xref: /xnu-10002.61.3/bsd/net/ndrv.c (revision 0f4c859e951fba394238ab619495c4e1d54d0f34)
1 /*
2  * Copyright (c) 1997-2021 Apple Inc. All rights reserved.
3  *
4  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
5  *
6  * This file contains Original Code and/or Modifications of Original Code
7  * as defined in and that are subject to the Apple Public Source License
8  * Version 2.0 (the 'License'). You may not use this file except in
9  * compliance with the License. The rights granted to you under the License
10  * may not be used to create, or enable the creation or redistribution of,
11  * unlawful or unlicensed copies of an Apple operating system, or to
12  * circumvent, violate, or enable the circumvention or violation of, any
13  * terms of an Apple operating system software license agreement.
14  *
15  * Please obtain a copy of the License at
16  * http://www.opensource.apple.com/apsl/ and read it before using this file.
17  *
18  * The Original Code and all software distributed under the License are
19  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
20  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
21  * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
22  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
23  * Please see the License for the specific language governing rights and
24  * limitations under the License.
25  *
26  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
27  */
28 /*
29  *	@(#)ndrv.c	1.1 (MacOSX) 6/10/43
30  * Justin Walker, 970604
31  *   AF_NDRV support
32  * 980130 - Cleanup, reorg, performance improvemements
33  * 000816 - Removal of Y adapter cruft
34  */
35 
36 /*
37  * PF_NDRV allows raw access to a specified network device, directly
38  *  with a socket.  Expected use involves a socket option to request
39  *  protocol packets.  This lets ndrv_output() call ifnet_output(), and
40  *  lets DLIL find the proper recipient for incoming packets.
41  *  The purpose here is for user-mode protocol implementation.
42  * Note that "pure raw access" will still be accomplished with BPF.
43  *
44  * In addition to the former use, when combined with socket NKEs,
45  * PF_NDRV permits a fairly flexible mechanism for implementing
46  * strange protocol support.
47  */
48 #include <mach/mach_types.h>
49 
50 #include <sys/param.h>
51 #include <sys/systm.h>
52 #include <sys/kernel.h>
53 #include <sys/malloc.h>
54 #include <sys/mbuf.h>
55 #include <sys/protosw.h>
56 #include <sys/domain.h>
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/ioctl.h>
60 #include <sys/sysctl.h>
61 #include <sys/errno.h>
62 #include <sys/syslog.h>
63 #include <sys/proc.h>
64 
65 #include <kern/queue.h>
66 #include <kern/assert.h>
67 
68 #include <net/ndrv.h>
69 #include <net/route.h>
70 #include <net/if_llc.h>
71 #include <net/if_dl.h>
72 #include <net/if_types.h>
73 #include <net/ndrv_var.h>
74 #include <net/dlil.h>
75 
76 #if INET
77 #include <netinet/in.h>
78 #include <netinet/in_var.h>
79 #endif
80 #include <netinet/if_ether.h>
81 
82 static unsigned int ndrv_multi_max_count = NDRV_DMUX_MAX_DESCR;
83 SYSCTL_UINT(_net, OID_AUTO, ndrv_multi_max_count, CTLFLAG_RW | CTLFLAG_LOCKED,
84     &ndrv_multi_max_count, 0, "Number of allowed multicast addresses per NRDV socket");
85 
86 /*
87  * The locking strategy relies on the PF_NRDRV domain mutex that protects both the
88  * PCB list "ndrvl" and the sockets themselves
89  */
90 
91 static int ndrv_do_detach(struct ndrv_cb *);
92 static int ndrv_do_disconnect(struct ndrv_cb *);
93 static struct ndrv_cb *ndrv_find_inbound(struct ifnet *ifp, u_int32_t protocol_family);
94 static int ndrv_setspec(struct ndrv_cb *np, struct sockopt *sopt);
95 static int ndrv_delspec(struct ndrv_cb *);
96 static int ndrv_to_ifnet_demux(struct ndrv_demux_desc* ndrv, struct ifnet_demux_desc* ifdemux);
97 static void ndrv_handle_ifp_detach(u_int32_t family, short unit);
98 static int ndrv_do_add_multicast(struct ndrv_cb *np, struct sockopt *sopt);
99 static int ndrv_do_remove_multicast(struct ndrv_cb *np, struct sockopt *sopt);
100 static struct ndrv_multiaddr* ndrv_have_multicast(struct ndrv_cb *np, struct sockaddr* addr);
101 static void ndrv_remove_all_multicast(struct ndrv_cb *np);
102 static void ndrv_dominit(struct domain *);
103 
104 u_int32_t  ndrv_sendspace = NDRVSNDQ;
105 u_int32_t  ndrv_recvspace = NDRVRCVQ;
106 TAILQ_HEAD(, ndrv_cb)   ndrvl = TAILQ_HEAD_INITIALIZER(ndrvl);
107 
108 static struct domain *ndrvdomain = NULL;
109 extern struct domain ndrvdomain_s;
110 
111 #define NDRV_PROTODEMUX_COUNT   10
112 
113 /*
114  * Verify these values match.
115  * To keep clients from including dlil.h, we define
116  * these values independently in ndrv.h. They must
117  * match or a conversion function must be written.
118  */
119 #if NDRV_DEMUXTYPE_ETHERTYPE != DLIL_DESC_ETYPE2
120 #error NDRV_DEMUXTYPE_ETHERTYPE must match DLIL_DESC_ETYPE2
121 #endif
122 #if NDRV_DEMUXTYPE_SAP != DLIL_DESC_SAP
123 #error NDRV_DEMUXTYPE_SAP must match DLIL_DESC_SAP
124 #endif
125 #if NDRV_DEMUXTYPE_SNAP != DLIL_DESC_SNAP
126 #error NDRV_DEMUXTYPE_SNAP must match DLIL_DESC_SNAP
127 #endif
128 
129 /*
130  * Protocol output - Called to output a raw network packet directly
131  *  to the driver.
132  */
133 static int
ndrv_output(struct mbuf * m,struct socket * so)134 ndrv_output(struct mbuf *m, struct socket *so)
135 {
136 	struct ndrv_cb *np = sotondrvcb(so);
137 	struct ifnet *ifp = np->nd_if;
138 	int result = 0;
139 
140 #if NDRV_DEBUG
141 	printf("NDRV output: %x, %x, %x\n", m, so, np);
142 #endif
143 
144 	/*
145 	 * No header is a format error
146 	 */
147 	if ((m->m_flags & M_PKTHDR) == 0) {
148 		return EINVAL;
149 	}
150 
151 	/* Unlock before calling ifnet_output */
152 	socket_unlock(so, 0);
153 
154 	/*
155 	 * Call DLIL if we can. DLIL is much safer than calling the
156 	 * ifp directly.
157 	 */
158 	result = ifnet_output_raw(ifp, np->nd_proto_family, m);
159 
160 	socket_lock(so, 0);
161 
162 	return result;
163 }
164 
165 /* Our input routine called from DLIL */
166 static errno_t
ndrv_input(ifnet_t ifp,protocol_family_t proto_family,mbuf_t m,char * frame_header)167 ndrv_input(
168 	ifnet_t                         ifp,
169 	protocol_family_t       proto_family,
170 	mbuf_t                          m,
171 	char                            *frame_header)
172 {
173 	struct socket *so;
174 	struct sockaddr_dl ndrvsrc = {};
175 	struct ndrv_cb *np;
176 	int error = 0;
177 
178 	ndrvsrc.sdl_len = sizeof(struct sockaddr_dl);
179 	ndrvsrc.sdl_family = AF_NDRV;
180 	ndrvsrc.sdl_index = 0;
181 
182 	/* move packet from if queue to socket */
183 	/* Should be media-independent */
184 	ndrvsrc.sdl_type = IFT_ETHER;
185 	ndrvsrc.sdl_nlen = 0;
186 	ndrvsrc.sdl_alen = 6;
187 	ndrvsrc.sdl_slen = 0;
188 	bcopy(frame_header, &ndrvsrc.sdl_data, 6);
189 
190 	/* prepend the frame header */
191 	m = m_prepend(m, ifnet_hdrlen(ifp), M_NOWAIT);
192 	if (m == NULL) {
193 		return EJUSTRETURN;
194 	}
195 	bcopy(frame_header, m->m_data, ifnet_hdrlen(ifp));
196 
197 	/*
198 	 * We need to take the domain mutex before the list RW lock
199 	 */
200 	LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED);
201 	lck_mtx_lock(ndrvdomain->dom_mtx);
202 
203 	np = ndrv_find_inbound(ifp, proto_family);
204 	if (np == NULL) {
205 		lck_mtx_unlock(ndrvdomain->dom_mtx);
206 		return ENOENT;
207 	}
208 
209 	so = np->nd_socket;
210 
211 	if (sbappendaddr(&(so->so_rcv), (struct sockaddr *)&ndrvsrc,
212 	    m, NULL, &error) != 0) {
213 		sorwakeup(so);
214 	}
215 
216 	lck_mtx_unlock(ndrvdomain->dom_mtx);
217 
218 	return 0; /* radar 4030377 - always return 0 */
219 }
220 
221 /*
222  * Allocate an ndrv control block and some buffer space for the socket
223  */
224 static int
ndrv_attach(struct socket * so,int proto,__unused struct proc * p)225 ndrv_attach(struct socket *so, int proto, __unused struct proc *p)
226 {
227 	int error;
228 	struct ndrv_cb *np = sotondrvcb(so);
229 
230 	if ((so->so_state & SS_PRIV) == 0) {
231 		return EPERM;
232 	}
233 
234 #if NDRV_DEBUG
235 	printf("NDRV attach: %x, %x, %x\n", so, proto, np);
236 #endif
237 
238 	if ((error = soreserve(so, ndrv_sendspace, ndrv_recvspace))) {
239 		return error;
240 	}
241 
242 	np = kalloc_type(struct ndrv_cb, Z_WAITOK | Z_ZERO | Z_NOFAIL);
243 	so->so_pcb = (caddr_t)np;
244 #if NDRV_DEBUG
245 	printf("NDRV attach: %x, %x, %x\n", so, proto, np);
246 #endif
247 	TAILQ_INIT(&np->nd_dlist);
248 	np->nd_signature = NDRV_SIGNATURE;
249 	np->nd_socket = so;
250 	np->nd_proto.sp_family = (uint16_t)SOCK_DOM(so);
251 	np->nd_proto.sp_protocol = (uint16_t)proto;
252 	np->nd_if = NULL;
253 	np->nd_proto_family = 0;
254 	np->nd_family = 0;
255 	np->nd_unit = 0;
256 
257 	/*
258 	 * Use the domain mutex to protect the list
259 	 */
260 	LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED);
261 	lck_mtx_lock(ndrvdomain->dom_mtx);
262 
263 	TAILQ_INSERT_TAIL(&ndrvl, np, nd_next);
264 
265 	lck_mtx_unlock(ndrvdomain->dom_mtx);
266 
267 	return 0;
268 }
269 
270 /*
271  * Destroy state just before socket deallocation.
272  * Flush data or not depending on the options.
273  */
274 
275 static int
ndrv_detach(struct socket * so)276 ndrv_detach(struct socket *so)
277 {
278 	struct ndrv_cb *np = sotondrvcb(so);
279 
280 	if (np == 0) {
281 		return EINVAL;
282 	}
283 	return ndrv_do_detach(np);
284 }
285 
286 
287 /*
288  * If a socket isn't bound to a single address,
289  * the ndrv input routine will hand it anything
290  * within that protocol family (assuming there's
291  * nothing else around it should go to).
292  *
293  * Don't expect this to be used.
294  */
295 
296 static int
ndrv_connect(struct socket * so,struct sockaddr * nam,__unused struct proc * p)297 ndrv_connect(struct socket *so, struct sockaddr *nam, __unused struct proc *p)
298 {
299 	struct ndrv_cb *np = sotondrvcb(so);
300 
301 	if (np == 0) {
302 		return EINVAL;
303 	}
304 
305 	if (np->nd_faddr) {
306 		return EISCONN;
307 	}
308 
309 	if (nam->sa_len < sizeof(struct sockaddr_ndrv)) {
310 		return EINVAL;
311 	}
312 
313 	/* Allocate memory to store the remote address */
314 	np->nd_faddr = kalloc_type(struct sockaddr_ndrv, Z_WAITOK | Z_NOFAIL | Z_ZERO);
315 
316 	bcopy((caddr_t) nam, (caddr_t) np->nd_faddr, MIN(sizeof(struct sockaddr_ndrv), nam->sa_len));
317 	np->nd_faddr->snd_len = sizeof(struct sockaddr_ndrv);
318 	soisconnected(so);
319 	return 0;
320 }
321 
322 static void
ndrv_event(struct ifnet * ifp,__unused protocol_family_t protocol,const struct kev_msg * event)323 ndrv_event(struct ifnet *ifp, __unused protocol_family_t protocol,
324     const struct kev_msg *event)
325 {
326 	if (event->vendor_code == KEV_VENDOR_APPLE &&
327 	    event->kev_class == KEV_NETWORK_CLASS &&
328 	    event->kev_subclass == KEV_DL_SUBCLASS &&
329 	    event->event_code == KEV_DL_IF_DETACHING) {
330 		LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_NOTOWNED);
331 		lck_mtx_lock(ndrvdomain->dom_mtx);
332 		ndrv_handle_ifp_detach(ifnet_family(ifp), ifp->if_unit);
333 		lck_mtx_unlock(ndrvdomain->dom_mtx);
334 	}
335 }
336 
337 /*
338  * This is the "driver open" hook - we 'bind' to the
339  *  named driver.
340  * Here's where we latch onto the driver.
341  */
342 static int
ndrv_bind(struct socket * so,struct sockaddr * nam,__unused struct proc * p)343 ndrv_bind(struct socket *so, struct sockaddr *nam, __unused struct proc *p)
344 {
345 	struct sockaddr_ndrv *sa = (struct sockaddr_ndrv *) nam;
346 	char *dname;
347 	struct ndrv_cb *np;
348 	struct ifnet *ifp;
349 	int result;
350 
351 	if (TAILQ_EMPTY(&ifnet_head)) {
352 		return EADDRNOTAVAIL;        /* Quick sanity check */
353 	}
354 	np = sotondrvcb(so);
355 	if (np == 0) {
356 		return EINVAL;
357 	}
358 
359 	if (np->nd_laddr) {
360 		return EINVAL;                  /* XXX */
361 	}
362 	/* I think we just latch onto a copy here; the caller frees */
363 	np->nd_laddr = kalloc_type(struct sockaddr_ndrv, Z_WAITOK | Z_NOFAIL | Z_ZERO);
364 	bcopy((caddr_t) sa, (caddr_t) np->nd_laddr, MIN(sizeof(struct sockaddr_ndrv), sa->snd_len));
365 	np->nd_laddr->snd_len = sizeof(struct sockaddr_ndrv);
366 	dname = (char *) sa->snd_name;
367 	if (*dname == '\0') {
368 		return EINVAL;
369 	}
370 #if NDRV_DEBUG
371 	printf("NDRV bind: %x, %x, %s\n", so, np, dname);
372 #endif
373 	/* Track down the driver and its ifnet structure.
374 	 * There's no internal call for this so we have to dup the code
375 	 *  in if.c/ifconf()
376 	 */
377 	ifnet_head_lock_shared();
378 	TAILQ_FOREACH(ifp, &ifnet_head, if_link) {
379 		if (strncmp(ifp->if_xname, dname, IFNAMSIZ) == 0) {
380 			break;
381 		}
382 	}
383 	ifnet_head_done();
384 
385 	if (ifp == NULL) {
386 		return EADDRNOTAVAIL;
387 	}
388 
389 	// PPP doesn't support PF_NDRV.
390 	if (ifnet_family(ifp) != APPLE_IF_FAM_PPP) {
391 		/* NDRV on this interface */
392 		struct ifnet_attach_proto_param ndrv_proto;
393 		result = 0;
394 		bzero(&ndrv_proto, sizeof(ndrv_proto));
395 		ndrv_proto.event = ndrv_event;
396 
397 		/* We aren't worried about double attaching, that should just return an error */
398 		socket_unlock(so, 0);
399 		result = ifnet_attach_protocol(ifp, PF_NDRV, &ndrv_proto);
400 		socket_lock(so, 0);
401 		if (result && result != EEXIST) {
402 			return result;
403 		}
404 		np->nd_proto_family = PF_NDRV;
405 	} else {
406 		np->nd_proto_family = 0;
407 	}
408 
409 	np->nd_if = ifp;
410 	np->nd_family = ifnet_family(ifp);
411 	np->nd_unit = ifp->if_unit;
412 
413 	return 0;
414 }
415 
416 static int
ndrv_disconnect(struct socket * so)417 ndrv_disconnect(struct socket *so)
418 {
419 	struct ndrv_cb *np = sotondrvcb(so);
420 
421 	if (np == 0) {
422 		return EINVAL;
423 	}
424 
425 	if (np->nd_faddr == 0) {
426 		return ENOTCONN;
427 	}
428 
429 	ndrv_do_disconnect(np);
430 	return 0;
431 }
432 
433 /*
434  * Mark the connection as being incapable of further input.
435  */
436 static int
ndrv_shutdown(struct socket * so)437 ndrv_shutdown(struct socket *so)
438 {
439 	LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_OWNED);
440 	socantsendmore(so);
441 	return 0;
442 }
443 
444 /*
445  * Ship a packet out.  The ndrv output will pass it
446  *  to the appropriate driver.  The really tricky part
447  *  is the destination address...
448  */
449 static int
ndrv_send(struct socket * so,__unused int flags,struct mbuf * m,__unused struct sockaddr * addr,struct mbuf * control,__unused struct proc * p)450 ndrv_send(struct socket *so, __unused int flags, struct mbuf *m,
451     __unused struct sockaddr *addr, struct mbuf *control,
452     __unused struct proc *p)
453 {
454 	int error;
455 
456 	if (control != NULL) {
457 		m_freem(control);
458 		return EOPNOTSUPP;
459 	}
460 
461 	error = ndrv_output(m, so);
462 	m = NULL;
463 	return error;
464 }
465 
466 
467 static int
ndrv_abort(struct socket * so)468 ndrv_abort(struct socket *so)
469 {
470 	struct ndrv_cb *np = sotondrvcb(so);
471 
472 	if (np == 0) {
473 		return EINVAL;
474 	}
475 
476 	ndrv_do_disconnect(np);
477 	return 0;
478 }
479 
480 static int
ndrv_sockaddr(struct socket * so,struct sockaddr ** nam)481 ndrv_sockaddr(struct socket *so, struct sockaddr **nam)
482 {
483 	struct ndrv_cb *np = sotondrvcb(so);
484 	int len;
485 
486 	if (np == 0) {
487 		return EINVAL;
488 	}
489 
490 	if (np->nd_laddr == 0) {
491 		return EINVAL;
492 	}
493 
494 	len = np->nd_laddr->snd_len;
495 	*nam = (struct sockaddr *)alloc_sockaddr(len,
496 	    Z_WAITOK | Z_NOFAIL);
497 
498 	bcopy((caddr_t)np->nd_laddr, *nam,
499 	    (unsigned)len);
500 	return 0;
501 }
502 
503 
504 static int
ndrv_peeraddr(struct socket * so,struct sockaddr ** nam)505 ndrv_peeraddr(struct socket *so, struct sockaddr **nam)
506 {
507 	struct ndrv_cb *np = sotondrvcb(so);
508 	int len;
509 
510 	if (np == 0) {
511 		return EINVAL;
512 	}
513 
514 	if (np->nd_faddr == 0) {
515 		return ENOTCONN;
516 	}
517 
518 	len = np->nd_faddr->snd_len;
519 	*nam = (struct sockaddr *)alloc_sockaddr(len,
520 	    Z_WAITOK | Z_NOFAIL);
521 
522 	bcopy((caddr_t)np->nd_faddr, *nam,
523 	    (unsigned)len);
524 	return 0;
525 }
526 
527 
528 /* Control output */
529 
530 static int
ndrv_ctloutput(struct socket * so,struct sockopt * sopt)531 ndrv_ctloutput(struct socket *so, struct sockopt *sopt)
532 {
533 	struct ndrv_cb *np = sotondrvcb(so);
534 	int error = 0;
535 
536 	switch (sopt->sopt_name) {
537 	case NDRV_DELDMXSPEC: /* Delete current spec */
538 		/* Verify no parameter was passed */
539 		if (sopt->sopt_val != 0 || sopt->sopt_valsize != 0) {
540 			/*
541 			 * We don't support deleting a specific demux, it's
542 			 * all or nothing.
543 			 */
544 			return EINVAL;
545 		}
546 		error = ndrv_delspec(np);
547 		break;
548 	case NDRV_SETDMXSPEC: /* Set protocol spec */
549 		error = ndrv_setspec(np, sopt);
550 		break;
551 	case NDRV_ADDMULTICAST:
552 		error = ndrv_do_add_multicast(np, sopt);
553 		break;
554 	case NDRV_DELMULTICAST:
555 		error = ndrv_do_remove_multicast(np, sopt);
556 		break;
557 	default:
558 		error = ENOTSUP;
559 	}
560 #ifdef NDRV_DEBUG
561 	log(LOG_WARNING, "NDRV CTLOUT: %x returns %d\n", sopt->sopt_name,
562 	    error);
563 #endif
564 	return error;
565 }
566 
567 static int
ndrv_do_detach(struct ndrv_cb * np)568 ndrv_do_detach(struct ndrv_cb *np)
569 {
570 	struct ndrv_cb*     cur_np = NULL;
571 	struct socket *so = np->nd_socket;
572 	int error = 0;
573 	struct ifnet * ifp;
574 
575 #if NDRV_DEBUG
576 	printf("NDRV detach: %x, %x\n", so, np);
577 #endif
578 	ndrv_remove_all_multicast(np);
579 
580 	/* Remove from the linked list of control blocks */
581 	LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_OWNED);
582 	TAILQ_REMOVE(&ndrvl, np, nd_next);
583 
584 	ifp = np->nd_if;
585 	if (ifp != NULL) {
586 		u_int32_t proto_family = np->nd_proto_family;
587 
588 		if (proto_family != PF_NDRV && proto_family != 0) {
589 			socket_unlock(so, 0);
590 			ifnet_detach_protocol(ifp, proto_family);
591 			socket_lock(so, 0);
592 		}
593 
594 		/* Check if this is the last socket attached to this interface */
595 		LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_OWNED);
596 		TAILQ_FOREACH(cur_np, &ndrvl, nd_next) {
597 			if (cur_np->nd_family == np->nd_family &&
598 			    cur_np->nd_unit == np->nd_unit) {
599 				break;
600 			}
601 		}
602 
603 		/* If there are no other interfaces, detach PF_NDRV from the interface */
604 		if (cur_np == NULL) {
605 			socket_unlock(so, 0);
606 			ifnet_detach_protocol(ifp, PF_NDRV);
607 			socket_lock(so, 0);
608 		}
609 	}
610 	if (np->nd_laddr != NULL) {
611 		kfree_type(struct sockaddr_ndrv, np->nd_laddr);
612 	}
613 	kfree_type(struct ndrv_cb, np);
614 	so->so_pcb = 0;
615 	so->so_flags |= SOF_PCBCLEARING;
616 	sofree(so);
617 	return error;
618 }
619 
620 static int
ndrv_do_disconnect(struct ndrv_cb * np)621 ndrv_do_disconnect(struct ndrv_cb *np)
622 {
623 	struct socket * so = np->nd_socket;
624 #if NDRV_DEBUG
625 	printf("NDRV disconnect: %x\n", np);
626 #endif
627 	if (np->nd_faddr) {
628 		kfree_type(struct sockaddr_ndrv, np->nd_faddr);
629 	}
630 	/*
631 	 * A multipath subflow socket would have its SS_NOFDREF set by default,
632 	 * so check for SOF_MP_SUBFLOW socket flag before detaching the PCB;
633 	 * when the socket is closed for real, SOF_MP_SUBFLOW would be cleared.
634 	 */
635 	if (!(so->so_flags & SOF_MP_SUBFLOW) && (so->so_state & SS_NOFDREF)) {
636 		ndrv_do_detach(np);
637 	}
638 	soisdisconnected(so);
639 	return 0;
640 }
641 
642 #if 0
643 //### Not used
644 /*
645  * When closing, dump any enqueued mbufs.
646  */
647 void
648 ndrv_flushq(struct ifqueue *q)
649 {
650 	struct mbuf *m;
651 	for (;;) {
652 		IF_DEQUEUE(q, m);
653 		if (m == NULL) {
654 			break;
655 		}
656 		IF_DROP(q);
657 		if (m) {
658 			m_freem(m);
659 		}
660 	}
661 }
662 #endif
663 
664 int
ndrv_setspec(struct ndrv_cb * np,struct sockopt * sopt)665 ndrv_setspec(struct ndrv_cb *np, struct sockopt *sopt)
666 {
667 	struct ifnet_attach_proto_param proto_param;
668 	struct ndrv_protocol_desc       ndrvSpec;
669 	struct ndrv_demux_desc*         ndrvDemux = NULL;
670 	size_t                          ndrvDemuxSize = 0;
671 	int                                                     error = 0;
672 	struct socket *                         so = np->nd_socket;
673 	user_addr_t                                     user_addr;
674 
675 	/* Sanity checking */
676 	if (np->nd_proto_family != PF_NDRV) {
677 		return EBUSY;
678 	}
679 	if (np->nd_if == NULL) {
680 		return EINVAL;
681 	}
682 
683 	/* Copy the ndrvSpec */
684 	if (proc_is64bit(sopt->sopt_p)) {
685 		struct ndrv_protocol_desc64     ndrvSpec64;
686 
687 		if (sopt->sopt_valsize != sizeof(ndrvSpec64)) {
688 			return EINVAL;
689 		}
690 
691 		error = sooptcopyin(sopt, &ndrvSpec64, sizeof(ndrvSpec64), sizeof(ndrvSpec64));
692 		if (error != 0) {
693 			return error;
694 		}
695 
696 		ndrvSpec.version         = ndrvSpec64.version;
697 		ndrvSpec.protocol_family = ndrvSpec64.protocol_family;
698 		ndrvSpec.demux_count     = ndrvSpec64.demux_count;
699 
700 		user_addr = CAST_USER_ADDR_T(ndrvSpec64.demux_list);
701 	} else {
702 		struct ndrv_protocol_desc32     ndrvSpec32;
703 
704 		if (sopt->sopt_valsize != sizeof(ndrvSpec32)) {
705 			return EINVAL;
706 		}
707 
708 		error = sooptcopyin(sopt, &ndrvSpec32, sizeof(ndrvSpec32), sizeof(ndrvSpec32));
709 		if (error != 0) {
710 			return error;
711 		}
712 
713 		ndrvSpec.version         = ndrvSpec32.version;
714 		ndrvSpec.protocol_family = ndrvSpec32.protocol_family;
715 		ndrvSpec.demux_count     = ndrvSpec32.demux_count;
716 
717 		user_addr = CAST_USER_ADDR_T(ndrvSpec32.demux_list);
718 	}
719 
720 	/*
721 	 * Do not allow PF_NDRV as it's non-sensical and most importantly because
722 	 * we use PF_NDRV to see if the protocol family has already been set
723 	 */
724 	if (ndrvSpec.protocol_family == PF_NDRV) {
725 		return EINVAL;
726 	}
727 
728 	/* Verify the parameter */
729 	if (ndrvSpec.version > NDRV_PROTOCOL_DESC_VERS) {
730 		return ENOTSUP; // version is too new!
731 	} else if (ndrvSpec.version < 1) {
732 		return EINVAL; // version is not valid
733 	} else if (ndrvSpec.demux_count > NDRV_PROTODEMUX_COUNT || ndrvSpec.demux_count == 0) {
734 		return EINVAL; // demux_count is not valid
735 	}
736 	bzero(&proto_param, sizeof(proto_param));
737 	proto_param.demux_count = ndrvSpec.demux_count;
738 
739 	/* Allocate storage for demux array */
740 	ndrvDemuxSize = proto_param.demux_count * sizeof(struct ndrv_demux_desc);
741 	ndrvDemux = (struct ndrv_demux_desc*) kalloc_data(ndrvDemuxSize, Z_WAITOK);
742 	if (ndrvDemux == NULL) {
743 		return ENOMEM;
744 	}
745 
746 	/* Allocate enough ifnet_demux_descs */
747 	proto_param.demux_array = kalloc_type(struct ifnet_demux_desc,
748 	    ndrvSpec.demux_count, Z_WAITOK | Z_ZERO);
749 	if (proto_param.demux_array == NULL) {
750 		error = ENOMEM;
751 	}
752 
753 	if (error == 0) {
754 		/* Copy the ndrv demux array from userland */
755 		error = copyin(user_addr, ndrvDemux,
756 		    ndrvSpec.demux_count * sizeof(struct ndrv_demux_desc));
757 		ndrvSpec.demux_list = ndrvDemux;
758 	}
759 
760 	if (error == 0) {
761 		/* At this point, we've at least got enough bytes to start looking around */
762 		u_int32_t       demuxOn = 0;
763 
764 		proto_param.demux_count = ndrvSpec.demux_count;
765 		proto_param.input = ndrv_input;
766 		proto_param.event = ndrv_event;
767 
768 		for (demuxOn = 0; demuxOn < ndrvSpec.demux_count; demuxOn++) {
769 			/* Convert an ndrv_demux_desc to a ifnet_demux_desc */
770 			error = ndrv_to_ifnet_demux(&ndrvSpec.demux_list[demuxOn],
771 			    &proto_param.demux_array[demuxOn]);
772 			if (error) {
773 				break;
774 			}
775 		}
776 	}
777 
778 	if (error == 0) {
779 		/*
780 		 * Set the protocol family to prevent other threads from
781 		 * attaching a protocol while the socket is unlocked
782 		 */
783 		np->nd_proto_family = ndrvSpec.protocol_family;
784 		socket_unlock(so, 0);
785 		error = ifnet_attach_protocol(np->nd_if, ndrvSpec.protocol_family,
786 		    &proto_param);
787 		socket_lock(so, 0);
788 		/*
789 		 * Upon failure, indicate that no protocol is attached
790 		 */
791 		if (error != 0) {
792 			np->nd_proto_family = PF_NDRV;
793 		}
794 	}
795 
796 	/* Free any memory we've allocated */
797 	if (proto_param.demux_array) {
798 		kfree_type(struct ifnet_demux_desc, ndrvSpec.demux_count,
799 		    proto_param.demux_array);
800 	}
801 	if (ndrvDemux) {
802 		kfree_data(ndrvDemux, ndrvDemuxSize);
803 	}
804 
805 	return error;
806 }
807 
808 
809 int
ndrv_to_ifnet_demux(struct ndrv_demux_desc * ndrv,struct ifnet_demux_desc * ifdemux)810 ndrv_to_ifnet_demux(struct ndrv_demux_desc* ndrv, struct ifnet_demux_desc* ifdemux)
811 {
812 	bzero(ifdemux, sizeof(*ifdemux));
813 
814 	if (ndrv->type < DLIL_DESC_ETYPE2) {
815 		/* using old "type", not supported */
816 		return ENOTSUP;
817 	}
818 
819 	if (ndrv->length > 28) {
820 		return EINVAL;
821 	}
822 
823 	ifdemux->type = ndrv->type;
824 	ifdemux->data = ndrv->data.other;
825 	ifdemux->datalen = ndrv->length;
826 
827 	return 0;
828 }
829 
830 int
ndrv_delspec(struct ndrv_cb * np)831 ndrv_delspec(struct ndrv_cb *np)
832 {
833 	int result = 0;
834 
835 	if (np->nd_proto_family == PF_NDRV ||
836 	    np->nd_proto_family == 0) {
837 		return EINVAL;
838 	}
839 
840 	/* Detach the protocol */
841 	result = ifnet_detach_protocol(np->nd_if, np->nd_proto_family);
842 	np->nd_proto_family = PF_NDRV;
843 
844 	return result;
845 }
846 
847 struct ndrv_cb *
ndrv_find_inbound(struct ifnet * ifp,u_int32_t protocol)848 ndrv_find_inbound(struct ifnet *ifp, u_int32_t protocol)
849 {
850 	struct ndrv_cb* np;
851 
852 	LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_OWNED);
853 
854 	if (protocol == PF_NDRV) {
855 		return NULL;
856 	}
857 
858 	TAILQ_FOREACH(np, &ndrvl, nd_next) {
859 		if (np->nd_proto_family == protocol &&
860 		    np->nd_if == ifp) {
861 			return np;
862 		}
863 	}
864 
865 	return NULL;
866 }
867 
868 static void
ndrv_handle_ifp_detach(u_int32_t family,short unit)869 ndrv_handle_ifp_detach(u_int32_t family, short unit)
870 {
871 	struct ndrv_cb* np;
872 	struct ifnet        *ifp = NULL;
873 	struct socket *so;
874 
875 	/* Find all sockets using this interface. */
876 	TAILQ_FOREACH(np, &ndrvl, nd_next) {
877 		if (np->nd_family == family &&
878 		    np->nd_unit == unit) {
879 			/* This cb is using the detaching interface, but not for long. */
880 			/* Let the protocol go */
881 			ifp = np->nd_if;
882 			if (np->nd_proto_family != 0) {
883 				ndrv_delspec(np);
884 			}
885 
886 			/* Delete the multicasts first */
887 			ndrv_remove_all_multicast(np);
888 
889 			/* Disavow all knowledge of the ifp */
890 			np->nd_if = NULL;
891 			np->nd_unit = 0;
892 			np->nd_family = 0;
893 
894 			so = np->nd_socket;
895 			/* Make sure sending returns an error */
896 			LCK_MTX_ASSERT(ndrvdomain->dom_mtx, LCK_MTX_ASSERT_OWNED);
897 			socantsendmore(so);
898 			socantrcvmore(so);
899 		}
900 	}
901 
902 	/* Unregister our protocol */
903 	if (ifp) {
904 		ifnet_detach_protocol(ifp, PF_NDRV);
905 	}
906 }
907 
908 static struct ndrv_multiaddr *
ndrv_multiaddr_alloc(size_t size)909 ndrv_multiaddr_alloc(size_t size)
910 {
911 	struct ndrv_multiaddr *ndrv_multi;
912 
913 	ndrv_multi = kalloc_type(struct ndrv_multiaddr, Z_WAITOK_ZERO_NOFAIL);
914 	ndrv_multi->addr = kalloc_data(size, Z_WAITOK_ZERO_NOFAIL);
915 	return ndrv_multi;
916 }
917 
918 static void
ndrv_multiaddr_free(struct ndrv_multiaddr * ndrv_multi,size_t size)919 ndrv_multiaddr_free(struct ndrv_multiaddr *ndrv_multi, size_t size)
920 {
921 	kfree_data(ndrv_multi->addr, size);
922 	kfree_type(struct ndrv_multiaddr, ndrv_multi);
923 }
924 
925 static int
ndrv_do_add_multicast(struct ndrv_cb * np,struct sockopt * sopt)926 ndrv_do_add_multicast(struct ndrv_cb *np, struct sockopt *sopt)
927 {
928 	struct ndrv_multiaddr *ndrv_multi;
929 	int                    result;
930 
931 	if (sopt->sopt_val == 0 || sopt->sopt_valsize < 2 ||
932 	    sopt->sopt_level != SOL_NDRVPROTO || sopt->sopt_valsize > SOCK_MAXADDRLEN) {
933 		return EINVAL;
934 	}
935 	if (np->nd_if == NULL) {
936 		return ENXIO;
937 	}
938 	if (!(np->nd_dlist_cnt < ndrv_multi_max_count)) {
939 		return EPERM;
940 	}
941 
942 	ndrv_multi = ndrv_multiaddr_alloc(sopt->sopt_valsize);
943 
944 	// Copy in the address
945 	result = copyin(sopt->sopt_val, ndrv_multi->addr, sopt->sopt_valsize);
946 
947 	// Validate the sockaddr
948 	if (result == 0 && sopt->sopt_valsize != ndrv_multi->addr->sa_len) {
949 		result = EINVAL;
950 	}
951 
952 	if (result == 0 && ndrv_have_multicast(np, ndrv_multi->addr)) {
953 		result = EEXIST;
954 	}
955 
956 	if (result == 0) {
957 		// Try adding the multicast
958 		result = ifnet_add_multicast(np->nd_if, ndrv_multi->addr,
959 		    &ndrv_multi->ifma);
960 	}
961 
962 	if (result == 0) {
963 		// Add to our linked list
964 		ndrv_multi->next = np->nd_multiaddrs;
965 		np->nd_multiaddrs = ndrv_multi;
966 		np->nd_dlist_cnt++;
967 	} else {
968 		// Free up the memory, something went wrong
969 		ndrv_multiaddr_free(ndrv_multi, sopt->sopt_valsize);
970 	}
971 
972 	return result;
973 }
974 
975 static void
ndrv_cb_remove_multiaddr(struct ndrv_cb * np,struct ndrv_multiaddr * ndrv_entry)976 ndrv_cb_remove_multiaddr(struct ndrv_cb *np, struct ndrv_multiaddr *ndrv_entry)
977 {
978 	struct ndrv_multiaddr   *cur = np->nd_multiaddrs;
979 	bool                    removed = false;
980 
981 	if (cur == ndrv_entry) {
982 		/* we were the head */
983 		np->nd_multiaddrs = cur->next;
984 		removed = true;
985 	} else {
986 		/* find our entry */
987 		struct ndrv_multiaddr  *cur_next = NULL;
988 
989 		for (; cur != NULL; cur = cur_next) {
990 			cur_next = cur->next;
991 			if (cur_next == ndrv_entry) {
992 				cur->next = cur_next->next;
993 				removed = true;
994 				break;
995 			}
996 		}
997 	}
998 	ASSERT(removed);
999 }
1000 
1001 static int
ndrv_do_remove_multicast(struct ndrv_cb * np,struct sockopt * sopt)1002 ndrv_do_remove_multicast(struct ndrv_cb *np, struct sockopt *sopt)
1003 {
1004 	struct sockaddr*            multi_addr;
1005 	struct ndrv_multiaddr*      ndrv_entry = NULL;
1006 	int                                 result;
1007 
1008 	if (sopt->sopt_val == 0 || sopt->sopt_valsize < 2 ||
1009 	    sopt->sopt_valsize > SOCK_MAXADDRLEN ||
1010 	    sopt->sopt_level != SOL_NDRVPROTO) {
1011 		return EINVAL;
1012 	}
1013 	if (np->nd_if == NULL || np->nd_dlist_cnt == 0) {
1014 		return ENXIO;
1015 	}
1016 
1017 	// Allocate storage
1018 	multi_addr = (struct sockaddr*) kalloc_data(sopt->sopt_valsize, Z_WAITOK);
1019 	if (multi_addr == NULL) {
1020 		return ENOMEM;
1021 	}
1022 
1023 	// Copy in the address
1024 	result = copyin(sopt->sopt_val, multi_addr, sopt->sopt_valsize);
1025 
1026 	// Validate the sockaddr
1027 	if (result == 0 && sopt->sopt_valsize != multi_addr->sa_len) {
1028 		result = EINVAL;
1029 	}
1030 
1031 	if (result == 0) {
1032 		/* Find the old entry */
1033 		ndrv_entry = ndrv_have_multicast(np, multi_addr);
1034 
1035 		if (ndrv_entry == NULL) {
1036 			result = ENOENT;
1037 		}
1038 	}
1039 
1040 	if (result == 0) {
1041 		// Try deleting the multicast
1042 		result = ifnet_remove_multicast(ndrv_entry->ifma);
1043 	}
1044 
1045 	if (result == 0) {
1046 		// Remove from our linked list
1047 		ifmaddr_release(ndrv_entry->ifma);
1048 
1049 		ndrv_cb_remove_multiaddr(np, ndrv_entry);
1050 		np->nd_dlist_cnt--;
1051 
1052 		ndrv_multiaddr_free(ndrv_entry, ndrv_entry->addr->sa_len);
1053 	}
1054 	kfree_data(multi_addr, sopt->sopt_valsize);
1055 
1056 	return result;
1057 }
1058 
1059 static struct ndrv_multiaddr*
ndrv_have_multicast(struct ndrv_cb * np,struct sockaddr * inAddr)1060 ndrv_have_multicast(struct ndrv_cb *np, struct sockaddr* inAddr)
1061 {
1062 	struct ndrv_multiaddr*      cur;
1063 	for (cur = np->nd_multiaddrs; cur != NULL; cur = cur->next) {
1064 		if ((inAddr->sa_len == cur->addr->sa_len) &&
1065 		    (bcmp(cur->addr, inAddr, inAddr->sa_len) == 0)) {
1066 			// Found a match
1067 			return cur;
1068 		}
1069 	}
1070 
1071 	return NULL;
1072 }
1073 
1074 static void
ndrv_remove_all_multicast(struct ndrv_cb * np)1075 ndrv_remove_all_multicast(struct ndrv_cb* np)
1076 {
1077 	struct ndrv_multiaddr*      cur;
1078 
1079 	if (np->nd_if != NULL) {
1080 		while (np->nd_multiaddrs != NULL) {
1081 			cur = np->nd_multiaddrs;
1082 			np->nd_multiaddrs = cur->next;
1083 
1084 			ifnet_remove_multicast(cur->ifma);
1085 			ifmaddr_release(cur->ifma);
1086 			ndrv_multiaddr_free(cur, cur->addr->sa_len);
1087 		}
1088 	}
1089 }
1090 
1091 static struct pr_usrreqs ndrv_usrreqs = {
1092 	.pru_abort =            ndrv_abort,
1093 	.pru_attach =           ndrv_attach,
1094 	.pru_bind =             ndrv_bind,
1095 	.pru_connect =          ndrv_connect,
1096 	.pru_detach =           ndrv_detach,
1097 	.pru_disconnect =       ndrv_disconnect,
1098 	.pru_peeraddr =         ndrv_peeraddr,
1099 	.pru_send =             ndrv_send,
1100 	.pru_shutdown =         ndrv_shutdown,
1101 	.pru_sockaddr =         ndrv_sockaddr,
1102 	.pru_sosend =           sosend,
1103 	.pru_soreceive =        soreceive,
1104 };
1105 
1106 static struct protosw ndrvsw[] = {
1107 	{
1108 		.pr_type =              SOCK_RAW,
1109 		.pr_protocol =          NDRVPROTO_NDRV,
1110 		.pr_flags =             PR_ATOMIC | PR_ADDR,
1111 		.pr_output =            ndrv_output,
1112 		.pr_ctloutput =         ndrv_ctloutput,
1113 		.pr_usrreqs =           &ndrv_usrreqs,
1114 	}
1115 };
1116 
1117 static int ndrv_proto_count = (sizeof(ndrvsw) / sizeof(struct protosw));
1118 
1119 struct domain ndrvdomain_s = {
1120 	.dom_family =           PF_NDRV,
1121 	.dom_name =             "NetDriver",
1122 	.dom_init =             ndrv_dominit,
1123 };
1124 
1125 static void
ndrv_dominit(struct domain * dp)1126 ndrv_dominit(struct domain *dp)
1127 {
1128 	struct protosw *pr;
1129 	int i;
1130 
1131 	VERIFY(!(dp->dom_flags & DOM_INITIALIZED));
1132 	VERIFY(ndrvdomain == NULL);
1133 
1134 	ndrvdomain = dp;
1135 
1136 	for (i = 0, pr = &ndrvsw[0]; i < ndrv_proto_count; i++, pr++) {
1137 		net_add_proto(pr, dp, 1);
1138 	}
1139 }
1140