xref: /freebsd/sys/netgraph/ng_socket.c (revision 1d386b48)
1 /*
2  * ng_socket.c
3  */
4 
5 /*-
6  * Copyright (c) 1996-1999 Whistle Communications, Inc.
7  * All rights reserved.
8  *
9  * Subject to the following obligations and disclaimer of warranty, use and
10  * redistribution of this software, in source or object code forms, with or
11  * without modifications are expressly permitted by Whistle Communications;
12  * provided, however, that:
13  * 1. Any and all reproductions of the source or object code must include the
14  *    copyright notice above and the following disclaimer of warranties; and
15  * 2. No rights are granted, in any manner or form, to use Whistle
16  *    Communications, Inc. trademarks, including the mark "WHISTLE
17  *    COMMUNICATIONS" on advertising, endorsements, or otherwise except as
18  *    such appears in the above copyright notice or in the software.
19  *
20  * THIS SOFTWARE IS BEING PROVIDED BY WHISTLE COMMUNICATIONS "AS IS", AND
21  * TO THE MAXIMUM EXTENT PERMITTED BY LAW, WHISTLE COMMUNICATIONS MAKES NO
22  * REPRESENTATIONS OR WARRANTIES, EXPRESS OR IMPLIED, REGARDING THIS SOFTWARE,
23  * INCLUDING WITHOUT LIMITATION, ANY AND ALL IMPLIED WARRANTIES OF
24  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT.
25  * WHISTLE COMMUNICATIONS DOES NOT WARRANT, GUARANTEE, OR MAKE ANY
26  * REPRESENTATIONS REGARDING THE USE OF, OR THE RESULTS OF THE USE OF THIS
27  * SOFTWARE IN TERMS OF ITS CORRECTNESS, ACCURACY, RELIABILITY OR OTHERWISE.
28  * IN NO EVENT SHALL WHISTLE COMMUNICATIONS BE LIABLE FOR ANY DAMAGES
29  * RESULTING FROM OR ARISING OUT OF ANY USE OF THIS SOFTWARE, INCLUDING
30  * WITHOUT LIMITATION, ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY,
31  * PUNITIVE, OR CONSEQUENTIAL DAMAGES, PROCUREMENT OF SUBSTITUTE GOODS OR
32  * SERVICES, LOSS OF USE, DATA OR PROFITS, HOWEVER CAUSED AND UNDER ANY
33  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
34  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
35  * THIS SOFTWARE, EVEN IF WHISTLE COMMUNICATIONS IS ADVISED OF THE POSSIBILITY
36  * OF SUCH DAMAGE.
37  *
38  * Author: Julian Elischer <julian@freebsd.org>
39  * $Whistle: ng_socket.c,v 1.28 1999/11/01 09:24:52 julian Exp $
40  */
41 
42 /*
43  * Netgraph socket nodes
44  *
45  * There are two types of netgraph sockets, control and data.
46  * Control sockets have a netgraph node, but data sockets are
47  * parasitic on control sockets, and have no node of their own.
48  */
49 
50 #include <sys/param.h>
51 #include <sys/domain.h>
52 #include <sys/hash.h>
53 #include <sys/kernel.h>
54 #include <sys/linker.h>
55 #include <sys/lock.h>
56 #include <sys/malloc.h>
57 #include <sys/mbuf.h>
58 #include <sys/mutex.h>
59 #include <sys/proc.h>
60 #include <sys/epoch.h>
61 #include <sys/priv.h>
62 #include <sys/protosw.h>
63 #include <sys/queue.h>
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <sys/syscallsubr.h>
67 #include <sys/sysctl.h>
68 
69 #include <net/vnet.h>
70 
71 #include <netgraph/ng_message.h>
72 #include <netgraph/netgraph.h>
73 #include <netgraph/ng_socketvar.h>
74 #include <netgraph/ng_socket.h>
75 
76 #ifdef NG_SEPARATE_MALLOC
77 static MALLOC_DEFINE(M_NETGRAPH_PATH, "netgraph_path", "netgraph path info");
78 static MALLOC_DEFINE(M_NETGRAPH_SOCK, "netgraph_sock", "netgraph socket info");
79 #else
80 #define M_NETGRAPH_PATH M_NETGRAPH
81 #define M_NETGRAPH_SOCK M_NETGRAPH
82 #endif
83 
84 /*
85  * It's Ascii-art time!
86  *   +-------------+   +-------------+
87  *   |socket  (ctl)|   |socket (data)|
88  *   +-------------+   +-------------+
89  *          ^                 ^
90  *          |                 |
91  *          v                 v
92  *    +-----------+     +-----------+
93  *    |pcb   (ctl)|     |pcb  (data)|
94  *    +-----------+     +-----------+
95  *          ^                 ^
96  *          |                 |
97  *          v                 v
98  *      +--------------------------+
99  *      |   Socket type private    |
100  *      |       data               |
101  *      +--------------------------+
102  *                   ^
103  *                   |
104  *                   v
105  *           +----------------+
106  *           | struct ng_node |
107  *           +----------------+
108  */
109 
110 /* Netgraph node methods */
111 static ng_constructor_t	ngs_constructor;
112 static ng_rcvmsg_t	ngs_rcvmsg;
113 static ng_shutdown_t	ngs_shutdown;
114 static ng_newhook_t	ngs_newhook;
115 static ng_connect_t	ngs_connect;
116 static ng_findhook_t	ngs_findhook;
117 static ng_rcvdata_t	ngs_rcvdata;
118 static ng_disconnect_t	ngs_disconnect;
119 
120 /* Internal methods */
121 static int	ng_attach_data(struct socket *so);
122 static int	ng_attach_cntl(struct socket *so);
123 static int	ng_attach_common(struct socket *so, int type);
124 static void	ng_detach_common(struct ngpcb *pcbp, int type);
125 static void	ng_socket_free_priv(struct ngsock *priv);
126 static int	ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp);
127 static int	ng_bind(struct sockaddr *nam, struct ngpcb *pcbp);
128 
129 static int	ngs_mod_event(module_t mod, int event, void *data);
130 static void	ng_socket_item_applied(void *context, int error);
131 
132 /* Netgraph type descriptor */
133 static struct ng_type typestruct = {
134 	.version =	NG_ABI_VERSION,
135 	.name =		NG_SOCKET_NODE_TYPE,
136 	.mod_event =	ngs_mod_event,
137 	.constructor =	ngs_constructor,
138 	.rcvmsg =	ngs_rcvmsg,
139 	.shutdown =	ngs_shutdown,
140 	.newhook =	ngs_newhook,
141 	.connect =	ngs_connect,
142 	.findhook =	ngs_findhook,
143 	.rcvdata =	ngs_rcvdata,
144 	.disconnect =	ngs_disconnect,
145 };
146 NETGRAPH_INIT_ORDERED(socket, &typestruct, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
147 
148 /* Buffer space */
149 static u_long ngpdg_sendspace = 20 * 1024;	/* really max datagram size */
150 SYSCTL_ULONG(_net_graph, OID_AUTO, maxdgram, CTLFLAG_RW,
151     &ngpdg_sendspace , 0, "Maximum outgoing Netgraph datagram size");
152 static u_long ngpdg_recvspace = 20 * 1024;
153 SYSCTL_ULONG(_net_graph, OID_AUTO, recvspace, CTLFLAG_RW,
154     &ngpdg_recvspace , 0, "Maximum space for incoming Netgraph datagrams");
155 
156 /* List of all sockets (for netstat -f netgraph) */
157 static LIST_HEAD(, ngpcb) ngsocklist;
158 
159 static struct mtx	ngsocketlist_mtx;
160 
161 #define sotongpcb(so) ((struct ngpcb *)(so)->so_pcb)
162 
163 /* If getting unexplained errors returned, set this to "kdb_enter("X"); */
164 #ifndef TRAP_ERROR
165 #define TRAP_ERROR
166 #endif
167 
168 struct hookpriv {
169 	LIST_ENTRY(hookpriv)	next;
170 	hook_p			hook;
171 };
172 LIST_HEAD(ngshash, hookpriv);
173 
174 /* Per-node private data */
175 struct ngsock {
176 	struct ng_node	*node;		/* the associated netgraph node */
177 	struct ngpcb	*datasock;	/* optional data socket */
178 	struct ngpcb	*ctlsock;	/* optional control socket */
179 	struct ngshash	*hash;		/* hash for hook names */
180 	u_long		hmask;		/* hash mask */
181 	int	flags;
182 	int	refs;
183 	struct mtx	mtx;		/* mtx to wait on */
184 	int		error;		/* place to store error */
185 };
186 
187 #define	NGS_FLAG_NOLINGER	1	/* close with last hook */
188 
189 /***************************************************************
190 	Control sockets
191 ***************************************************************/
192 
193 static int
194 ngc_attach(struct socket *so, int proto, struct thread *td)
195 {
196 	struct ngpcb *const pcbp = sotongpcb(so);
197 	int error;
198 
199 	error = priv_check(td, PRIV_NETGRAPH_CONTROL);
200 	if (error)
201 		return (error);
202 	if (pcbp != NULL)
203 		return (EISCONN);
204 	return (ng_attach_cntl(so));
205 }
206 
207 static void
208 ngc_detach(struct socket *so)
209 {
210 	struct ngpcb *const pcbp = sotongpcb(so);
211 
212 	KASSERT(pcbp != NULL, ("ngc_detach: pcbp == NULL"));
213 	ng_detach_common(pcbp, NG_CONTROL);
214 }
215 
216 static int
217 ngc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
218 	 struct mbuf *control, struct thread *td)
219 {
220 	struct ngpcb *const pcbp = sotongpcb(so);
221 	struct ngsock *const priv = NG_NODE_PRIVATE(pcbp->sockdata->node);
222 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) addr;
223 	struct ng_mesg *msg;
224 	struct mbuf *m0;
225 	item_p item;
226 	char *path = NULL;
227 	int len, error = 0;
228 	struct ng_apply_info apply;
229 
230 	if (control) {
231 		error = EINVAL;
232 		goto release;
233 	}
234 
235 	/* Require destination as there may be >= 1 hooks on this node. */
236 	if (addr == NULL) {
237 		error = EDESTADDRREQ;
238 		goto release;
239 	}
240 
241 	if (sap->sg_len > NG_NODESIZ + offsetof(struct sockaddr_ng, sg_data)) {
242 		error = EINVAL;
243 		goto release;
244 	}
245 
246 	/*
247 	 * Allocate an expendable buffer for the path, chop off
248 	 * the sockaddr header, and make sure it's NUL terminated.
249 	 */
250 	len = sap->sg_len - offsetof(struct sockaddr_ng, sg_data);
251 	path = malloc(len + 1, M_NETGRAPH_PATH, M_WAITOK);
252 	bcopy(sap->sg_data, path, len);
253 	path[len] = '\0';
254 
255 	/*
256 	 * Move the actual message out of mbufs into a linear buffer.
257 	 * Start by adding up the size of the data. (could use mh_len?)
258 	 */
259 	for (len = 0, m0 = m; m0 != NULL; m0 = m0->m_next)
260 		len += m0->m_len;
261 
262 	/*
263 	 * Move the data into a linear buffer as well.
264 	 * Messages are not delivered in mbufs.
265 	 */
266 	msg = malloc(len + 1, M_NETGRAPH_MSG, M_WAITOK);
267 	m_copydata(m, 0, len, (char *)msg);
268 
269 	if (msg->header.version != NG_VERSION) {
270 		free(msg, M_NETGRAPH_MSG);
271 		error = EINVAL;
272 		goto release;
273 	}
274 
275 	/*
276 	 * Hack alert!
277 	 * We look into the message and if it mkpeers a node of unknown type, we
278 	 * try to load it. We need to do this now, in syscall thread, because if
279 	 * message gets queued and applied later we will get panic.
280 	 */
281 	if (msg->header.typecookie == NGM_GENERIC_COOKIE &&
282 	    msg->header.cmd == NGM_MKPEER) {
283 		struct ngm_mkpeer *const mkp = (struct ngm_mkpeer *) msg->data;
284 
285 		if (ng_findtype(mkp->type) == NULL) {
286 			char filename[NG_TYPESIZ + 3];
287 			int fileid;
288 
289 			/* Not found, try to load it as a loadable module. */
290 			snprintf(filename, sizeof(filename), "ng_%s",
291 			    mkp->type);
292 			error = kern_kldload(curthread, filename, &fileid);
293 			if (error != 0) {
294 				free(msg, M_NETGRAPH_MSG);
295 				goto release;
296 			}
297 
298 			/* See if type has been loaded successfully. */
299 			if (ng_findtype(mkp->type) == NULL) {
300 				free(msg, M_NETGRAPH_MSG);
301 				(void)kern_kldunload(curthread, fileid,
302 				    LINKER_UNLOAD_NORMAL);
303 				error =  ENXIO;
304 				goto release;
305 			}
306 		}
307 	}
308 
309 	item = ng_package_msg(msg, NG_WAITOK);
310 	if ((error = ng_address_path((pcbp->sockdata->node), item, path, 0))
311 	    != 0) {
312 #ifdef TRACE_MESSAGES
313 		printf("ng_address_path: errx=%d\n", error);
314 #endif
315 		goto release;
316 	}
317 
318 #ifdef TRACE_MESSAGES
319 	printf("[%x]:<---------[socket]: c=<%d>cmd=%x(%s) f=%x #%d (%s)\n",
320 		item->el_dest->nd_ID,
321 		msg->header.typecookie,
322 		msg->header.cmd,
323 		msg->header.cmdstr,
324 		msg->header.flags,
325 		msg->header.token,
326 		item->el_dest->nd_type->name);
327 #endif
328 	SAVE_LINE(item);
329 	/*
330 	 * We do not want to return from syscall until the item
331 	 * is processed by destination node. We register callback
332 	 * on the item, which will update priv->error when item
333 	 * was applied.
334 	 * If ng_snd_item() has queued item, we sleep until
335 	 * callback wakes us up.
336 	 */
337 	bzero(&apply, sizeof(apply));
338 	apply.apply = ng_socket_item_applied;
339 	apply.context = priv;
340 	item->apply = &apply;
341 	priv->error = -1;
342 
343 	error = ng_snd_item(item, 0);
344 
345 	mtx_lock(&priv->mtx);
346 	if (priv->error == -1)
347 		msleep(priv, &priv->mtx, 0, "ngsock", 0);
348 	mtx_unlock(&priv->mtx);
349 	KASSERT(priv->error != -1,
350 	    ("ng_socket: priv->error wasn't updated"));
351 	error = priv->error;
352 
353 release:
354 	if (path != NULL)
355 		free(path, M_NETGRAPH_PATH);
356 	if (control != NULL)
357 		m_freem(control);
358 	if (m != NULL)
359 		m_freem(m);
360 	return (error);
361 }
362 
363 static int
364 ngc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
365 {
366 	struct ngpcb *const pcbp = sotongpcb(so);
367 
368 	if (pcbp == NULL)
369 		return (EINVAL);
370 	return (ng_bind(nam, pcbp));
371 }
372 
373 static int
374 ngc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
375 {
376 	/*
377 	 * At this time refuse to do this.. it used to
378 	 * do something but it was undocumented and not used.
379 	 */
380 	printf("program tried to connect control socket to remote node\n");
381 	return (EINVAL);
382 }
383 
384 /***************************************************************
385 	Data sockets
386 ***************************************************************/
387 
388 static int
389 ngd_attach(struct socket *so, int proto, struct thread *td)
390 {
391 	struct ngpcb *const pcbp = sotongpcb(so);
392 
393 	if (pcbp != NULL)
394 		return (EISCONN);
395 	return (ng_attach_data(so));
396 }
397 
398 static void
399 ngd_detach(struct socket *so)
400 {
401 	struct ngpcb *const pcbp = sotongpcb(so);
402 
403 	KASSERT(pcbp != NULL, ("ngd_detach: pcbp == NULL"));
404 	ng_detach_common(pcbp, NG_DATA);
405 }
406 
407 static int
408 ngd_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
409 	 struct mbuf *control, struct thread *td)
410 {
411 	struct epoch_tracker et;
412 	struct ngpcb *const pcbp = sotongpcb(so);
413 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) addr;
414 	int	len, error;
415 	hook_p  hook = NULL;
416 	item_p	item;
417 	char	hookname[NG_HOOKSIZ];
418 
419 	if ((pcbp == NULL) || (control != NULL)) {
420 		error = EINVAL;
421 		goto release;
422 	}
423 	if (pcbp->sockdata == NULL) {
424 		error = ENOTCONN;
425 		goto release;
426 	}
427 
428 	if (sap == NULL) {
429 		len = 0;		/* Make compiler happy. */
430 	} else {
431 		if (sap->sg_len > NG_NODESIZ +
432 		    offsetof(struct sockaddr_ng, sg_data)) {
433 			error = EINVAL;
434 			goto release;
435 		}
436 		len = sap->sg_len - offsetof(struct sockaddr_ng, sg_data);
437 	}
438 
439 	/*
440 	 * If the user used any of these ways to not specify an address
441 	 * then handle specially.
442 	 */
443 	if ((sap == NULL) || (len <= 0) || (*sap->sg_data == '\0')) {
444 		if (NG_NODE_NUMHOOKS(pcbp->sockdata->node) != 1) {
445 			error = EDESTADDRREQ;
446 			goto release;
447 		}
448 		/*
449 		 * If exactly one hook exists, just use it.
450 		 * Special case to allow write(2) to work on an ng_socket.
451 		 */
452 		hook = LIST_FIRST(&pcbp->sockdata->node->nd_hooks);
453 	} else {
454 		if (len >= NG_HOOKSIZ) {
455 			error = EINVAL;
456 			goto release;
457 		}
458 
459 		/*
460 		 * chop off the sockaddr header, and make sure it's NUL
461 		 * terminated
462 		 */
463 		bcopy(sap->sg_data, hookname, len);
464 		hookname[len] = '\0';
465 
466 		/* Find the correct hook from 'hookname' */
467 		hook = ng_findhook(pcbp->sockdata->node, hookname);
468 		if (hook == NULL) {
469 			error = EHOSTUNREACH;
470 			goto release;
471 		}
472 	}
473 
474 	/* Send data. */
475 	item = ng_package_data(m, NG_WAITOK);
476 	m = NULL;
477 	NET_EPOCH_ENTER(et);
478 	NG_FWD_ITEM_HOOK(error, item, hook);
479 	NET_EPOCH_EXIT(et);
480 
481 release:
482 	if (control != NULL)
483 		m_freem(control);
484 	if (m != NULL)
485 		m_freem(m);
486 	return (error);
487 }
488 
489 static int
490 ngd_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
491 {
492 	struct ngpcb *const pcbp = sotongpcb(so);
493 
494 	if (pcbp == NULL)
495 		return (EINVAL);
496 	return (ng_connect_data(nam, pcbp));
497 }
498 
499 /*
500  * Used for both data and control sockets
501  */
502 static int
503 ng_getsockaddr(struct socket *so, struct sockaddr **addr)
504 {
505 	struct ngpcb *pcbp;
506 	struct sockaddr_ng *sg;
507 	int sg_len;
508 	int error = 0;
509 
510 	pcbp = sotongpcb(so);
511 	if ((pcbp == NULL) || (pcbp->sockdata == NULL))
512 		/* XXXGL: can this still happen? */
513 		return (EINVAL);
514 
515 	sg_len = sizeof(struct sockaddr_ng) + NG_NODESIZ -
516 	    sizeof(sg->sg_data);
517 	sg = malloc(sg_len, M_SONAME, M_WAITOK | M_ZERO);
518 
519 	mtx_lock(&pcbp->sockdata->mtx);
520 	if (pcbp->sockdata->node != NULL) {
521 		node_p node = pcbp->sockdata->node;
522 
523 		if (NG_NODE_HAS_NAME(node))
524 			bcopy(NG_NODE_NAME(node), sg->sg_data,
525 			    strlen(NG_NODE_NAME(node)));
526 		mtx_unlock(&pcbp->sockdata->mtx);
527 
528 		sg->sg_len = sg_len;
529 		sg->sg_family = AF_NETGRAPH;
530 		*addr = (struct sockaddr *)sg;
531 	} else {
532 		mtx_unlock(&pcbp->sockdata->mtx);
533 		free(sg, M_SONAME);
534 		error = EINVAL;
535 	}
536 
537 	return (error);
538 }
539 
540 /*
541  * Attach a socket to it's protocol specific partner.
542  * For a control socket, actually create a netgraph node and attach
543  * to it as well.
544  */
545 
546 static int
547 ng_attach_cntl(struct socket *so)
548 {
549 	struct ngsock *priv;
550 	struct ngpcb *pcbp;
551 	node_p node;
552 	int error;
553 
554 	/* Setup protocol control block */
555 	if ((error = ng_attach_common(so, NG_CONTROL)) != 0)
556 		return (error);
557 	pcbp = sotongpcb(so);
558 
559 	/* Make the generic node components */
560 	if ((error = ng_make_node_common(&typestruct, &node)) != 0) {
561 		ng_detach_common(pcbp, NG_CONTROL);
562 		return (error);
563 	}
564 
565 	/*
566 	 * Allocate node private info and hash. We start
567 	 * with 16 hash entries, however we may grow later
568 	 * in ngs_newhook(). We can't predict how much hooks
569 	 * does this node plan to have.
570 	 */
571 	priv = malloc(sizeof(*priv), M_NETGRAPH_SOCK, M_WAITOK | M_ZERO);
572 	priv->hash = hashinit(16, M_NETGRAPH_SOCK, &priv->hmask);
573 
574 	/* Initialize mutex. */
575 	mtx_init(&priv->mtx, "ng_socket", NULL, MTX_DEF);
576 
577 	/* Link the pcb the private data. */
578 	priv->ctlsock = pcbp;
579 	pcbp->sockdata = priv;
580 	priv->refs++;
581 	priv->node = node;
582 	pcbp->node_id = node->nd_ID;	/* hint for netstat(1) */
583 
584 	/* Link the node and the private data. */
585 	NG_NODE_SET_PRIVATE(priv->node, priv);
586 	NG_NODE_REF(priv->node);
587 	priv->refs++;
588 
589 	return (0);
590 }
591 
592 static int
593 ng_attach_data(struct socket *so)
594 {
595 	return (ng_attach_common(so, NG_DATA));
596 }
597 
598 /*
599  * Set up a socket protocol control block.
600  * This code is shared between control and data sockets.
601  */
602 static int
603 ng_attach_common(struct socket *so, int type)
604 {
605 	struct ngpcb *pcbp;
606 	int error;
607 
608 	/* Standard socket setup stuff. */
609 	error = soreserve(so, ngpdg_sendspace, ngpdg_recvspace);
610 	if (error)
611 		return (error);
612 
613 	/* Allocate the pcb. */
614 	pcbp = malloc(sizeof(struct ngpcb), M_PCB, M_WAITOK | M_ZERO);
615 	pcbp->type = type;
616 
617 	/* Link the pcb and the socket. */
618 	so->so_pcb = (caddr_t)pcbp;
619 	pcbp->ng_socket = so;
620 
621 	/* Add the socket to linked list */
622 	mtx_lock(&ngsocketlist_mtx);
623 	LIST_INSERT_HEAD(&ngsocklist, pcbp, socks);
624 	mtx_unlock(&ngsocketlist_mtx);
625 	return (0);
626 }
627 
628 /*
629  * Disassociate the socket from it's protocol specific
630  * partner. If it's attached to a node's private data structure,
631  * then unlink from that too. If we were the last socket attached to it,
632  * then shut down the entire node. Shared code for control and data sockets.
633  */
634 static void
635 ng_detach_common(struct ngpcb *pcbp, int which)
636 {
637 	struct ngsock *priv = pcbp->sockdata;
638 
639 	if (priv != NULL) {
640 		mtx_lock(&priv->mtx);
641 
642 		switch (which) {
643 		case NG_CONTROL:
644 			priv->ctlsock = NULL;
645 			break;
646 		case NG_DATA:
647 			priv->datasock = NULL;
648 			break;
649 		default:
650 			panic("%s", __func__);
651 		}
652 		pcbp->sockdata = NULL;
653 		pcbp->node_id = 0;
654 
655 		ng_socket_free_priv(priv);
656 	}
657 
658 	pcbp->ng_socket->so_pcb = NULL;
659 	mtx_lock(&ngsocketlist_mtx);
660 	LIST_REMOVE(pcbp, socks);
661 	mtx_unlock(&ngsocketlist_mtx);
662 	free(pcbp, M_PCB);
663 }
664 
665 /*
666  * Remove a reference from node private data.
667  */
668 static void
669 ng_socket_free_priv(struct ngsock *priv)
670 {
671 	mtx_assert(&priv->mtx, MA_OWNED);
672 
673 	priv->refs--;
674 
675 	if (priv->refs == 0) {
676 		mtx_destroy(&priv->mtx);
677 		hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
678 		free(priv, M_NETGRAPH_SOCK);
679 		return;
680 	}
681 
682 	if ((priv->refs == 1) && (priv->node != NULL)) {
683 		node_p node = priv->node;
684 
685 		priv->node = NULL;
686 		mtx_unlock(&priv->mtx);
687 		NG_NODE_UNREF(node);
688 		ng_rmnode_self(node);
689 	} else
690 		mtx_unlock(&priv->mtx);
691 }
692 
693 /*
694  * Connect the data socket to a named control socket node.
695  */
696 static int
697 ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp)
698 {
699 	struct sockaddr_ng *sap;
700 	node_p farnode;
701 	struct ngsock *priv;
702 	int error;
703 	item_p item;
704 
705 	/* If we are already connected, don't do it again. */
706 	if (pcbp->sockdata != NULL)
707 		return (EISCONN);
708 
709 	/*
710 	 * Find the target (victim) and check it doesn't already have
711 	 * a data socket. Also check it is a 'socket' type node.
712 	 * Use ng_package_data() and ng_address_path() to do this.
713 	 */
714 
715 	sap = (struct sockaddr_ng *) nam;
716 	/* The item will hold the node reference. */
717 	item = ng_package_data(NULL, NG_WAITOK);
718 
719 	if ((error = ng_address_path(NULL, item,  sap->sg_data, 0)))
720 		return (error); /* item is freed on failure */
721 
722 	/*
723 	 * Extract node from item and free item. Remember we now have
724 	 * a reference on the node. The item holds it for us.
725 	 * when we free the item we release the reference.
726 	 */
727 	farnode = item->el_dest; /* shortcut */
728 	if (strcmp(farnode->nd_type->name, NG_SOCKET_NODE_TYPE) != 0) {
729 		NG_FREE_ITEM(item); /* drop the reference to the node */
730 		return (EINVAL);
731 	}
732 	priv = NG_NODE_PRIVATE(farnode);
733 	if (priv->datasock != NULL) {
734 		NG_FREE_ITEM(item);	/* drop the reference to the node */
735 		return (EADDRINUSE);
736 	}
737 
738 	/*
739 	 * Link the PCB and the private data struct. and note the extra
740 	 * reference. Drop the extra reference on the node.
741 	 */
742 	mtx_lock(&priv->mtx);
743 	priv->datasock = pcbp;
744 	pcbp->sockdata = priv;
745 	pcbp->node_id = priv->node->nd_ID;	/* hint for netstat(1) */
746 	priv->refs++;
747 	mtx_unlock(&priv->mtx);
748 	NG_FREE_ITEM(item);	/* drop the reference to the node */
749 	return (0);
750 }
751 
752 /*
753  * Binding a socket means giving the corresponding node a name
754  */
755 static int
756 ng_bind(struct sockaddr *nam, struct ngpcb *pcbp)
757 {
758 	struct ngsock *const priv = pcbp->sockdata;
759 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) nam;
760 
761 	if (priv == NULL) {
762 		TRAP_ERROR;
763 		return (EINVAL);
764 	}
765 	if ((sap->sg_len < 4) || (sap->sg_len > (NG_NODESIZ + 2)) ||
766 	    (sap->sg_data[0] == '\0') ||
767 	    (sap->sg_data[sap->sg_len - 3] != '\0')) {
768 		TRAP_ERROR;
769 		return (EINVAL);
770 	}
771 	return (ng_name_node(priv->node, sap->sg_data));
772 }
773 
774 /***************************************************************
775 	Netgraph node
776 ***************************************************************/
777 
778 /*
779  * You can only create new nodes from the socket end of things.
780  */
781 static int
782 ngs_constructor(node_p nodep)
783 {
784 	return (EINVAL);
785 }
786 
787 static void
788 ngs_rehash(node_p node)
789 {
790 	struct ngsock *priv = NG_NODE_PRIVATE(node);
791 	struct ngshash *new;
792 	struct hookpriv *hp;
793 	hook_p hook;
794 	uint32_t h;
795 	u_long hmask;
796 
797 	new = hashinit_flags((priv->hmask + 1) * 2, M_NETGRAPH_SOCK, &hmask,
798 	    HASH_NOWAIT);
799 	if (new == NULL)
800 		return;
801 
802 	LIST_FOREACH(hook, &node->nd_hooks, hk_hooks) {
803 		hp = NG_HOOK_PRIVATE(hook);
804 #ifdef INVARIANTS
805 		LIST_REMOVE(hp, next);
806 #endif
807 		h = hash32_str(NG_HOOK_NAME(hook), HASHINIT) & hmask;
808 		LIST_INSERT_HEAD(&new[h], hp, next);
809 	}
810 
811 	hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
812 	priv->hash = new;
813 	priv->hmask = hmask;
814 }
815 
816 /*
817  * We allow any hook to be connected to the node.
818  * There is no per-hook private information though.
819  */
820 static int
821 ngs_newhook(node_p node, hook_p hook, const char *name)
822 {
823 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
824 	struct hookpriv *hp;
825 	uint32_t h;
826 
827 	hp = malloc(sizeof(*hp), M_NETGRAPH_SOCK, M_NOWAIT);
828 	if (hp == NULL)
829 		return (ENOMEM);
830 	if (node->nd_numhooks * 2 > priv->hmask)
831 		ngs_rehash(node);
832 	hp->hook = hook;
833 	h = hash32_str(name, HASHINIT) & priv->hmask;
834 	LIST_INSERT_HEAD(&priv->hash[h], hp, next);
835 	NG_HOOK_SET_PRIVATE(hook, hp);
836 
837 	return (0);
838 }
839 
840 /*
841  * If only one hook, allow read(2) and write(2) to work.
842  */
843 static int
844 ngs_connect(hook_p hook)
845 {
846 	node_p node = NG_HOOK_NODE(hook);
847 	struct ngsock *priv = NG_NODE_PRIVATE(node);
848 
849 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
850 		if (NG_NODE_NUMHOOKS(node) == 1)
851 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
852 		else
853 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
854 	}
855 	return (0);
856 }
857 
858 /* Look up hook by name */
859 static hook_p
860 ngs_findhook(node_p node, const char *name)
861 {
862 	struct ngsock *priv = NG_NODE_PRIVATE(node);
863 	struct hookpriv *hp;
864 	uint32_t h;
865 
866 	/*
867 	 * Microoptimisation for an ng_socket with
868 	 * a single hook, which is a common case.
869 	 */
870 	if (node->nd_numhooks == 1) {
871 		hook_p hook;
872 
873 		hook = LIST_FIRST(&node->nd_hooks);
874 
875 		if (strcmp(NG_HOOK_NAME(hook), name) == 0)
876 			return (hook);
877 		else
878 			return (NULL);
879 	}
880 
881 	h = hash32_str(name, HASHINIT) & priv->hmask;
882 
883 	LIST_FOREACH(hp, &priv->hash[h], next)
884 		if (strcmp(NG_HOOK_NAME(hp->hook), name) == 0)
885 			return (hp->hook);
886 
887 	return (NULL);
888 }
889 
890 /*
891  * Incoming messages get passed up to the control socket.
892  * Unless they are for us specifically (socket_type)
893  */
894 static int
895 ngs_rcvmsg(node_p node, item_p item, hook_p lasthook)
896 {
897 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
898 	struct ngpcb *pcbp;
899 	struct socket *so;
900 	struct sockaddr_ng addr;
901 	struct ng_mesg *msg;
902 	struct mbuf *m;
903 	ng_ID_t	retaddr = NGI_RETADDR(item);
904 	int addrlen;
905 	int error = 0;
906 
907 	NGI_GET_MSG(item, msg);
908 	NG_FREE_ITEM(item);
909 
910 	/*
911 	 * Grab priv->mtx here to prevent destroying of control socket
912 	 * after checking that priv->ctlsock is not NULL.
913 	 */
914 	mtx_lock(&priv->mtx);
915 	pcbp = priv->ctlsock;
916 
917 	/*
918 	 * Only allow mesgs to be passed if we have the control socket.
919 	 * Data sockets can only support the generic messages.
920 	 */
921 	if (pcbp == NULL) {
922 		mtx_unlock(&priv->mtx);
923 		TRAP_ERROR;
924 		NG_FREE_MSG(msg);
925 		return (EINVAL);
926 	}
927 	so = pcbp->ng_socket;
928 	SOCKBUF_LOCK(&so->so_rcv);
929 
930 	/* As long as the race is handled, priv->mtx may be unlocked now. */
931 	mtx_unlock(&priv->mtx);
932 
933 #ifdef TRACE_MESSAGES
934 	printf("[%x]:---------->[socket]: c=<%d>cmd=%x(%s) f=%x #%d\n",
935 		retaddr,
936 		msg->header.typecookie,
937 		msg->header.cmd,
938 		msg->header.cmdstr,
939 		msg->header.flags,
940 		msg->header.token);
941 #endif
942 
943 	if (msg->header.typecookie == NGM_SOCKET_COOKIE) {
944 		switch (msg->header.cmd) {
945 		case NGM_SOCK_CMD_NOLINGER:
946 			priv->flags |= NGS_FLAG_NOLINGER;
947 			break;
948 		case NGM_SOCK_CMD_LINGER:
949 			priv->flags &= ~NGS_FLAG_NOLINGER;
950 			break;
951 		default:
952 			error = EINVAL;		/* unknown command */
953 		}
954 		SOCKBUF_UNLOCK(&so->so_rcv);
955 
956 		/* Free the message and return. */
957 		NG_FREE_MSG(msg);
958 		return (error);
959 	}
960 
961 	/* Get the return address into a sockaddr. */
962 	bzero(&addr, sizeof(addr));
963 	addr.sg_len = sizeof(addr);
964 	addr.sg_family = AF_NETGRAPH;
965 	addrlen = snprintf((char *)&addr.sg_data, sizeof(addr.sg_data),
966 	    "[%x]:", retaddr);
967 	if (addrlen < 0 || addrlen > sizeof(addr.sg_data)) {
968 		SOCKBUF_UNLOCK(&so->so_rcv);
969 		printf("%s: snprintf([%x]) failed - %d\n", __func__, retaddr,
970 		    addrlen);
971 		NG_FREE_MSG(msg);
972 		return (EINVAL);
973 	}
974 
975 	/* Copy the message itself into an mbuf chain. */
976 	m = m_devget((caddr_t)msg, sizeof(struct ng_mesg) + msg->header.arglen,
977 	    0, NULL, NULL);
978 
979 	/*
980 	 * Here we free the message. We need to do that
981 	 * regardless of whether we got mbufs.
982 	 */
983 	NG_FREE_MSG(msg);
984 
985 	if (m == NULL) {
986 		SOCKBUF_UNLOCK(&so->so_rcv);
987 		TRAP_ERROR;
988 		return (ENOBUFS);
989 	}
990 
991 	/* Send it up to the socket. */
992 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&addr, m,
993 	    NULL) == 0) {
994 		soroverflow_locked(so);
995 		TRAP_ERROR;
996 		m_freem(m);
997 		return (ENOBUFS);
998 	}
999 
1000 	/* sorwakeup_locked () releases the lock internally. */
1001 	sorwakeup_locked(so);
1002 
1003 	return (error);
1004 }
1005 
1006 /*
1007  * Receive data on a hook
1008  */
1009 static int
1010 ngs_rcvdata(hook_p hook, item_p item)
1011 {
1012 	struct ngsock *const priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
1013 	struct ngpcb *const pcbp = priv->datasock;
1014 	struct socket *so;
1015 	struct sockaddr_ng *addr;
1016 	char *addrbuf[NG_HOOKSIZ + 4];
1017 	int addrlen;
1018 	struct mbuf *m;
1019 
1020 	NGI_GET_M(item, m);
1021 	NG_FREE_ITEM(item);
1022 
1023 	/* If there is no data socket, black-hole it. */
1024 	if (pcbp == NULL) {
1025 		NG_FREE_M(m);
1026 		return (0);
1027 	}
1028 	so = pcbp->ng_socket;
1029 
1030 	/* Get the return address into a sockaddr. */
1031 	addrlen = strlen(NG_HOOK_NAME(hook));	/* <= NG_HOOKSIZ - 1 */
1032 	addr = (struct sockaddr_ng *) addrbuf;
1033 	addr->sg_len = addrlen + 3;
1034 	addr->sg_family = AF_NETGRAPH;
1035 	bcopy(NG_HOOK_NAME(hook), addr->sg_data, addrlen);
1036 	addr->sg_data[addrlen] = '\0';
1037 
1038 	/* Try to tell the socket which hook it came in on. */
1039 	SOCKBUF_LOCK(&so->so_rcv);
1040 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)addr, m,
1041 	    NULL) == 0) {
1042 		SOCKBUF_UNLOCK(&so->so_rcv);
1043 		m_freem(m);
1044 		TRAP_ERROR;
1045 		return (ENOBUFS);
1046 	}
1047 
1048 	/* sorwakeup_locked () releases the lock internally. */
1049 	sorwakeup_locked(so);
1050 	return (0);
1051 }
1052 
1053 /*
1054  * Hook disconnection
1055  *
1056  * For this type, removal of the last link destroys the node
1057  * if the NOLINGER flag is set.
1058  */
1059 static int
1060 ngs_disconnect(hook_p hook)
1061 {
1062 	node_p node = NG_HOOK_NODE(hook);
1063 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1064 	struct hookpriv *hp = NG_HOOK_PRIVATE(hook);
1065 
1066 	LIST_REMOVE(hp, next);
1067 	free(hp, M_NETGRAPH_SOCK);
1068 
1069 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
1070 		if (NG_NODE_NUMHOOKS(node) == 1)
1071 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
1072 		else
1073 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
1074 	}
1075 
1076 	if ((priv->flags & NGS_FLAG_NOLINGER) &&
1077 	    (NG_NODE_NUMHOOKS(node) == 0) && (NG_NODE_IS_VALID(node)))
1078 		ng_rmnode_self(node);
1079 
1080 	return (0);
1081 }
1082 
1083 /*
1084  * Do local shutdown processing.
1085  * In this case, that involves making sure the socket
1086  * knows we should be shutting down.
1087  */
1088 static int
1089 ngs_shutdown(node_p node)
1090 {
1091 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1092 	struct ngpcb *dpcbp, *pcbp;
1093 
1094 	mtx_lock(&priv->mtx);
1095 	dpcbp = priv->datasock;
1096 	pcbp = priv->ctlsock;
1097 
1098 	if (dpcbp != NULL)
1099 		soisdisconnected(dpcbp->ng_socket);
1100 
1101 	if (pcbp != NULL)
1102 		soisdisconnected(pcbp->ng_socket);
1103 
1104 	priv->node = NULL;
1105 	NG_NODE_SET_PRIVATE(node, NULL);
1106 	ng_socket_free_priv(priv);
1107 
1108 	NG_NODE_UNREF(node);
1109 	return (0);
1110 }
1111 
1112 static void
1113 ng_socket_item_applied(void *context, int error)
1114 {
1115 	struct ngsock *const priv = (struct ngsock *)context;
1116 
1117 	mtx_lock(&priv->mtx);
1118 	priv->error = error;
1119 	wakeup(priv);
1120 	mtx_unlock(&priv->mtx);
1121 
1122 }
1123 
1124 static	int
1125 dummy_disconnect(struct socket *so)
1126 {
1127 	return (0);
1128 }
1129 
1130 /*
1131  * Definitions of protocols supported in the NETGRAPH domain.
1132  * Control and data socket type descriptors
1133  *
1134  * XXXRW: Perhaps _close should do something?
1135  */
1136 static struct protosw ngcontrol_protosw = {
1137 	.pr_type =		SOCK_DGRAM,
1138 	.pr_protocol =		NG_CONTROL,
1139 	.pr_flags =		PR_ATOMIC | PR_ADDR /* | PR_RIGHTS */,
1140 	.pr_attach =		ngc_attach,
1141 	.pr_bind =		ngc_bind,
1142 	.pr_connect =		ngc_connect,
1143 	.pr_detach =		ngc_detach,
1144 	.pr_disconnect =	dummy_disconnect,
1145 	.pr_send =		ngc_send,
1146 	.pr_sockaddr =		ng_getsockaddr,
1147 };
1148 static struct protosw ngdata_protosw = {
1149 	.pr_type =		SOCK_DGRAM,
1150 	.pr_protocol =		NG_DATA,
1151 	.pr_flags =		PR_ATOMIC | PR_ADDR,
1152 	.pr_attach =		ngd_attach,
1153 	.pr_connect =		ngd_connect,
1154 	.pr_detach =		ngd_detach,
1155 	.pr_disconnect =	dummy_disconnect,
1156 	.pr_send =		ngd_send,
1157 	.pr_sockaddr =		ng_getsockaddr,
1158 };
1159 
1160 static struct domain ngdomain = {
1161 	.dom_family =		AF_NETGRAPH,
1162 	.dom_name =		"netgraph",
1163 	.dom_nprotosw =		2,
1164 	.dom_protosw =		{ &ngcontrol_protosw, &ngdata_protosw },
1165 };
1166 
1167 /*
1168  * Handle loading and unloading for this node type.
1169  * This is to handle auxiliary linkages (e.g protocol domain addition).
1170  */
1171 static int
1172 ngs_mod_event(module_t mod, int event, void *data)
1173 {
1174 	int error = 0;
1175 
1176 	switch (event) {
1177 	case MOD_LOAD:
1178 		mtx_init(&ngsocketlist_mtx, "ng_socketlist", NULL, MTX_DEF);
1179 		break;
1180 	case MOD_UNLOAD:
1181 		/* Ensure there are no open netgraph sockets. */
1182 		if (!LIST_EMPTY(&ngsocklist)) {
1183 			error = EBUSY;
1184 			break;
1185 		}
1186 #ifdef NOTYET
1187 		/* Unregister protocol domain XXX can't do this yet.. */
1188 #endif
1189 		error = EBUSY;
1190 		break;
1191 	default:
1192 		error = EOPNOTSUPP;
1193 		break;
1194 	}
1195 	return (error);
1196 }
1197 
1198 DOMAIN_SET(ng);
1199 
1200 SYSCTL_INT(_net_graph, OID_AUTO, family, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, AF_NETGRAPH, "");
1201 static SYSCTL_NODE(_net_graph, OID_AUTO, data, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1202     "DATA");
1203 SYSCTL_INT(_net_graph_data, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_DATA, "");
1204 static SYSCTL_NODE(_net_graph, OID_AUTO, control, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1205     "CONTROL");
1206 SYSCTL_INT(_net_graph_control, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_CONTROL, "");
1207