xref: /freebsd/sys/netgraph/ng_socket.c (revision e0c4386e)
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 *sa)
504 {
505 	struct sockaddr_ng *sg = (struct sockaddr_ng *)sa;
506 	struct ngpcb *pcbp;
507 	int error = 0;
508 
509 	pcbp = sotongpcb(so);
510 	if ((pcbp == NULL) || (pcbp->sockdata == NULL))
511 		/* XXXGL: can this still happen? */
512 		return (EINVAL);
513 
514 	*sg = (struct sockaddr_ng ){
515 		.sg_len = sizeof(struct sockaddr_ng),
516 		.sg_family = AF_NETGRAPH,
517 	};
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 		else
527 			snprintf(sg->sg_data, sizeof(sg->sg_data), "[%x]",
528 			    ng_node2ID(node));
529 	} else
530 		error = EINVAL;
531 	mtx_unlock(&pcbp->sockdata->mtx);
532 
533 	return (error);
534 }
535 
536 /*
537  * Attach a socket to it's protocol specific partner.
538  * For a control socket, actually create a netgraph node and attach
539  * to it as well.
540  */
541 
542 static int
543 ng_attach_cntl(struct socket *so)
544 {
545 	struct ngsock *priv;
546 	struct ngpcb *pcbp;
547 	node_p node;
548 	int error;
549 
550 	/* Setup protocol control block */
551 	if ((error = ng_attach_common(so, NG_CONTROL)) != 0)
552 		return (error);
553 	pcbp = sotongpcb(so);
554 
555 	/* Make the generic node components */
556 	if ((error = ng_make_node_common(&typestruct, &node)) != 0) {
557 		ng_detach_common(pcbp, NG_CONTROL);
558 		return (error);
559 	}
560 
561 	/*
562 	 * Allocate node private info and hash. We start
563 	 * with 16 hash entries, however we may grow later
564 	 * in ngs_newhook(). We can't predict how much hooks
565 	 * does this node plan to have.
566 	 */
567 	priv = malloc(sizeof(*priv), M_NETGRAPH_SOCK, M_WAITOK | M_ZERO);
568 	priv->hash = hashinit(16, M_NETGRAPH_SOCK, &priv->hmask);
569 
570 	/* Initialize mutex. */
571 	mtx_init(&priv->mtx, "ng_socket", NULL, MTX_DEF);
572 
573 	/* Link the pcb the private data. */
574 	priv->ctlsock = pcbp;
575 	pcbp->sockdata = priv;
576 	priv->refs++;
577 	priv->node = node;
578 	pcbp->node_id = node->nd_ID;	/* hint for netstat(1) */
579 
580 	/* Link the node and the private data. */
581 	NG_NODE_SET_PRIVATE(priv->node, priv);
582 	NG_NODE_REF(priv->node);
583 	priv->refs++;
584 
585 	return (0);
586 }
587 
588 static int
589 ng_attach_data(struct socket *so)
590 {
591 	return (ng_attach_common(so, NG_DATA));
592 }
593 
594 /*
595  * Set up a socket protocol control block.
596  * This code is shared between control and data sockets.
597  */
598 static int
599 ng_attach_common(struct socket *so, int type)
600 {
601 	struct ngpcb *pcbp;
602 	int error;
603 
604 	/* Standard socket setup stuff. */
605 	error = soreserve(so, ngpdg_sendspace, ngpdg_recvspace);
606 	if (error)
607 		return (error);
608 
609 	/* Allocate the pcb. */
610 	pcbp = malloc(sizeof(struct ngpcb), M_PCB, M_WAITOK | M_ZERO);
611 	pcbp->type = type;
612 
613 	/* Link the pcb and the socket. */
614 	so->so_pcb = (caddr_t)pcbp;
615 	pcbp->ng_socket = so;
616 
617 	/* Add the socket to linked list */
618 	mtx_lock(&ngsocketlist_mtx);
619 	LIST_INSERT_HEAD(&ngsocklist, pcbp, socks);
620 	mtx_unlock(&ngsocketlist_mtx);
621 	return (0);
622 }
623 
624 /*
625  * Disassociate the socket from it's protocol specific
626  * partner. If it's attached to a node's private data structure,
627  * then unlink from that too. If we were the last socket attached to it,
628  * then shut down the entire node. Shared code for control and data sockets.
629  */
630 static void
631 ng_detach_common(struct ngpcb *pcbp, int which)
632 {
633 	struct ngsock *priv = pcbp->sockdata;
634 
635 	if (priv != NULL) {
636 		mtx_lock(&priv->mtx);
637 
638 		switch (which) {
639 		case NG_CONTROL:
640 			priv->ctlsock = NULL;
641 			break;
642 		case NG_DATA:
643 			priv->datasock = NULL;
644 			break;
645 		default:
646 			panic("%s", __func__);
647 		}
648 		pcbp->sockdata = NULL;
649 		pcbp->node_id = 0;
650 
651 		ng_socket_free_priv(priv);
652 	}
653 
654 	pcbp->ng_socket->so_pcb = NULL;
655 	mtx_lock(&ngsocketlist_mtx);
656 	LIST_REMOVE(pcbp, socks);
657 	mtx_unlock(&ngsocketlist_mtx);
658 	free(pcbp, M_PCB);
659 }
660 
661 /*
662  * Remove a reference from node private data.
663  */
664 static void
665 ng_socket_free_priv(struct ngsock *priv)
666 {
667 	mtx_assert(&priv->mtx, MA_OWNED);
668 
669 	priv->refs--;
670 
671 	if (priv->refs == 0) {
672 		mtx_destroy(&priv->mtx);
673 		hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
674 		free(priv, M_NETGRAPH_SOCK);
675 		return;
676 	}
677 
678 	if ((priv->refs == 1) && (priv->node != NULL)) {
679 		node_p node = priv->node;
680 
681 		priv->node = NULL;
682 		mtx_unlock(&priv->mtx);
683 		NG_NODE_UNREF(node);
684 		ng_rmnode_self(node);
685 	} else
686 		mtx_unlock(&priv->mtx);
687 }
688 
689 /*
690  * Connect the data socket to a named control socket node.
691  */
692 static int
693 ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp)
694 {
695 	struct sockaddr_ng *sap;
696 	node_p farnode;
697 	struct ngsock *priv;
698 	int error;
699 	item_p item;
700 
701 	/* If we are already connected, don't do it again. */
702 	if (pcbp->sockdata != NULL)
703 		return (EISCONN);
704 
705 	/*
706 	 * Find the target (victim) and check it doesn't already have
707 	 * a data socket. Also check it is a 'socket' type node.
708 	 * Use ng_package_data() and ng_address_path() to do this.
709 	 */
710 
711 	sap = (struct sockaddr_ng *) nam;
712 	/* The item will hold the node reference. */
713 	item = ng_package_data(NULL, NG_WAITOK);
714 
715 	if ((error = ng_address_path(NULL, item,  sap->sg_data, 0)))
716 		return (error); /* item is freed on failure */
717 
718 	/*
719 	 * Extract node from item and free item. Remember we now have
720 	 * a reference on the node. The item holds it for us.
721 	 * when we free the item we release the reference.
722 	 */
723 	farnode = item->el_dest; /* shortcut */
724 	if (strcmp(farnode->nd_type->name, NG_SOCKET_NODE_TYPE) != 0) {
725 		NG_FREE_ITEM(item); /* drop the reference to the node */
726 		return (EINVAL);
727 	}
728 	priv = NG_NODE_PRIVATE(farnode);
729 	if (priv->datasock != NULL) {
730 		NG_FREE_ITEM(item);	/* drop the reference to the node */
731 		return (EADDRINUSE);
732 	}
733 
734 	/*
735 	 * Link the PCB and the private data struct. and note the extra
736 	 * reference. Drop the extra reference on the node.
737 	 */
738 	mtx_lock(&priv->mtx);
739 	priv->datasock = pcbp;
740 	pcbp->sockdata = priv;
741 	pcbp->node_id = priv->node->nd_ID;	/* hint for netstat(1) */
742 	priv->refs++;
743 	mtx_unlock(&priv->mtx);
744 	NG_FREE_ITEM(item);	/* drop the reference to the node */
745 	return (0);
746 }
747 
748 /*
749  * Binding a socket means giving the corresponding node a name
750  */
751 static int
752 ng_bind(struct sockaddr *nam, struct ngpcb *pcbp)
753 {
754 	struct ngsock *const priv = pcbp->sockdata;
755 	struct sockaddr_ng *const sap = (struct sockaddr_ng *) nam;
756 
757 	if (priv == NULL) {
758 		TRAP_ERROR;
759 		return (EINVAL);
760 	}
761 	if ((sap->sg_len < 4) || (sap->sg_len > (NG_NODESIZ + 2)) ||
762 	    (sap->sg_data[0] == '\0') ||
763 	    (sap->sg_data[sap->sg_len - 3] != '\0')) {
764 		TRAP_ERROR;
765 		return (EINVAL);
766 	}
767 	return (ng_name_node(priv->node, sap->sg_data));
768 }
769 
770 /***************************************************************
771 	Netgraph node
772 ***************************************************************/
773 
774 /*
775  * You can only create new nodes from the socket end of things.
776  */
777 static int
778 ngs_constructor(node_p nodep)
779 {
780 	return (EINVAL);
781 }
782 
783 static void
784 ngs_rehash(node_p node)
785 {
786 	struct ngsock *priv = NG_NODE_PRIVATE(node);
787 	struct ngshash *new;
788 	struct hookpriv *hp;
789 	hook_p hook;
790 	uint32_t h;
791 	u_long hmask;
792 
793 	new = hashinit_flags((priv->hmask + 1) * 2, M_NETGRAPH_SOCK, &hmask,
794 	    HASH_NOWAIT);
795 	if (new == NULL)
796 		return;
797 
798 	LIST_FOREACH(hook, &node->nd_hooks, hk_hooks) {
799 		hp = NG_HOOK_PRIVATE(hook);
800 #ifdef INVARIANTS
801 		LIST_REMOVE(hp, next);
802 #endif
803 		h = hash32_str(NG_HOOK_NAME(hook), HASHINIT) & hmask;
804 		LIST_INSERT_HEAD(&new[h], hp, next);
805 	}
806 
807 	hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
808 	priv->hash = new;
809 	priv->hmask = hmask;
810 }
811 
812 /*
813  * We allow any hook to be connected to the node.
814  * There is no per-hook private information though.
815  */
816 static int
817 ngs_newhook(node_p node, hook_p hook, const char *name)
818 {
819 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
820 	struct hookpriv *hp;
821 	uint32_t h;
822 
823 	hp = malloc(sizeof(*hp), M_NETGRAPH_SOCK, M_NOWAIT);
824 	if (hp == NULL)
825 		return (ENOMEM);
826 	if (node->nd_numhooks * 2 > priv->hmask)
827 		ngs_rehash(node);
828 	hp->hook = hook;
829 	h = hash32_str(name, HASHINIT) & priv->hmask;
830 	LIST_INSERT_HEAD(&priv->hash[h], hp, next);
831 	NG_HOOK_SET_PRIVATE(hook, hp);
832 
833 	return (0);
834 }
835 
836 /*
837  * If only one hook, allow read(2) and write(2) to work.
838  */
839 static int
840 ngs_connect(hook_p hook)
841 {
842 	node_p node = NG_HOOK_NODE(hook);
843 	struct ngsock *priv = NG_NODE_PRIVATE(node);
844 
845 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
846 		if (NG_NODE_NUMHOOKS(node) == 1)
847 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
848 		else
849 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
850 	}
851 	return (0);
852 }
853 
854 /* Look up hook by name */
855 static hook_p
856 ngs_findhook(node_p node, const char *name)
857 {
858 	struct ngsock *priv = NG_NODE_PRIVATE(node);
859 	struct hookpriv *hp;
860 	uint32_t h;
861 
862 	/*
863 	 * Microoptimisation for an ng_socket with
864 	 * a single hook, which is a common case.
865 	 */
866 	if (node->nd_numhooks == 1) {
867 		hook_p hook;
868 
869 		hook = LIST_FIRST(&node->nd_hooks);
870 
871 		if (strcmp(NG_HOOK_NAME(hook), name) == 0)
872 			return (hook);
873 		else
874 			return (NULL);
875 	}
876 
877 	h = hash32_str(name, HASHINIT) & priv->hmask;
878 
879 	LIST_FOREACH(hp, &priv->hash[h], next)
880 		if (strcmp(NG_HOOK_NAME(hp->hook), name) == 0)
881 			return (hp->hook);
882 
883 	return (NULL);
884 }
885 
886 /*
887  * Incoming messages get passed up to the control socket.
888  * Unless they are for us specifically (socket_type)
889  */
890 static int
891 ngs_rcvmsg(node_p node, item_p item, hook_p lasthook)
892 {
893 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
894 	struct ngpcb *pcbp;
895 	struct socket *so;
896 	struct sockaddr_ng addr;
897 	struct ng_mesg *msg;
898 	struct mbuf *m;
899 	ng_ID_t	retaddr = NGI_RETADDR(item);
900 	int addrlen;
901 	int error = 0;
902 
903 	NGI_GET_MSG(item, msg);
904 	NG_FREE_ITEM(item);
905 
906 	/*
907 	 * Grab priv->mtx here to prevent destroying of control socket
908 	 * after checking that priv->ctlsock is not NULL.
909 	 */
910 	mtx_lock(&priv->mtx);
911 	pcbp = priv->ctlsock;
912 
913 	/*
914 	 * Only allow mesgs to be passed if we have the control socket.
915 	 * Data sockets can only support the generic messages.
916 	 */
917 	if (pcbp == NULL) {
918 		mtx_unlock(&priv->mtx);
919 		TRAP_ERROR;
920 		NG_FREE_MSG(msg);
921 		return (EINVAL);
922 	}
923 	so = pcbp->ng_socket;
924 	SOCKBUF_LOCK(&so->so_rcv);
925 
926 	/* As long as the race is handled, priv->mtx may be unlocked now. */
927 	mtx_unlock(&priv->mtx);
928 
929 #ifdef TRACE_MESSAGES
930 	printf("[%x]:---------->[socket]: c=<%d>cmd=%x(%s) f=%x #%d\n",
931 		retaddr,
932 		msg->header.typecookie,
933 		msg->header.cmd,
934 		msg->header.cmdstr,
935 		msg->header.flags,
936 		msg->header.token);
937 #endif
938 
939 	if (msg->header.typecookie == NGM_SOCKET_COOKIE) {
940 		switch (msg->header.cmd) {
941 		case NGM_SOCK_CMD_NOLINGER:
942 			priv->flags |= NGS_FLAG_NOLINGER;
943 			break;
944 		case NGM_SOCK_CMD_LINGER:
945 			priv->flags &= ~NGS_FLAG_NOLINGER;
946 			break;
947 		default:
948 			error = EINVAL;		/* unknown command */
949 		}
950 		SOCKBUF_UNLOCK(&so->so_rcv);
951 
952 		/* Free the message and return. */
953 		NG_FREE_MSG(msg);
954 		return (error);
955 	}
956 
957 	/* Get the return address into a sockaddr. */
958 	bzero(&addr, sizeof(addr));
959 	addr.sg_len = sizeof(addr);
960 	addr.sg_family = AF_NETGRAPH;
961 	addrlen = snprintf((char *)&addr.sg_data, sizeof(addr.sg_data),
962 	    "[%x]:", retaddr);
963 	if (addrlen < 0 || addrlen > sizeof(addr.sg_data)) {
964 		SOCKBUF_UNLOCK(&so->so_rcv);
965 		printf("%s: snprintf([%x]) failed - %d\n", __func__, retaddr,
966 		    addrlen);
967 		NG_FREE_MSG(msg);
968 		return (EINVAL);
969 	}
970 
971 	/* Copy the message itself into an mbuf chain. */
972 	m = m_devget((caddr_t)msg, sizeof(struct ng_mesg) + msg->header.arglen,
973 	    0, NULL, NULL);
974 
975 	/*
976 	 * Here we free the message. We need to do that
977 	 * regardless of whether we got mbufs.
978 	 */
979 	NG_FREE_MSG(msg);
980 
981 	if (m == NULL) {
982 		SOCKBUF_UNLOCK(&so->so_rcv);
983 		TRAP_ERROR;
984 		return (ENOBUFS);
985 	}
986 
987 	/* Send it up to the socket. */
988 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&addr, m,
989 	    NULL) == 0) {
990 		soroverflow_locked(so);
991 		TRAP_ERROR;
992 		m_freem(m);
993 		return (ENOBUFS);
994 	}
995 
996 	/* sorwakeup_locked () releases the lock internally. */
997 	sorwakeup_locked(so);
998 
999 	return (error);
1000 }
1001 
1002 /*
1003  * Receive data on a hook
1004  */
1005 static int
1006 ngs_rcvdata(hook_p hook, item_p item)
1007 {
1008 	struct ngsock *const priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
1009 	struct ngpcb *const pcbp = priv->datasock;
1010 	struct socket *so;
1011 	struct sockaddr_ng *addr;
1012 	char *addrbuf[NG_HOOKSIZ + 4];
1013 	int addrlen;
1014 	struct mbuf *m;
1015 
1016 	NGI_GET_M(item, m);
1017 	NG_FREE_ITEM(item);
1018 
1019 	/* If there is no data socket, black-hole it. */
1020 	if (pcbp == NULL) {
1021 		NG_FREE_M(m);
1022 		return (0);
1023 	}
1024 	so = pcbp->ng_socket;
1025 
1026 	/* Get the return address into a sockaddr. */
1027 	addrlen = strlen(NG_HOOK_NAME(hook));	/* <= NG_HOOKSIZ - 1 */
1028 	addr = (struct sockaddr_ng *) addrbuf;
1029 	addr->sg_len = addrlen + 3;
1030 	addr->sg_family = AF_NETGRAPH;
1031 	bcopy(NG_HOOK_NAME(hook), addr->sg_data, addrlen);
1032 	addr->sg_data[addrlen] = '\0';
1033 
1034 	/* Try to tell the socket which hook it came in on. */
1035 	SOCKBUF_LOCK(&so->so_rcv);
1036 	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)addr, m,
1037 	    NULL) == 0) {
1038 		SOCKBUF_UNLOCK(&so->so_rcv);
1039 		m_freem(m);
1040 		TRAP_ERROR;
1041 		return (ENOBUFS);
1042 	}
1043 
1044 	/* sorwakeup_locked () releases the lock internally. */
1045 	sorwakeup_locked(so);
1046 	return (0);
1047 }
1048 
1049 /*
1050  * Hook disconnection
1051  *
1052  * For this type, removal of the last link destroys the node
1053  * if the NOLINGER flag is set.
1054  */
1055 static int
1056 ngs_disconnect(hook_p hook)
1057 {
1058 	node_p node = NG_HOOK_NODE(hook);
1059 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1060 	struct hookpriv *hp = NG_HOOK_PRIVATE(hook);
1061 
1062 	LIST_REMOVE(hp, next);
1063 	free(hp, M_NETGRAPH_SOCK);
1064 
1065 	if ((priv->datasock) && (priv->datasock->ng_socket)) {
1066 		if (NG_NODE_NUMHOOKS(node) == 1)
1067 			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
1068 		else
1069 			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
1070 	}
1071 
1072 	if ((priv->flags & NGS_FLAG_NOLINGER) &&
1073 	    (NG_NODE_NUMHOOKS(node) == 0) && (NG_NODE_IS_VALID(node)))
1074 		ng_rmnode_self(node);
1075 
1076 	return (0);
1077 }
1078 
1079 /*
1080  * Do local shutdown processing.
1081  * In this case, that involves making sure the socket
1082  * knows we should be shutting down.
1083  */
1084 static int
1085 ngs_shutdown(node_p node)
1086 {
1087 	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1088 	struct ngpcb *dpcbp, *pcbp;
1089 
1090 	mtx_lock(&priv->mtx);
1091 	dpcbp = priv->datasock;
1092 	pcbp = priv->ctlsock;
1093 
1094 	if (dpcbp != NULL)
1095 		soisdisconnected(dpcbp->ng_socket);
1096 
1097 	if (pcbp != NULL)
1098 		soisdisconnected(pcbp->ng_socket);
1099 
1100 	priv->node = NULL;
1101 	NG_NODE_SET_PRIVATE(node, NULL);
1102 	ng_socket_free_priv(priv);
1103 
1104 	NG_NODE_UNREF(node);
1105 	return (0);
1106 }
1107 
1108 static void
1109 ng_socket_item_applied(void *context, int error)
1110 {
1111 	struct ngsock *const priv = (struct ngsock *)context;
1112 
1113 	mtx_lock(&priv->mtx);
1114 	priv->error = error;
1115 	wakeup(priv);
1116 	mtx_unlock(&priv->mtx);
1117 
1118 }
1119 
1120 static	int
1121 dummy_disconnect(struct socket *so)
1122 {
1123 	return (0);
1124 }
1125 
1126 /*
1127  * Definitions of protocols supported in the NETGRAPH domain.
1128  * Control and data socket type descriptors
1129  *
1130  * XXXRW: Perhaps _close should do something?
1131  */
1132 static struct protosw ngcontrol_protosw = {
1133 	.pr_type =		SOCK_DGRAM,
1134 	.pr_protocol =		NG_CONTROL,
1135 	.pr_flags =		PR_ATOMIC | PR_ADDR /* | PR_RIGHTS */,
1136 	.pr_attach =		ngc_attach,
1137 	.pr_bind =		ngc_bind,
1138 	.pr_connect =		ngc_connect,
1139 	.pr_detach =		ngc_detach,
1140 	.pr_disconnect =	dummy_disconnect,
1141 	.pr_send =		ngc_send,
1142 	.pr_sockaddr =		ng_getsockaddr,
1143 };
1144 static struct protosw ngdata_protosw = {
1145 	.pr_type =		SOCK_DGRAM,
1146 	.pr_protocol =		NG_DATA,
1147 	.pr_flags =		PR_ATOMIC | PR_ADDR,
1148 	.pr_attach =		ngd_attach,
1149 	.pr_connect =		ngd_connect,
1150 	.pr_detach =		ngd_detach,
1151 	.pr_disconnect =	dummy_disconnect,
1152 	.pr_send =		ngd_send,
1153 	.pr_sockaddr =		ng_getsockaddr,
1154 };
1155 
1156 static struct domain ngdomain = {
1157 	.dom_family =		AF_NETGRAPH,
1158 	.dom_name =		"netgraph",
1159 	.dom_nprotosw =		2,
1160 	.dom_protosw =		{ &ngcontrol_protosw, &ngdata_protosw },
1161 };
1162 
1163 /*
1164  * Handle loading and unloading for this node type.
1165  * This is to handle auxiliary linkages (e.g protocol domain addition).
1166  */
1167 static int
1168 ngs_mod_event(module_t mod, int event, void *data)
1169 {
1170 	int error = 0;
1171 
1172 	switch (event) {
1173 	case MOD_LOAD:
1174 		mtx_init(&ngsocketlist_mtx, "ng_socketlist", NULL, MTX_DEF);
1175 		break;
1176 	case MOD_UNLOAD:
1177 		/* Ensure there are no open netgraph sockets. */
1178 		if (!LIST_EMPTY(&ngsocklist)) {
1179 			error = EBUSY;
1180 			break;
1181 		}
1182 #ifdef NOTYET
1183 		/* Unregister protocol domain XXX can't do this yet.. */
1184 #endif
1185 		error = EBUSY;
1186 		break;
1187 	default:
1188 		error = EOPNOTSUPP;
1189 		break;
1190 	}
1191 	return (error);
1192 }
1193 
1194 DOMAIN_SET(ng);
1195 
1196 SYSCTL_INT(_net_graph, OID_AUTO, family, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, AF_NETGRAPH, "");
1197 static SYSCTL_NODE(_net_graph, OID_AUTO, data, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1198     "DATA");
1199 SYSCTL_INT(_net_graph_data, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_DATA, "");
1200 static SYSCTL_NODE(_net_graph, OID_AUTO, control, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1201     "CONTROL");
1202 SYSCTL_INT(_net_graph_control, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_CONTROL, "");
1203