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
77static MALLOC_DEFINE(M_NETGRAPH_PATH, "netgraph_path", "netgraph path info");
78static 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 */
111static ng_constructor_t	ngs_constructor;
112static ng_rcvmsg_t	ngs_rcvmsg;
113static ng_shutdown_t	ngs_shutdown;
114static ng_newhook_t	ngs_newhook;
115static ng_connect_t	ngs_connect;
116static ng_findhook_t	ngs_findhook;
117static ng_rcvdata_t	ngs_rcvdata;
118static ng_disconnect_t	ngs_disconnect;
119
120/* Internal methods */
121static int	ng_attach_data(struct socket *so);
122static int	ng_attach_cntl(struct socket *so);
123static int	ng_attach_common(struct socket *so, int type);
124static void	ng_detach_common(struct ngpcb *pcbp, int type);
125static void	ng_socket_free_priv(struct ngsock *priv);
126static int	ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp);
127static int	ng_bind(struct sockaddr *nam, struct ngpcb *pcbp);
128
129static int	ngs_mod_event(module_t mod, int event, void *data);
130static void	ng_socket_item_applied(void *context, int error);
131
132/* Netgraph type descriptor */
133static 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};
146NETGRAPH_INIT_ORDERED(socket, &typestruct, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
147
148/* Buffer space */
149static u_long ngpdg_sendspace = 20 * 1024;	/* really max datagram size */
150SYSCTL_ULONG(_net_graph, OID_AUTO, maxdgram, CTLFLAG_RW,
151    &ngpdg_sendspace , 0, "Maximum outgoing Netgraph datagram size");
152static u_long ngpdg_recvspace = 20 * 1024;
153SYSCTL_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) */
157static LIST_HEAD(, ngpcb) ngsocklist;
158
159static 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
168struct hookpriv {
169	LIST_ENTRY(hookpriv)	next;
170	hook_p			hook;
171};
172LIST_HEAD(ngshash, hookpriv);
173
174/* Per-node private data */
175struct 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
193static int
194ngc_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
207static void
208ngc_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
216static int
217ngc_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			bool loaded;
289
290			/* Not found, try to load it as a loadable module. */
291			snprintf(filename, sizeof(filename), "ng_%s",
292			    mkp->type);
293			error = kern_kldload(curthread, filename, &fileid);
294			loaded = (error == 0);
295			if (error == EEXIST)
296				error = 0;
297			if (error != 0) {
298				free(msg, M_NETGRAPH_MSG);
299				goto release;
300			}
301
302			/* See if type has been loaded successfully. */
303			if (ng_findtype(mkp->type) == NULL) {
304				free(msg, M_NETGRAPH_MSG);
305				if (loaded)
306					(void)kern_kldunload(curthread, fileid,
307					    LINKER_UNLOAD_NORMAL);
308				error = ENXIO;
309				goto release;
310			}
311		}
312	}
313
314	item = ng_package_msg(msg, NG_WAITOK);
315	if ((error = ng_address_path((pcbp->sockdata->node), item, path, 0))
316	    != 0) {
317#ifdef TRACE_MESSAGES
318		printf("ng_address_path: errx=%d\n", error);
319#endif
320		goto release;
321	}
322
323#ifdef TRACE_MESSAGES
324	printf("[%x]:<---------[socket]: c=<%d>cmd=%x(%s) f=%x #%d (%s)\n",
325		item->el_dest->nd_ID,
326		msg->header.typecookie,
327		msg->header.cmd,
328		msg->header.cmdstr,
329		msg->header.flags,
330		msg->header.token,
331		item->el_dest->nd_type->name);
332#endif
333	SAVE_LINE(item);
334	/*
335	 * We do not want to return from syscall until the item
336	 * is processed by destination node. We register callback
337	 * on the item, which will update priv->error when item
338	 * was applied.
339	 * If ng_snd_item() has queued item, we sleep until
340	 * callback wakes us up.
341	 */
342	bzero(&apply, sizeof(apply));
343	apply.apply = ng_socket_item_applied;
344	apply.context = priv;
345	item->apply = &apply;
346	priv->error = -1;
347
348	error = ng_snd_item(item, 0);
349
350	mtx_lock(&priv->mtx);
351	if (priv->error == -1)
352		msleep(priv, &priv->mtx, 0, "ngsock", 0);
353	mtx_unlock(&priv->mtx);
354	KASSERT(priv->error != -1,
355	    ("ng_socket: priv->error wasn't updated"));
356	error = priv->error;
357
358release:
359	if (path != NULL)
360		free(path, M_NETGRAPH_PATH);
361	if (control != NULL)
362		m_freem(control);
363	if (m != NULL)
364		m_freem(m);
365	return (error);
366}
367
368static int
369ngc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
370{
371	struct ngpcb *const pcbp = sotongpcb(so);
372
373	if (pcbp == NULL)
374		return (EINVAL);
375	return (ng_bind(nam, pcbp));
376}
377
378static int
379ngc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
380{
381	/*
382	 * At this time refuse to do this.. it used to
383	 * do something but it was undocumented and not used.
384	 */
385	printf("program tried to connect control socket to remote node\n");
386	return (EINVAL);
387}
388
389/***************************************************************
390	Data sockets
391***************************************************************/
392
393static int
394ngd_attach(struct socket *so, int proto, struct thread *td)
395{
396	struct ngpcb *const pcbp = sotongpcb(so);
397
398	if (pcbp != NULL)
399		return (EISCONN);
400	return (ng_attach_data(so));
401}
402
403static void
404ngd_detach(struct socket *so)
405{
406	struct ngpcb *const pcbp = sotongpcb(so);
407
408	KASSERT(pcbp != NULL, ("ngd_detach: pcbp == NULL"));
409	ng_detach_common(pcbp, NG_DATA);
410}
411
412static int
413ngd_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *addr,
414	 struct mbuf *control, struct thread *td)
415{
416	struct epoch_tracker et;
417	struct ngpcb *const pcbp = sotongpcb(so);
418	struct sockaddr_ng *const sap = (struct sockaddr_ng *) addr;
419	int	len, error;
420	hook_p  hook = NULL;
421	item_p	item;
422	char	hookname[NG_HOOKSIZ];
423
424	if ((pcbp == NULL) || (control != NULL)) {
425		error = EINVAL;
426		goto release;
427	}
428	if (pcbp->sockdata == NULL) {
429		error = ENOTCONN;
430		goto release;
431	}
432
433	if (sap == NULL) {
434		len = 0;		/* Make compiler happy. */
435	} else {
436		if (sap->sg_len > NG_NODESIZ +
437		    offsetof(struct sockaddr_ng, sg_data)) {
438			error = EINVAL;
439			goto release;
440		}
441		len = sap->sg_len - offsetof(struct sockaddr_ng, sg_data);
442	}
443
444	/*
445	 * If the user used any of these ways to not specify an address
446	 * then handle specially.
447	 */
448	if ((sap == NULL) || (len <= 0) || (*sap->sg_data == '\0')) {
449		if (NG_NODE_NUMHOOKS(pcbp->sockdata->node) != 1) {
450			error = EDESTADDRREQ;
451			goto release;
452		}
453		/*
454		 * If exactly one hook exists, just use it.
455		 * Special case to allow write(2) to work on an ng_socket.
456		 */
457		hook = LIST_FIRST(&pcbp->sockdata->node->nd_hooks);
458	} else {
459		if (len >= NG_HOOKSIZ) {
460			error = EINVAL;
461			goto release;
462		}
463
464		/*
465		 * chop off the sockaddr header, and make sure it's NUL
466		 * terminated
467		 */
468		bcopy(sap->sg_data, hookname, len);
469		hookname[len] = '\0';
470
471		/* Find the correct hook from 'hookname' */
472		hook = ng_findhook(pcbp->sockdata->node, hookname);
473		if (hook == NULL) {
474			error = EHOSTUNREACH;
475			goto release;
476		}
477	}
478
479	/* Send data. */
480	item = ng_package_data(m, NG_WAITOK);
481	m = NULL;
482	NET_EPOCH_ENTER(et);
483	NG_FWD_ITEM_HOOK(error, item, hook);
484	NET_EPOCH_EXIT(et);
485
486release:
487	if (control != NULL)
488		m_freem(control);
489	if (m != NULL)
490		m_freem(m);
491	return (error);
492}
493
494static int
495ngd_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
496{
497	struct ngpcb *const pcbp = sotongpcb(so);
498
499	if (pcbp == NULL)
500		return (EINVAL);
501	return (ng_connect_data(nam, pcbp));
502}
503
504/*
505 * Used for both data and control sockets
506 */
507static int
508ng_getsockaddr(struct socket *so, struct sockaddr *sa)
509{
510	struct sockaddr_ng *sg = (struct sockaddr_ng *)sa;
511	struct ngpcb *pcbp;
512	int error = 0;
513
514	pcbp = sotongpcb(so);
515	if ((pcbp == NULL) || (pcbp->sockdata == NULL))
516		/* XXXGL: can this still happen? */
517		return (EINVAL);
518
519	*sg = (struct sockaddr_ng ){
520		.sg_len = sizeof(struct sockaddr_ng),
521		.sg_family = AF_NETGRAPH,
522	};
523
524	mtx_lock(&pcbp->sockdata->mtx);
525	if (pcbp->sockdata->node != NULL) {
526		node_p node = pcbp->sockdata->node;
527
528		if (NG_NODE_HAS_NAME(node))
529			bcopy(NG_NODE_NAME(node), sg->sg_data,
530			    strlen(NG_NODE_NAME(node)));
531		else
532			snprintf(sg->sg_data, sizeof(sg->sg_data), "[%x]",
533			    ng_node2ID(node));
534	} else
535		error = EINVAL;
536	mtx_unlock(&pcbp->sockdata->mtx);
537
538	return (error);
539}
540
541/*
542 * Attach a socket to it's protocol specific partner.
543 * For a control socket, actually create a netgraph node and attach
544 * to it as well.
545 */
546
547static int
548ng_attach_cntl(struct socket *so)
549{
550	struct ngsock *priv;
551	struct ngpcb *pcbp;
552	node_p node;
553	int error;
554
555	/* Setup protocol control block */
556	if ((error = ng_attach_common(so, NG_CONTROL)) != 0)
557		return (error);
558	pcbp = sotongpcb(so);
559
560	/* Make the generic node components */
561	if ((error = ng_make_node_common(&typestruct, &node)) != 0) {
562		ng_detach_common(pcbp, NG_CONTROL);
563		return (error);
564	}
565
566	/*
567	 * Allocate node private info and hash. We start
568	 * with 16 hash entries, however we may grow later
569	 * in ngs_newhook(). We can't predict how much hooks
570	 * does this node plan to have.
571	 */
572	priv = malloc(sizeof(*priv), M_NETGRAPH_SOCK, M_WAITOK | M_ZERO);
573	priv->hash = hashinit(16, M_NETGRAPH_SOCK, &priv->hmask);
574
575	/* Initialize mutex. */
576	mtx_init(&priv->mtx, "ng_socket", NULL, MTX_DEF);
577
578	/* Link the pcb the private data. */
579	priv->ctlsock = pcbp;
580	pcbp->sockdata = priv;
581	priv->refs++;
582	priv->node = node;
583	pcbp->node_id = node->nd_ID;	/* hint for netstat(1) */
584
585	/* Link the node and the private data. */
586	NG_NODE_SET_PRIVATE(priv->node, priv);
587	NG_NODE_REF(priv->node);
588	priv->refs++;
589
590	return (0);
591}
592
593static int
594ng_attach_data(struct socket *so)
595{
596	return (ng_attach_common(so, NG_DATA));
597}
598
599/*
600 * Set up a socket protocol control block.
601 * This code is shared between control and data sockets.
602 */
603static int
604ng_attach_common(struct socket *so, int type)
605{
606	struct ngpcb *pcbp;
607	int error;
608
609	/* Standard socket setup stuff. */
610	error = soreserve(so, ngpdg_sendspace, ngpdg_recvspace);
611	if (error)
612		return (error);
613
614	/* Allocate the pcb. */
615	pcbp = malloc(sizeof(struct ngpcb), M_PCB, M_WAITOK | M_ZERO);
616	pcbp->type = type;
617
618	/* Link the pcb and the socket. */
619	so->so_pcb = (caddr_t)pcbp;
620	pcbp->ng_socket = so;
621
622	/* Add the socket to linked list */
623	mtx_lock(&ngsocketlist_mtx);
624	LIST_INSERT_HEAD(&ngsocklist, pcbp, socks);
625	mtx_unlock(&ngsocketlist_mtx);
626	return (0);
627}
628
629/*
630 * Disassociate the socket from it's protocol specific
631 * partner. If it's attached to a node's private data structure,
632 * then unlink from that too. If we were the last socket attached to it,
633 * then shut down the entire node. Shared code for control and data sockets.
634 */
635static void
636ng_detach_common(struct ngpcb *pcbp, int which)
637{
638	struct ngsock *priv = pcbp->sockdata;
639
640	if (priv != NULL) {
641		mtx_lock(&priv->mtx);
642
643		switch (which) {
644		case NG_CONTROL:
645			priv->ctlsock = NULL;
646			break;
647		case NG_DATA:
648			priv->datasock = NULL;
649			break;
650		default:
651			panic("%s", __func__);
652		}
653		pcbp->sockdata = NULL;
654		pcbp->node_id = 0;
655
656		ng_socket_free_priv(priv);
657	}
658
659	pcbp->ng_socket->so_pcb = NULL;
660	mtx_lock(&ngsocketlist_mtx);
661	LIST_REMOVE(pcbp, socks);
662	mtx_unlock(&ngsocketlist_mtx);
663	free(pcbp, M_PCB);
664}
665
666/*
667 * Remove a reference from node private data.
668 */
669static void
670ng_socket_free_priv(struct ngsock *priv)
671{
672	mtx_assert(&priv->mtx, MA_OWNED);
673
674	priv->refs--;
675
676	if (priv->refs == 0) {
677		mtx_destroy(&priv->mtx);
678		hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
679		free(priv, M_NETGRAPH_SOCK);
680		return;
681	}
682
683	if ((priv->refs == 1) && (priv->node != NULL)) {
684		node_p node = priv->node;
685
686		priv->node = NULL;
687		mtx_unlock(&priv->mtx);
688		NG_NODE_UNREF(node);
689		ng_rmnode_self(node);
690	} else
691		mtx_unlock(&priv->mtx);
692}
693
694/*
695 * Connect the data socket to a named control socket node.
696 */
697static int
698ng_connect_data(struct sockaddr *nam, struct ngpcb *pcbp)
699{
700	struct sockaddr_ng *sap;
701	node_p farnode;
702	struct ngsock *priv;
703	int error;
704	item_p item;
705
706	/* If we are already connected, don't do it again. */
707	if (pcbp->sockdata != NULL)
708		return (EISCONN);
709
710	/*
711	 * Find the target (victim) and check it doesn't already have
712	 * a data socket. Also check it is a 'socket' type node.
713	 * Use ng_package_data() and ng_address_path() to do this.
714	 */
715
716	sap = (struct sockaddr_ng *) nam;
717	/* The item will hold the node reference. */
718	item = ng_package_data(NULL, NG_WAITOK);
719
720	if ((error = ng_address_path(NULL, item,  sap->sg_data, 0)))
721		return (error); /* item is freed on failure */
722
723	/*
724	 * Extract node from item and free item. Remember we now have
725	 * a reference on the node. The item holds it for us.
726	 * when we free the item we release the reference.
727	 */
728	farnode = item->el_dest; /* shortcut */
729	if (strcmp(farnode->nd_type->name, NG_SOCKET_NODE_TYPE) != 0) {
730		NG_FREE_ITEM(item); /* drop the reference to the node */
731		return (EINVAL);
732	}
733	priv = NG_NODE_PRIVATE(farnode);
734	if (priv->datasock != NULL) {
735		NG_FREE_ITEM(item);	/* drop the reference to the node */
736		return (EADDRINUSE);
737	}
738
739	/*
740	 * Link the PCB and the private data struct. and note the extra
741	 * reference. Drop the extra reference on the node.
742	 */
743	mtx_lock(&priv->mtx);
744	priv->datasock = pcbp;
745	pcbp->sockdata = priv;
746	pcbp->node_id = priv->node->nd_ID;	/* hint for netstat(1) */
747	priv->refs++;
748	mtx_unlock(&priv->mtx);
749	NG_FREE_ITEM(item);	/* drop the reference to the node */
750	return (0);
751}
752
753/*
754 * Binding a socket means giving the corresponding node a name
755 */
756static int
757ng_bind(struct sockaddr *nam, struct ngpcb *pcbp)
758{
759	struct ngsock *const priv = pcbp->sockdata;
760	struct sockaddr_ng *const sap = (struct sockaddr_ng *) nam;
761
762	if (priv == NULL) {
763		TRAP_ERROR;
764		return (EINVAL);
765	}
766	if ((sap->sg_len < 4) || (sap->sg_len > (NG_NODESIZ + 2)) ||
767	    (sap->sg_data[0] == '\0') ||
768	    (sap->sg_data[sap->sg_len - 3] != '\0')) {
769		TRAP_ERROR;
770		return (EINVAL);
771	}
772	return (ng_name_node(priv->node, sap->sg_data));
773}
774
775/***************************************************************
776	Netgraph node
777***************************************************************/
778
779/*
780 * You can only create new nodes from the socket end of things.
781 */
782static int
783ngs_constructor(node_p nodep)
784{
785	return (EINVAL);
786}
787
788static void
789ngs_rehash(node_p node)
790{
791	struct ngsock *priv = NG_NODE_PRIVATE(node);
792	struct ngshash *new;
793	struct hookpriv *hp;
794	hook_p hook;
795	uint32_t h;
796	u_long hmask;
797
798	new = hashinit_flags((priv->hmask + 1) * 2, M_NETGRAPH_SOCK, &hmask,
799	    HASH_NOWAIT);
800	if (new == NULL)
801		return;
802
803	LIST_FOREACH(hook, &node->nd_hooks, hk_hooks) {
804		hp = NG_HOOK_PRIVATE(hook);
805#ifdef INVARIANTS
806		LIST_REMOVE(hp, next);
807#endif
808		h = hash32_str(NG_HOOK_NAME(hook), HASHINIT) & hmask;
809		LIST_INSERT_HEAD(&new[h], hp, next);
810	}
811
812	hashdestroy(priv->hash, M_NETGRAPH_SOCK, priv->hmask);
813	priv->hash = new;
814	priv->hmask = hmask;
815}
816
817/*
818 * We allow any hook to be connected to the node.
819 * There is no per-hook private information though.
820 */
821static int
822ngs_newhook(node_p node, hook_p hook, const char *name)
823{
824	struct ngsock *const priv = NG_NODE_PRIVATE(node);
825	struct hookpriv *hp;
826	uint32_t h;
827
828	hp = malloc(sizeof(*hp), M_NETGRAPH_SOCK, M_NOWAIT);
829	if (hp == NULL)
830		return (ENOMEM);
831	if (node->nd_numhooks * 2 > priv->hmask)
832		ngs_rehash(node);
833	hp->hook = hook;
834	h = hash32_str(name, HASHINIT) & priv->hmask;
835	LIST_INSERT_HEAD(&priv->hash[h], hp, next);
836	NG_HOOK_SET_PRIVATE(hook, hp);
837
838	return (0);
839}
840
841/*
842 * If only one hook, allow read(2) and write(2) to work.
843 */
844static int
845ngs_connect(hook_p hook)
846{
847	node_p node = NG_HOOK_NODE(hook);
848	struct ngsock *priv = NG_NODE_PRIVATE(node);
849
850	if ((priv->datasock) && (priv->datasock->ng_socket)) {
851		if (NG_NODE_NUMHOOKS(node) == 1)
852			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
853		else
854			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
855	}
856	return (0);
857}
858
859/* Look up hook by name */
860static hook_p
861ngs_findhook(node_p node, const char *name)
862{
863	struct ngsock *priv = NG_NODE_PRIVATE(node);
864	struct hookpriv *hp;
865	uint32_t h;
866
867	/*
868	 * Microoptimisation for an ng_socket with
869	 * a single hook, which is a common case.
870	 */
871	if (node->nd_numhooks == 1) {
872		hook_p hook;
873
874		hook = LIST_FIRST(&node->nd_hooks);
875
876		if (strcmp(NG_HOOK_NAME(hook), name) == 0)
877			return (hook);
878		else
879			return (NULL);
880	}
881
882	h = hash32_str(name, HASHINIT) & priv->hmask;
883
884	LIST_FOREACH(hp, &priv->hash[h], next)
885		if (strcmp(NG_HOOK_NAME(hp->hook), name) == 0)
886			return (hp->hook);
887
888	return (NULL);
889}
890
891/*
892 * Incoming messages get passed up to the control socket.
893 * Unless they are for us specifically (socket_type)
894 */
895static int
896ngs_rcvmsg(node_p node, item_p item, hook_p lasthook)
897{
898	struct ngsock *const priv = NG_NODE_PRIVATE(node);
899	struct ngpcb *pcbp;
900	struct socket *so;
901	struct sockaddr_ng addr;
902	struct ng_mesg *msg;
903	struct mbuf *m;
904	ng_ID_t	retaddr = NGI_RETADDR(item);
905	int addrlen;
906	int error = 0;
907
908	NGI_GET_MSG(item, msg);
909	NG_FREE_ITEM(item);
910
911	/*
912	 * Grab priv->mtx here to prevent destroying of control socket
913	 * after checking that priv->ctlsock is not NULL.
914	 */
915	mtx_lock(&priv->mtx);
916	pcbp = priv->ctlsock;
917
918	/*
919	 * Only allow mesgs to be passed if we have the control socket.
920	 * Data sockets can only support the generic messages.
921	 */
922	if (pcbp == NULL) {
923		mtx_unlock(&priv->mtx);
924		TRAP_ERROR;
925		NG_FREE_MSG(msg);
926		return (EINVAL);
927	}
928	so = pcbp->ng_socket;
929	SOCKBUF_LOCK(&so->so_rcv);
930
931	/* As long as the race is handled, priv->mtx may be unlocked now. */
932	mtx_unlock(&priv->mtx);
933
934#ifdef TRACE_MESSAGES
935	printf("[%x]:---------->[socket]: c=<%d>cmd=%x(%s) f=%x #%d\n",
936		retaddr,
937		msg->header.typecookie,
938		msg->header.cmd,
939		msg->header.cmdstr,
940		msg->header.flags,
941		msg->header.token);
942#endif
943
944	if (msg->header.typecookie == NGM_SOCKET_COOKIE) {
945		switch (msg->header.cmd) {
946		case NGM_SOCK_CMD_NOLINGER:
947			priv->flags |= NGS_FLAG_NOLINGER;
948			break;
949		case NGM_SOCK_CMD_LINGER:
950			priv->flags &= ~NGS_FLAG_NOLINGER;
951			break;
952		default:
953			error = EINVAL;		/* unknown command */
954		}
955		SOCKBUF_UNLOCK(&so->so_rcv);
956
957		/* Free the message and return. */
958		NG_FREE_MSG(msg);
959		return (error);
960	}
961
962	/* Get the return address into a sockaddr. */
963	bzero(&addr, sizeof(addr));
964	addr.sg_len = sizeof(addr);
965	addr.sg_family = AF_NETGRAPH;
966	addrlen = snprintf((char *)&addr.sg_data, sizeof(addr.sg_data),
967	    "[%x]:", retaddr);
968	if (addrlen < 0 || addrlen > sizeof(addr.sg_data)) {
969		SOCKBUF_UNLOCK(&so->so_rcv);
970		printf("%s: snprintf([%x]) failed - %d\n", __func__, retaddr,
971		    addrlen);
972		NG_FREE_MSG(msg);
973		return (EINVAL);
974	}
975
976	/* Copy the message itself into an mbuf chain. */
977	m = m_devget((caddr_t)msg, sizeof(struct ng_mesg) + msg->header.arglen,
978	    0, NULL, NULL);
979
980	/*
981	 * Here we free the message. We need to do that
982	 * regardless of whether we got mbufs.
983	 */
984	NG_FREE_MSG(msg);
985
986	if (m == NULL) {
987		SOCKBUF_UNLOCK(&so->so_rcv);
988		TRAP_ERROR;
989		return (ENOBUFS);
990	}
991
992	/* Send it up to the socket. */
993	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)&addr, m,
994	    NULL) == 0) {
995		soroverflow_locked(so);
996		TRAP_ERROR;
997		m_freem(m);
998		return (ENOBUFS);
999	}
1000
1001	/* sorwakeup_locked () releases the lock internally. */
1002	sorwakeup_locked(so);
1003
1004	return (error);
1005}
1006
1007/*
1008 * Receive data on a hook
1009 */
1010static int
1011ngs_rcvdata(hook_p hook, item_p item)
1012{
1013	struct ngsock *const priv = NG_NODE_PRIVATE(NG_HOOK_NODE(hook));
1014	struct ngpcb *const pcbp = priv->datasock;
1015	struct socket *so;
1016	struct sockaddr_ng *addr;
1017	char *addrbuf[NG_HOOKSIZ + 4];
1018	int addrlen;
1019	struct mbuf *m;
1020
1021	NGI_GET_M(item, m);
1022	NG_FREE_ITEM(item);
1023
1024	/* If there is no data socket, black-hole it. */
1025	if (pcbp == NULL) {
1026		NG_FREE_M(m);
1027		return (0);
1028	}
1029	so = pcbp->ng_socket;
1030
1031	/* Get the return address into a sockaddr. */
1032	addrlen = strlen(NG_HOOK_NAME(hook));	/* <= NG_HOOKSIZ - 1 */
1033	addr = (struct sockaddr_ng *) addrbuf;
1034	addr->sg_len = addrlen + 3;
1035	addr->sg_family = AF_NETGRAPH;
1036	bcopy(NG_HOOK_NAME(hook), addr->sg_data, addrlen);
1037	addr->sg_data[addrlen] = '\0';
1038
1039	/* Try to tell the socket which hook it came in on. */
1040	SOCKBUF_LOCK(&so->so_rcv);
1041	if (sbappendaddr_locked(&so->so_rcv, (struct sockaddr *)addr, m,
1042	    NULL) == 0) {
1043		SOCKBUF_UNLOCK(&so->so_rcv);
1044		m_freem(m);
1045		TRAP_ERROR;
1046		return (ENOBUFS);
1047	}
1048
1049	/* sorwakeup_locked () releases the lock internally. */
1050	sorwakeup_locked(so);
1051	return (0);
1052}
1053
1054/*
1055 * Hook disconnection
1056 *
1057 * For this type, removal of the last link destroys the node
1058 * if the NOLINGER flag is set.
1059 */
1060static int
1061ngs_disconnect(hook_p hook)
1062{
1063	node_p node = NG_HOOK_NODE(hook);
1064	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1065	struct hookpriv *hp = NG_HOOK_PRIVATE(hook);
1066
1067	LIST_REMOVE(hp, next);
1068	free(hp, M_NETGRAPH_SOCK);
1069
1070	if ((priv->datasock) && (priv->datasock->ng_socket)) {
1071		if (NG_NODE_NUMHOOKS(node) == 1)
1072			priv->datasock->ng_socket->so_state |= SS_ISCONNECTED;
1073		else
1074			priv->datasock->ng_socket->so_state &= ~SS_ISCONNECTED;
1075	}
1076
1077	if ((priv->flags & NGS_FLAG_NOLINGER) &&
1078	    (NG_NODE_NUMHOOKS(node) == 0) && (NG_NODE_IS_VALID(node)))
1079		ng_rmnode_self(node);
1080
1081	return (0);
1082}
1083
1084/*
1085 * Do local shutdown processing.
1086 * In this case, that involves making sure the socket
1087 * knows we should be shutting down.
1088 */
1089static int
1090ngs_shutdown(node_p node)
1091{
1092	struct ngsock *const priv = NG_NODE_PRIVATE(node);
1093	struct ngpcb *dpcbp, *pcbp;
1094
1095	mtx_lock(&priv->mtx);
1096	dpcbp = priv->datasock;
1097	pcbp = priv->ctlsock;
1098
1099	if (dpcbp != NULL)
1100		soisdisconnected(dpcbp->ng_socket);
1101
1102	if (pcbp != NULL)
1103		soisdisconnected(pcbp->ng_socket);
1104
1105	priv->node = NULL;
1106	NG_NODE_SET_PRIVATE(node, NULL);
1107	ng_socket_free_priv(priv);
1108
1109	NG_NODE_UNREF(node);
1110	return (0);
1111}
1112
1113static void
1114ng_socket_item_applied(void *context, int error)
1115{
1116	struct ngsock *const priv = (struct ngsock *)context;
1117
1118	mtx_lock(&priv->mtx);
1119	priv->error = error;
1120	wakeup(priv);
1121	mtx_unlock(&priv->mtx);
1122
1123}
1124
1125static	int
1126dummy_disconnect(struct socket *so)
1127{
1128	return (0);
1129}
1130
1131/*
1132 * Definitions of protocols supported in the NETGRAPH domain.
1133 * Control and data socket type descriptors
1134 *
1135 * XXXRW: Perhaps _close should do something?
1136 */
1137static struct protosw ngcontrol_protosw = {
1138	.pr_type =		SOCK_DGRAM,
1139	.pr_protocol =		NG_CONTROL,
1140	.pr_flags =		PR_ATOMIC | PR_ADDR /* | PR_RIGHTS */,
1141	.pr_attach =		ngc_attach,
1142	.pr_bind =		ngc_bind,
1143	.pr_connect =		ngc_connect,
1144	.pr_detach =		ngc_detach,
1145	.pr_disconnect =	dummy_disconnect,
1146	.pr_send =		ngc_send,
1147	.pr_sockaddr =		ng_getsockaddr,
1148};
1149static struct protosw ngdata_protosw = {
1150	.pr_type =		SOCK_DGRAM,
1151	.pr_protocol =		NG_DATA,
1152	.pr_flags =		PR_ATOMIC | PR_ADDR,
1153	.pr_attach =		ngd_attach,
1154	.pr_connect =		ngd_connect,
1155	.pr_detach =		ngd_detach,
1156	.pr_disconnect =	dummy_disconnect,
1157	.pr_send =		ngd_send,
1158	.pr_sockaddr =		ng_getsockaddr,
1159};
1160
1161static struct domain ngdomain = {
1162	.dom_family =		AF_NETGRAPH,
1163	.dom_name =		"netgraph",
1164	.dom_nprotosw =		2,
1165	.dom_protosw =		{ &ngcontrol_protosw, &ngdata_protosw },
1166};
1167
1168/*
1169 * Handle loading and unloading for this node type.
1170 * This is to handle auxiliary linkages (e.g protocol domain addition).
1171 */
1172static int
1173ngs_mod_event(module_t mod, int event, void *data)
1174{
1175	int error = 0;
1176
1177	switch (event) {
1178	case MOD_LOAD:
1179		mtx_init(&ngsocketlist_mtx, "ng_socketlist", NULL, MTX_DEF);
1180		break;
1181	case MOD_UNLOAD:
1182		/* Ensure there are no open netgraph sockets. */
1183		if (!LIST_EMPTY(&ngsocklist)) {
1184			error = EBUSY;
1185			break;
1186		}
1187#ifdef NOTYET
1188		/* Unregister protocol domain XXX can't do this yet.. */
1189#endif
1190		error = EBUSY;
1191		break;
1192	default:
1193		error = EOPNOTSUPP;
1194		break;
1195	}
1196	return (error);
1197}
1198
1199DOMAIN_SET(ng);
1200
1201SYSCTL_INT(_net_graph, OID_AUTO, family, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, AF_NETGRAPH, "");
1202static SYSCTL_NODE(_net_graph, OID_AUTO, data, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1203    "DATA");
1204SYSCTL_INT(_net_graph_data, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_DATA, "");
1205static SYSCTL_NODE(_net_graph, OID_AUTO, control, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
1206    "CONTROL");
1207SYSCTL_INT(_net_graph_control, OID_AUTO, proto, CTLFLAG_RD, SYSCTL_NULL_INT_PTR, NG_CONTROL, "");
1208