xref: /freebsd/sys/kern/uipc_socket.c (revision ce3adf4362fcca6a43e500b2531f0038adbfbd21)
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 SO_ISCONFIRMING, or SO_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 	struct socket *so;
490 	int over;
491 
492 	ACCEPT_LOCK();
493 	over = (head->so_qlen > 3 * head->so_qlimit / 2);
494 	ACCEPT_UNLOCK();
495 #ifdef REGRESSION
496 	if (regression_sonewconn_earlytest && over) {
497 #else
498 	if (over) {
499 #endif
500 		log(LOG_DEBUG, "%s: pcb %p: Listen queue overflow: "
501 		    "%i already in queue awaiting acceptance\n",
502 		    __func__, head->so_pcb, head->so_qlen);
503 		return (NULL);
504 	}
505 	VNET_ASSERT(head->so_vnet != NULL, ("%s:%d so_vnet is NULL, head=%p",
506 	    __func__, __LINE__, head));
507 	so = soalloc(head->so_vnet);
508 	if (so == NULL) {
509 		log(LOG_DEBUG, "%s: pcb %p: New socket allocation failure: "
510 		    "limit reached or out of memory\n",
511 		    __func__, head->so_pcb);
512 		return (NULL);
513 	}
514 	if ((head->so_options & SO_ACCEPTFILTER) != 0)
515 		connstatus = 0;
516 	so->so_head = head;
517 	so->so_type = head->so_type;
518 	so->so_options = head->so_options &~ SO_ACCEPTCONN;
519 	so->so_linger = head->so_linger;
520 	so->so_state = head->so_state | SS_NOFDREF;
521 	so->so_fibnum = head->so_fibnum;
522 	so->so_proto = head->so_proto;
523 	so->so_cred = crhold(head->so_cred);
524 #ifdef MAC
525 	mac_socket_newconn(head, so);
526 #endif
527 	knlist_init_mtx(&so->so_rcv.sb_sel.si_note, SOCKBUF_MTX(&so->so_rcv));
528 	knlist_init_mtx(&so->so_snd.sb_sel.si_note, SOCKBUF_MTX(&so->so_snd));
529 	VNET_SO_ASSERT(head);
530 	if (soreserve(so, head->so_snd.sb_hiwat, head->so_rcv.sb_hiwat)) {
531 		sodealloc(so);
532 		log(LOG_DEBUG, "%s: pcb %p: soreserve() failed\n",
533 		    __func__, head->so_pcb);
534 		return (NULL);
535 	}
536 	if ((*so->so_proto->pr_usrreqs->pru_attach)(so, 0, NULL)) {
537 		sodealloc(so);
538 		log(LOG_DEBUG, "%s: pcb %p: pru_attach() failed\n",
539 		    __func__, head->so_pcb);
540 		return (NULL);
541 	}
542 	so->so_rcv.sb_lowat = head->so_rcv.sb_lowat;
543 	so->so_snd.sb_lowat = head->so_snd.sb_lowat;
544 	so->so_rcv.sb_timeo = head->so_rcv.sb_timeo;
545 	so->so_snd.sb_timeo = head->so_snd.sb_timeo;
546 	so->so_rcv.sb_flags |= head->so_rcv.sb_flags & SB_AUTOSIZE;
547 	so->so_snd.sb_flags |= head->so_snd.sb_flags & SB_AUTOSIZE;
548 	so->so_state |= connstatus;
549 	ACCEPT_LOCK();
550 	/*
551 	 * The accept socket may be tearing down but we just
552 	 * won a race on the ACCEPT_LOCK.
553 	 * However, if sctp_peeloff() is called on a 1-to-many
554 	 * style socket, the SO_ACCEPTCONN doesn't need to be set.
555 	 */
556 	if (!(head->so_options & SO_ACCEPTCONN) &&
557 	    ((head->so_proto->pr_protocol != IPPROTO_SCTP) ||
558 	     (head->so_type != SOCK_SEQPACKET))) {
559 		SOCK_LOCK(so);
560 		so->so_head = NULL;
561 		sofree(so);		/* NB: returns ACCEPT_UNLOCK'ed. */
562 		return (NULL);
563 	}
564 	if (connstatus) {
565 		TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
566 		so->so_qstate |= SQ_COMP;
567 		head->so_qlen++;
568 	} else {
569 		/*
570 		 * Keep removing sockets from the head until there's room for
571 		 * us to insert on the tail.  In pre-locking revisions, this
572 		 * was a simple if(), but as we could be racing with other
573 		 * threads and soabort() requires dropping locks, we must
574 		 * loop waiting for the condition to be true.
575 		 */
576 		while (head->so_incqlen > head->so_qlimit) {
577 			struct socket *sp;
578 			sp = TAILQ_FIRST(&head->so_incomp);
579 			TAILQ_REMOVE(&head->so_incomp, sp, so_list);
580 			head->so_incqlen--;
581 			sp->so_qstate &= ~SQ_INCOMP;
582 			sp->so_head = NULL;
583 			ACCEPT_UNLOCK();
584 			soabort(sp);
585 			ACCEPT_LOCK();
586 		}
587 		TAILQ_INSERT_TAIL(&head->so_incomp, so, so_list);
588 		so->so_qstate |= SQ_INCOMP;
589 		head->so_incqlen++;
590 	}
591 	ACCEPT_UNLOCK();
592 	if (connstatus) {
593 		sorwakeup(head);
594 		wakeup_one(&head->so_timeo);
595 	}
596 	return (so);
597 }
598 
599 int
600 sobind(struct socket *so, struct sockaddr *nam, struct thread *td)
601 {
602 	int error;
603 
604 	CURVNET_SET(so->so_vnet);
605 	error = (*so->so_proto->pr_usrreqs->pru_bind)(so, nam, td);
606 	CURVNET_RESTORE();
607 	return (error);
608 }
609 
610 int
611 sobindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
612 {
613 	int error;
614 
615 	CURVNET_SET(so->so_vnet);
616 	error = (*so->so_proto->pr_usrreqs->pru_bindat)(fd, so, nam, td);
617 	CURVNET_RESTORE();
618 	return (error);
619 }
620 
621 /*
622  * solisten() transitions a socket from a non-listening state to a listening
623  * state, but can also be used to update the listen queue depth on an
624  * existing listen socket.  The protocol will call back into the sockets
625  * layer using solisten_proto_check() and solisten_proto() to check and set
626  * socket-layer listen state.  Call backs are used so that the protocol can
627  * acquire both protocol and socket layer locks in whatever order is required
628  * by the protocol.
629  *
630  * Protocol implementors are advised to hold the socket lock across the
631  * socket-layer test and set to avoid races at the socket layer.
632  */
633 int
634 solisten(struct socket *so, int backlog, struct thread *td)
635 {
636 	int error;
637 
638 	CURVNET_SET(so->so_vnet);
639 	error = (*so->so_proto->pr_usrreqs->pru_listen)(so, backlog, td);
640 	CURVNET_RESTORE();
641 	return (error);
642 }
643 
644 int
645 solisten_proto_check(struct socket *so)
646 {
647 
648 	SOCK_LOCK_ASSERT(so);
649 
650 	if (so->so_state & (SS_ISCONNECTED | SS_ISCONNECTING |
651 	    SS_ISDISCONNECTING))
652 		return (EINVAL);
653 	return (0);
654 }
655 
656 void
657 solisten_proto(struct socket *so, int backlog)
658 {
659 
660 	SOCK_LOCK_ASSERT(so);
661 
662 	if (backlog < 0 || backlog > somaxconn)
663 		backlog = somaxconn;
664 	so->so_qlimit = backlog;
665 	so->so_options |= SO_ACCEPTCONN;
666 }
667 
668 /*
669  * Evaluate the reference count and named references on a socket; if no
670  * references remain, free it.  This should be called whenever a reference is
671  * released, such as in sorele(), but also when named reference flags are
672  * cleared in socket or protocol code.
673  *
674  * sofree() will free the socket if:
675  *
676  * - There are no outstanding file descriptor references or related consumers
677  *   (so_count == 0).
678  *
679  * - The socket has been closed by user space, if ever open (SS_NOFDREF).
680  *
681  * - The protocol does not have an outstanding strong reference on the socket
682  *   (SS_PROTOREF).
683  *
684  * - The socket is not in a completed connection queue, so a process has been
685  *   notified that it is present.  If it is removed, the user process may
686  *   block in accept() despite select() saying the socket was ready.
687  */
688 void
689 sofree(struct socket *so)
690 {
691 	struct protosw *pr = so->so_proto;
692 	struct socket *head;
693 
694 	ACCEPT_LOCK_ASSERT();
695 	SOCK_LOCK_ASSERT(so);
696 
697 	if ((so->so_state & SS_NOFDREF) == 0 || so->so_count != 0 ||
698 	    (so->so_state & SS_PROTOREF) || (so->so_qstate & SQ_COMP)) {
699 		SOCK_UNLOCK(so);
700 		ACCEPT_UNLOCK();
701 		return;
702 	}
703 
704 	head = so->so_head;
705 	if (head != NULL) {
706 		KASSERT((so->so_qstate & SQ_COMP) != 0 ||
707 		    (so->so_qstate & SQ_INCOMP) != 0,
708 		    ("sofree: so_head != NULL, but neither SQ_COMP nor "
709 		    "SQ_INCOMP"));
710 		KASSERT((so->so_qstate & SQ_COMP) == 0 ||
711 		    (so->so_qstate & SQ_INCOMP) == 0,
712 		    ("sofree: so->so_qstate is SQ_COMP and also SQ_INCOMP"));
713 		TAILQ_REMOVE(&head->so_incomp, so, so_list);
714 		head->so_incqlen--;
715 		so->so_qstate &= ~SQ_INCOMP;
716 		so->so_head = NULL;
717 	}
718 	KASSERT((so->so_qstate & SQ_COMP) == 0 &&
719 	    (so->so_qstate & SQ_INCOMP) == 0,
720 	    ("sofree: so_head == NULL, but still SQ_COMP(%d) or SQ_INCOMP(%d)",
721 	    so->so_qstate & SQ_COMP, so->so_qstate & SQ_INCOMP));
722 	if (so->so_options & SO_ACCEPTCONN) {
723 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
724 		    ("sofree: so_comp populated"));
725 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
726 		    ("sofree: so_incomp populated"));
727 	}
728 	SOCK_UNLOCK(so);
729 	ACCEPT_UNLOCK();
730 
731 	VNET_SO_ASSERT(so);
732 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
733 		(*pr->pr_domain->dom_dispose)(so->so_rcv.sb_mb);
734 	if (pr->pr_usrreqs->pru_detach != NULL)
735 		(*pr->pr_usrreqs->pru_detach)(so);
736 
737 	/*
738 	 * From this point on, we assume that no other references to this
739 	 * socket exist anywhere else in the stack.  Therefore, no locks need
740 	 * to be acquired or held.
741 	 *
742 	 * We used to do a lot of socket buffer and socket locking here, as
743 	 * well as invoke sorflush() and perform wakeups.  The direct call to
744 	 * dom_dispose() and sbrelease_internal() are an inlining of what was
745 	 * necessary from sorflush().
746 	 *
747 	 * Notice that the socket buffer and kqueue state are torn down
748 	 * before calling pru_detach.  This means that protocols shold not
749 	 * assume they can perform socket wakeups, etc, in their detach code.
750 	 */
751 	sbdestroy(&so->so_snd, so);
752 	sbdestroy(&so->so_rcv, so);
753 	seldrain(&so->so_snd.sb_sel);
754 	seldrain(&so->so_rcv.sb_sel);
755 	knlist_destroy(&so->so_rcv.sb_sel.si_note);
756 	knlist_destroy(&so->so_snd.sb_sel.si_note);
757 	sodealloc(so);
758 }
759 
760 /*
761  * Close a socket on last file table reference removal.  Initiate disconnect
762  * if connected.  Free socket when disconnect complete.
763  *
764  * This function will sorele() the socket.  Note that soclose() may be called
765  * prior to the ref count reaching zero.  The actual socket structure will
766  * not be freed until the ref count reaches zero.
767  */
768 int
769 soclose(struct socket *so)
770 {
771 	int error = 0;
772 
773 	KASSERT(!(so->so_state & SS_NOFDREF), ("soclose: SS_NOFDREF on enter"));
774 
775 	CURVNET_SET(so->so_vnet);
776 	funsetown(&so->so_sigio);
777 	if (so->so_state & SS_ISCONNECTED) {
778 		if ((so->so_state & SS_ISDISCONNECTING) == 0) {
779 			error = sodisconnect(so);
780 			if (error) {
781 				if (error == ENOTCONN)
782 					error = 0;
783 				goto drop;
784 			}
785 		}
786 		if (so->so_options & SO_LINGER) {
787 			if ((so->so_state & SS_ISDISCONNECTING) &&
788 			    (so->so_state & SS_NBIO))
789 				goto drop;
790 			while (so->so_state & SS_ISCONNECTED) {
791 				error = tsleep(&so->so_timeo,
792 				    PSOCK | PCATCH, "soclos",
793 				    so->so_linger * hz);
794 				if (error)
795 					break;
796 			}
797 		}
798 	}
799 
800 drop:
801 	if (so->so_proto->pr_usrreqs->pru_close != NULL)
802 		(*so->so_proto->pr_usrreqs->pru_close)(so);
803 	ACCEPT_LOCK();
804 	if (so->so_options & SO_ACCEPTCONN) {
805 		struct socket *sp;
806 		/*
807 		 * Prevent new additions to the accept queues due
808 		 * to ACCEPT_LOCK races while we are draining them.
809 		 */
810 		so->so_options &= ~SO_ACCEPTCONN;
811 		while ((sp = TAILQ_FIRST(&so->so_incomp)) != NULL) {
812 			TAILQ_REMOVE(&so->so_incomp, sp, so_list);
813 			so->so_incqlen--;
814 			sp->so_qstate &= ~SQ_INCOMP;
815 			sp->so_head = NULL;
816 			ACCEPT_UNLOCK();
817 			soabort(sp);
818 			ACCEPT_LOCK();
819 		}
820 		while ((sp = TAILQ_FIRST(&so->so_comp)) != NULL) {
821 			TAILQ_REMOVE(&so->so_comp, sp, so_list);
822 			so->so_qlen--;
823 			sp->so_qstate &= ~SQ_COMP;
824 			sp->so_head = NULL;
825 			ACCEPT_UNLOCK();
826 			soabort(sp);
827 			ACCEPT_LOCK();
828 		}
829 		KASSERT((TAILQ_EMPTY(&so->so_comp)),
830 		    ("%s: so_comp populated", __func__));
831 		KASSERT((TAILQ_EMPTY(&so->so_incomp)),
832 		    ("%s: so_incomp populated", __func__));
833 	}
834 	SOCK_LOCK(so);
835 	KASSERT((so->so_state & SS_NOFDREF) == 0, ("soclose: NOFDREF"));
836 	so->so_state |= SS_NOFDREF;
837 	sorele(so);			/* NB: Returns with ACCEPT_UNLOCK(). */
838 	CURVNET_RESTORE();
839 	return (error);
840 }
841 
842 /*
843  * soabort() is used to abruptly tear down a connection, such as when a
844  * resource limit is reached (listen queue depth exceeded), or if a listen
845  * socket is closed while there are sockets waiting to be accepted.
846  *
847  * This interface is tricky, because it is called on an unreferenced socket,
848  * and must be called only by a thread that has actually removed the socket
849  * from the listen queue it was on, or races with other threads are risked.
850  *
851  * This interface will call into the protocol code, so must not be called
852  * with any socket locks held.  Protocols do call it while holding their own
853  * recursible protocol mutexes, but this is something that should be subject
854  * to review in the future.
855  */
856 void
857 soabort(struct socket *so)
858 {
859 
860 	/*
861 	 * In as much as is possible, assert that no references to this
862 	 * socket are held.  This is not quite the same as asserting that the
863 	 * current thread is responsible for arranging for no references, but
864 	 * is as close as we can get for now.
865 	 */
866 	KASSERT(so->so_count == 0, ("soabort: so_count"));
867 	KASSERT((so->so_state & SS_PROTOREF) == 0, ("soabort: SS_PROTOREF"));
868 	KASSERT(so->so_state & SS_NOFDREF, ("soabort: !SS_NOFDREF"));
869 	KASSERT((so->so_state & SQ_COMP) == 0, ("soabort: SQ_COMP"));
870 	KASSERT((so->so_state & SQ_INCOMP) == 0, ("soabort: SQ_INCOMP"));
871 	VNET_SO_ASSERT(so);
872 
873 	if (so->so_proto->pr_usrreqs->pru_abort != NULL)
874 		(*so->so_proto->pr_usrreqs->pru_abort)(so);
875 	ACCEPT_LOCK();
876 	SOCK_LOCK(so);
877 	sofree(so);
878 }
879 
880 int
881 soaccept(struct socket *so, struct sockaddr **nam)
882 {
883 	int error;
884 
885 	SOCK_LOCK(so);
886 	KASSERT((so->so_state & SS_NOFDREF) != 0, ("soaccept: !NOFDREF"));
887 	so->so_state &= ~SS_NOFDREF;
888 	SOCK_UNLOCK(so);
889 
890 	CURVNET_SET(so->so_vnet);
891 	error = (*so->so_proto->pr_usrreqs->pru_accept)(so, nam);
892 	CURVNET_RESTORE();
893 	return (error);
894 }
895 
896 int
897 soconnect(struct socket *so, struct sockaddr *nam, struct thread *td)
898 {
899 
900 	return (soconnectat(AT_FDCWD, so, nam, td));
901 }
902 
903 int
904 soconnectat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
905 {
906 	int error;
907 
908 	if (so->so_options & SO_ACCEPTCONN)
909 		return (EOPNOTSUPP);
910 
911 	CURVNET_SET(so->so_vnet);
912 	/*
913 	 * If protocol is connection-based, can only connect once.
914 	 * Otherwise, if connected, try to disconnect first.  This allows
915 	 * user to disconnect by connecting to, e.g., a null address.
916 	 */
917 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING) &&
918 	    ((so->so_proto->pr_flags & PR_CONNREQUIRED) ||
919 	    (error = sodisconnect(so)))) {
920 		error = EISCONN;
921 	} else {
922 		/*
923 		 * Prevent accumulated error from previous connection from
924 		 * biting us.
925 		 */
926 		so->so_error = 0;
927 		if (fd == AT_FDCWD) {
928 			error = (*so->so_proto->pr_usrreqs->pru_connect)(so,
929 			    nam, td);
930 		} else {
931 			error = (*so->so_proto->pr_usrreqs->pru_connectat)(fd,
932 			    so, nam, td);
933 		}
934 	}
935 	CURVNET_RESTORE();
936 
937 	return (error);
938 }
939 
940 int
941 soconnect2(struct socket *so1, struct socket *so2)
942 {
943 	int error;
944 
945 	CURVNET_SET(so1->so_vnet);
946 	error = (*so1->so_proto->pr_usrreqs->pru_connect2)(so1, so2);
947 	CURVNET_RESTORE();
948 	return (error);
949 }
950 
951 int
952 sodisconnect(struct socket *so)
953 {
954 	int error;
955 
956 	if ((so->so_state & SS_ISCONNECTED) == 0)
957 		return (ENOTCONN);
958 	if (so->so_state & SS_ISDISCONNECTING)
959 		return (EALREADY);
960 	VNET_SO_ASSERT(so);
961 	error = (*so->so_proto->pr_usrreqs->pru_disconnect)(so);
962 	return (error);
963 }
964 
965 #define	SBLOCKWAIT(f)	(((f) & MSG_DONTWAIT) ? 0 : SBL_WAIT)
966 
967 int
968 sosend_dgram(struct socket *so, struct sockaddr *addr, struct uio *uio,
969     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
970 {
971 	long space;
972 	ssize_t resid;
973 	int clen = 0, error, dontroute;
974 
975 	KASSERT(so->so_type == SOCK_DGRAM, ("sosend_dgram: !SOCK_DGRAM"));
976 	KASSERT(so->so_proto->pr_flags & PR_ATOMIC,
977 	    ("sosend_dgram: !PR_ATOMIC"));
978 
979 	if (uio != NULL)
980 		resid = uio->uio_resid;
981 	else
982 		resid = top->m_pkthdr.len;
983 	/*
984 	 * In theory resid should be unsigned.  However, space must be
985 	 * signed, as it might be less than 0 if we over-committed, and we
986 	 * must use a signed comparison of space and resid.  On the other
987 	 * hand, a negative resid causes us to loop sending 0-length
988 	 * segments to the protocol.
989 	 */
990 	if (resid < 0) {
991 		error = EINVAL;
992 		goto out;
993 	}
994 
995 	dontroute =
996 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0;
997 	if (td != NULL)
998 		td->td_ru.ru_msgsnd++;
999 	if (control != NULL)
1000 		clen = control->m_len;
1001 
1002 	SOCKBUF_LOCK(&so->so_snd);
1003 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1004 		SOCKBUF_UNLOCK(&so->so_snd);
1005 		error = EPIPE;
1006 		goto out;
1007 	}
1008 	if (so->so_error) {
1009 		error = so->so_error;
1010 		so->so_error = 0;
1011 		SOCKBUF_UNLOCK(&so->so_snd);
1012 		goto out;
1013 	}
1014 	if ((so->so_state & SS_ISCONNECTED) == 0) {
1015 		/*
1016 		 * `sendto' and `sendmsg' is allowed on a connection-based
1017 		 * socket if it supports implied connect.  Return ENOTCONN if
1018 		 * not connected and no address is supplied.
1019 		 */
1020 		if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1021 		    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1022 			if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1023 			    !(resid == 0 && clen != 0)) {
1024 				SOCKBUF_UNLOCK(&so->so_snd);
1025 				error = ENOTCONN;
1026 				goto out;
1027 			}
1028 		} else if (addr == NULL) {
1029 			if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1030 				error = ENOTCONN;
1031 			else
1032 				error = EDESTADDRREQ;
1033 			SOCKBUF_UNLOCK(&so->so_snd);
1034 			goto out;
1035 		}
1036 	}
1037 
1038 	/*
1039 	 * Do we need MSG_OOB support in SOCK_DGRAM?  Signs here may be a
1040 	 * problem and need fixing.
1041 	 */
1042 	space = sbspace(&so->so_snd);
1043 	if (flags & MSG_OOB)
1044 		space += 1024;
1045 	space -= clen;
1046 	SOCKBUF_UNLOCK(&so->so_snd);
1047 	if (resid > space) {
1048 		error = EMSGSIZE;
1049 		goto out;
1050 	}
1051 	if (uio == NULL) {
1052 		resid = 0;
1053 		if (flags & MSG_EOR)
1054 			top->m_flags |= M_EOR;
1055 	} else {
1056 		/*
1057 		 * Copy the data from userland into a mbuf chain.
1058 		 * If no data is to be copied in, a single empty mbuf
1059 		 * is returned.
1060 		 */
1061 		top = m_uiotombuf(uio, M_WAITOK, space, max_hdr,
1062 		    (M_PKTHDR | ((flags & MSG_EOR) ? M_EOR : 0)));
1063 		if (top == NULL) {
1064 			error = EFAULT;	/* only possible error */
1065 			goto out;
1066 		}
1067 		space -= resid - uio->uio_resid;
1068 		resid = uio->uio_resid;
1069 	}
1070 	KASSERT(resid == 0, ("sosend_dgram: resid != 0"));
1071 	/*
1072 	 * XXXRW: Frobbing SO_DONTROUTE here is even worse without sblock
1073 	 * than with.
1074 	 */
1075 	if (dontroute) {
1076 		SOCK_LOCK(so);
1077 		so->so_options |= SO_DONTROUTE;
1078 		SOCK_UNLOCK(so);
1079 	}
1080 	/*
1081 	 * XXX all the SBS_CANTSENDMORE checks previously done could be out
1082 	 * of date.  We could have recieved a reset packet in an interrupt or
1083 	 * maybe we slept while doing page faults in uiomove() etc.  We could
1084 	 * probably recheck again inside the locking protection here, but
1085 	 * there are probably other places that this also happens.  We must
1086 	 * rethink this.
1087 	 */
1088 	VNET_SO_ASSERT(so);
1089 	error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1090 	    (flags & MSG_OOB) ? PRUS_OOB :
1091 	/*
1092 	 * If the user set MSG_EOF, the protocol understands this flag and
1093 	 * nothing left to send then use PRU_SEND_EOF instead of PRU_SEND.
1094 	 */
1095 	    ((flags & MSG_EOF) &&
1096 	     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1097 	     (resid <= 0)) ?
1098 		PRUS_EOF :
1099 		/* If there is more to send set PRUS_MORETOCOME */
1100 		(resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1101 		top, addr, control, td);
1102 	if (dontroute) {
1103 		SOCK_LOCK(so);
1104 		so->so_options &= ~SO_DONTROUTE;
1105 		SOCK_UNLOCK(so);
1106 	}
1107 	clen = 0;
1108 	control = NULL;
1109 	top = NULL;
1110 out:
1111 	if (top != NULL)
1112 		m_freem(top);
1113 	if (control != NULL)
1114 		m_freem(control);
1115 	return (error);
1116 }
1117 
1118 /*
1119  * Send on a socket.  If send must go all at once and message is larger than
1120  * send buffering, then hard error.  Lock against other senders.  If must go
1121  * all at once and not enough room now, then inform user that this would
1122  * block and do nothing.  Otherwise, if nonblocking, send as much as
1123  * possible.  The data to be sent is described by "uio" if nonzero, otherwise
1124  * by the mbuf chain "top" (which must be null if uio is not).  Data provided
1125  * in mbuf chain must be small enough to send all at once.
1126  *
1127  * Returns nonzero on error, timeout or signal; callers must check for short
1128  * counts if EINTR/ERESTART are returned.  Data and control buffers are freed
1129  * on return.
1130  */
1131 int
1132 sosend_generic(struct socket *so, struct sockaddr *addr, struct uio *uio,
1133     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1134 {
1135 	long space;
1136 	ssize_t resid;
1137 	int clen = 0, error, dontroute;
1138 	int atomic = sosendallatonce(so) || top;
1139 
1140 	if (uio != NULL)
1141 		resid = uio->uio_resid;
1142 	else
1143 		resid = top->m_pkthdr.len;
1144 	/*
1145 	 * In theory resid should be unsigned.  However, space must be
1146 	 * signed, as it might be less than 0 if we over-committed, and we
1147 	 * must use a signed comparison of space and resid.  On the other
1148 	 * hand, a negative resid causes us to loop sending 0-length
1149 	 * segments to the protocol.
1150 	 *
1151 	 * Also check to make sure that MSG_EOR isn't used on SOCK_STREAM
1152 	 * type sockets since that's an error.
1153 	 */
1154 	if (resid < 0 || (so->so_type == SOCK_STREAM && (flags & MSG_EOR))) {
1155 		error = EINVAL;
1156 		goto out;
1157 	}
1158 
1159 	dontroute =
1160 	    (flags & MSG_DONTROUTE) && (so->so_options & SO_DONTROUTE) == 0 &&
1161 	    (so->so_proto->pr_flags & PR_ATOMIC);
1162 	if (td != NULL)
1163 		td->td_ru.ru_msgsnd++;
1164 	if (control != NULL)
1165 		clen = control->m_len;
1166 
1167 	error = sblock(&so->so_snd, SBLOCKWAIT(flags));
1168 	if (error)
1169 		goto out;
1170 
1171 restart:
1172 	do {
1173 		SOCKBUF_LOCK(&so->so_snd);
1174 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
1175 			SOCKBUF_UNLOCK(&so->so_snd);
1176 			error = EPIPE;
1177 			goto release;
1178 		}
1179 		if (so->so_error) {
1180 			error = so->so_error;
1181 			so->so_error = 0;
1182 			SOCKBUF_UNLOCK(&so->so_snd);
1183 			goto release;
1184 		}
1185 		if ((so->so_state & SS_ISCONNECTED) == 0) {
1186 			/*
1187 			 * `sendto' and `sendmsg' is allowed on a connection-
1188 			 * based socket if it supports implied connect.
1189 			 * Return ENOTCONN if not connected and no address is
1190 			 * supplied.
1191 			 */
1192 			if ((so->so_proto->pr_flags & PR_CONNREQUIRED) &&
1193 			    (so->so_proto->pr_flags & PR_IMPLOPCL) == 0) {
1194 				if ((so->so_state & SS_ISCONFIRMING) == 0 &&
1195 				    !(resid == 0 && clen != 0)) {
1196 					SOCKBUF_UNLOCK(&so->so_snd);
1197 					error = ENOTCONN;
1198 					goto release;
1199 				}
1200 			} else if (addr == NULL) {
1201 				SOCKBUF_UNLOCK(&so->so_snd);
1202 				if (so->so_proto->pr_flags & PR_CONNREQUIRED)
1203 					error = ENOTCONN;
1204 				else
1205 					error = EDESTADDRREQ;
1206 				goto release;
1207 			}
1208 		}
1209 		space = sbspace(&so->so_snd);
1210 		if (flags & MSG_OOB)
1211 			space += 1024;
1212 		if ((atomic && resid > so->so_snd.sb_hiwat) ||
1213 		    clen > so->so_snd.sb_hiwat) {
1214 			SOCKBUF_UNLOCK(&so->so_snd);
1215 			error = EMSGSIZE;
1216 			goto release;
1217 		}
1218 		if (space < resid + clen &&
1219 		    (atomic || space < so->so_snd.sb_lowat || space < clen)) {
1220 			if ((so->so_state & SS_NBIO) || (flags & MSG_NBIO)) {
1221 				SOCKBUF_UNLOCK(&so->so_snd);
1222 				error = EWOULDBLOCK;
1223 				goto release;
1224 			}
1225 			error = sbwait(&so->so_snd);
1226 			SOCKBUF_UNLOCK(&so->so_snd);
1227 			if (error)
1228 				goto release;
1229 			goto restart;
1230 		}
1231 		SOCKBUF_UNLOCK(&so->so_snd);
1232 		space -= clen;
1233 		do {
1234 			if (uio == NULL) {
1235 				resid = 0;
1236 				if (flags & MSG_EOR)
1237 					top->m_flags |= M_EOR;
1238 			} else {
1239 				/*
1240 				 * Copy the data from userland into a mbuf
1241 				 * chain.  If no data is to be copied in,
1242 				 * a single empty mbuf is returned.
1243 				 */
1244 				top = m_uiotombuf(uio, M_WAITOK, space,
1245 				    (atomic ? max_hdr : 0),
1246 				    (atomic ? M_PKTHDR : 0) |
1247 				    ((flags & MSG_EOR) ? M_EOR : 0));
1248 				if (top == NULL) {
1249 					error = EFAULT; /* only possible error */
1250 					goto release;
1251 				}
1252 				space -= resid - uio->uio_resid;
1253 				resid = uio->uio_resid;
1254 			}
1255 			if (dontroute) {
1256 				SOCK_LOCK(so);
1257 				so->so_options |= SO_DONTROUTE;
1258 				SOCK_UNLOCK(so);
1259 			}
1260 			/*
1261 			 * XXX all the SBS_CANTSENDMORE checks previously
1262 			 * done could be out of date.  We could have recieved
1263 			 * a reset packet in an interrupt or maybe we slept
1264 			 * while doing page faults in uiomove() etc.  We
1265 			 * could probably recheck again inside the locking
1266 			 * protection here, but there are probably other
1267 			 * places that this also happens.  We must rethink
1268 			 * this.
1269 			 */
1270 			VNET_SO_ASSERT(so);
1271 			error = (*so->so_proto->pr_usrreqs->pru_send)(so,
1272 			    (flags & MSG_OOB) ? PRUS_OOB :
1273 			/*
1274 			 * If the user set MSG_EOF, the protocol understands
1275 			 * this flag and nothing left to send then use
1276 			 * PRU_SEND_EOF instead of PRU_SEND.
1277 			 */
1278 			    ((flags & MSG_EOF) &&
1279 			     (so->so_proto->pr_flags & PR_IMPLOPCL) &&
1280 			     (resid <= 0)) ?
1281 				PRUS_EOF :
1282 			/* If there is more to send set PRUS_MORETOCOME. */
1283 			    (resid > 0 && space > 0) ? PRUS_MORETOCOME : 0,
1284 			    top, addr, control, td);
1285 			if (dontroute) {
1286 				SOCK_LOCK(so);
1287 				so->so_options &= ~SO_DONTROUTE;
1288 				SOCK_UNLOCK(so);
1289 			}
1290 			clen = 0;
1291 			control = NULL;
1292 			top = NULL;
1293 			if (error)
1294 				goto release;
1295 		} while (resid && space > 0);
1296 	} while (resid);
1297 
1298 release:
1299 	sbunlock(&so->so_snd);
1300 out:
1301 	if (top != NULL)
1302 		m_freem(top);
1303 	if (control != NULL)
1304 		m_freem(control);
1305 	return (error);
1306 }
1307 
1308 int
1309 sosend(struct socket *so, struct sockaddr *addr, struct uio *uio,
1310     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
1311 {
1312 	int error;
1313 
1314 	CURVNET_SET(so->so_vnet);
1315 	error = so->so_proto->pr_usrreqs->pru_sosend(so, addr, uio, top,
1316 	    control, flags, td);
1317 	CURVNET_RESTORE();
1318 	return (error);
1319 }
1320 
1321 /*
1322  * The part of soreceive() that implements reading non-inline out-of-band
1323  * data from a socket.  For more complete comments, see soreceive(), from
1324  * which this code originated.
1325  *
1326  * Note that soreceive_rcvoob(), unlike the remainder of soreceive(), is
1327  * unable to return an mbuf chain to the caller.
1328  */
1329 static int
1330 soreceive_rcvoob(struct socket *so, struct uio *uio, int flags)
1331 {
1332 	struct protosw *pr = so->so_proto;
1333 	struct mbuf *m;
1334 	int error;
1335 
1336 	KASSERT(flags & MSG_OOB, ("soreceive_rcvoob: (flags & MSG_OOB) == 0"));
1337 	VNET_SO_ASSERT(so);
1338 
1339 	m = m_get(M_WAITOK, MT_DATA);
1340 	error = (*pr->pr_usrreqs->pru_rcvoob)(so, m, flags & MSG_PEEK);
1341 	if (error)
1342 		goto bad;
1343 	do {
1344 		error = uiomove(mtod(m, void *),
1345 		    (int) min(uio->uio_resid, m->m_len), uio);
1346 		m = m_free(m);
1347 	} while (uio->uio_resid && error == 0 && m);
1348 bad:
1349 	if (m != NULL)
1350 		m_freem(m);
1351 	return (error);
1352 }
1353 
1354 /*
1355  * Following replacement or removal of the first mbuf on the first mbuf chain
1356  * of a socket buffer, push necessary state changes back into the socket
1357  * buffer so that other consumers see the values consistently.  'nextrecord'
1358  * is the callers locally stored value of the original value of
1359  * sb->sb_mb->m_nextpkt which must be restored when the lead mbuf changes.
1360  * NOTE: 'nextrecord' may be NULL.
1361  */
1362 static __inline void
1363 sockbuf_pushsync(struct sockbuf *sb, struct mbuf *nextrecord)
1364 {
1365 
1366 	SOCKBUF_LOCK_ASSERT(sb);
1367 	/*
1368 	 * First, update for the new value of nextrecord.  If necessary, make
1369 	 * it the first record.
1370 	 */
1371 	if (sb->sb_mb != NULL)
1372 		sb->sb_mb->m_nextpkt = nextrecord;
1373 	else
1374 		sb->sb_mb = nextrecord;
1375 
1376 	/*
1377 	 * Now update any dependent socket buffer fields to reflect the new
1378 	 * state.  This is an expanded inline of SB_EMPTY_FIXUP(), with the
1379 	 * addition of a second clause that takes care of the case where
1380 	 * sb_mb has been updated, but remains the last record.
1381 	 */
1382 	if (sb->sb_mb == NULL) {
1383 		sb->sb_mbtail = NULL;
1384 		sb->sb_lastrecord = NULL;
1385 	} else if (sb->sb_mb->m_nextpkt == NULL)
1386 		sb->sb_lastrecord = sb->sb_mb;
1387 }
1388 
1389 /*
1390  * Implement receive operations on a socket.  We depend on the way that
1391  * records are added to the sockbuf by sbappend.  In particular, each record
1392  * (mbufs linked through m_next) must begin with an address if the protocol
1393  * so specifies, followed by an optional mbuf or mbufs containing ancillary
1394  * data, and then zero or more mbufs of data.  In order to allow parallelism
1395  * between network receive and copying to user space, as well as avoid
1396  * sleeping with a mutex held, we release the socket buffer mutex during the
1397  * user space copy.  Although the sockbuf is locked, new data may still be
1398  * appended, and thus we must maintain consistency of the sockbuf during that
1399  * time.
1400  *
1401  * The caller may receive the data as a single mbuf chain by supplying an
1402  * mbuf **mp0 for use in returning the chain.  The uio is then used only for
1403  * the count in uio_resid.
1404  */
1405 int
1406 soreceive_generic(struct socket *so, struct sockaddr **psa, struct uio *uio,
1407     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1408 {
1409 	struct mbuf *m, **mp;
1410 	int flags, error, offset;
1411 	ssize_t len;
1412 	struct protosw *pr = so->so_proto;
1413 	struct mbuf *nextrecord;
1414 	int moff, type = 0;
1415 	ssize_t orig_resid = uio->uio_resid;
1416 
1417 	mp = mp0;
1418 	if (psa != NULL)
1419 		*psa = NULL;
1420 	if (controlp != NULL)
1421 		*controlp = NULL;
1422 	if (flagsp != NULL)
1423 		flags = *flagsp &~ MSG_EOR;
1424 	else
1425 		flags = 0;
1426 	if (flags & MSG_OOB)
1427 		return (soreceive_rcvoob(so, uio, flags));
1428 	if (mp != NULL)
1429 		*mp = NULL;
1430 	if ((pr->pr_flags & PR_WANTRCVD) && (so->so_state & SS_ISCONFIRMING)
1431 	    && uio->uio_resid) {
1432 		VNET_SO_ASSERT(so);
1433 		(*pr->pr_usrreqs->pru_rcvd)(so, 0);
1434 	}
1435 
1436 	error = sblock(&so->so_rcv, SBLOCKWAIT(flags));
1437 	if (error)
1438 		return (error);
1439 
1440 restart:
1441 	SOCKBUF_LOCK(&so->so_rcv);
1442 	m = so->so_rcv.sb_mb;
1443 	/*
1444 	 * If we have less data than requested, block awaiting more (subject
1445 	 * to any timeout) if:
1446 	 *   1. the current count is less than the low water mark, or
1447 	 *   2. MSG_DONTWAIT is not set
1448 	 */
1449 	if (m == NULL || (((flags & MSG_DONTWAIT) == 0 &&
1450 	    so->so_rcv.sb_cc < uio->uio_resid) &&
1451 	    so->so_rcv.sb_cc < so->so_rcv.sb_lowat &&
1452 	    m->m_nextpkt == NULL && (pr->pr_flags & PR_ATOMIC) == 0)) {
1453 		KASSERT(m != NULL || !so->so_rcv.sb_cc,
1454 		    ("receive: m == %p so->so_rcv.sb_cc == %u",
1455 		    m, so->so_rcv.sb_cc));
1456 		if (so->so_error) {
1457 			if (m != NULL)
1458 				goto dontblock;
1459 			error = so->so_error;
1460 			if ((flags & MSG_PEEK) == 0)
1461 				so->so_error = 0;
1462 			SOCKBUF_UNLOCK(&so->so_rcv);
1463 			goto release;
1464 		}
1465 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1466 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
1467 			if (m == NULL) {
1468 				SOCKBUF_UNLOCK(&so->so_rcv);
1469 				goto release;
1470 			} else
1471 				goto dontblock;
1472 		}
1473 		for (; m != NULL; m = m->m_next)
1474 			if (m->m_type == MT_OOBDATA  || (m->m_flags & M_EOR)) {
1475 				m = so->so_rcv.sb_mb;
1476 				goto dontblock;
1477 			}
1478 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) == 0 &&
1479 		    (so->so_proto->pr_flags & PR_CONNREQUIRED)) {
1480 			SOCKBUF_UNLOCK(&so->so_rcv);
1481 			error = ENOTCONN;
1482 			goto release;
1483 		}
1484 		if (uio->uio_resid == 0) {
1485 			SOCKBUF_UNLOCK(&so->so_rcv);
1486 			goto release;
1487 		}
1488 		if ((so->so_state & SS_NBIO) ||
1489 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
1490 			SOCKBUF_UNLOCK(&so->so_rcv);
1491 			error = EWOULDBLOCK;
1492 			goto release;
1493 		}
1494 		SBLASTRECORDCHK(&so->so_rcv);
1495 		SBLASTMBUFCHK(&so->so_rcv);
1496 		error = sbwait(&so->so_rcv);
1497 		SOCKBUF_UNLOCK(&so->so_rcv);
1498 		if (error)
1499 			goto release;
1500 		goto restart;
1501 	}
1502 dontblock:
1503 	/*
1504 	 * From this point onward, we maintain 'nextrecord' as a cache of the
1505 	 * pointer to the next record in the socket buffer.  We must keep the
1506 	 * various socket buffer pointers and local stack versions of the
1507 	 * pointers in sync, pushing out modifications before dropping the
1508 	 * socket buffer mutex, and re-reading them when picking it up.
1509 	 *
1510 	 * Otherwise, we will race with the network stack appending new data
1511 	 * or records onto the socket buffer by using inconsistent/stale
1512 	 * versions of the field, possibly resulting in socket buffer
1513 	 * corruption.
1514 	 *
1515 	 * By holding the high-level sblock(), we prevent simultaneous
1516 	 * readers from pulling off the front of the socket buffer.
1517 	 */
1518 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1519 	if (uio->uio_td)
1520 		uio->uio_td->td_ru.ru_msgrcv++;
1521 	KASSERT(m == so->so_rcv.sb_mb, ("soreceive: m != so->so_rcv.sb_mb"));
1522 	SBLASTRECORDCHK(&so->so_rcv);
1523 	SBLASTMBUFCHK(&so->so_rcv);
1524 	nextrecord = m->m_nextpkt;
1525 	if (pr->pr_flags & PR_ADDR) {
1526 		KASSERT(m->m_type == MT_SONAME,
1527 		    ("m->m_type == %d", m->m_type));
1528 		orig_resid = 0;
1529 		if (psa != NULL)
1530 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
1531 			    M_NOWAIT);
1532 		if (flags & MSG_PEEK) {
1533 			m = m->m_next;
1534 		} else {
1535 			sbfree(&so->so_rcv, m);
1536 			so->so_rcv.sb_mb = m_free(m);
1537 			m = so->so_rcv.sb_mb;
1538 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1539 		}
1540 	}
1541 
1542 	/*
1543 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
1544 	 * in the first mbuf chain on the socket buffer.  If MSG_PEEK, we
1545 	 * just copy the data; if !MSG_PEEK, we call into the protocol to
1546 	 * perform externalization (or freeing if controlp == NULL).
1547 	 */
1548 	if (m != NULL && m->m_type == MT_CONTROL) {
1549 		struct mbuf *cm = NULL, *cmn;
1550 		struct mbuf **cme = &cm;
1551 
1552 		do {
1553 			if (flags & MSG_PEEK) {
1554 				if (controlp != NULL) {
1555 					*controlp = m_copy(m, 0, m->m_len);
1556 					controlp = &(*controlp)->m_next;
1557 				}
1558 				m = m->m_next;
1559 			} else {
1560 				sbfree(&so->so_rcv, m);
1561 				so->so_rcv.sb_mb = m->m_next;
1562 				m->m_next = NULL;
1563 				*cme = m;
1564 				cme = &(*cme)->m_next;
1565 				m = so->so_rcv.sb_mb;
1566 			}
1567 		} while (m != NULL && m->m_type == MT_CONTROL);
1568 		if ((flags & MSG_PEEK) == 0)
1569 			sockbuf_pushsync(&so->so_rcv, nextrecord);
1570 		while (cm != NULL) {
1571 			cmn = cm->m_next;
1572 			cm->m_next = NULL;
1573 			if (pr->pr_domain->dom_externalize != NULL) {
1574 				SOCKBUF_UNLOCK(&so->so_rcv);
1575 				VNET_SO_ASSERT(so);
1576 				error = (*pr->pr_domain->dom_externalize)
1577 				    (cm, controlp, flags);
1578 				SOCKBUF_LOCK(&so->so_rcv);
1579 			} else if (controlp != NULL)
1580 				*controlp = cm;
1581 			else
1582 				m_freem(cm);
1583 			if (controlp != NULL) {
1584 				orig_resid = 0;
1585 				while (*controlp != NULL)
1586 					controlp = &(*controlp)->m_next;
1587 			}
1588 			cm = cmn;
1589 		}
1590 		if (m != NULL)
1591 			nextrecord = so->so_rcv.sb_mb->m_nextpkt;
1592 		else
1593 			nextrecord = so->so_rcv.sb_mb;
1594 		orig_resid = 0;
1595 	}
1596 	if (m != NULL) {
1597 		if ((flags & MSG_PEEK) == 0) {
1598 			KASSERT(m->m_nextpkt == nextrecord,
1599 			    ("soreceive: post-control, nextrecord !sync"));
1600 			if (nextrecord == NULL) {
1601 				KASSERT(so->so_rcv.sb_mb == m,
1602 				    ("soreceive: post-control, sb_mb!=m"));
1603 				KASSERT(so->so_rcv.sb_lastrecord == m,
1604 				    ("soreceive: post-control, lastrecord!=m"));
1605 			}
1606 		}
1607 		type = m->m_type;
1608 		if (type == MT_OOBDATA)
1609 			flags |= MSG_OOB;
1610 	} else {
1611 		if ((flags & MSG_PEEK) == 0) {
1612 			KASSERT(so->so_rcv.sb_mb == nextrecord,
1613 			    ("soreceive: sb_mb != nextrecord"));
1614 			if (so->so_rcv.sb_mb == NULL) {
1615 				KASSERT(so->so_rcv.sb_lastrecord == NULL,
1616 				    ("soreceive: sb_lastercord != NULL"));
1617 			}
1618 		}
1619 	}
1620 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1621 	SBLASTRECORDCHK(&so->so_rcv);
1622 	SBLASTMBUFCHK(&so->so_rcv);
1623 
1624 	/*
1625 	 * Now continue to read any data mbufs off of the head of the socket
1626 	 * buffer until the read request is satisfied.  Note that 'type' is
1627 	 * used to store the type of any mbuf reads that have happened so far
1628 	 * such that soreceive() can stop reading if the type changes, which
1629 	 * causes soreceive() to return only one of regular data and inline
1630 	 * out-of-band data in a single socket receive operation.
1631 	 */
1632 	moff = 0;
1633 	offset = 0;
1634 	while (m != NULL && uio->uio_resid > 0 && error == 0) {
1635 		/*
1636 		 * If the type of mbuf has changed since the last mbuf
1637 		 * examined ('type'), end the receive operation.
1638 		 */
1639 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1640 		if (m->m_type == MT_OOBDATA || m->m_type == MT_CONTROL) {
1641 			if (type != m->m_type)
1642 				break;
1643 		} else if (type == MT_OOBDATA)
1644 			break;
1645 		else
1646 		    KASSERT(m->m_type == MT_DATA,
1647 			("m->m_type == %d", m->m_type));
1648 		so->so_rcv.sb_state &= ~SBS_RCVATMARK;
1649 		len = uio->uio_resid;
1650 		if (so->so_oobmark && len > so->so_oobmark - offset)
1651 			len = so->so_oobmark - offset;
1652 		if (len > m->m_len - moff)
1653 			len = m->m_len - moff;
1654 		/*
1655 		 * If mp is set, just pass back the mbufs.  Otherwise copy
1656 		 * them out via the uio, then free.  Sockbuf must be
1657 		 * consistent here (points to current mbuf, it points to next
1658 		 * record) when we drop priority; we must note any additions
1659 		 * to the sockbuf when we block interrupts again.
1660 		 */
1661 		if (mp == NULL) {
1662 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1663 			SBLASTRECORDCHK(&so->so_rcv);
1664 			SBLASTMBUFCHK(&so->so_rcv);
1665 			SOCKBUF_UNLOCK(&so->so_rcv);
1666 			error = uiomove(mtod(m, char *) + moff, (int)len, uio);
1667 			SOCKBUF_LOCK(&so->so_rcv);
1668 			if (error) {
1669 				/*
1670 				 * The MT_SONAME mbuf has already been removed
1671 				 * from the record, so it is necessary to
1672 				 * remove the data mbufs, if any, to preserve
1673 				 * the invariant in the case of PR_ADDR that
1674 				 * requires MT_SONAME mbufs at the head of
1675 				 * each record.
1676 				 */
1677 				if (m && pr->pr_flags & PR_ATOMIC &&
1678 				    ((flags & MSG_PEEK) == 0))
1679 					(void)sbdroprecord_locked(&so->so_rcv);
1680 				SOCKBUF_UNLOCK(&so->so_rcv);
1681 				goto release;
1682 			}
1683 		} else
1684 			uio->uio_resid -= len;
1685 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1686 		if (len == m->m_len - moff) {
1687 			if (m->m_flags & M_EOR)
1688 				flags |= MSG_EOR;
1689 			if (flags & MSG_PEEK) {
1690 				m = m->m_next;
1691 				moff = 0;
1692 			} else {
1693 				nextrecord = m->m_nextpkt;
1694 				sbfree(&so->so_rcv, m);
1695 				if (mp != NULL) {
1696 					m->m_nextpkt = NULL;
1697 					*mp = m;
1698 					mp = &m->m_next;
1699 					so->so_rcv.sb_mb = m = m->m_next;
1700 					*mp = NULL;
1701 				} else {
1702 					so->so_rcv.sb_mb = m_free(m);
1703 					m = so->so_rcv.sb_mb;
1704 				}
1705 				sockbuf_pushsync(&so->so_rcv, nextrecord);
1706 				SBLASTRECORDCHK(&so->so_rcv);
1707 				SBLASTMBUFCHK(&so->so_rcv);
1708 			}
1709 		} else {
1710 			if (flags & MSG_PEEK)
1711 				moff += len;
1712 			else {
1713 				if (mp != NULL) {
1714 					int copy_flag;
1715 
1716 					if (flags & MSG_DONTWAIT)
1717 						copy_flag = M_NOWAIT;
1718 					else
1719 						copy_flag = M_WAIT;
1720 					if (copy_flag == M_WAITOK)
1721 						SOCKBUF_UNLOCK(&so->so_rcv);
1722 					*mp = m_copym(m, 0, len, copy_flag);
1723 					if (copy_flag == M_WAITOK)
1724 						SOCKBUF_LOCK(&so->so_rcv);
1725 					if (*mp == NULL) {
1726 						/*
1727 						 * m_copym() couldn't
1728 						 * allocate an mbuf.  Adjust
1729 						 * uio_resid back (it was
1730 						 * adjusted down by len
1731 						 * bytes, which we didn't end
1732 						 * up "copying" over).
1733 						 */
1734 						uio->uio_resid += len;
1735 						break;
1736 					}
1737 				}
1738 				m->m_data += len;
1739 				m->m_len -= len;
1740 				so->so_rcv.sb_cc -= len;
1741 			}
1742 		}
1743 		SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1744 		if (so->so_oobmark) {
1745 			if ((flags & MSG_PEEK) == 0) {
1746 				so->so_oobmark -= len;
1747 				if (so->so_oobmark == 0) {
1748 					so->so_rcv.sb_state |= SBS_RCVATMARK;
1749 					break;
1750 				}
1751 			} else {
1752 				offset += len;
1753 				if (offset == so->so_oobmark)
1754 					break;
1755 			}
1756 		}
1757 		if (flags & MSG_EOR)
1758 			break;
1759 		/*
1760 		 * If the MSG_WAITALL flag is set (for non-atomic socket), we
1761 		 * must not quit until "uio->uio_resid == 0" or an error
1762 		 * termination.  If a signal/timeout occurs, return with a
1763 		 * short count but without error.  Keep sockbuf locked
1764 		 * against other readers.
1765 		 */
1766 		while (flags & MSG_WAITALL && m == NULL && uio->uio_resid > 0 &&
1767 		    !sosendallatonce(so) && nextrecord == NULL) {
1768 			SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1769 			if (so->so_error ||
1770 			    so->so_rcv.sb_state & SBS_CANTRCVMORE)
1771 				break;
1772 			/*
1773 			 * Notify the protocol that some data has been
1774 			 * drained before blocking.
1775 			 */
1776 			if (pr->pr_flags & PR_WANTRCVD) {
1777 				SOCKBUF_UNLOCK(&so->so_rcv);
1778 				VNET_SO_ASSERT(so);
1779 				(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1780 				SOCKBUF_LOCK(&so->so_rcv);
1781 			}
1782 			SBLASTRECORDCHK(&so->so_rcv);
1783 			SBLASTMBUFCHK(&so->so_rcv);
1784 			/*
1785 			 * We could receive some data while was notifying
1786 			 * the protocol. Skip blocking in this case.
1787 			 */
1788 			if (so->so_rcv.sb_mb == NULL) {
1789 				error = sbwait(&so->so_rcv);
1790 				if (error) {
1791 					SOCKBUF_UNLOCK(&so->so_rcv);
1792 					goto release;
1793 				}
1794 			}
1795 			m = so->so_rcv.sb_mb;
1796 			if (m != NULL)
1797 				nextrecord = m->m_nextpkt;
1798 		}
1799 	}
1800 
1801 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1802 	if (m != NULL && pr->pr_flags & PR_ATOMIC) {
1803 		flags |= MSG_TRUNC;
1804 		if ((flags & MSG_PEEK) == 0)
1805 			(void) sbdroprecord_locked(&so->so_rcv);
1806 	}
1807 	if ((flags & MSG_PEEK) == 0) {
1808 		if (m == NULL) {
1809 			/*
1810 			 * First part is an inline SB_EMPTY_FIXUP().  Second
1811 			 * part makes sure sb_lastrecord is up-to-date if
1812 			 * there is still data in the socket buffer.
1813 			 */
1814 			so->so_rcv.sb_mb = nextrecord;
1815 			if (so->so_rcv.sb_mb == NULL) {
1816 				so->so_rcv.sb_mbtail = NULL;
1817 				so->so_rcv.sb_lastrecord = NULL;
1818 			} else if (nextrecord->m_nextpkt == NULL)
1819 				so->so_rcv.sb_lastrecord = nextrecord;
1820 		}
1821 		SBLASTRECORDCHK(&so->so_rcv);
1822 		SBLASTMBUFCHK(&so->so_rcv);
1823 		/*
1824 		 * If soreceive() is being done from the socket callback,
1825 		 * then don't need to generate ACK to peer to update window,
1826 		 * since ACK will be generated on return to TCP.
1827 		 */
1828 		if (!(flags & MSG_SOCALLBCK) &&
1829 		    (pr->pr_flags & PR_WANTRCVD)) {
1830 			SOCKBUF_UNLOCK(&so->so_rcv);
1831 			VNET_SO_ASSERT(so);
1832 			(*pr->pr_usrreqs->pru_rcvd)(so, flags);
1833 			SOCKBUF_LOCK(&so->so_rcv);
1834 		}
1835 	}
1836 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1837 	if (orig_resid == uio->uio_resid && orig_resid &&
1838 	    (flags & MSG_EOR) == 0 && (so->so_rcv.sb_state & SBS_CANTRCVMORE) == 0) {
1839 		SOCKBUF_UNLOCK(&so->so_rcv);
1840 		goto restart;
1841 	}
1842 	SOCKBUF_UNLOCK(&so->so_rcv);
1843 
1844 	if (flagsp != NULL)
1845 		*flagsp |= flags;
1846 release:
1847 	sbunlock(&so->so_rcv);
1848 	return (error);
1849 }
1850 
1851 /*
1852  * Optimized version of soreceive() for stream (TCP) sockets.
1853  * XXXAO: (MSG_WAITALL | MSG_PEEK) isn't properly handled.
1854  */
1855 int
1856 soreceive_stream(struct socket *so, struct sockaddr **psa, struct uio *uio,
1857     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
1858 {
1859 	int len = 0, error = 0, flags, oresid;
1860 	struct sockbuf *sb;
1861 	struct mbuf *m, *n = NULL;
1862 
1863 	/* We only do stream sockets. */
1864 	if (so->so_type != SOCK_STREAM)
1865 		return (EINVAL);
1866 	if (psa != NULL)
1867 		*psa = NULL;
1868 	if (controlp != NULL)
1869 		return (EINVAL);
1870 	if (flagsp != NULL)
1871 		flags = *flagsp &~ MSG_EOR;
1872 	else
1873 		flags = 0;
1874 	if (flags & MSG_OOB)
1875 		return (soreceive_rcvoob(so, uio, flags));
1876 	if (mp0 != NULL)
1877 		*mp0 = NULL;
1878 
1879 	sb = &so->so_rcv;
1880 
1881 	/* Prevent other readers from entering the socket. */
1882 	error = sblock(sb, SBLOCKWAIT(flags));
1883 	if (error)
1884 		goto out;
1885 	SOCKBUF_LOCK(sb);
1886 
1887 	/* Easy one, no space to copyout anything. */
1888 	if (uio->uio_resid == 0) {
1889 		error = EINVAL;
1890 		goto out;
1891 	}
1892 	oresid = uio->uio_resid;
1893 
1894 	/* We will never ever get anything unless we are or were connected. */
1895 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISDISCONNECTED))) {
1896 		error = ENOTCONN;
1897 		goto out;
1898 	}
1899 
1900 restart:
1901 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1902 
1903 	/* Abort if socket has reported problems. */
1904 	if (so->so_error) {
1905 		if (sb->sb_cc > 0)
1906 			goto deliver;
1907 		if (oresid > uio->uio_resid)
1908 			goto out;
1909 		error = so->so_error;
1910 		if (!(flags & MSG_PEEK))
1911 			so->so_error = 0;
1912 		goto out;
1913 	}
1914 
1915 	/* Door is closed.  Deliver what is left, if any. */
1916 	if (sb->sb_state & SBS_CANTRCVMORE) {
1917 		if (sb->sb_cc > 0)
1918 			goto deliver;
1919 		else
1920 			goto out;
1921 	}
1922 
1923 	/* Socket buffer is empty and we shall not block. */
1924 	if (sb->sb_cc == 0 &&
1925 	    ((so->so_state & SS_NBIO) || (flags & (MSG_DONTWAIT|MSG_NBIO)))) {
1926 		error = EAGAIN;
1927 		goto out;
1928 	}
1929 
1930 	/* Socket buffer got some data that we shall deliver now. */
1931 	if (sb->sb_cc > 0 && !(flags & MSG_WAITALL) &&
1932 	    ((sb->sb_flags & SS_NBIO) ||
1933 	     (flags & (MSG_DONTWAIT|MSG_NBIO)) ||
1934 	     sb->sb_cc >= sb->sb_lowat ||
1935 	     sb->sb_cc >= uio->uio_resid ||
1936 	     sb->sb_cc >= sb->sb_hiwat) ) {
1937 		goto deliver;
1938 	}
1939 
1940 	/* On MSG_WAITALL we must wait until all data or error arrives. */
1941 	if ((flags & MSG_WAITALL) &&
1942 	    (sb->sb_cc >= uio->uio_resid || sb->sb_cc >= sb->sb_hiwat))
1943 		goto deliver;
1944 
1945 	/*
1946 	 * Wait and block until (more) data comes in.
1947 	 * NB: Drops the sockbuf lock during wait.
1948 	 */
1949 	error = sbwait(sb);
1950 	if (error)
1951 		goto out;
1952 	goto restart;
1953 
1954 deliver:
1955 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
1956 	KASSERT(sb->sb_cc > 0, ("%s: sockbuf empty", __func__));
1957 	KASSERT(sb->sb_mb != NULL, ("%s: sb_mb == NULL", __func__));
1958 
1959 	/* Statistics. */
1960 	if (uio->uio_td)
1961 		uio->uio_td->td_ru.ru_msgrcv++;
1962 
1963 	/* Fill uio until full or current end of socket buffer is reached. */
1964 	len = min(uio->uio_resid, sb->sb_cc);
1965 	if (mp0 != NULL) {
1966 		/* Dequeue as many mbufs as possible. */
1967 		if (!(flags & MSG_PEEK) && len >= sb->sb_mb->m_len) {
1968 			if (*mp0 == NULL)
1969 				*mp0 = sb->sb_mb;
1970 			else
1971 				m_cat(*mp0, sb->sb_mb);
1972 			for (m = sb->sb_mb;
1973 			     m != NULL && m->m_len <= len;
1974 			     m = m->m_next) {
1975 				len -= m->m_len;
1976 				uio->uio_resid -= m->m_len;
1977 				sbfree(sb, m);
1978 				n = m;
1979 			}
1980 			n->m_next = NULL;
1981 			sb->sb_mb = m;
1982 			sb->sb_lastrecord = sb->sb_mb;
1983 			if (sb->sb_mb == NULL)
1984 				SB_EMPTY_FIXUP(sb);
1985 		}
1986 		/* Copy the remainder. */
1987 		if (len > 0) {
1988 			KASSERT(sb->sb_mb != NULL,
1989 			    ("%s: len > 0 && sb->sb_mb empty", __func__));
1990 
1991 			m = m_copym(sb->sb_mb, 0, len, M_NOWAIT);
1992 			if (m == NULL)
1993 				len = 0;	/* Don't flush data from sockbuf. */
1994 			else
1995 				uio->uio_resid -= len;
1996 			if (*mp0 != NULL)
1997 				m_cat(*mp0, m);
1998 			else
1999 				*mp0 = m;
2000 			if (*mp0 == NULL) {
2001 				error = ENOBUFS;
2002 				goto out;
2003 			}
2004 		}
2005 	} else {
2006 		/* NB: Must unlock socket buffer as uiomove may sleep. */
2007 		SOCKBUF_UNLOCK(sb);
2008 		error = m_mbuftouio(uio, sb->sb_mb, len);
2009 		SOCKBUF_LOCK(sb);
2010 		if (error)
2011 			goto out;
2012 	}
2013 	SBLASTRECORDCHK(sb);
2014 	SBLASTMBUFCHK(sb);
2015 
2016 	/*
2017 	 * Remove the delivered data from the socket buffer unless we
2018 	 * were only peeking.
2019 	 */
2020 	if (!(flags & MSG_PEEK)) {
2021 		if (len > 0)
2022 			sbdrop_locked(sb, len);
2023 
2024 		/* Notify protocol that we drained some data. */
2025 		if ((so->so_proto->pr_flags & PR_WANTRCVD) &&
2026 		    (((flags & MSG_WAITALL) && uio->uio_resid > 0) ||
2027 		     !(flags & MSG_SOCALLBCK))) {
2028 			SOCKBUF_UNLOCK(sb);
2029 			VNET_SO_ASSERT(so);
2030 			(*so->so_proto->pr_usrreqs->pru_rcvd)(so, flags);
2031 			SOCKBUF_LOCK(sb);
2032 		}
2033 	}
2034 
2035 	/*
2036 	 * For MSG_WAITALL we may have to loop again and wait for
2037 	 * more data to come in.
2038 	 */
2039 	if ((flags & MSG_WAITALL) && uio->uio_resid > 0)
2040 		goto restart;
2041 out:
2042 	SOCKBUF_LOCK_ASSERT(sb);
2043 	SBLASTRECORDCHK(sb);
2044 	SBLASTMBUFCHK(sb);
2045 	SOCKBUF_UNLOCK(sb);
2046 	sbunlock(sb);
2047 	return (error);
2048 }
2049 
2050 /*
2051  * Optimized version of soreceive() for simple datagram cases from userspace.
2052  * Unlike in the stream case, we're able to drop a datagram if copyout()
2053  * fails, and because we handle datagrams atomically, we don't need to use a
2054  * sleep lock to prevent I/O interlacing.
2055  */
2056 int
2057 soreceive_dgram(struct socket *so, struct sockaddr **psa, struct uio *uio,
2058     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2059 {
2060 	struct mbuf *m, *m2;
2061 	int flags, error;
2062 	ssize_t len;
2063 	struct protosw *pr = so->so_proto;
2064 	struct mbuf *nextrecord;
2065 
2066 	if (psa != NULL)
2067 		*psa = NULL;
2068 	if (controlp != NULL)
2069 		*controlp = NULL;
2070 	if (flagsp != NULL)
2071 		flags = *flagsp &~ MSG_EOR;
2072 	else
2073 		flags = 0;
2074 
2075 	/*
2076 	 * For any complicated cases, fall back to the full
2077 	 * soreceive_generic().
2078 	 */
2079 	if (mp0 != NULL || (flags & MSG_PEEK) || (flags & MSG_OOB))
2080 		return (soreceive_generic(so, psa, uio, mp0, controlp,
2081 		    flagsp));
2082 
2083 	/*
2084 	 * Enforce restrictions on use.
2085 	 */
2086 	KASSERT((pr->pr_flags & PR_WANTRCVD) == 0,
2087 	    ("soreceive_dgram: wantrcvd"));
2088 	KASSERT(pr->pr_flags & PR_ATOMIC, ("soreceive_dgram: !atomic"));
2089 	KASSERT((so->so_rcv.sb_state & SBS_RCVATMARK) == 0,
2090 	    ("soreceive_dgram: SBS_RCVATMARK"));
2091 	KASSERT((so->so_proto->pr_flags & PR_CONNREQUIRED) == 0,
2092 	    ("soreceive_dgram: P_CONNREQUIRED"));
2093 
2094 	/*
2095 	 * Loop blocking while waiting for a datagram.
2096 	 */
2097 	SOCKBUF_LOCK(&so->so_rcv);
2098 	while ((m = so->so_rcv.sb_mb) == NULL) {
2099 		KASSERT(so->so_rcv.sb_cc == 0,
2100 		    ("soreceive_dgram: sb_mb NULL but sb_cc %u",
2101 		    so->so_rcv.sb_cc));
2102 		if (so->so_error) {
2103 			error = so->so_error;
2104 			so->so_error = 0;
2105 			SOCKBUF_UNLOCK(&so->so_rcv);
2106 			return (error);
2107 		}
2108 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE ||
2109 		    uio->uio_resid == 0) {
2110 			SOCKBUF_UNLOCK(&so->so_rcv);
2111 			return (0);
2112 		}
2113 		if ((so->so_state & SS_NBIO) ||
2114 		    (flags & (MSG_DONTWAIT|MSG_NBIO))) {
2115 			SOCKBUF_UNLOCK(&so->so_rcv);
2116 			return (EWOULDBLOCK);
2117 		}
2118 		SBLASTRECORDCHK(&so->so_rcv);
2119 		SBLASTMBUFCHK(&so->so_rcv);
2120 		error = sbwait(&so->so_rcv);
2121 		if (error) {
2122 			SOCKBUF_UNLOCK(&so->so_rcv);
2123 			return (error);
2124 		}
2125 	}
2126 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
2127 
2128 	if (uio->uio_td)
2129 		uio->uio_td->td_ru.ru_msgrcv++;
2130 	SBLASTRECORDCHK(&so->so_rcv);
2131 	SBLASTMBUFCHK(&so->so_rcv);
2132 	nextrecord = m->m_nextpkt;
2133 	if (nextrecord == NULL) {
2134 		KASSERT(so->so_rcv.sb_lastrecord == m,
2135 		    ("soreceive_dgram: lastrecord != m"));
2136 	}
2137 
2138 	KASSERT(so->so_rcv.sb_mb->m_nextpkt == nextrecord,
2139 	    ("soreceive_dgram: m_nextpkt != nextrecord"));
2140 
2141 	/*
2142 	 * Pull 'm' and its chain off the front of the packet queue.
2143 	 */
2144 	so->so_rcv.sb_mb = NULL;
2145 	sockbuf_pushsync(&so->so_rcv, nextrecord);
2146 
2147 	/*
2148 	 * Walk 'm's chain and free that many bytes from the socket buffer.
2149 	 */
2150 	for (m2 = m; m2 != NULL; m2 = m2->m_next)
2151 		sbfree(&so->so_rcv, m2);
2152 
2153 	/*
2154 	 * Do a few last checks before we let go of the lock.
2155 	 */
2156 	SBLASTRECORDCHK(&so->so_rcv);
2157 	SBLASTMBUFCHK(&so->so_rcv);
2158 	SOCKBUF_UNLOCK(&so->so_rcv);
2159 
2160 	if (pr->pr_flags & PR_ADDR) {
2161 		KASSERT(m->m_type == MT_SONAME,
2162 		    ("m->m_type == %d", m->m_type));
2163 		if (psa != NULL)
2164 			*psa = sodupsockaddr(mtod(m, struct sockaddr *),
2165 			    M_NOWAIT);
2166 		m = m_free(m);
2167 	}
2168 	if (m == NULL) {
2169 		/* XXXRW: Can this happen? */
2170 		return (0);
2171 	}
2172 
2173 	/*
2174 	 * Packet to copyout() is now in 'm' and it is disconnected from the
2175 	 * queue.
2176 	 *
2177 	 * Process one or more MT_CONTROL mbufs present before any data mbufs
2178 	 * in the first mbuf chain on the socket buffer.  We call into the
2179 	 * protocol to perform externalization (or freeing if controlp ==
2180 	 * NULL).
2181 	 */
2182 	if (m->m_type == MT_CONTROL) {
2183 		struct mbuf *cm = NULL, *cmn;
2184 		struct mbuf **cme = &cm;
2185 
2186 		do {
2187 			m2 = m->m_next;
2188 			m->m_next = NULL;
2189 			*cme = m;
2190 			cme = &(*cme)->m_next;
2191 			m = m2;
2192 		} while (m != NULL && m->m_type == MT_CONTROL);
2193 		while (cm != NULL) {
2194 			cmn = cm->m_next;
2195 			cm->m_next = NULL;
2196 			if (pr->pr_domain->dom_externalize != NULL) {
2197 				error = (*pr->pr_domain->dom_externalize)
2198 				    (cm, controlp, flags);
2199 			} else if (controlp != NULL)
2200 				*controlp = cm;
2201 			else
2202 				m_freem(cm);
2203 			if (controlp != NULL) {
2204 				while (*controlp != NULL)
2205 					controlp = &(*controlp)->m_next;
2206 			}
2207 			cm = cmn;
2208 		}
2209 	}
2210 	KASSERT(m->m_type == MT_DATA, ("soreceive_dgram: !data"));
2211 
2212 	while (m != NULL && uio->uio_resid > 0) {
2213 		len = uio->uio_resid;
2214 		if (len > m->m_len)
2215 			len = m->m_len;
2216 		error = uiomove(mtod(m, char *), (int)len, uio);
2217 		if (error) {
2218 			m_freem(m);
2219 			return (error);
2220 		}
2221 		if (len == m->m_len)
2222 			m = m_free(m);
2223 		else {
2224 			m->m_data += len;
2225 			m->m_len -= len;
2226 		}
2227 	}
2228 	if (m != NULL)
2229 		flags |= MSG_TRUNC;
2230 	m_freem(m);
2231 	if (flagsp != NULL)
2232 		*flagsp |= flags;
2233 	return (0);
2234 }
2235 
2236 int
2237 soreceive(struct socket *so, struct sockaddr **psa, struct uio *uio,
2238     struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
2239 {
2240 	int error;
2241 
2242 	CURVNET_SET(so->so_vnet);
2243 	error = (so->so_proto->pr_usrreqs->pru_soreceive(so, psa, uio, mp0,
2244 	    controlp, flagsp));
2245 	CURVNET_RESTORE();
2246 	return (error);
2247 }
2248 
2249 int
2250 soshutdown(struct socket *so, int how)
2251 {
2252 	struct protosw *pr = so->so_proto;
2253 	int error;
2254 
2255 	if (!(how == SHUT_RD || how == SHUT_WR || how == SHUT_RDWR))
2256 		return (EINVAL);
2257 
2258 	CURVNET_SET(so->so_vnet);
2259 	if (pr->pr_usrreqs->pru_flush != NULL)
2260 		(*pr->pr_usrreqs->pru_flush)(so, how);
2261 	if (how != SHUT_WR)
2262 		sorflush(so);
2263 	if (how != SHUT_RD) {
2264 		error = (*pr->pr_usrreqs->pru_shutdown)(so);
2265 		wakeup(&so->so_timeo);
2266 		CURVNET_RESTORE();
2267 		return (error);
2268 	}
2269 	wakeup(&so->so_timeo);
2270 	CURVNET_RESTORE();
2271 	return (0);
2272 }
2273 
2274 void
2275 sorflush(struct socket *so)
2276 {
2277 	struct sockbuf *sb = &so->so_rcv;
2278 	struct protosw *pr = so->so_proto;
2279 	struct sockbuf asb;
2280 
2281 	VNET_SO_ASSERT(so);
2282 
2283 	/*
2284 	 * In order to avoid calling dom_dispose with the socket buffer mutex
2285 	 * held, and in order to generally avoid holding the lock for a long
2286 	 * time, we make a copy of the socket buffer and clear the original
2287 	 * (except locks, state).  The new socket buffer copy won't have
2288 	 * initialized locks so we can only call routines that won't use or
2289 	 * assert those locks.
2290 	 *
2291 	 * Dislodge threads currently blocked in receive and wait to acquire
2292 	 * a lock against other simultaneous readers before clearing the
2293 	 * socket buffer.  Don't let our acquire be interrupted by a signal
2294 	 * despite any existing socket disposition on interruptable waiting.
2295 	 */
2296 	socantrcvmore(so);
2297 	(void) sblock(sb, SBL_WAIT | SBL_NOINTR);
2298 
2299 	/*
2300 	 * Invalidate/clear most of the sockbuf structure, but leave selinfo
2301 	 * and mutex data unchanged.
2302 	 */
2303 	SOCKBUF_LOCK(sb);
2304 	bzero(&asb, offsetof(struct sockbuf, sb_startzero));
2305 	bcopy(&sb->sb_startzero, &asb.sb_startzero,
2306 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2307 	bzero(&sb->sb_startzero,
2308 	    sizeof(*sb) - offsetof(struct sockbuf, sb_startzero));
2309 	SOCKBUF_UNLOCK(sb);
2310 	sbunlock(sb);
2311 
2312 	/*
2313 	 * Dispose of special rights and flush the socket buffer.  Don't call
2314 	 * any unsafe routines (that rely on locks being initialized) on asb.
2315 	 */
2316 	if (pr->pr_flags & PR_RIGHTS && pr->pr_domain->dom_dispose != NULL)
2317 		(*pr->pr_domain->dom_dispose)(asb.sb_mb);
2318 	sbrelease_internal(&asb, so);
2319 }
2320 
2321 /*
2322  * Perhaps this routine, and sooptcopyout(), below, ought to come in an
2323  * additional variant to handle the case where the option value needs to be
2324  * some kind of integer, but not a specific size.  In addition to their use
2325  * here, these functions are also called by the protocol-level pr_ctloutput()
2326  * routines.
2327  */
2328 int
2329 sooptcopyin(struct sockopt *sopt, void *buf, size_t len, size_t minlen)
2330 {
2331 	size_t	valsize;
2332 
2333 	/*
2334 	 * If the user gives us more than we wanted, we ignore it, but if we
2335 	 * don't get the minimum length the caller wants, we return EINVAL.
2336 	 * On success, sopt->sopt_valsize is set to however much we actually
2337 	 * retrieved.
2338 	 */
2339 	if ((valsize = sopt->sopt_valsize) < minlen)
2340 		return EINVAL;
2341 	if (valsize > len)
2342 		sopt->sopt_valsize = valsize = len;
2343 
2344 	if (sopt->sopt_td != NULL)
2345 		return (copyin(sopt->sopt_val, buf, valsize));
2346 
2347 	bcopy(sopt->sopt_val, buf, valsize);
2348 	return (0);
2349 }
2350 
2351 /*
2352  * Kernel version of setsockopt(2).
2353  *
2354  * XXX: optlen is size_t, not socklen_t
2355  */
2356 int
2357 so_setsockopt(struct socket *so, int level, int optname, void *optval,
2358     size_t optlen)
2359 {
2360 	struct sockopt sopt;
2361 
2362 	sopt.sopt_level = level;
2363 	sopt.sopt_name = optname;
2364 	sopt.sopt_dir = SOPT_SET;
2365 	sopt.sopt_val = optval;
2366 	sopt.sopt_valsize = optlen;
2367 	sopt.sopt_td = NULL;
2368 	return (sosetopt(so, &sopt));
2369 }
2370 
2371 int
2372 sosetopt(struct socket *so, struct sockopt *sopt)
2373 {
2374 	int	error, optval;
2375 	struct	linger l;
2376 	struct	timeval tv;
2377 	sbintime_t val;
2378 	uint32_t val32;
2379 #ifdef MAC
2380 	struct mac extmac;
2381 #endif
2382 
2383 	CURVNET_SET(so->so_vnet);
2384 	error = 0;
2385 	if (sopt->sopt_level != SOL_SOCKET) {
2386 		if (so->so_proto->pr_ctloutput != NULL) {
2387 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2388 			CURVNET_RESTORE();
2389 			return (error);
2390 		}
2391 		error = ENOPROTOOPT;
2392 	} else {
2393 		switch (sopt->sopt_name) {
2394 #ifdef INET
2395 		case SO_ACCEPTFILTER:
2396 			error = do_setopt_accept_filter(so, sopt);
2397 			if (error)
2398 				goto bad;
2399 			break;
2400 #endif
2401 		case SO_LINGER:
2402 			error = sooptcopyin(sopt, &l, sizeof l, sizeof l);
2403 			if (error)
2404 				goto bad;
2405 
2406 			SOCK_LOCK(so);
2407 			so->so_linger = l.l_linger;
2408 			if (l.l_onoff)
2409 				so->so_options |= SO_LINGER;
2410 			else
2411 				so->so_options &= ~SO_LINGER;
2412 			SOCK_UNLOCK(so);
2413 			break;
2414 
2415 		case SO_DEBUG:
2416 		case SO_KEEPALIVE:
2417 		case SO_DONTROUTE:
2418 		case SO_USELOOPBACK:
2419 		case SO_BROADCAST:
2420 		case SO_REUSEADDR:
2421 		case SO_REUSEPORT:
2422 		case SO_OOBINLINE:
2423 		case SO_TIMESTAMP:
2424 		case SO_BINTIME:
2425 		case SO_NOSIGPIPE:
2426 		case SO_NO_DDP:
2427 		case SO_NO_OFFLOAD:
2428 			error = sooptcopyin(sopt, &optval, sizeof optval,
2429 			    sizeof optval);
2430 			if (error)
2431 				goto bad;
2432 			SOCK_LOCK(so);
2433 			if (optval)
2434 				so->so_options |= sopt->sopt_name;
2435 			else
2436 				so->so_options &= ~sopt->sopt_name;
2437 			SOCK_UNLOCK(so);
2438 			break;
2439 
2440 		case SO_SETFIB:
2441 			error = sooptcopyin(sopt, &optval, sizeof optval,
2442 			    sizeof optval);
2443 			if (error)
2444 				goto bad;
2445 
2446 			if (optval < 0 || optval >= rt_numfibs) {
2447 				error = EINVAL;
2448 				goto bad;
2449 			}
2450 			if (((so->so_proto->pr_domain->dom_family == PF_INET) ||
2451 			   (so->so_proto->pr_domain->dom_family == PF_INET6) ||
2452 			   (so->so_proto->pr_domain->dom_family == PF_ROUTE)))
2453 				so->so_fibnum = optval;
2454 			else
2455 				so->so_fibnum = 0;
2456 			break;
2457 
2458 		case SO_USER_COOKIE:
2459 			error = sooptcopyin(sopt, &val32, sizeof val32,
2460 			    sizeof val32);
2461 			if (error)
2462 				goto bad;
2463 			so->so_user_cookie = val32;
2464 			break;
2465 
2466 		case SO_SNDBUF:
2467 		case SO_RCVBUF:
2468 		case SO_SNDLOWAT:
2469 		case SO_RCVLOWAT:
2470 			error = sooptcopyin(sopt, &optval, sizeof optval,
2471 			    sizeof optval);
2472 			if (error)
2473 				goto bad;
2474 
2475 			/*
2476 			 * Values < 1 make no sense for any of these options,
2477 			 * so disallow them.
2478 			 */
2479 			if (optval < 1) {
2480 				error = EINVAL;
2481 				goto bad;
2482 			}
2483 
2484 			switch (sopt->sopt_name) {
2485 			case SO_SNDBUF:
2486 			case SO_RCVBUF:
2487 				if (sbreserve(sopt->sopt_name == SO_SNDBUF ?
2488 				    &so->so_snd : &so->so_rcv, (u_long)optval,
2489 				    so, curthread) == 0) {
2490 					error = ENOBUFS;
2491 					goto bad;
2492 				}
2493 				(sopt->sopt_name == SO_SNDBUF ? &so->so_snd :
2494 				    &so->so_rcv)->sb_flags &= ~SB_AUTOSIZE;
2495 				break;
2496 
2497 			/*
2498 			 * Make sure the low-water is never greater than the
2499 			 * high-water.
2500 			 */
2501 			case SO_SNDLOWAT:
2502 				SOCKBUF_LOCK(&so->so_snd);
2503 				so->so_snd.sb_lowat =
2504 				    (optval > so->so_snd.sb_hiwat) ?
2505 				    so->so_snd.sb_hiwat : optval;
2506 				SOCKBUF_UNLOCK(&so->so_snd);
2507 				break;
2508 			case SO_RCVLOWAT:
2509 				SOCKBUF_LOCK(&so->so_rcv);
2510 				so->so_rcv.sb_lowat =
2511 				    (optval > so->so_rcv.sb_hiwat) ?
2512 				    so->so_rcv.sb_hiwat : optval;
2513 				SOCKBUF_UNLOCK(&so->so_rcv);
2514 				break;
2515 			}
2516 			break;
2517 
2518 		case SO_SNDTIMEO:
2519 		case SO_RCVTIMEO:
2520 #ifdef COMPAT_FREEBSD32
2521 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2522 				struct timeval32 tv32;
2523 
2524 				error = sooptcopyin(sopt, &tv32, sizeof tv32,
2525 				    sizeof tv32);
2526 				CP(tv32, tv, tv_sec);
2527 				CP(tv32, tv, tv_usec);
2528 			} else
2529 #endif
2530 				error = sooptcopyin(sopt, &tv, sizeof tv,
2531 				    sizeof tv);
2532 			if (error)
2533 				goto bad;
2534 			if (tv.tv_sec < 0 || tv.tv_usec < 0 ||
2535 			    tv.tv_usec >= 1000000) {
2536 				error = EDOM;
2537 				goto bad;
2538 			}
2539 			val = tvtosbt(tv);
2540 
2541 			switch (sopt->sopt_name) {
2542 			case SO_SNDTIMEO:
2543 				so->so_snd.sb_timeo = val;
2544 				break;
2545 			case SO_RCVTIMEO:
2546 				so->so_rcv.sb_timeo = val;
2547 				break;
2548 			}
2549 			break;
2550 
2551 		case SO_LABEL:
2552 #ifdef MAC
2553 			error = sooptcopyin(sopt, &extmac, sizeof extmac,
2554 			    sizeof extmac);
2555 			if (error)
2556 				goto bad;
2557 			error = mac_setsockopt_label(sopt->sopt_td->td_ucred,
2558 			    so, &extmac);
2559 #else
2560 			error = EOPNOTSUPP;
2561 #endif
2562 			break;
2563 
2564 		default:
2565 			error = ENOPROTOOPT;
2566 			break;
2567 		}
2568 		if (error == 0 && so->so_proto->pr_ctloutput != NULL)
2569 			(void)(*so->so_proto->pr_ctloutput)(so, sopt);
2570 	}
2571 bad:
2572 	CURVNET_RESTORE();
2573 	return (error);
2574 }
2575 
2576 /*
2577  * Helper routine for getsockopt.
2578  */
2579 int
2580 sooptcopyout(struct sockopt *sopt, const void *buf, size_t len)
2581 {
2582 	int	error;
2583 	size_t	valsize;
2584 
2585 	error = 0;
2586 
2587 	/*
2588 	 * Documented get behavior is that we always return a value, possibly
2589 	 * truncated to fit in the user's buffer.  Traditional behavior is
2590 	 * that we always tell the user precisely how much we copied, rather
2591 	 * than something useful like the total amount we had available for
2592 	 * her.  Note that this interface is not idempotent; the entire
2593 	 * answer must generated ahead of time.
2594 	 */
2595 	valsize = min(len, sopt->sopt_valsize);
2596 	sopt->sopt_valsize = valsize;
2597 	if (sopt->sopt_val != NULL) {
2598 		if (sopt->sopt_td != NULL)
2599 			error = copyout(buf, sopt->sopt_val, valsize);
2600 		else
2601 			bcopy(buf, sopt->sopt_val, valsize);
2602 	}
2603 	return (error);
2604 }
2605 
2606 int
2607 sogetopt(struct socket *so, struct sockopt *sopt)
2608 {
2609 	int	error, optval;
2610 	struct	linger l;
2611 	struct	timeval tv;
2612 #ifdef MAC
2613 	struct mac extmac;
2614 #endif
2615 
2616 	CURVNET_SET(so->so_vnet);
2617 	error = 0;
2618 	if (sopt->sopt_level != SOL_SOCKET) {
2619 		if (so->so_proto->pr_ctloutput != NULL)
2620 			error = (*so->so_proto->pr_ctloutput)(so, sopt);
2621 		else
2622 			error = ENOPROTOOPT;
2623 		CURVNET_RESTORE();
2624 		return (error);
2625 	} else {
2626 		switch (sopt->sopt_name) {
2627 #ifdef INET
2628 		case SO_ACCEPTFILTER:
2629 			error = do_getopt_accept_filter(so, sopt);
2630 			break;
2631 #endif
2632 		case SO_LINGER:
2633 			SOCK_LOCK(so);
2634 			l.l_onoff = so->so_options & SO_LINGER;
2635 			l.l_linger = so->so_linger;
2636 			SOCK_UNLOCK(so);
2637 			error = sooptcopyout(sopt, &l, sizeof l);
2638 			break;
2639 
2640 		case SO_USELOOPBACK:
2641 		case SO_DONTROUTE:
2642 		case SO_DEBUG:
2643 		case SO_KEEPALIVE:
2644 		case SO_REUSEADDR:
2645 		case SO_REUSEPORT:
2646 		case SO_BROADCAST:
2647 		case SO_OOBINLINE:
2648 		case SO_ACCEPTCONN:
2649 		case SO_TIMESTAMP:
2650 		case SO_BINTIME:
2651 		case SO_NOSIGPIPE:
2652 			optval = so->so_options & sopt->sopt_name;
2653 integer:
2654 			error = sooptcopyout(sopt, &optval, sizeof optval);
2655 			break;
2656 
2657 		case SO_TYPE:
2658 			optval = so->so_type;
2659 			goto integer;
2660 
2661 		case SO_PROTOCOL:
2662 			optval = so->so_proto->pr_protocol;
2663 			goto integer;
2664 
2665 		case SO_ERROR:
2666 			SOCK_LOCK(so);
2667 			optval = so->so_error;
2668 			so->so_error = 0;
2669 			SOCK_UNLOCK(so);
2670 			goto integer;
2671 
2672 		case SO_SNDBUF:
2673 			optval = so->so_snd.sb_hiwat;
2674 			goto integer;
2675 
2676 		case SO_RCVBUF:
2677 			optval = so->so_rcv.sb_hiwat;
2678 			goto integer;
2679 
2680 		case SO_SNDLOWAT:
2681 			optval = so->so_snd.sb_lowat;
2682 			goto integer;
2683 
2684 		case SO_RCVLOWAT:
2685 			optval = so->so_rcv.sb_lowat;
2686 			goto integer;
2687 
2688 		case SO_SNDTIMEO:
2689 		case SO_RCVTIMEO:
2690 			optval = (sopt->sopt_name == SO_SNDTIMEO ?
2691 				  so->so_snd.sb_timeo : so->so_rcv.sb_timeo);
2692 
2693 			tv = sbttotv(optval);
2694 #ifdef COMPAT_FREEBSD32
2695 			if (SV_CURPROC_FLAG(SV_ILP32)) {
2696 				struct timeval32 tv32;
2697 
2698 				CP(tv, tv32, tv_sec);
2699 				CP(tv, tv32, tv_usec);
2700 				error = sooptcopyout(sopt, &tv32, sizeof tv32);
2701 			} else
2702 #endif
2703 				error = sooptcopyout(sopt, &tv, sizeof tv);
2704 			break;
2705 
2706 		case SO_LABEL:
2707 #ifdef MAC
2708 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2709 			    sizeof(extmac));
2710 			if (error)
2711 				goto bad;
2712 			error = mac_getsockopt_label(sopt->sopt_td->td_ucred,
2713 			    so, &extmac);
2714 			if (error)
2715 				goto bad;
2716 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2717 #else
2718 			error = EOPNOTSUPP;
2719 #endif
2720 			break;
2721 
2722 		case SO_PEERLABEL:
2723 #ifdef MAC
2724 			error = sooptcopyin(sopt, &extmac, sizeof(extmac),
2725 			    sizeof(extmac));
2726 			if (error)
2727 				goto bad;
2728 			error = mac_getsockopt_peerlabel(
2729 			    sopt->sopt_td->td_ucred, so, &extmac);
2730 			if (error)
2731 				goto bad;
2732 			error = sooptcopyout(sopt, &extmac, sizeof extmac);
2733 #else
2734 			error = EOPNOTSUPP;
2735 #endif
2736 			break;
2737 
2738 		case SO_LISTENQLIMIT:
2739 			optval = so->so_qlimit;
2740 			goto integer;
2741 
2742 		case SO_LISTENQLEN:
2743 			optval = so->so_qlen;
2744 			goto integer;
2745 
2746 		case SO_LISTENINCQLEN:
2747 			optval = so->so_incqlen;
2748 			goto integer;
2749 
2750 		default:
2751 			error = ENOPROTOOPT;
2752 			break;
2753 		}
2754 	}
2755 #ifdef MAC
2756 bad:
2757 #endif
2758 	CURVNET_RESTORE();
2759 	return (error);
2760 }
2761 
2762 int
2763 soopt_getm(struct sockopt *sopt, struct mbuf **mp)
2764 {
2765 	struct mbuf *m, *m_prev;
2766 	int sopt_size = sopt->sopt_valsize;
2767 
2768 	MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2769 	if (m == NULL)
2770 		return ENOBUFS;
2771 	if (sopt_size > MLEN) {
2772 		MCLGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT);
2773 		if ((m->m_flags & M_EXT) == 0) {
2774 			m_free(m);
2775 			return ENOBUFS;
2776 		}
2777 		m->m_len = min(MCLBYTES, sopt_size);
2778 	} else {
2779 		m->m_len = min(MLEN, sopt_size);
2780 	}
2781 	sopt_size -= m->m_len;
2782 	*mp = m;
2783 	m_prev = m;
2784 
2785 	while (sopt_size) {
2786 		MGET(m, sopt->sopt_td ? M_WAITOK : M_NOWAIT, MT_DATA);
2787 		if (m == NULL) {
2788 			m_freem(*mp);
2789 			return ENOBUFS;
2790 		}
2791 		if (sopt_size > MLEN) {
2792 			MCLGET(m, sopt->sopt_td != NULL ? M_WAITOK :
2793 			    M_NOWAIT);
2794 			if ((m->m_flags & M_EXT) == 0) {
2795 				m_freem(m);
2796 				m_freem(*mp);
2797 				return ENOBUFS;
2798 			}
2799 			m->m_len = min(MCLBYTES, sopt_size);
2800 		} else {
2801 			m->m_len = min(MLEN, sopt_size);
2802 		}
2803 		sopt_size -= m->m_len;
2804 		m_prev->m_next = m;
2805 		m_prev = m;
2806 	}
2807 	return (0);
2808 }
2809 
2810 int
2811 soopt_mcopyin(struct sockopt *sopt, struct mbuf *m)
2812 {
2813 	struct mbuf *m0 = m;
2814 
2815 	if (sopt->sopt_val == NULL)
2816 		return (0);
2817 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2818 		if (sopt->sopt_td != NULL) {
2819 			int error;
2820 
2821 			error = copyin(sopt->sopt_val, mtod(m, char *),
2822 			    m->m_len);
2823 			if (error != 0) {
2824 				m_freem(m0);
2825 				return(error);
2826 			}
2827 		} else
2828 			bcopy(sopt->sopt_val, mtod(m, char *), m->m_len);
2829 		sopt->sopt_valsize -= m->m_len;
2830 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2831 		m = m->m_next;
2832 	}
2833 	if (m != NULL) /* should be allocated enoughly at ip6_sooptmcopyin() */
2834 		panic("ip6_sooptmcopyin");
2835 	return (0);
2836 }
2837 
2838 int
2839 soopt_mcopyout(struct sockopt *sopt, struct mbuf *m)
2840 {
2841 	struct mbuf *m0 = m;
2842 	size_t valsize = 0;
2843 
2844 	if (sopt->sopt_val == NULL)
2845 		return (0);
2846 	while (m != NULL && sopt->sopt_valsize >= m->m_len) {
2847 		if (sopt->sopt_td != NULL) {
2848 			int error;
2849 
2850 			error = copyout(mtod(m, char *), sopt->sopt_val,
2851 			    m->m_len);
2852 			if (error != 0) {
2853 				m_freem(m0);
2854 				return(error);
2855 			}
2856 		} else
2857 			bcopy(mtod(m, char *), sopt->sopt_val, m->m_len);
2858 		sopt->sopt_valsize -= m->m_len;
2859 		sopt->sopt_val = (char *)sopt->sopt_val + m->m_len;
2860 		valsize += m->m_len;
2861 		m = m->m_next;
2862 	}
2863 	if (m != NULL) {
2864 		/* enough soopt buffer should be given from user-land */
2865 		m_freem(m0);
2866 		return(EINVAL);
2867 	}
2868 	sopt->sopt_valsize = valsize;
2869 	return (0);
2870 }
2871 
2872 /*
2873  * sohasoutofband(): protocol notifies socket layer of the arrival of new
2874  * out-of-band data, which will then notify socket consumers.
2875  */
2876 void
2877 sohasoutofband(struct socket *so)
2878 {
2879 
2880 	if (so->so_sigio != NULL)
2881 		pgsigio(&so->so_sigio, SIGURG, 0);
2882 	selwakeuppri(&so->so_rcv.sb_sel, PSOCK);
2883 }
2884 
2885 int
2886 sopoll(struct socket *so, int events, struct ucred *active_cred,
2887     struct thread *td)
2888 {
2889 
2890 	/*
2891 	 * We do not need to set or assert curvnet as long as everyone uses
2892 	 * sopoll_generic().
2893 	 */
2894 	return (so->so_proto->pr_usrreqs->pru_sopoll(so, events, active_cred,
2895 	    td));
2896 }
2897 
2898 int
2899 sopoll_generic(struct socket *so, int events, struct ucred *active_cred,
2900     struct thread *td)
2901 {
2902 	int revents = 0;
2903 
2904 	SOCKBUF_LOCK(&so->so_snd);
2905 	SOCKBUF_LOCK(&so->so_rcv);
2906 	if (events & (POLLIN | POLLRDNORM))
2907 		if (soreadabledata(so))
2908 			revents |= events & (POLLIN | POLLRDNORM);
2909 
2910 	if (events & (POLLOUT | POLLWRNORM))
2911 		if (sowriteable(so))
2912 			revents |= events & (POLLOUT | POLLWRNORM);
2913 
2914 	if (events & (POLLPRI | POLLRDBAND))
2915 		if (so->so_oobmark || (so->so_rcv.sb_state & SBS_RCVATMARK))
2916 			revents |= events & (POLLPRI | POLLRDBAND);
2917 
2918 	if ((events & POLLINIGNEOF) == 0) {
2919 		if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
2920 			revents |= events & (POLLIN | POLLRDNORM);
2921 			if (so->so_snd.sb_state & SBS_CANTSENDMORE)
2922 				revents |= POLLHUP;
2923 		}
2924 	}
2925 
2926 	if (revents == 0) {
2927 		if (events & (POLLIN | POLLPRI | POLLRDNORM | POLLRDBAND)) {
2928 			selrecord(td, &so->so_rcv.sb_sel);
2929 			so->so_rcv.sb_flags |= SB_SEL;
2930 		}
2931 
2932 		if (events & (POLLOUT | POLLWRNORM)) {
2933 			selrecord(td, &so->so_snd.sb_sel);
2934 			so->so_snd.sb_flags |= SB_SEL;
2935 		}
2936 	}
2937 
2938 	SOCKBUF_UNLOCK(&so->so_rcv);
2939 	SOCKBUF_UNLOCK(&so->so_snd);
2940 	return (revents);
2941 }
2942 
2943 int
2944 soo_kqfilter(struct file *fp, struct knote *kn)
2945 {
2946 	struct socket *so = kn->kn_fp->f_data;
2947 	struct sockbuf *sb;
2948 
2949 	switch (kn->kn_filter) {
2950 	case EVFILT_READ:
2951 		if (so->so_options & SO_ACCEPTCONN)
2952 			kn->kn_fop = &solisten_filtops;
2953 		else
2954 			kn->kn_fop = &soread_filtops;
2955 		sb = &so->so_rcv;
2956 		break;
2957 	case EVFILT_WRITE:
2958 		kn->kn_fop = &sowrite_filtops;
2959 		sb = &so->so_snd;
2960 		break;
2961 	default:
2962 		return (EINVAL);
2963 	}
2964 
2965 	SOCKBUF_LOCK(sb);
2966 	knlist_add(&sb->sb_sel.si_note, kn, 1);
2967 	sb->sb_flags |= SB_KNOTE;
2968 	SOCKBUF_UNLOCK(sb);
2969 	return (0);
2970 }
2971 
2972 /*
2973  * Some routines that return EOPNOTSUPP for entry points that are not
2974  * supported by a protocol.  Fill in as needed.
2975  */
2976 int
2977 pru_accept_notsupp(struct socket *so, struct sockaddr **nam)
2978 {
2979 
2980 	return EOPNOTSUPP;
2981 }
2982 
2983 int
2984 pru_attach_notsupp(struct socket *so, int proto, struct thread *td)
2985 {
2986 
2987 	return EOPNOTSUPP;
2988 }
2989 
2990 int
2991 pru_bind_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
2992 {
2993 
2994 	return EOPNOTSUPP;
2995 }
2996 
2997 int
2998 pru_bindat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
2999     struct thread *td)
3000 {
3001 
3002 	return EOPNOTSUPP;
3003 }
3004 
3005 int
3006 pru_connect_notsupp(struct socket *so, struct sockaddr *nam, struct thread *td)
3007 {
3008 
3009 	return EOPNOTSUPP;
3010 }
3011 
3012 int
3013 pru_connectat_notsupp(int fd, struct socket *so, struct sockaddr *nam,
3014     struct thread *td)
3015 {
3016 
3017 	return EOPNOTSUPP;
3018 }
3019 
3020 int
3021 pru_connect2_notsupp(struct socket *so1, struct socket *so2)
3022 {
3023 
3024 	return EOPNOTSUPP;
3025 }
3026 
3027 int
3028 pru_control_notsupp(struct socket *so, u_long cmd, caddr_t data,
3029     struct ifnet *ifp, struct thread *td)
3030 {
3031 
3032 	return EOPNOTSUPP;
3033 }
3034 
3035 int
3036 pru_disconnect_notsupp(struct socket *so)
3037 {
3038 
3039 	return EOPNOTSUPP;
3040 }
3041 
3042 int
3043 pru_listen_notsupp(struct socket *so, int backlog, struct thread *td)
3044 {
3045 
3046 	return EOPNOTSUPP;
3047 }
3048 
3049 int
3050 pru_peeraddr_notsupp(struct socket *so, struct sockaddr **nam)
3051 {
3052 
3053 	return EOPNOTSUPP;
3054 }
3055 
3056 int
3057 pru_rcvd_notsupp(struct socket *so, int flags)
3058 {
3059 
3060 	return EOPNOTSUPP;
3061 }
3062 
3063 int
3064 pru_rcvoob_notsupp(struct socket *so, struct mbuf *m, int flags)
3065 {
3066 
3067 	return EOPNOTSUPP;
3068 }
3069 
3070 int
3071 pru_send_notsupp(struct socket *so, int flags, struct mbuf *m,
3072     struct sockaddr *addr, struct mbuf *control, struct thread *td)
3073 {
3074 
3075 	return EOPNOTSUPP;
3076 }
3077 
3078 /*
3079  * This isn't really a ``null'' operation, but it's the default one and
3080  * doesn't do anything destructive.
3081  */
3082 int
3083 pru_sense_null(struct socket *so, struct stat *sb)
3084 {
3085 
3086 	sb->st_blksize = so->so_snd.sb_hiwat;
3087 	return 0;
3088 }
3089 
3090 int
3091 pru_shutdown_notsupp(struct socket *so)
3092 {
3093 
3094 	return EOPNOTSUPP;
3095 }
3096 
3097 int
3098 pru_sockaddr_notsupp(struct socket *so, struct sockaddr **nam)
3099 {
3100 
3101 	return EOPNOTSUPP;
3102 }
3103 
3104 int
3105 pru_sosend_notsupp(struct socket *so, struct sockaddr *addr, struct uio *uio,
3106     struct mbuf *top, struct mbuf *control, int flags, struct thread *td)
3107 {
3108 
3109 	return EOPNOTSUPP;
3110 }
3111 
3112 int
3113 pru_soreceive_notsupp(struct socket *so, struct sockaddr **paddr,
3114     struct uio *uio, struct mbuf **mp0, struct mbuf **controlp, int *flagsp)
3115 {
3116 
3117 	return EOPNOTSUPP;
3118 }
3119 
3120 int
3121 pru_sopoll_notsupp(struct socket *so, int events, struct ucred *cred,
3122     struct thread *td)
3123 {
3124 
3125 	return EOPNOTSUPP;
3126 }
3127 
3128 static void
3129 filt_sordetach(struct knote *kn)
3130 {
3131 	struct socket *so = kn->kn_fp->f_data;
3132 
3133 	SOCKBUF_LOCK(&so->so_rcv);
3134 	knlist_remove(&so->so_rcv.sb_sel.si_note, kn, 1);
3135 	if (knlist_empty(&so->so_rcv.sb_sel.si_note))
3136 		so->so_rcv.sb_flags &= ~SB_KNOTE;
3137 	SOCKBUF_UNLOCK(&so->so_rcv);
3138 }
3139 
3140 /*ARGSUSED*/
3141 static int
3142 filt_soread(struct knote *kn, long hint)
3143 {
3144 	struct socket *so;
3145 
3146 	so = kn->kn_fp->f_data;
3147 	SOCKBUF_LOCK_ASSERT(&so->so_rcv);
3148 
3149 	kn->kn_data = so->so_rcv.sb_cc - so->so_rcv.sb_ctl;
3150 	if (so->so_rcv.sb_state & SBS_CANTRCVMORE) {
3151 		kn->kn_flags |= EV_EOF;
3152 		kn->kn_fflags = so->so_error;
3153 		return (1);
3154 	} else if (so->so_error)	/* temporary udp error */
3155 		return (1);
3156 	else if (kn->kn_sfflags & NOTE_LOWAT)
3157 		return (kn->kn_data >= kn->kn_sdata);
3158 	else
3159 		return (so->so_rcv.sb_cc >= so->so_rcv.sb_lowat);
3160 }
3161 
3162 static void
3163 filt_sowdetach(struct knote *kn)
3164 {
3165 	struct socket *so = kn->kn_fp->f_data;
3166 
3167 	SOCKBUF_LOCK(&so->so_snd);
3168 	knlist_remove(&so->so_snd.sb_sel.si_note, kn, 1);
3169 	if (knlist_empty(&so->so_snd.sb_sel.si_note))
3170 		so->so_snd.sb_flags &= ~SB_KNOTE;
3171 	SOCKBUF_UNLOCK(&so->so_snd);
3172 }
3173 
3174 /*ARGSUSED*/
3175 static int
3176 filt_sowrite(struct knote *kn, long hint)
3177 {
3178 	struct socket *so;
3179 
3180 	so = kn->kn_fp->f_data;
3181 	SOCKBUF_LOCK_ASSERT(&so->so_snd);
3182 	kn->kn_data = sbspace(&so->so_snd);
3183 	if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
3184 		kn->kn_flags |= EV_EOF;
3185 		kn->kn_fflags = so->so_error;
3186 		return (1);
3187 	} else if (so->so_error)	/* temporary udp error */
3188 		return (1);
3189 	else if (((so->so_state & SS_ISCONNECTED) == 0) &&
3190 	    (so->so_proto->pr_flags & PR_CONNREQUIRED))
3191 		return (0);
3192 	else if (kn->kn_sfflags & NOTE_LOWAT)
3193 		return (kn->kn_data >= kn->kn_sdata);
3194 	else
3195 		return (kn->kn_data >= so->so_snd.sb_lowat);
3196 }
3197 
3198 /*ARGSUSED*/
3199 static int
3200 filt_solisten(struct knote *kn, long hint)
3201 {
3202 	struct socket *so = kn->kn_fp->f_data;
3203 
3204 	kn->kn_data = so->so_qlen;
3205 	return (!TAILQ_EMPTY(&so->so_comp));
3206 }
3207 
3208 int
3209 socheckuid(struct socket *so, uid_t uid)
3210 {
3211 
3212 	if (so == NULL)
3213 		return (EPERM);
3214 	if (so->so_cred->cr_uid != uid)
3215 		return (EPERM);
3216 	return (0);
3217 }
3218 
3219 /*
3220  * These functions are used by protocols to notify the socket layer (and its
3221  * consumers) of state changes in the sockets driven by protocol-side events.
3222  */
3223 
3224 /*
3225  * Procedures to manipulate state flags of socket and do appropriate wakeups.
3226  *
3227  * Normal sequence from the active (originating) side is that
3228  * soisconnecting() is called during processing of connect() call, resulting
3229  * in an eventual call to soisconnected() if/when the connection is
3230  * established.  When the connection is torn down soisdisconnecting() is
3231  * called during processing of disconnect() call, and soisdisconnected() is
3232  * called when the connection to the peer is totally severed.  The semantics
3233  * of these routines are such that connectionless protocols can call
3234  * soisconnected() and soisdisconnected() only, bypassing the in-progress
3235  * calls when setting up a ``connection'' takes no time.
3236  *
3237  * From the passive side, a socket is created with two queues of sockets:
3238  * so_incomp for connections in progress and so_comp for connections already
3239  * made and awaiting user acceptance.  As a protocol is preparing incoming
3240  * connections, it creates a socket structure queued on so_incomp by calling
3241  * sonewconn().  When the connection is established, soisconnected() is
3242  * called, and transfers the socket structure to so_comp, making it available
3243  * to accept().
3244  *
3245  * If a socket is closed with sockets on either so_incomp or so_comp, these
3246  * sockets are dropped.
3247  *
3248  * If higher-level protocols are implemented in the kernel, the wakeups done
3249  * here will sometimes cause software-interrupt process scheduling.
3250  */
3251 void
3252 soisconnecting(struct socket *so)
3253 {
3254 
3255 	SOCK_LOCK(so);
3256 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
3257 	so->so_state |= SS_ISCONNECTING;
3258 	SOCK_UNLOCK(so);
3259 }
3260 
3261 void
3262 soisconnected(struct socket *so)
3263 {
3264 	struct socket *head;
3265 	int ret;
3266 
3267 restart:
3268 	ACCEPT_LOCK();
3269 	SOCK_LOCK(so);
3270 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING|SS_ISCONFIRMING);
3271 	so->so_state |= SS_ISCONNECTED;
3272 	head = so->so_head;
3273 	if (head != NULL && (so->so_qstate & SQ_INCOMP)) {
3274 		if ((so->so_options & SO_ACCEPTFILTER) == 0) {
3275 			SOCK_UNLOCK(so);
3276 			TAILQ_REMOVE(&head->so_incomp, so, so_list);
3277 			head->so_incqlen--;
3278 			so->so_qstate &= ~SQ_INCOMP;
3279 			TAILQ_INSERT_TAIL(&head->so_comp, so, so_list);
3280 			head->so_qlen++;
3281 			so->so_qstate |= SQ_COMP;
3282 			ACCEPT_UNLOCK();
3283 			sorwakeup(head);
3284 			wakeup_one(&head->so_timeo);
3285 		} else {
3286 			ACCEPT_UNLOCK();
3287 			soupcall_set(so, SO_RCV,
3288 			    head->so_accf->so_accept_filter->accf_callback,
3289 			    head->so_accf->so_accept_filter_arg);
3290 			so->so_options &= ~SO_ACCEPTFILTER;
3291 			ret = head->so_accf->so_accept_filter->accf_callback(so,
3292 			    head->so_accf->so_accept_filter_arg, M_NOWAIT);
3293 			if (ret == SU_ISCONNECTED)
3294 				soupcall_clear(so, SO_RCV);
3295 			SOCK_UNLOCK(so);
3296 			if (ret == SU_ISCONNECTED)
3297 				goto restart;
3298 		}
3299 		return;
3300 	}
3301 	SOCK_UNLOCK(so);
3302 	ACCEPT_UNLOCK();
3303 	wakeup(&so->so_timeo);
3304 	sorwakeup(so);
3305 	sowwakeup(so);
3306 }
3307 
3308 void
3309 soisdisconnecting(struct socket *so)
3310 {
3311 
3312 	/*
3313 	 * Note: This code assumes that SOCK_LOCK(so) and
3314 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3315 	 */
3316 	SOCKBUF_LOCK(&so->so_rcv);
3317 	so->so_state &= ~SS_ISCONNECTING;
3318 	so->so_state |= SS_ISDISCONNECTING;
3319 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3320 	sorwakeup_locked(so);
3321 	SOCKBUF_LOCK(&so->so_snd);
3322 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3323 	sowwakeup_locked(so);
3324 	wakeup(&so->so_timeo);
3325 }
3326 
3327 void
3328 soisdisconnected(struct socket *so)
3329 {
3330 
3331 	/*
3332 	 * Note: This code assumes that SOCK_LOCK(so) and
3333 	 * SOCKBUF_LOCK(&so->so_rcv) are the same.
3334 	 */
3335 	SOCKBUF_LOCK(&so->so_rcv);
3336 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
3337 	so->so_state |= SS_ISDISCONNECTED;
3338 	so->so_rcv.sb_state |= SBS_CANTRCVMORE;
3339 	sorwakeup_locked(so);
3340 	SOCKBUF_LOCK(&so->so_snd);
3341 	so->so_snd.sb_state |= SBS_CANTSENDMORE;
3342 	sbdrop_locked(&so->so_snd, so->so_snd.sb_cc);
3343 	sowwakeup_locked(so);
3344 	wakeup(&so->so_timeo);
3345 }
3346 
3347 /*
3348  * Make a copy of a sockaddr in a malloced buffer of type M_SONAME.
3349  */
3350 struct sockaddr *
3351 sodupsockaddr(const struct sockaddr *sa, int mflags)
3352 {
3353 	struct sockaddr *sa2;
3354 
3355 	sa2 = malloc(sa->sa_len, M_SONAME, mflags);
3356 	if (sa2)
3357 		bcopy(sa, sa2, sa->sa_len);
3358 	return sa2;
3359 }
3360 
3361 /*
3362  * Register per-socket buffer upcalls.
3363  */
3364 void
3365 soupcall_set(struct socket *so, int which,
3366     int (*func)(struct socket *, void *, int), void *arg)
3367 {
3368 	struct sockbuf *sb;
3369 
3370 	switch (which) {
3371 	case SO_RCV:
3372 		sb = &so->so_rcv;
3373 		break;
3374 	case SO_SND:
3375 		sb = &so->so_snd;
3376 		break;
3377 	default:
3378 		panic("soupcall_set: bad which");
3379 	}
3380 	SOCKBUF_LOCK_ASSERT(sb);
3381 #if 0
3382 	/* XXX: accf_http actually wants to do this on purpose. */
3383 	KASSERT(sb->sb_upcall == NULL, ("soupcall_set: overwriting upcall"));
3384 #endif
3385 	sb->sb_upcall = func;
3386 	sb->sb_upcallarg = arg;
3387 	sb->sb_flags |= SB_UPCALL;
3388 }
3389 
3390 void
3391 soupcall_clear(struct socket *so, int which)
3392 {
3393 	struct sockbuf *sb;
3394 
3395 	switch (which) {
3396 	case SO_RCV:
3397 		sb = &so->so_rcv;
3398 		break;
3399 	case SO_SND:
3400 		sb = &so->so_snd;
3401 		break;
3402 	default:
3403 		panic("soupcall_clear: bad which");
3404 	}
3405 	SOCKBUF_LOCK_ASSERT(sb);
3406 	KASSERT(sb->sb_upcall != NULL, ("soupcall_clear: no upcall to clear"));
3407 	sb->sb_upcall = NULL;
3408 	sb->sb_upcallarg = NULL;
3409 	sb->sb_flags &= ~SB_UPCALL;
3410 }
3411 
3412 /*
3413  * Create an external-format (``xsocket'') structure using the information in
3414  * the kernel-format socket structure pointed to by so.  This is done to
3415  * reduce the spew of irrelevant information over this interface, to isolate
3416  * user code from changes in the kernel structure, and potentially to provide
3417  * information-hiding if we decide that some of this information should be
3418  * hidden from users.
3419  */
3420 void
3421 sotoxsocket(struct socket *so, struct xsocket *xso)
3422 {
3423 
3424 	xso->xso_len = sizeof *xso;
3425 	xso->xso_so = so;
3426 	xso->so_type = so->so_type;
3427 	xso->so_options = so->so_options;
3428 	xso->so_linger = so->so_linger;
3429 	xso->so_state = so->so_state;
3430 	xso->so_pcb = so->so_pcb;
3431 	xso->xso_protocol = so->so_proto->pr_protocol;
3432 	xso->xso_family = so->so_proto->pr_domain->dom_family;
3433 	xso->so_qlen = so->so_qlen;
3434 	xso->so_incqlen = so->so_incqlen;
3435 	xso->so_qlimit = so->so_qlimit;
3436 	xso->so_timeo = so->so_timeo;
3437 	xso->so_error = so->so_error;
3438 	xso->so_pgid = so->so_sigio ? so->so_sigio->sio_pgid : 0;
3439 	xso->so_oobmark = so->so_oobmark;
3440 	sbtoxsockbuf(&so->so_snd, &xso->so_snd);
3441 	sbtoxsockbuf(&so->so_rcv, &xso->so_rcv);
3442 	xso->so_uid = so->so_cred->cr_uid;
3443 }
3444 
3445 
3446 /*
3447  * Socket accessor functions to provide external consumers with
3448  * a safe interface to socket state
3449  *
3450  */
3451 
3452 void
3453 so_listeners_apply_all(struct socket *so, void (*func)(struct socket *, void *),
3454     void *arg)
3455 {
3456 
3457 	TAILQ_FOREACH(so, &so->so_comp, so_list)
3458 		func(so, arg);
3459 }
3460 
3461 struct sockbuf *
3462 so_sockbuf_rcv(struct socket *so)
3463 {
3464 
3465 	return (&so->so_rcv);
3466 }
3467 
3468 struct sockbuf *
3469 so_sockbuf_snd(struct socket *so)
3470 {
3471 
3472 	return (&so->so_snd);
3473 }
3474 
3475 int
3476 so_state_get(const struct socket *so)
3477 {
3478 
3479 	return (so->so_state);
3480 }
3481 
3482 void
3483 so_state_set(struct socket *so, int val)
3484 {
3485 
3486 	so->so_state = val;
3487 }
3488 
3489 int
3490 so_options_get(const struct socket *so)
3491 {
3492 
3493 	return (so->so_options);
3494 }
3495 
3496 void
3497 so_options_set(struct socket *so, int val)
3498 {
3499 
3500 	so->so_options = val;
3501 }
3502 
3503 int
3504 so_error_get(const struct socket *so)
3505 {
3506 
3507 	return (so->so_error);
3508 }
3509 
3510 void
3511 so_error_set(struct socket *so, int val)
3512 {
3513 
3514 	so->so_error = val;
3515 }
3516 
3517 int
3518 so_linger_get(const struct socket *so)
3519 {
3520 
3521 	return (so->so_linger);
3522 }
3523 
3524 void
3525 so_linger_set(struct socket *so, int val)
3526 {
3527 
3528 	so->so_linger = val;
3529 }
3530 
3531 struct protosw *
3532 so_protosw_get(const struct socket *so)
3533 {
3534 
3535 	return (so->so_proto);
3536 }
3537 
3538 void
3539 so_protosw_set(struct socket *so, struct protosw *val)
3540 {
3541 
3542 	so->so_proto = val;
3543 }
3544 
3545 void
3546 so_sorwakeup(struct socket *so)
3547 {
3548 
3549 	sorwakeup(so);
3550 }
3551 
3552 void
3553 so_sowwakeup(struct socket *so)
3554 {
3555 
3556 	sowwakeup(so);
3557 }
3558 
3559 void
3560 so_sorwakeup_locked(struct socket *so)
3561 {
3562 
3563 	sorwakeup_locked(so);
3564 }
3565 
3566 void
3567 so_sowwakeup_locked(struct socket *so)
3568 {
3569 
3570 	sowwakeup_locked(so);
3571 }
3572 
3573 void
3574 so_lock(struct socket *so)
3575 {
3576 
3577 	SOCK_LOCK(so);
3578 }
3579 
3580 void
3581 so_unlock(struct socket *so)
3582 {
3583 
3584 	SOCK_UNLOCK(so);
3585 }
3586