the option value needs to be
2336  * some kind of integer, but not a specific size.  In addition to their use
2337  * here, these functions are also called by the protocol-level pr_ctloutput()
2338  * routines.
2339  */
2340 int
2341 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2342 {
2343 	size_t	valsize;
2344 
2345 	/*
2346 	 * If the user gives us more than we wanted, we ignore it, but if we
2347 	 * don't get the minimum length the caller wants, we return EINVAL.
2348 	 * On success, sopt->sopt_valsize is set to however much we actually
2349 	 * retrieved.
2350 	 */
2351 	if ((valsize = sopt->sopt_valsize) < minlen)
2352 		return EINVAL;
2353 	if (valsize > len)
2354 		sopt->sopt_valsize = valsize = len;
2355 
2356 	if (sopt->sopt_td != NULL)
2357 		return (copyin(sopt->sopt_val, buf, valsize));
2358 
2359 	bcopy(sopt->sopt_val, buf, valsize);
2360 	return (0);
2361 }
2362 
2363 /*
2364  * Kernel version of setsockopt(2).
2365  *
2366  * XXX: optlen is size_t, not socklen_t
2367  */
2368 int
2369 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2370     size_t optlen)
2371 {
2372 	struct sockopt sopt;
2373 
2374 	sopt.sopt_level = level;
2375 	sopt.sopt_name = optname;
2376 	sopt.sopt_dir = SOPT_SET;
2377 	sopt.sopt_val = optval;
2378 	sopt.sopt_valsize = optlen;
2379 	sopt.sopt_td = NULL;
2380 	return (sosetopt(so, &sopt));
2381 }
2382 
2383 int
2384 sosetopt(struct socket *so, struct sockopt *sopt)
2385 {
2386 	int	error, optval;
2387 	struct	linger l;
2388 	struct	timeval tv;
2389 	sbintime_t val;
2390 	uint32_t val32;
2391 #ifdef MAC
2392 	struct mac extmac;
2393 #endif
2394 
2395 	CURVNET_SET(so->so_vnet);
2396 	error = 0;
2397 	if (sopt->sopt_level != SOL_SOCKET) {
2398 		if (so->so_proto->pr_ctloutput != NULL) {
2399 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2400 			CURVNET_RESTORE();
2401 			return (error);
2402 		}
2403 		error = ENOPROTOOPT;
2404 	} else {
2405 		switch (sopt->sopt_name) {
2406 #ifdef INET
2407 		case SO_ACCEPTFILTER:
2408 			error = do_setopt_accept_filter(so, sopt);
2409 			if (error)
2410 				goto bad;
2411 			break;
2412 #endif
2413 		case SO_LINGER:
2414 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2415 			if (error)
2416 				goto bad;
2417 
2418 			SOCK_LOCK(so);
2419 			so->so_linger = l.l_linger;
2420 			if (l.l_onoff)
2421 				so->so_options |= SO_LINGER;
2422 			else
2423 				so->so_options &= ~SO_LINGER;
2424 			SOCK_UNLOCK(so);
2425 			break;
2426 
2427 		case SO_DEBUG:
2428 		case SO_KEEPALIVE:
2429 		case SO_DONTROUTE:
2430 		case SO_USELOOPBACK:
2431 		case SO_BROADCAST:
2432 		case SO_REUSEADDR:
2433 		case SO_REUSEPORT:
2434 		case SO_OOBINLINE:
2435 		case SO_TIMESTAMP:
2436 		case SO_BINTIME:
2437 		case SO_NOSIGPIPE:
2438 		case SO_NO_DDP:
2439 		case SO_NO_OFFLOAD:
2440 			error = sooptcopyin(sopt, &optval, sizeof optval,
2441 			    sizeof optval);
2442 			if (error)
2443 				goto bad;
2444 			SOCK_LOCK(so);
2445 			if (optval)
2446 				so->so_options |= sopt->sopt_name;
2447 			else
2448 				so->so_options &= ~sopt->sopt_name;
2449 			SOCK_UNLOCK(so);
2450 			break;
2451 
2452 		case SO_SETFIB:
2453 			error = sooptcopyin(sopt, &optval, sizeof optval,
2454 			    sizeof optval);
2455 			if (error)
2456 				goto bad;
2457 
2458 			if (optval < 0 || optval >= rt_numfibs) {
2459 				error = EINVAL;
2460 				goto bad;
2461 			}
2462 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2463 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2464 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2465 				so->so_fibnum = optval;
2466 			else
2467 				so->so_fibnum = 0;
2468 			break;
2469 
2470 		case SO_USER_COOKIE:
2471 			error = sooptcopyin(sopt, &val32, sizeof val32,
2472 			    sizeof val32);
2473 			if (error)
2474 				goto bad;
2475 			so->so_user_cookie = val32;
2476 			break;
2477 
2478 		case SO_SNDBUF:
2479 		case SO_RCVBUF:
2480 		case SO_SNDLOWAT:
2481 		case SO_RCVLOWAT:
2482 			error = sooptcopyin(sopt, &optval, sizeof optval,
2483 			    sizeof optval);
2484 			if (error)
2485 				goto bad;
2486 
2487 			/*
2488 			 * Values < 1 make no sense for any of these options,
2489 			 * so disallow them.
2490 			 */
2491 			if (optval < 1) {
2492 				error = EINVAL;
2493 				goto bad;
2494 			}
2495 
2496 			switch (sopt->sopt_name) {
2497 			case SO_SNDBUF:
2498 			case SO_RCVBUF:
2499 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2500 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2501 				    so, curthread) == 0) {
2502 					error = ENOBUFS;
2503 					goto bad;
2504 				}
2505 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2506 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2507 				break;
2508 
2509 			/*
2510 			 * Make sure the low-water is never greater than the
2511 			 * high-water.
2512 			 */
2513 			case SO_SNDLOWAT:
2514 				SOCKBUF_LOCK(&so->so_snd);
2515 				so->so_snd.sb_lowat =
2516 				    (optval > so->so_snd.sb_hiwat) ?
2517 				    so->so_snd.sb_hiwat : optval;
2518 				SOCKBUF_UNLOCK(&so->so_snd);
2519 				break;
2520 			case SO_RCVLOWAT:
2521 				SOCKBUF_LOCK(&so->so_rcv);
2522 				so->so_rcv.sb_lowat =
2523 				    (optval > so->so_rcv.sb_hiwat) ?
2524 				    so->so_rcv.sb_hiwat : optval;
2525 				SOCKBUF_UNLOCK(&so->so_rcv);
2526 				break;
2527 			}
2528 			break;
2529 
2530 		case SO_SNDTIMEO:
2531 		case SO_RCVTIMEO:
2532 #ifdef COMPAT_FREEBSD32
2533 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2534 				struct timeval32 tv32;
2535 
2536 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2537 				    sizeof tv32);
2538 				CP(tv32, tv, tv_sec);
2539 				CP(tv32, tv, tv_usec);
2540 			} else
2541 #endif
2542 				error = sooptcopyin(sopt, &tv, sizeof tv,
2543 				    sizeof tv);
2544 			if (error)
2545 				goto bad;
2546 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2547 			    tv.tv_usec >= 1000000) {
2548 				error = EDOM;
2549 				goto bad;
2550 			}
2551 			val = tvtosbt(tv);
2552 
2553 			switch (sopt->sopt_name) {
2554 			case SO_SNDTIMEO:
2555 				so->so_snd.sb_timeo = val;
2556 				break;
2557 			case SO_RCVTIMEO:
2558 				so->so_rcv.sb_timeo = val;
2559 				break;
2560 			}
2561 			break;
2562 
2563 		case SO_LABEL:
2564 #ifdef MAC
2565 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2566 			    sizeof extmac);
2567 			if (error)
2568 				goto bad;
2569 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2570 			    so, &extmac);
2571 #else
2572 			error = EOPNOTSUPP;
2573 #endif
2574 			break;
2575 
2576 		default:
2577 			error = ENOPROTOOPT;
2578 			break;
2579 		}
2580 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2581 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2582 	}
2583 bad:
2584 	CURVNET_RESTORE();
2585 	return (error);
2586 }
2587 
2588 /*
2589  * Helper routine for getsockopt.
2590  */
2591 int
2592 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2593 {
2594 	int	error;
2595 	size_t	valsize;
2596 
2597 	error = 0;
2598 
2599 	/*
2600 	 * Documented get behavior is that we always return a value, possibly
2601 	 * truncated to fit in the user's buffer.  Traditional behavior is
2602 	 * that we always tell the user precisely how much we copied, rather
2603 	 * than something useful like the total amount we had available for
2604 	 * her.  Note that this interface is not idempotent; the entire
2605 	 * answer must generated ahead of time.
2606 	 */
2607 	valsize = min(len, sopt->sopt_valsize);
2608 	sopt->sopt_valsize = valsize;
2609 	if (sopt->sopt_val != NULL) {
2610 		if (sopt->sopt_td != NULL)
2611 			error = copyout(buf, sopt->sopt_val, valsize);
2612 		else
2613 			bcopy(buf, sopt->sopt_val, valsize);
2614 	}
2615 	return (error);
2616 }
2617 
2618 int
2619 sogetopt(struct socket *so, struct sockopt *sopt)
2620 {
2621 	int	error, optval;
2622 	struct	linger l;
2623 	struct	timeval tv;
2624 #ifdef MAC
2625 	struct mac extmac;
2626 #endif
2627 
2628 	CURVNET_SET(so->so_vnet);
2629 	error = 0;
2630 	if (sopt->sopt_level != SOL_SOCKET) {
2631 		if (so->so_proto->pr_ctloutput != NULL)
2632 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2633 		else
2634 			error = ENOPROTOOPT;
2635 		CURVNET_RESTORE();
2636 		return (error);
2637 	} else {
2638 		switch (sopt->sopt_name) {
2639 #ifdef INET
2640 		case SO_ACCEPTFILTER:
2641 			error = do_getopt_accept_filter(so, sopt);
2642 			break;
2643 #endif
2644 		case SO_LINGER:
2645 			SOCK_LOCK(so);
2646 			l.l_onoff = so->so_options & SO_LINGER;
2647 			l.l_linger = so->so_linger;
2648 			SOCK_UNLOCK(so);
2649 			error = sooptcopyout(sopt, &l, sizeof l);
2650 			break;
2651 
2652 		case SO_USELOOPBACK:
2653 		case SO_DONTROUTE:
2654 		case SO_DEBUG:
2655 		case SO_KEEPALIVE:
2656 		case SO_REUSEADDR:
2657 		case SO_REUSEPORT:
2658 		case SO_BROADCAST:
2659 		case SO_OOBINLINE:
2660 		case SO_ACCEPTCONN:
2661 		case SO_TIMESTAMP:
2662 		case SO_BINTIME:
2663 		case SO_NOSIGPIPE:
2664 			optval = so->so_options & sopt->sopt_name;
2665 integer:
2666 			error = sooptcopyout(sopt, &optval, sizeof optval);
2667 			break;
2668 
2669 		case SO_TYPE:
2670 			optval = so->so_type;
2671 			goto integer;
2672 
2673 		case SO_PROTOCOL:
2674 			optval = so->so_proto->pr_protocol;
2675 			goto integer;
2676 
2677 		case SO_ERROR:
2678 			SOCK_LOCK(so);
2679 			optval = so->so_error;
2680 			so->so_error = 0;
2681 			SOCK_UNLOCK(so);
2682 			goto integer;
2683 
2684 		case SO_SNDBUF:
2685 			optval = so->so_snd.sb_hiwat;
2686 			goto integer;
2687 
2688 		case SO_RCVBUF:
2689 			optval = so->so_rcv.sb_hiwat;
2690 			goto integer;
2691 
2692 		case SO_SNDLOWAT:
2693 			optval = so->so_snd.sb_lowat;
2694 			goto integer;
2695 
2696 		case SO_RCVLOWAT:
2697 			optval = so->so_rcv.sb_lowat;
2698 			goto integer;
2699 
2700 		case SO_SNDTIMEO:
2701 		case SO_RCVTIMEO:
2702 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2703 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2704 
2705 			tv = sbttotv(optval);
2706 #ifdef COMPAT_FREEBSD32
2707 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2708 				struct timeval32 tv32;
2709 
2710 				CP(tv, tv32, tv_sec);
2711 				CP(tv, tv32, tv_usec);
2712 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2713 			} else
2714 #endif
2715 				error = sooptcopyout(sopt, &tv, sizeof tv);
2716 			break;
2717 
2718 		case SO_LABEL:
2719 #ifdef MAC
2720 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2721 			    sizeof(extmac));
2722 			if (error)
2723 				goto bad;
2724 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2725 			    so, &extmac);
2726 			if (error)
2727 				goto bad;
2728 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2729 #else
2730 			error = EOPNOTSUPP;
2731 #endif
2732 			break;
2733 
2734 		case SO_PEERLABEL:
2735 #ifdef MAC
2736 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2737 			    sizeof(extmac));
2738 			if (error)
2739 				goto bad;
2740 			error = mac_getsockopt_peerlabel(
2741 			    sopt->sopt_td->td_ucred, so, &extmac);
2742 			if (error)
2743 				goto bad;
2744 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2745 #else
2746 			error = EOPNOTSUPP;
2747 #endif
2748 			break;
2749 
2750 		case SO_LISTENQLIMIT:
2751 			optval = so->so_qlimit;
2752 			goto integer;
2753 
2754 		case SO_LISTENQLEN:
2755 			optval = so->so_qlen;
2756 			goto integer;
2757 
2758 		case SO_LISTENINCQLEN:
2759 			optval = so->so_incqlen;
2760 			goto integer;
2761 
2762 		default:
2763 			error = ENOPROTOOPT;
2764 			break;
2765 		}
2766 	}
2767 #ifdef MAC
2768 bad:
2769 #endif
2770 	CURVNET_RESTORE();
2771 	return (error);
2772 }
2773 
2774 int
2775 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2776 {
2777 	struct mbuf *m, *m_prev;
2778 	int sopt_size = sopt->sopt_valsize;
2779 
2780 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2781 	if (m == NULL)
2782 		return ENOBUFS;
2783 	if (sopt_size > MLEN) {
2784 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2785 		if ((m->m_flags & M_EXT) == 0) {
2786 			m_free(m);
2787 			return ENOBUFS;
2788 		}
2789 		m->m_len = min(MCLBYTES, sopt_size);
2790 	} else {
2791 		m->m_len = min(MLEN, sopt_size);
2792 	}
2793 	sopt_size -= m->m_len;
2794 	*mp = m;
2795 	m_prev = m;
2796 
2797 	while (sopt_size) {
2798 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2799 		if (m == NULL) {
2800 			m_freem(*mp);
2801 			return ENOBUFS;
2802 		}
2803 		if (sopt_size > MLEN) {
2804 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2805 			    M_NOWAIT);
2806 			if ((m->m_flags & M_EXT) == 0) {
2807 				m_freem(m);
2808 				m_freem(*mp);
2809 				return ENOBUFS;
2810 			}
2811 			m->m_len = min(MCLBYTES, sopt_size);
2812 		} else {
2813 			m->m_len = min(MLEN, sopt_size);
2814 		}
2815 		sopt_size -= m->m_len;
2816 		m_prev->m_next = m;
2817 		m_prev = m;
2818 	}
2819 	return (0);
2820 }
2821 
2822 int
2823 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2824 {
2825 	struct mbuf *m0 = m;
2826 
2827 	if (sopt->sopt_val == NULL)
2828 		return (0);
2829 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2830 		if (sopt->sopt_td != NULL) {
2831 			int error;
2832 
2833 			error = copyin(sopt->sopt_val, mtod(m, char *),
2834 			    m->m_len);
2835 			if (error != 0) {
2836 				m_freem(m0);
2837 				return(error);
2838 			}
2839 		} else
2840 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2841 		sopt->sopt_valsize -= m->m_len;
2842 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2843 		m = m->m_next;
2844 	}
2845 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2846 		panic("ip6_sooptmcopyin");
2847 	return (0);
2848 }
2849 
2850 int
2851 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2852 {
2853 	struct mbuf *m0 = m;
2854 	size_t valsize = 0;
2855 
2856 	if (sopt->sopt_val == NULL)
2857 		return (0);
2858 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2859 		if (sopt->sopt_td != NULL) {
2860 			int error;
2861 
2862 			error = copyout(mtod(m, char *), sopt->sopt_val,
2863 			    m->m_len);
2864 			if (error != 0) {
2865 				m_freem(m0);
2866 				return(error);
2867 			}
2868 		} else
2869 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2870 		sopt->sopt_valsize -= m->m_len;
2871 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2872 		valsize += m->m_len;
2873 		m = m->m_next;
2874 	}
2875 	if (m != NULL) {
2876 		/* enough soopt buffer should be given from user-land */
2877 		m_freem(m0);
2878 		return(EINVAL);
2879 	}
2880 	sopt->sopt_valsize = valsize;
2881 	return (0);
2882 }
2883 
2884 /*
2885  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2886  * out-of-band data, which will then notify socket consumers.
2887  */
2888 void
2889 sohasoutofband(struct socket *so)
2890 {
2891 
2892 	if (so->so_sigio != NULL)
2893 		pgsigio(&so->so_sigio, SIGURG, 0);
2894 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2895 }
2896 
2897 int
2898 sopoll(struct socket *so, int events, struct ucred *active_cred,
2899     struct thread *td)
2900 {
2901 
2902 	/*
2903 	 * We do not need to set or assert curvnet as long as everyone uses
2904 	 * sopoll_generic().
2905 	 */
2906 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2907 	    td));
2908 }
2909 
2910 int
2911 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2912     struct thread *td)
2913 {
2914 	int revents = 0;
2915 
2916 	SOCKBUF_LOCK(&so->so_snd);
2917 	SOCKBUF_LOCK(&so->so_rcv);
2918 	if (events & (POLLIN | POLLRDNORM))
2919 		if (soreadabledata(so))
2920 			revents |= events & (POLLIN | POLLRDNORM);
2921 
2922 	if (events & (POLLOUT | POLLWRNORM))
2923 		if (sowriteable(so))
2924 			revents |= events & (POLLOUT | POLLWRNORM);
2925 
2926 	if (events & (POLLPRI | POLLRDBAND))
2927 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2928 			revents |= events & (POLLPRI | POLLRDBAND);
2929 
2930 	if ((events & POLLINIGNEOF) == 0) {
2931 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2932 			revents |= events & (POLLIN | POLLRDNORM);
2933 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2934 				revents |= POLLHUP;
2935 		}
2936 	}
2937 
2938 	if (revents == 0) {
2939 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2940 			selrecord(td, &so->so_rcv.sb_sel);
2941 			so->so_rcv.sb_flags |= SB_SEL;
2942 		}
2943 
2944 		if (events & (POLLOUT | POLLWRNORM)) {
2945 			selrecord(td, &so->so_snd.sb_sel);
2946 			so->so_snd.sb_flags |= SB_SEL;
2947 		}
2948 	}
2949 
2950 	SOCKBUF_UNLOCK(&so->so_rcv);
2951 	SOCKBUF_UNLOCK(&so->so_snd);
2952 	return (revents);
2953 }
2954 
2955 int
2956 soo_kqfilter(struct file *fp, struct knote *kn)
2957 {
2958 	struct socket *so = kn->kn_fp->f_data;
2959 	struct sockbuf *sb;
2960 
2961 	switch (kn->kn_filter) {
2962 	case EVFILT_READ:
2963 		if (so->so_options & SO_ACCEPTCONN)
2964 			kn->kn_fop = &solisten_filtops;
2965 		else
2966 			kn->kn_fop = &soread_filtops;
2967 		sb = &so->so_rcv;
2968 		break;
2969 	case EVFILT_WRITE:
2970 		kn->kn_fop = &sowrite_filtops;
2971 		sb = &so->so_snd;
2972 		break;
2973 	default:
2974 		return (EINVAL);
2975 	}
2976 
2977 	SOCKBUF_LOCK(sb);
2978 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2979 	sb->sb_flags |= SB_KNOTE;
2980 	SOCKBUF_UNLOCK(sb);
2981 	return (0);
2982 }
2983 
2984 /*
2985  * Some routines that return EOPNOTSUPP for entry points that are not
2986  * supported by a protocol.  Fill in as needed.
2987  */
2988 int
2989 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2990 {
2991 
2992 	return EOPNOTSUPP;
2993 }
2994 
2995 int
2996 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2997 {
2998 
2999 	return EOPNOTSUPP;
3000 }
3001 
3002 int
3003 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3004 {
3005 
3006 	return EOPNOTSUPP;
3007 }
3008 
3009 int
3010 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3011     struct thread *td)
3012 {
3013 
3014 	return EOPNOTSUPP;
3015 }
3016 
3017 int
3018 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3019 {
3020 
3021 	return EOPNOTSUPP;
3022 }
3023 
3024 int
3025 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3026     struct thread *td)
3027 {
3028 
3029 	return EOPNOTSUPP;
3030 }
3031 
3032 int
3033 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3034 {
3035 
3036 	return EOPNOTSUPP;
3037 }
3038 
3039 int
3040 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3041     struct ifnet *ifp, struct thread *td)
3042 {
3043 
3044 	return EOPNOTSUPP;
3045 }
3046 
3047 int
3048 pru_disconnect_notsupp(struct socket *so)
3049 {
3050 
3051 	return EOPNOTSUPP;
3052 }
3053 
3054 int
3055 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3056 {
3057 
3058 	return EOPNOTSUPP;
3059 }
3060 
3061 int
3062 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3063 {
3064 
3065 	return EOPNOTSUPP;
3066 }
3067 
3068 int
3069 pru_rcvd_notsupp(struct socket *so, int flags)
3070 {
3071 
3072 	return EOPNOTSUPP;
3073 }
3074 
3075 int
3076 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3077 {
3078 
3079 	return EOPNOTSUPP;
3080 }
3081 
3082 int
3083 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3084     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3085 {
3086 
3087 	return EOPNOTSUPP;
3088 }
3089 
3090 /*
3091  * This isn't really a ``null'' operation, but it's the default one and
3092  * doesn't do anything destructive.
3093  */
3094 int
3095 pru_sense_null(struct socket *so, struct stat *sb)
3096 {
3097 
3098 	sb->st_blksize = so->so_snd.sb_hiwat;
3099 	return 0;
3100 }
3101 
3102 int
3103 pru_shutdown_notsupp(struct socket *so)
3104 {
3105 
3106 	return EOPNOTSUPP;
3107 }
3108 
3109 int
3110 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3111 {
3112 
3113 	return EOPNOTSUPP;
3114 }
3115 
3116 int
3117 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3118     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3119 {
3120 
3121 	return EOPNOTSUPP;
3122 }
3123 
3124 int
3125 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3126     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3127 {
3128 
3129 	return EOPNOTSUPP;
3130 }
3131 
3132 int
3133 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3134     struct thread *td)
3135 {
3136 
3137 	return EOPNOTSUPP;
3138 }
3139 
3140 static void
3141 filt_sordetach(struct knote *kn)
3142 {
3143 	struct socket *so = kn->kn_fp->f_data;
3144 
3145 	SOCKBUF_LOCK(&so->so_rcv);
3146 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3147 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3148 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3149 	SOCKBUF_UNLOCK(&so->so_rcv);
3150 }
3151 
3152 /*ARGSUSED*/
3153 static int
3154 filt_soread(struct knote *kn, long hint)
3155 {
3156 	struct socket *so;
3157 
3158 	so = kn->kn_fp->f_data;
3159 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3160 
3161 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3162 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3163 		kn->kn_flags |= EV_EOF;
3164 		kn->kn_fflags = so->so_error;
3165 		return (1);
3166 	} else if (so->so_error)	/* temporary udp error */
3167 		return (1);
3168 	else if (kn->kn_sfflags & NOTE_LOWAT)
3169 		return (kn->kn_data >= kn->kn_sdata);
3170 	else
3171 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3172 }
3173 
3174 static void
3175 filt_sowdetach(struct knote *kn)
3176 {
3177 	struct socket *so = kn->kn_fp->f_data;
3178 
3179 	SOCKBUF_LOCK(&so->so_snd);
3180 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3181 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3182 		so->so_snd.sb_flags &= ~SB_KNOTE;
3183 	SOCKBUF_UNLOCK(&so->so_snd);
3184 }
3185 
3186 /*ARGSUSED*/
3187 static int
3188 filt_sowrite(struct knote *kn, long hint)
3189 {
3190 	struct socket *so;
3191 
3192 	so = kn->kn_fp->f_data;
3193 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3194 	kn->kn_data = sbspace(&so->so_snd);
3195 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3196 		kn->kn_flags |= EV_EOF;
3197 		kn->kn_fflags = so->so_error;
3198 		return (1);
3199 	} else if (so->so_error)	/* temporary udp error */
3200 		return (1);
3201 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3202 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3203 		return (0);
3204 	else if (kn->kn_sfflags & NOTE_LOWAT)
3205 		return (kn->kn_data >= kn->kn_sdata);
3206 	else
3207 		return (kn->kn_data >= so->so_snd.sb_lowat);
3208 }
3209 
3210 /*ARGSUSED*/
3211 static int
3212 filt_solisten(struct knote *kn, long hint)
3213 {
3214 	struct socket *so = kn->kn_fp->f_data;
3215 
3216 	kn->kn_data = so->so_qlen;
3217 	return (!TAILQ_EMPTY(&so->so_comp));
3218 }
3219 
3220 int
3221 socheckuid(struct socket *so, uid_t uid)
3222 {
3223 
3224 	if (so == NULL)
3225 		return (EPERM);
3226 	if (so->so_cred->cr_uid != uid)
3227 		return (EPERM);
3228 	return (0);
3229 }
3230 
3231 /*
3232  * These functions are used by protocols to notify the socket layer (and its
3233  * consumers) of state changes in the sockets driven by protocol-side events.
3234  */
3235 
3236 /*
3237  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3238  *
3239  * Normal sequence from the active (originating) side is that
3240  * soisconnecting() is called during processing of connect() call, resulting
3241  * in an eventual call to soisconnected() if/when the connection is
3242  * established.  When the connection is torn down soisdisconnecting() is
3243  * called during processing of disconnect() call, and soisdisconnected() is
3244  * called when the connection to the peer is totally severed.  The semantics
3245  * of these routines are such that connectionless protocols can call
3246  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3247  * calls when setting up a ``connection'' takes no time.
3248  *
3249  * From the passive side, a socket is created with two queues of sockets:
3250  * so_incomp for connections in progress and so_comp for connections already
3251  * made and awaiting user acceptance.  As a protocol is preparing incoming
3252  * connections, it creates a socket structure queued on so_incomp by calling
3253  * sonewconn().  When the connection is established, soisconnected() is
3254  * called, and transfers the socket structure to so_comp, making it available
3255  * to accept().
3256  *
3257  * If a socket is closed with sockets on either so_incomp or so_comp, these
3258  * sockets are dropped.
3259  *
3260  * If higher-level protocols are implemented in the kernel, the wakeups done
3261  * here will sometimes cause software-interrupt process scheduling.
3262  */
3263 void
3264 soisconnecting(struct socket *so)
3265 {
3266 
3267 	SOCK_LOCK(so);
3268 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3269 	so->so_state |= SS_ISCONNECTING;
3270 	SOCK_UNLOCK(so);
3271 }
3272 
3273 void
3274 soisconnected(struct socket *so)
3275 {
3276 	struct socket *head;
3277 	int ret;
3278 
3279 restart:
3280 	ACCEPT_LOCK();
3281 	SOCK_LOCK(so);
3282 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3283 	so->so_state |= SS_ISCONNECTED;
3284 	head = so->so_head;
3285 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3286 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3287 			SOCK_UNLOCK(so);
3288 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3289 			head->so_incqlen--;
3290 			so->so_qstate &= ~SQ_INCOMP;
3291 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3292 			head->so_qlen++;
3293 			so->so_qstate |= SQ_COMP;
3294 			ACCEPT_UNLOCK();
3295 			sorwakeup(head);
3296 			wakeup_one(&head->so_timeo);
3297 		} else {
3298 			ACCEPT_UNLOCK();
3299 			soupcall_set(so, SO_RCV,
3300 			    head->so_accf->so_accept_filter->accf_callback,
3301 			    head->so_accf->so_accept_filter_arg);
3302 			so->so_options &= ~SO_ACCEPTFILTER;
3303 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3304 			    head->so_accf->so_accept_filter_arg, M_NOWAIT);
3305 			if (ret == SU_ISCONNECTED)
3306 				soupcall_clear(so, SO_RCV);
3307 			SOCK_UNLOCK(so);
3308 			if (ret == SU_ISCONNECTED)
3309 				goto restart;
3310 		}
3311 		return;
3312 	}
3313 	SOCK_UNLOCK(so);
3314 	ACCEPT_UNLOCK();
3315 	wakeup(&so->so_timeo);
3316 	sorwakeup(so);
3317 	sowwakeup(so);
3318 }
3319 
3320 void
3321 soisdisconnecting(struct socket *so)
3322 {
3323 
3324 	/*
3325 	 * Note: This code assumes that SOCK_LOCK(so) and
3326 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3327 	 */
3328 	SOCKBUF_LOCK(&so->so_rcv);
3329 	so->so_state &= ~SS_ISCONNECTING;
3330 	so->so_state |= SS_ISDISCONNECTING;
3331 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3332 	sorwakeup_locked(so);
3333 	SOCKBUF_LOCK(&so->so_snd);
3334 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3335 	sowwakeup_locked(so);
3336 	wakeup(&so->so_timeo);
3337 }
3338 
3339 void
3340 soisdisconnected(struct socket *so)
3341 {
3342 
3343 	/*
3344 	 * Note: This code assumes that SOCK_LOCK(so) and
3345 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3346 	 */
3347 	SOCKBUF_LOCK(&so->so_rcv);
3348 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3349 	so->so_state |= SS_ISDISCONNECTED;
3350 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3351 	sorwakeup_locked(so);
3352 	SOCKBUF_LOCK(&so->so_snd);
3353 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3354 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3355 	sowwakeup_locked(so);
3356 	wakeup(&so->so_timeo);
3357 }
3358 
3359 /*
3360  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3361  */
3362 struct sockaddr *
3363 sodupsockaddr(const struct sockaddr *sa, int mflags)
3364 {
3365 	struct sockaddr *sa2;
3366 
3367 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3368 	if (sa2)
3369 		bcopy(sa, sa2, sa->sa_len);
3370 	return sa2;
3371 }
3372 
3373 /*
3374  * Register per-socket buffer upcalls.
3375  */
3376 void
3377 soupcall_set(struct socket *so, int which,
3378     int (*func)(struct socket *, void *, int), void *arg)
3379 {
3380 	struct sockbuf *sb;
3381 
3382 	switch (which) {
3383 	case SO_RCV:
3384 		sb = &so->so_rcv;
3385 		break;
3386 	case SO_SND:
3387 		sb = &so->so_snd;
3388 		break;
3389 	default:
3390 		panic("soupcall_set: bad which");
3391 	}
3392 	SOCKBUF_LOCK_ASSERT(sb);
3393 #if 0
3394 	/* XXX: accf_http actually wants to do this on purpose. */
3395 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3396 #endif
3397 	sb->sb_upcall = func;
3398 	sb->sb_upcallarg = arg;
3399 	sb->sb_flags |= SB_UPCALL;
3400 }
3401 
3402 void
3403 soupcall_clear(struct socket *so, int which)
3404 {
3405 	struct sockbuf *sb;
3406 
3407 	switch (which) {
3408 	case SO_RCV:
3409 		sb = &so->so_rcv;
3410 		break;
3411 	case SO_SND:
3412 		sb = &so->so_snd;
3413 		break;
3414 	default:
3415 		panic("soupcall_clear: bad which");
3416 	}
3417 	SOCKBUF_LOCK_ASSERT(sb);
3418 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3419 	sb->sb_upcall = NULL;
3420 	sb->sb_upcallarg = NULL;
3421 	sb->sb_flags &= ~SB_UPCALL;
3422 }
3423 
3424 /*
3425  * Create an external-format (``xsocket'') structure using the information in
3426  * the kernel-format socket structure pointed to by so.  This is done to
3427  * reduce the spew of irrelevant information over this interface, to isolate
3428  * user code from changes in the kernel structure, and potentially to provide
3429  * information-hiding if we decide that some of this information should be
3430  * hidden from users.
3431  */
3432 void
3433 sotoxsocket(struct socket *so, struct xsocket *xso)
3434 {
3435 
3436 	xso->xso_len = sizeof *xso;
3437 	xso->xso_so = so;
3438 	xso->so_type = so->so_type;
3439 	xso->so_options = so->so_options;
3440 	xso->so_linger = so->so_linger;
3441 	xso->so_state = so->so_state;
3442 	xso->so_pcb = so->so_pcb;
3443 	xso->xso_protocol = so->so_proto->pr_protocol;
3444 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3445 	xso->so_qlen = so->so_qlen;
3446 	xso->so_incqlen = so->so_incqlen;
3447 	xso->so_qlimit = so->so_qlimit;
3448 	xso->so_timeo = so->so_timeo;
3449 	xso->so_error = so->so_error;
3450 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3451 	xso->so_oobmark = so->so_oobmark;
3452 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3453 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3454 	xso->so_uid = so->so_cred->cr_uid;
3455 }
3456 
3457 
3458 /*
3459  * Socket accessor functions to provide external consumers with
3460  * a safe interface to socket state
3461  *
3462  */
3463 
3464 void
3465 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3466     void *arg)
3467 {
3468 
3469 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3470 		func(so, arg);
3471 }
3472 
3473 struct sockbuf *
3474 so_sockbuf_rcv(struct socket *so)
3475 {
3476 
3477 	return (&so->so_rcv);
3478 }
3479 
3480 struct sockbuf *
3481 so_sockbuf_snd(struct socket *so)
3482 {
3483 
3484 	return (&so->so_snd);
3485 }
3486 
3487 int
3488 so_state_get(const struct socket *so)
3489 {
3490 
3491 	return (so->so_state);
3492 }
3493 
3494 void
3495 so_state_set(struct socket *so, int val)
3496 {
3497 
3498 	so->so_state = val;
3499 }
3500 
3501 int
3502 so_options_get(const struct socket *so)
3503 {
3504 
3505 	return (so->so_options);
3506 }
3507 
3508 void
3509 so_options_set(struct socket *so, int val)
3510 {
3511 
3512 	so->so_options = val;
3513 }
3514 
3515 int
3516 so_error_get(const struct socket *so)
3517 {
3518 
3519 	return (so->so_error);
3520 }
3521 
3522 void
3523 so_error_set(struct socket *so, int val)
3524 {
3525 
3526 	so->so_error = val;
3527 }
3528 
3529 int
3530 so_linger_get(const struct socket *so)
3531 {
3532 
3533 	return (so->so_linger);
3534 }
3535 
3536 void
3537 so_linger_set(struct socket *so, int val)
3538 {
3539 
3540 	so->so_linger = val;
3541 }
3542 
3543 struct protosw *
3544 so_protosw_get(const struct socket *so)
3545 {
3546 
3547 	return (so->so_proto);
3548 }
3549 
3550 void
3551 so_protosw_set(struct socket *so, struct protosw *val)
3552 {
3553 
3554 	so->so_proto = val;
3555 }
3556 
3557 void
3558 so_sorwakeup(struct socket *so)
3559 {
3560 
3561 	sorwakeup(so);
3562 }
3563 
3564 void
3565 so_sowwakeup(struct socket *so)
3566 {
3567 
3568 	sowwakeup(so);
3569 }
3570 
3571 void
3572 so_sorwakeup_locked(struct socket *so)
3573 {
3574 
3575 	sorwakeup_locked(so);
3576 }
3577 
3578 void
3579 so_sowwakeup_locked(struct socket *so)
3580 {
3581 
3582 	sowwakeup_locked(so);
3583 }
3584 
3585 void
3586 so_lock(struct socket *so)
3587 {
3588 
3589 	SOCK_LOCK(so);
3590 }
3591 
3592 void
3593 so_unlock(struct socket *so)
3594 {
3595 
3596 	SOCK_UNLOCK(so);
3597 }
3598