xref: /freebsd/sys/netgraph/ng_socket.c (revision 9768746b)
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  *
40  * $FreeBSD$
41  * $Whistle: ng_socket.c,v 1.28 1999/11/01 09:24:52 julian Exp $
42  */
43 
44 /*
45  * Netgraph socket nodes
46  *
47  * There are two types of netgraph sockets, control and data.
48  * Control sockets have a netgraph node, but data sockets are
49  * parasitic on control sockets, and have no node of their own.
50  */
51 
52 #include <sys/param.h>
53 #include <sys/domain.h>
54 #include <sys/hash.h>
55 #include <sys/kernel.h>
56 #include <sys/linker.h>
57 #include <sys/lock.h>
58 #include <sys/malloc.h>
59 #include <sys/mbuf.h>
60 #include <sys/mutex.h>
61 #include <sys/proc.h>
62 #include <sys/epoch.h>
63 #include <sys/priv.h>
64 #include <sys/protosw.h>
65 #include <sys/queue.h>
66 #include <sys/socket.h>
67 #include <sys/socketvar.h>
68 #include <sys/syscallsubr.h>
69 #include <sys/sysctl.h>
70 
71 #include <net/vnet.h>
72 
73 #include <netgraph/ng_message.h>
74 #include <netgraph/netgraph.h>
75 #include <netgraph/ng_socketvar.h>
76 #include <netgraph/ng_socket.h>
77 
78 #ifdef NG_SEPARATE_MALLOC
79 static MALLOC_DEFINE(M_NETGRAPH_PATH, "netgraph_path", "netgraph path info");
80 static MALLOC_DEFINE(M_NETGRAPH_SOCK, "netgraph_sock", "netgraph socket info");
81 #else
82 #define M_NETGRAPH_PATH M_NETGRAPH
83 #define M_NETGRAPH_SOCK M_NETGRAPH
84 #endif
85 
86 /*
87  * It's Ascii-art time!
88  *   +-------------+   +-------------+
89  *   |socket  (ctl)|   |socket (data)|
90  *   +-------------+   +-------------+
91  *          ^                 ^
92  *          |                 |
93  *          v                 v
94  *    +-----------+     +-----------+
95  *    |pcb   (ctl)|     |pcb  (data)|
96  *    +-----------+     +-----------+
97  *          ^                 ^
98  *          |                 |
99  *          v                 v
100  *      +--------------------------+
101  *      |   Socket type private    |
102  *      |       data               |
103  *      +--------------------------+
104  *                   ^
105  *                   |
106  *                   v
107  *           +----------------+
108  *           | struct ng_node |
109  *           +----------------+
110  */
111 
112 /* Netgraph node methods */
113 static ng_constructor_t	ngs_constructor;
114 static ng_rcvmsg_t	ngs_rcvmsg;
115 static ng_shutdown_t	ngs_shutdown;
116 static ng_newhook_t	ngs_newhook;
117 static ng_connect_t	ngs_connect;
118 static ng_findhook_t	ngs_findhook;
119 static ng_rcvdata_t	ngs_rcvdata;
120 static ng_disconnect_t	ngs_disconnect;
121 
122 /* Internal methods */
123 static int	ng_attach_data(struct socket *so);
124 static int	ng_attach_cntl(struct socket *so);
125 static int	ng_attach_common(struct socket *so, int type);
126 static void	ng_detach_common(struct ngpcb *pcbp, int type);
127 static void	ng_socket_free_priv(struct ngsock *priv);
128 static int	ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp);
129 static int	ng_bind(struct sockaddr *nam, struct ngpcb *pcbp);
130 
131 static int	ngs_mod_event(module_t mod, int event, void *data);
132 static void	ng_socket_item_applied(void *context, int error);
133 
134 /* Netgraph type descriptor */
135 static struct ng_type typestruct = {
136 	.version =	NG_ABI_VERSION,
137 	.name =		NG_SOCKET_NODE_TYPE,
138 	.mod_event =	ngs_mod_event,
139 	.constructor =	ngs_constructor,
140 	.rcvmsg =	ngs_rcvmsg,
141 	.shutdown =	ngs_shutdown,
142 	.newhook =	ngs_newhook,
143 	.connect =	ngs_connect,
144 	.findhook =	ngs_findhook,
145 	.rcvdata =	ngs_rcvdata,
146 	.disconnect =	ngs_disconnect,
147 };
148 NETGRAPH_INIT_ORDERED(socket, &typestruct, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
149 
150 /* Buffer space */
151 static u_long ngpdg_sendspace = 20 * 1024;	/* really max datagram size */
152 SYSCTL_ULONG(_net_graph, OID_AUTO, maxdgram, CTLFLAG_RW,
153     &ngpdg_sendspace , 0, "Maximum outgoing Netgraph datagram size");
154 static u_long ngpdg_recvspace = 20 * 1024;
155 SYSCTL_ULONG(_net_graph, OID_AUTO, recvspace, CTLFLAG_RW,
156     &ngpdg_recvspace , 0, "Maximum space for incoming Netgraph datagrams");
157 
158 /* List of all sockets (for netstat -f netgraph) */
159 static LIST_HEAD(, ngpcb) ngsocklist;
160 
161 static struct mtx	ngsocketlist_mtx;
162 
163 #define sotongpcb(so) ((struct ngpcb *)(so)->so_pcb)
164 
165 /* If getting unexplained errors returned, set this to "kdb_enter("X"); */
166 #ifndef TRAP_ERROR
167 #define TRAP_ERROR
168 #endif
169 
170 struct hookpriv {
171 	LIST_ENTRY(hookpriv)	next;
172 	hook_p			hook;
173 };
174 LIST_HEAD(ngshash, hookpriv);
175 
176 /* Per-node private data */
177 struct ngsock {
178 	struct ng_node	*node;		/* the associated netgraph node */
179 	struct ngpcb	*datasock;	/* optional data socket */
180 	struct ngpcb	*ctlsock;	/* optional control socket */
181 	struct ngshash	*hash;		/* hash for hook names */
182 	u_long		hmask;		/* hash mask */
183 	int	flags;
184 	int	refs;
185 	struct mtx	mtx;		/* mtx to wait on */
186 	int		error;		/* place to store error */
187 };
188 
189 #define	NGS_FLAG_NOLINGER	1	/* close with last hook */
190 
191 /***************************************************************
192 	Control sockets
193 ***************************************************************/
194 
195 static int
196 ngc_attach(struct socket *so, int proto, struct thread *td)
197 {
198 	struct ngpcb *const pcbp = sotongpcb(so);
199 	int error;
200 
201 	error = priv_check(td, PRIV_NETGRAPH_CONTROL);
202 	if (error)
203 		return (error);
204 	if (pcbp != NULL)
205 		return (EISCONN);
206 	return (ng_attach_cntl(so));
207 }
208 
209 static void
210 ngc_detach(struct socket *so)
211 {
212 	struct ngpcb *const pcbp = sotongpcb(so);
213 
214 	KASSERT(pcbp != NULL, ("ngc_detach: pcbp == NULL"));
215 	ng_detach_common(pcbp, NG_CONTROL);
216 }
217 
218 static int
219 ngc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
220 	 struct mbuf *control, struct thread *td)
221 {
222 	struct ngpcb *const pcbp = sotongpcb(so);
223 	struct ngsock *const priv = NG_NODE_PRIVATE(pcbp->sockdata->node);
224 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) addr;
225 	struct ng_mesg *msg;
226 	struct mbuf *m0;
227 	item_p item;
228 	char *path = NULL;
229 	int len, error = 0;
230 	struct ng_apply_info apply;
231 
232 	if (control) {
233 		error = EINVAL;
234 		goto release;
235 	}
236 
237 	/* Require destination as there may be >= 1 hooks on this node. */
238 	if (addr == NULL) {
239 		error = EDESTADDRREQ;
240 		goto release;
241 	}
242 
243 	if (sap->sg_len > NG_NODESIZ + offsetof(struct sockaddr_ng, sg_data)) {
244 		error = EINVAL;
245 		goto release;
246 	}
247 
248 	/*
249 	 * Allocate an expendable buffer for the path, chop off
250 	 * the sockaddr header, and make sure it's NUL terminated.
251 	 */
252 	len = sap->sg_len - offsetof(struct sockaddr_ng, sg_data);
253 	path = malloc(len + 1, M_NETGRAPH_PATH, M_WAITOK);
254 	bcopy(sap->sg_data, path, len);
255 	path[len] = '\0';
256 
257 	/*
258 	 * Move the actual message out of mbufs into a linear buffer.
259 	 * Start by adding up the size of the data. (could use mh_len?)
260 	 */
261 	for (len = 0, m0 = m; m0 != NULL; m0 = m0->m_next)
262 		len += m0->m_len;
263 
264 	/*
265 	 * Move the data into a linear buffer as well.
266 	 * Messages are not delivered in mbufs.
267 	 */
268 	msg = malloc(len + 1, M_NETGRAPH_MSG, M_WAITOK);
269 	m_copydata(m, 0, len, (char *)msg);
270 
271 	if (msg->header.version != NG_VERSION) {
272 		free(msg, M_NETGRAPH_MSG);
273 		error = EINVAL;
274 		goto release;
275 	}
276 
277 	/*
278 	 * Hack alert!
279 	 * We look into the message and if it mkpeers a node of unknown type, we
280 	 * try to load it. We need to do this now, in syscall thread, because if
281 	 * message gets queued and applied later we will get panic.
282 	 */
283 	if (msg->header.typecookie == NGM_GENERIC_COOKIE &&
284 	    msg->header.cmd == NGM_MKPEER) {
285 		struct ngm_mkpeer *const mkp = (struct ngm_mkpeer *) msg->data;
286 
287 		if (ng_findtype(mkp->type) == NULL) {
288 			char filename[NG_TYPESIZ + 3];
289 			int fileid;
290 
291 			/* Not found, try to load it as a loadable module. */
292 			snprintf(filename, sizeof(filename), "ng_%s",
293 			    mkp->type);
294 			error = kern_kldload(curthread, filename, &fileid);
295 			if (error != 0) {
296 				free(msg, M_NETGRAPH_MSG);
297 				goto release;
298 			}
299 
300 			/* See if type has been loaded successfully. */
301 			if (ng_findtype(mkp->type) == NULL) {
302 				free(msg, M_NETGRAPH_MSG);
303 				(void)kern_kldunload(curthread, fileid,
304 				    LINKER_UNLOAD_NORMAL);
305 				error =  ENXIO;
306 				goto release;
307 			}
308 		}
309 	}
310 
311 	item = ng_package_msg(msg, NG_WAITOK);
312 	if ((error = ng_address_path((pcbp->sockdata->node), item, path, 0))
313 	    != 0) {
314 #ifdef TRACE_MESSAGES
315 		printf("ng_address_path: errx=%d\n", error);
316 #endif
317 		goto release;
318 	}
319 
320 #ifdef TRACE_MESSAGES
321 	printf("[%x]:<---------[socket]: c=<%d>cmd=%x(%s) f=%x #%d (%s)\n",
322 		item->el_dest->nd_ID,
323 		msg->header.typecookie,
324 		msg->header.cmd,
325 		msg->header.cmdstr,
326 		msg->header.flags,
327 		msg->header.token,
328 		item->el_dest->nd_type->name);
329 #endif
330 	SAVE_LINE(item);
331 	/*
332 	 * We do not want to return from syscall until the item
333 	 * is processed by destination node. We register callback
334 	 * on the item, which will update priv->error when item
335 	 * was applied.
336 	 * If ng_snd_item() has queued item, we sleep until
337 	 * callback wakes us up.
338 	 */
339 	bzero(&apply, sizeof(apply));
340 	apply.apply = ng_socket_item_applied;
341 	apply.context = priv;
342 	item->apply = &apply;
343 	priv->error = -1;
344 
345 	error = ng_snd_item(item, 0);
346 
347 	mtx_lock(&priv->mtx);
348 	if (priv->error == -1)
349 		msleep(priv, &priv->mtx, 0, "ngsock", 0);
350 	mtx_unlock(&priv->mtx);
351 	KASSERT(priv->error != -1,
352 	    ("ng_socket: priv->error wasn't updated"));
353 	error = priv->error;
354 
355 release:
356 	if (path != NULL)
357 		free(path, M_NETGRAPH_PATH);
358 	if (control != NULL)
359 		m_freem(control);
360 	if (m != NULL)
361 		m_freem(m);
362 	return (error);
363 }
364 
365 static int
366 ngc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
367 {
368 	struct ngpcb *const pcbp = sotongpcb(so);
369 
370 	if (pcbp == NULL)
371 		return (EINVAL);
372 	return (ng_bind(nam, pcbp));
373 }
374 
375 static int
376 ngc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
377 {
378 	/*
379 	 * At this time refuse to do this.. it used to
380 	 * do something but it was undocumented and not used.
381 	 */
382 	printf("program tried to connect control socket to remote node\n");
383 	return (EINVAL);
384 }
385 
386 /***************************************************************
387 	Data sockets
388 ***************************************************************/
389 
390 static int
391 ngd_attach(struct socket *so, int proto, struct thread *td)
392 {
393 	struct ngpcb *const pcbp = sotongpcb(so);
394 
395 	if (pcbp != NULL)
396 		return (EISCONN);
397 	return (ng_attach_data(so));
398 }
399 
400 static void
401 ngd_detach(struct socket *so)
402 {
403 	struct ngpcb *const pcbp = sotongpcb(so);
404 
405 	KASSERT(pcbp != NULL, ("ngd_detach: pcbp == NULL"));
406 	ng_detach_common(pcbp, NG_DATA);
407 }
408 
409 static int
410 ngd_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
411 	 struct mbuf *control, struct thread *td)
412 {
413 	struct epoch_tracker et;
414 	struct ngpcb *const pcbp = sotongpcb(so);
415 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) addr;
416 	int	len, error;
417 	hook_p  hook = NULL;
418 	item_p	item;
419 	char	hookname[NG_HOOKSIZ];
420 
421 	if ((pcbp == NULL) || (control != NULL)) {
422 		error = EINVAL;
423 		goto release;
424 	}
425 	if (pcbp->sockdata == NULL) {
426 		error = ENOTCONN;
427 		goto release;
428 	}
429 
430 	if (sap == NULL) {
431 		len = 0;		/* Make compiler happy. */
432 	} else {
433 		if (sap->sg_len > NG_NODESIZ +
434 		    offsetof(struct sockaddr_ng, sg_data)) {
435 			error = EINVAL;
436 			goto release;
437 		}
438 		len = sap->sg_len - offsetof(struct sockaddr_ng, sg_data);
439 	}
440 
441 	/*
442 	 * If the user used any of these ways to not specify an address
443 	 * then handle specially.
444 	 */
445 	if ((sap == NULL) || (len <= 0) || (*sap->sg_data == '\0')) {
446 		if (NG_NODE_NUMHOOKS(pcbp->sockdata->node) != 1) {
447 			error = EDESTADDRREQ;
448 			goto release;
449 		}
450 		/*
451 		 * If exactly one hook exists, just use it.
452 		 * Special case to allow write(2) to work on an ng_socket.
453 		 */
454 		hook = LIST_FIRST(&pcbp->sockdata->node->nd_hooks);
455 	} else {
456 		if (len >= NG_HOOKSIZ) {
457 			error = EINVAL;
458 			goto release;
459 		}
460 
461 		/*
462 		 * chop off the sockaddr header, and make sure it's NUL
463 		 * terminated
464 		 */
465 		bcopy(sap->sg_data, hookname, len);
466 		hookname[len] = '\0';
467 
468 		/* Find the correct hook from 'hookname' */
469 		hook = ng_findhook(pcbp->sockdata->node, hookname);
470 		if (hook == NULL) {
471 			error = EHOSTUNREACH;
472 			goto release;
473 		}
474 	}
475 
476 	/* Send data. */
477 	item = ng_package_data(m, NG_WAITOK);
478 	m = NULL;
479 	NET_EPOCH_ENTER(et);
480 	NG_FWD_ITEM_HOOK(error, item, hook);
481 	NET_EPOCH_EXIT(et);
482 
483 release:
484 	if (control != NULL)
485 		m_freem(control);
486 	if (m != NULL)
487 		m_freem(m);
488 	return (error);
489 }
490 
491 static int
492 ngd_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
493 {
494 	struct ngpcb *const pcbp = sotongpcb(so);
495 
496 	if (pcbp == NULL)
497 		return (EINVAL);
498 	return (ng_connect_data(nam, pcbp));
499 }
500 
501 /*
502  * Used for both data and control sockets
503  */
504 static int
505 ng_getsockaddr(struct socket *so, struct sockaddr **addr)
506 {
507 	struct ngpcb *pcbp;
508 	struct sockaddr_ng *sg;
509 	int sg_len;
510 	int error = 0;
511 
512 	pcbp = sotongpcb(so);
513 	if ((pcbp == NULL) || (pcbp->sockdata == NULL))
514 		/* XXXGL: can this still happen? */
515 		return (EINVAL);
516 
517 	sg_len = sizeof(struct sockaddr_ng) + NG_NODESIZ -
518 	    sizeof(sg->sg_data);
519 	sg = malloc(sg_len, M_SONAME, M_WAITOK | M_ZERO);
520 
521 	mtx_lock(&pcbp->sockdata->mtx);
522 	if (pcbp->sockdata->node != NULL) {
523 		node_p node = pcbp->sockdata->node;
524 
525 		if (NG_NODE_HAS_NAME(node))
526 			bcopy(NG_NODE_NAME(node), sg->sg_data,
527 			    strlen(NG_NODE_NAME(node)));
528 		mtx_unlock(&pcbp->sockdata->mtx);
529 
530 		sg->sg_len = sg_len;
531 		sg->sg_family = AF_NETGRAPH;
532 		*addr = (struct sockaddr *)sg;
533 	} else {
534 		mtx_unlock(&pcbp->sockdata->mtx);
535 		free(sg, M_SONAME);
536 		error = EINVAL;
537 	}
538 
539 	return (error);
540 }
541 
542 /*
543  * Attach a socket to it's protocol specific partner.
544  * For a control socket, actually create a netgraph node and attach
545  * to it as well.
546  */
547 
548 static int
549 ng_attach_cntl(struct socket *so)
550 {
551 	struct ngsock *priv;
552 	struct ngpcb *pcbp;
553 	node_p node;
554 	int error;
555 
556 	/* Setup protocol control block */
557 	if ((error = ng_attach_common(so, NG_CONTROL)) != 0)
558 		return (error);
559 	pcbp = sotongpcb(so);
560 
561 	/* Make the generic node components */
562 	if ((error = ng_make_node_common(&typestruct, &node)) != 0) {
563 		ng_detach_common(pcbp, NG_CONTROL);
564 		return (error);
565 	}
566 
567 	/*
568 	 * Allocate node private info and hash. We start
569 	 * with 16 hash entries, however we may grow later
570 	 * in ngs_newhook(). We can't predict how much hooks
571 	 * does this node plan to have.
572 	 */
573 	priv = malloc(sizeof(*priv), M_NETGRAPH_SOCK, M_WAITOK | M_ZERO);
574 	priv->hash = hashinit(16, M_NETGRAPH_SOCK, &priv->hmask);
575 
576 	/* Initialize mutex. */
577 	mtx_init(&priv->mtx, "ng_socket", NULL, MTX_DEF);
578 
579 	/* Link the pcb the private data. */
580 	priv->ctlsock = pcbp;
581 	pcbp->sockdata = priv;
582 	priv->refs++;
583 	priv->node = node;
584 	pcbp->node_id = node->nd_ID;	/* hint for netstat(1) */
585 
586 	/* Link the node and the private data. */
587 	NG_NODE_SET_PRIVATE(priv->node, priv);
588 	NG_NODE_REF(priv->node);
589 	priv->refs++;
590 
591 	return (0);
592 }
593 
594 static int
595 ng_attach_data(struct socket *so)
596 {
597 	return (ng_attach_common(so, NG_DATA));
598 }
599 
600 /*
601  * Set up a socket protocol control block.
602  * This code is shared between control and data sockets.
603  */
604 static int
605 ng_attach_common(struct socket *so, int type)
606 {
607 	struct ngpcb *pcbp;
608 	int error;
609 
610 	/* Standard socket setup stuff. */
611 	error = soreserve(so, ngpdg_sendspace, ngpdg_recvspace);
612 	if (error)
613 		return (error);
614 
615 	/* Allocate the pcb. */
616 	pcbp = malloc(sizeof(struct ngpcb), M_PCB, M_WAITOK | M_ZERO);
617 	pcbp->type = type;
618 
619 	/* Link the pcb and the socket. */
620 	so->so_pcb = (caddr_t)pcbp;
621 	pcbp->ng_socket = so;
622 
623 	/* Add the socket to linked list */
624 	mtx_lock(&ngsocketlist_mtx);
625 	LIST_INSERT_HEAD(&ngsocklist, pcbp, socks);
626 	mtx_unlock(&ngsocketlist_mtx);
627 	return (0);
628 }
629 
630 /*
631  * Disassociate the socket from it's protocol specific
632  * partner. If it's attached to a node's private data structure,
633  * then unlink from that too. If we were the last socket attached to it,
634  * then shut down the entire node. Shared code for control and data sockets.
635  */
636 static void
637 ng_detach_common(struct ngpcb *pcbp, int which)
638 {
639 	struct ngsock *priv = pcbp->sockdata;
640 
641 	if (priv != NULL) {
642 		mtx_lock(&priv->mtx);
643 
644 		switch (which) {
645 		case NG_CONTROL:
646 			priv->ctlsock = NULL;
647 			break;
648 		case NG_DATA:
649 			priv->datasock = NULL;
650 			break;
651 		default:
652 			panic("%s", __func__);
653 		}
654 		pcbp->sockdata = NULL;
655 		pcbp->node_id = 0;
656 
657 		ng_socket_free_priv(priv);
658 	}
659 
660 	pcbp->ng_socket->so_pcb = NULL;
661 	mtx_lock(&ngsocketlist_mtx);
662 	LIST_REMOVE(pcbp, socks);
663 	mtx_unlock(&ngsocketlist_mtx);
664 	free(pcbp, M_PCB);
665 }
666 
667 /*
668  * Remove a reference from node private data.
669  */
670 static void
671 ng_socket_free_priv(struct ngsock *priv)
672 {
673 	mtx_assert(&priv->mtx, MA_OWNED);
674 
675 	priv->refs--;
676 
677 	if (priv->refs == 0) {
678 		mtx_destroy(&priv->mtx);
679 		hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
680 		free(priv, M_NETGRAPH_SOCK);
681 		return;
682 	}
683 
684 	if ((priv->refs == 1) && (priv->node != NULL)) {
685 		node_p node = priv->node;
686 
687 		priv->node = NULL;
688 		mtx_unlock(&priv->mtx);
689 		NG_NODE_UNREF(node);
690 		ng_rmnode_self(node);
691 	} else
692 		mtx_unlock(&priv->mtx);
693 }
694 
695 /*
696  * Connect the data socket to a named control socket node.
697  */
698 static int
699 ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp)
700 {
701 	struct sockaddr_ng *sap;
702 	node_p farnode;
703 	struct ngsock *priv;
704 	int error;
705 	item_p item;
706 
707 	/* If we are already connected, don't do it again. */
708 	if (pcbp->sockdata != NULL)
709 		return (EISCONN);
710 
711 	/*
712 	 * Find the target (victim) and check it doesn't already have
713 	 * a data socket. Also check it is a 'socket' type node.
714 	 * Use ng_package_data() and ng_address_path() to do this.
715 	 */
716 
717 	sap = (struct sockaddr_ng *) nam;
718 	/* The item will hold the node reference. */
719 	item = ng_package_data(NULL, NG_WAITOK);
720 
721 	if ((error = ng_address_path(NULL, item,  sap->sg_data, 0)))
722 		return (error); /* item is freed on failure */
723 
724 	/*
725 	 * Extract node from item and free item. Remember we now have
726 	 * a reference on the node. The item holds it for us.
727 	 * when we free the item we release the reference.
728 	 */
729 	farnode = item->el_dest; /* shortcut */
730 	if (strcmp(farnode->nd_type->name, NG_SOCKET_NODE_TYPE) != 0) {
731 		NG_FREE_ITEM(item); /* drop the reference to the node */
732 		return (EINVAL);
733 	}
734 	priv = NG_NODE_PRIVATE(farnode);
735 	if (priv->datasock != NULL) {
736 		NG_FREE_ITEM(item);	/* drop the reference to the node */
737 		return (EADDRINUSE);
738 	}
739 
740 	/*
741 	 * Link the PCB and the private data struct. and note the extra
742 	 * reference. Drop the extra reference on the node.
743 	 */
744 	mtx_lock(&priv->mtx);
745 	priv->datasock = pcbp;
746 	pcbp->sockdata = priv;
747 	pcbp->node_id = priv->node->nd_ID;	/* hint for netstat(1) */
748 	priv->refs++;
749 	mtx_unlock(&priv->mtx);
750 	NG_FREE_ITEM(item);	/* drop the reference to the node */
751 	return (0);
752 }
753 
754 /*
755  * Binding a socket means giving the corresponding node a name
756  */
757 static int
758 ng_bind(struct sockaddr *nam, struct ngpcb *pcbp)
759 {
760 	struct ngsock *const priv = pcbp->sockdata;
761 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) nam;
762 
763 	if (priv == NULL) {
764 		TRAP_ERROR;
765 		return (EINVAL);
766 	}
767 	if ((sap->sg_len < 4) || (sap->sg_len > (NG_NODESIZ + 2)) ||
768 	    (sap->sg_data[0] == '\0') ||
769 	    (sap->sg_data[sap->sg_len - 3] != '\0')) {
770 		TRAP_ERROR;
771 		return (EINVAL);
772 	}
773 	return (ng_name_node(priv->node, sap->sg_data));
774 }
775 
776 /***************************************************************
777 	Netgraph node
778 ***************************************************************/
779 
780 /*
781  * You can only create new nodes from the socket end of things.
782  */
783 static int
784 ngs_constructor(node_p nodep)
785 {
786 	return (EINVAL);
787 }
788 
789 static void
790 ngs_rehash(node_p node)
791 {
792 	struct ngsock *priv = NG_NODE_PRIVATE(node);
793 	struct ngshash *new;
794 	struct hookpriv *hp;
795 	hook_p hook;
796 	uint32_t h;
797 	u_long hmask;
798 
799 	new = hashinit_flags((priv->hmask + 1) * 2, M_NETGRAPH_SOCK, &hmask,
800 	    HASH_NOWAIT);
801 	if (new == NULL)
802 		return;
803 
804 	LIST_FOREACH(hook, &node->nd_hooks, hk_hooks) {
805 		hp = NG_HOOK_PRIVATE(hook);
806 #ifdef INVARIANTS
807 		LIST_REMOVE(hp, next);
808 #endif
809 		h = hash32_str(NG_HOOK_NAME(hook), HASHINIT) & hmask;
810 		LIST_INSERT_HEAD(&new[h], hp, next);
811 	}
812 
813 	hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
814 	priv->hash = new;
815 	priv->hmask = hmask;
816 }
817 
818 /*
819  * We allow any hook to be connected to the node.
820  * There is no per-hook private information though.
821  */
822 static int
823 ngs_newhook(node_p node, hook_p hook, const char *name)
824 {
825 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
826 	struct hookpriv *hp;
827 	uint32_t h;
828 
829 	hp = malloc(sizeof(*hp), M_NETGRAPH_SOCK, M_NOWAIT);
830 	if (hp == NULL)
831 		return (ENOMEM);
832 	if (node->nd_numhooks * 2 > priv->hmask)
833 		ngs_rehash(node);
834 	hp->hook = hook;
835 	h = hash32_str(name, HASHINIT) & priv->hmask;
836 	LIST_INSERT_HEAD(&priv->hash[h], hp, next);
837 	NG_HOOK_SET_PRIVATE(hook, hp);
838 
839 	return (0);
840 }
841 
842 /*
843  * If only one hook, allow read(2) and write(2) to work.
844  */
845 static int
846 ngs_connect(hook_p hook)
847 {
848 	node_p node = NG_HOOK_NODE(hook);
849 	struct ngsock *priv = NG_NODE_PRIVATE(node);
850 
851 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
852 		if (NG_NODE_NUMHOOKS(node) == 1)
853 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
854 		else
855 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
856 	}
857 	return (0);
858 }
859 
860 /* Look up hook by name */
861 static hook_p
862 ngs_findhook(node_p node, const char *name)
863 {
864 	struct ngsock *priv = NG_NODE_PRIVATE(node);
865 	struct hookpriv *hp;
866 	uint32_t h;
867 
868 	/*
869 	 * Microoptimisation for an ng_socket with
870 	 * a single hook, which is a common case.
871 	 */
872 	if (node->nd_numhooks == 1) {
873 		hook_p hook;
874 
875 		hook = LIST_FIRST(&node->nd_hooks);
876 
877 		if (strcmp(NG_HOOK_NAME(hook), name) == 0)
878 			return (hook);
879 		else
880 			return (NULL);
881 	}
882 
883 	h = hash32_str(name, HASHINIT) & priv->hmask;
884 
885 	LIST_FOREACH(hp, &priv->hash[h], next)
886 		if (strcmp(NG_HOOK_NAME(hp->hook), name) == 0)
887 			return (hp->hook);
888 
889 	return (NULL);
890 }
891 
892 /*
893  * Incoming messages get passed up to the control socket.
894  * Unless they are for us specifically (socket_type)
895  */
896 static int
897 ngs_rcvmsg(node_p node, item_p item, hook_p lasthook)
898 {
899 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
900 	struct ngpcb *pcbp;
901 	struct socket *so;
902 	struct sockaddr_ng addr;
903 	struct ng_mesg *msg;
904 	struct mbuf *m;
905 	ng_ID_t	retaddr = NGI_RETADDR(item);
906 	int addrlen;
907 	int error = 0;
908 
909 	NGI_GET_MSG(item, msg);
910 	NG_FREE_ITEM(item);
911 
912 	/*
913 	 * Grab priv->mtx here to prevent destroying of control socket
914 	 * after checking that priv->ctlsock is not NULL.
915 	 */
916 	mtx_lock(&priv->mtx);
917 	pcbp = priv->ctlsock;
918 
919 	/*
920 	 * Only allow mesgs to be passed if we have the control socket.
921 	 * Data sockets can only support the generic messages.
922 	 */
923 	if (pcbp == NULL) {
924 		mtx_unlock(&priv->mtx);
925 		TRAP_ERROR;
926 		NG_FREE_MSG(msg);
927 		return (EINVAL);
928 	}
929 	so = pcbp->ng_socket;
930 	SOCKBUF_LOCK(&so->so_rcv);
931 
932 	/* As long as the race is handled, priv->mtx may be unlocked now. */
933 	mtx_unlock(&priv->mtx);
934 
935 #ifdef TRACE_MESSAGES
936 	printf("[%x]:---------->[socket]: c=<%d>cmd=%x(%s) f=%x #%d\n",
937 		retaddr,
938 		msg->header.typecookie,
939 		msg->header.cmd,
940 		msg->header.cmdstr,
941 		msg->header.flags,
942 		msg->header.token);
943 #endif
944 
945 	if (msg->header.typecookie == NGM_SOCKET_COOKIE) {
946 		switch (msg->header.cmd) {
947 		case NGM_SOCK_CMD_NOLINGER:
948 			priv->flags |= NGS_FLAG_NOLINGER;
949 			break;
950 		case NGM_SOCK_CMD_LINGER:
951 			priv->flags &= ~NGS_FLAG_NOLINGER;
952 			break;
953 		default:
954 			error = EINVAL;		/* unknown command */
955 		}
956 		SOCKBUF_UNLOCK(&so->so_rcv);
957 
958 		/* Free the message and return. */
959 		NG_FREE_MSG(msg);
960 		return (error);
961 	}
962 
963 	/* Get the return address into a sockaddr. */
964 	bzero(&addr, sizeof(addr));
965 	addr.sg_len = sizeof(addr);
966 	addr.sg_family = AF_NETGRAPH;
967 	addrlen = snprintf((char *)&addr.sg_data, sizeof(addr.sg_data),
968 	    "[%x]:", retaddr);
969 	if (addrlen < 0 || addrlen > sizeof(addr.sg_data)) {
970 		SOCKBUF_UNLOCK(&so->so_rcv);
971 		printf("%s: snprintf([%x]) failed - %d\n", __func__, retaddr,
972 		    addrlen);
973 		NG_FREE_MSG(msg);
974 		return (EINVAL);
975 	}
976 
977 	/* Copy the message itself into an mbuf chain. */
978 	m = m_devget((caddr_t)msg, sizeof(struct ng_mesg) + msg->header.arglen,
979 	    0, NULL, NULL);
980 
981 	/*
982 	 * Here we free the message. We need to do that
983 	 * regardless of whether we got mbufs.
984 	 */
985 	NG_FREE_MSG(msg);
986 
987 	if (m == NULL) {
988 		SOCKBUF_UNLOCK(&so->so_rcv);
989 		TRAP_ERROR;
990 		return (ENOBUFS);
991 	}
992 
993 	/* Send it up to the socket. */
994 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&addr, m,
995 	    NULL) == 0) {
996 		soroverflow_locked(so);
997 		TRAP_ERROR;
998 		m_freem(m);
999 		return (ENOBUFS);
1000 	}
1001 
1002 	/* sorwakeup_locked () releases the lock internally. */
1003 	sorwakeup_locked(so);
1004 
1005 	return (error);
1006 }
1007 
1008 /*
1009  * Receive data on a hook
1010  */
1011 static int
1012 ngs_rcvdata(hook_p hook, item_p item)
1013 {
1014 	struct ngsock *const priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
1015 	struct ngpcb *const pcbp = priv->datasock;
1016 	struct socket *so;
1017 	struct sockaddr_ng *addr;
1018 	char *addrbuf[NG_HOOKSIZ + 4];
1019 	int addrlen;
1020 	struct mbuf *m;
1021 
1022 	NGI_GET_M(item, m);
1023 	NG_FREE_ITEM(item);
1024 
1025 	/* If there is no data socket, black-hole it. */
1026 	if (pcbp == NULL) {
1027 		NG_FREE_M(m);
1028 		return (0);
1029 	}
1030 	so = pcbp->ng_socket;
1031 
1032 	/* Get the return address into a sockaddr. */
1033 	addrlen = strlen(NG_HOOK_NAME(hook));	/* <= NG_HOOKSIZ - 1 */
1034 	addr = (struct sockaddr_ng *) addrbuf;
1035 	addr->sg_len = addrlen + 3;
1036 	addr->sg_family = AF_NETGRAPH;
1037 	bcopy(NG_HOOK_NAME(hook), addr->sg_data, addrlen);
1038 	addr->sg_data[addrlen] = '\0';
1039 
1040 	/* Try to tell the socket which hook it came in on. */
1041 	SOCKBUF_LOCK(&so->so_rcv);
1042 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)addr, m,
1043 	    NULL) == 0) {
1044 		SOCKBUF_UNLOCK(&so->so_rcv);
1045 		m_freem(m);
1046 		TRAP_ERROR;
1047 		return (ENOBUFS);
1048 	}
1049 
1050 	/* sorwakeup_locked () releases the lock internally. */
1051 	sorwakeup_locked(so);
1052 	return (0);
1053 }
1054 
1055 /*
1056  * Hook disconnection
1057  *
1058  * For this type, removal of the last link destroys the node
1059  * if the NOLINGER flag is set.
1060  */
1061 static int
1062 ngs_disconnect(hook_p hook)
1063 {
1064 	node_p node = NG_HOOK_NODE(hook);
1065 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1066 	struct hookpriv *hp = NG_HOOK_PRIVATE(hook);
1067 
1068 	LIST_REMOVE(hp, next);
1069 	free(hp, M_NETGRAPH_SOCK);
1070 
1071 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
1072 		if (NG_NODE_NUMHOOKS(node) == 1)
1073 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
1074 		else
1075 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
1076 	}
1077 
1078 	if ((priv->flags & NGS_FLAG_NOLINGER) &&
1079 	    (NG_NODE_NUMHOOKS(node) == 0) && (NG_NODE_IS_VALID(node)))
1080 		ng_rmnode_self(node);
1081 
1082 	return (0);
1083 }
1084 
1085 /*
1086  * Do local shutdown processing.
1087  * In this case, that involves making sure the socket
1088  * knows we should be shutting down.
1089  */
1090 static int
1091 ngs_shutdown(node_p node)
1092 {
1093 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1094 	struct ngpcb *dpcbp, *pcbp;
1095 
1096 	mtx_lock(&priv->mtx);
1097 	dpcbp = priv->datasock;
1098 	pcbp = priv->ctlsock;
1099 
1100 	if (dpcbp != NULL)
1101 		soisdisconnected(dpcbp->ng_socket);
1102 
1103 	if (pcbp != NULL)
1104 		soisdisconnected(pcbp->ng_socket);
1105 
1106 	priv->node = NULL;
1107 	NG_NODE_SET_PRIVATE(node, NULL);
1108 	ng_socket_free_priv(priv);
1109 
1110 	NG_NODE_UNREF(node);
1111 	return (0);
1112 }
1113 
1114 static void
1115 ng_socket_item_applied(void *context, int error)
1116 {
1117 	struct ngsock *const priv = (struct ngsock *)context;
1118 
1119 	mtx_lock(&priv->mtx);
1120 	priv->error = error;
1121 	wakeup(priv);
1122 	mtx_unlock(&priv->mtx);
1123 
1124 }
1125 
1126 static	int
1127 dummy_disconnect(struct socket *so)
1128 {
1129 	return (0);
1130 }
1131 
1132 /*
1133  * Definitions of protocols supported in the NETGRAPH domain.
1134  * Control and data socket type descriptors
1135  *
1136  * XXXRW: Perhaps _close should do something?
1137  */
1138 static struct protosw ngcontrol_protosw = {
1139 	.pr_type =		SOCK_DGRAM,
1140 	.pr_protocol =		NG_CONTROL,
1141 	.pr_flags =		PR_ATOMIC | PR_ADDR /* | PR_RIGHTS */,
1142 	.pr_attach =		ngc_attach,
1143 	.pr_bind =		ngc_bind,
1144 	.pr_connect =		ngc_connect,
1145 	.pr_detach =		ngc_detach,
1146 	.pr_disconnect =	dummy_disconnect,
1147 	.pr_send =		ngc_send,
1148 	.pr_sockaddr =		ng_getsockaddr,
1149 };
1150 static struct protosw ngdata_protosw = {
1151 	.pr_type =		SOCK_DGRAM,
1152 	.pr_protocol =		NG_DATA,
1153 	.pr_flags =		PR_ATOMIC | PR_ADDR,
1154 	.pr_attach =		ngd_attach,
1155 	.pr_connect =		ngd_connect,
1156 	.pr_detach =		ngd_detach,
1157 	.pr_disconnect =	dummy_disconnect,
1158 	.pr_send =		ngd_send,
1159 	.pr_sockaddr =		ng_getsockaddr,
1160 };
1161 
1162 static struct domain ngdomain = {
1163 	.dom_family =		AF_NETGRAPH,
1164 	.dom_name =		"netgraph",
1165 	.dom_nprotosw =		2,
1166 	.dom_protosw =		{ &ngcontrol_protosw, &ngdata_protosw },
1167 };
1168 
1169 /*
1170  * Handle loading and unloading for this node type.
1171  * This is to handle auxiliary linkages (e.g protocol domain addition).
1172  */
1173 static int
1174 ngs_mod_event(module_t mod, int event, void *data)
1175 {
1176 	int error = 0;
1177 
1178 	switch (event) {
1179 	case MOD_LOAD:
1180 		mtx_init(&ngsocketlist_mtx, "ng_socketlist", NULL, MTX_DEF);
1181 		break;
1182 	case MOD_UNLOAD:
1183 		/* Ensure there are no open netgraph sockets. */
1184 		if (!LIST_EMPTY(&ngsocklist)) {
1185 			error = EBUSY;
1186 			break;
1187 		}
1188 #ifdef NOTYET
1189 		/* Unregister protocol domain XXX can't do this yet.. */
1190 #endif
1191 		error = EBUSY;
1192 		break;
1193 	default:
1194 		error = EOPNOTSUPP;
1195 		break;
1196 	}
1197 	return (error);
1198 }
1199 
1200 DOMAIN_SET(ng);
1201 
1202 SYSCTL_INT(_net_graph, OID_AUTO, family, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, AF_NETGRAPH, "");
1203 static SYSCTL_NODE(_net_graph, OID_AUTO, data, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1204     "DATA");
1205 SYSCTL_INT(_net_graph_data, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_DATA, "");
1206 static SYSCTL_NODE(_net_graph, OID_AUTO, control, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1207     "CONTROL");
1208 SYSCTL_INT(_net_graph_control, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_CONTROL, "");
1209