xref: /linux/drivers/net/ppp/pppoe.c (revision 8c749ce93ee69e789e46b3be98de9e0cbfcf8ed8)
1 /** -*- linux-c -*- ***********************************************************
2  * Linux PPP over Ethernet (PPPoX/PPPoE) Sockets
3  *
4  * PPPoX --- Generic PPP encapsulation socket family
5  * PPPoE --- PPP over Ethernet (RFC 2516)
6  *
7  *
8  * Version:	0.7.0
9  *
10  * 070228 :	Fix to allow multiple sessions with same remote MAC and same
11  *		session id by including the local device ifindex in the
12  *		tuple identifying a session. This also ensures packets can't
13  *		be injected into a session from interfaces other than the one
14  *		specified by userspace. Florian Zumbiehl <florz@florz.de>
15  *		(Oh, BTW, this one is YYMMDD, in case you were wondering ...)
16  * 220102 :	Fix module use count on failure in pppoe_create, pppox_sk -acme
17  * 030700 :	Fixed connect logic to allow for disconnect.
18  * 270700 :	Fixed potential SMP problems; we must protect against
19  *		simultaneous invocation of ppp_input
20  *		and ppp_unregister_channel.
21  * 040800 :	Respect reference count mechanisms on net-devices.
22  * 200800 :	fix kfree(skb) in pppoe_rcv (acme)
23  *		Module reference count is decremented in the right spot now,
24  *		guards against sock_put not actually freeing the sk
25  *		in pppoe_release.
26  * 051000 :	Initialization cleanup.
27  * 111100 :	Fix recvmsg.
28  * 050101 :	Fix PADT procesing.
29  * 140501 :	Use pppoe_rcv_core to handle all backlog. (Alexey)
30  * 170701 :	Do not lock_sock with rwlock held. (DaveM)
31  *		Ignore discovery frames if user has socket
32  *		locked. (DaveM)
33  *		Ignore return value of dev_queue_xmit in __pppoe_xmit
34  *		or else we may kfree an SKB twice. (DaveM)
35  * 190701 :	When doing copies of skb's in __pppoe_xmit, always delete
36  *		the original skb that was passed in on success, never on
37  *		failure.  Delete the copy of the skb on failure to avoid
38  *		a memory leak.
39  * 081001 :	Misc. cleanup (licence string, non-blocking, prevent
40  *		reference of device on close).
41  * 121301 :	New ppp channels interface; cannot unregister a channel
42  *		from interrupts.  Thus, we mark the socket as a ZOMBIE
43  *		and do the unregistration later.
44  * 081002 :	seq_file support for proc stuff -acme
45  * 111602 :	Merge all 2.4 fixes into 2.5/2.6 tree.  Label 2.5/2.6
46  *		as version 0.7.  Spacing cleanup.
47  * Author:	Michal Ostrowski <mostrows@speakeasy.net>
48  * Contributors:
49  * 		Arnaldo Carvalho de Melo <acme@conectiva.com.br>
50  *		David S. Miller (davem@redhat.com)
51  *
52  * License:
53  *		This program is free software; you can redistribute it and/or
54  *		modify it under the terms of the GNU General Public License
55  *		as published by the Free Software Foundation; either version
56  *		2 of the License, or (at your option) any later version.
57  *
58  */
59 
60 #include <linux/string.h>
61 #include <linux/module.h>
62 #include <linux/kernel.h>
63 #include <linux/slab.h>
64 #include <linux/errno.h>
65 #include <linux/netdevice.h>
66 #include <linux/net.h>
67 #include <linux/inetdevice.h>
68 #include <linux/etherdevice.h>
69 #include <linux/skbuff.h>
70 #include <linux/init.h>
71 #include <linux/if_ether.h>
72 #include <linux/if_pppox.h>
73 #include <linux/ppp_channel.h>
74 #include <linux/ppp_defs.h>
75 #include <linux/ppp-ioctl.h>
76 #include <linux/notifier.h>
77 #include <linux/file.h>
78 #include <linux/proc_fs.h>
79 #include <linux/seq_file.h>
80 
81 #include <linux/nsproxy.h>
82 #include <net/net_namespace.h>
83 #include <net/netns/generic.h>
84 #include <net/sock.h>
85 
86 #include <asm/uaccess.h>
87 
88 #define PPPOE_HASH_BITS 4
89 #define PPPOE_HASH_SIZE (1 << PPPOE_HASH_BITS)
90 #define PPPOE_HASH_MASK	(PPPOE_HASH_SIZE - 1)
91 
92 static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb);
93 
94 static const struct proto_ops pppoe_ops;
95 static const struct ppp_channel_ops pppoe_chan_ops;
96 
97 /* per-net private data for this module */
98 static int pppoe_net_id __read_mostly;
99 struct pppoe_net {
100 	/*
101 	 * we could use _single_ hash table for all
102 	 * nets by injecting net id into the hash but
103 	 * it would increase hash chains and add
104 	 * a few additional math comparations messy
105 	 * as well, moreover in case of SMP less locking
106 	 * controversy here
107 	 */
108 	struct pppox_sock *hash_table[PPPOE_HASH_SIZE];
109 	rwlock_t hash_lock;
110 };
111 
112 /*
113  * PPPoE could be in the following stages:
114  * 1) Discovery stage (to obtain remote MAC and Session ID)
115  * 2) Session stage (MAC and SID are known)
116  *
117  * Ethernet frames have a special tag for this but
118  * we use simpler approach based on session id
119  */
120 static inline bool stage_session(__be16 sid)
121 {
122 	return sid != 0;
123 }
124 
125 static inline struct pppoe_net *pppoe_pernet(struct net *net)
126 {
127 	BUG_ON(!net);
128 
129 	return net_generic(net, pppoe_net_id);
130 }
131 
132 static inline int cmp_2_addr(struct pppoe_addr *a, struct pppoe_addr *b)
133 {
134 	return a->sid == b->sid && ether_addr_equal(a->remote, b->remote);
135 }
136 
137 static inline int cmp_addr(struct pppoe_addr *a, __be16 sid, char *addr)
138 {
139 	return a->sid == sid && ether_addr_equal(a->remote, addr);
140 }
141 
142 #if 8 % PPPOE_HASH_BITS
143 #error 8 must be a multiple of PPPOE_HASH_BITS
144 #endif
145 
146 static int hash_item(__be16 sid, unsigned char *addr)
147 {
148 	unsigned char hash = 0;
149 	unsigned int i;
150 
151 	for (i = 0; i < ETH_ALEN; i++)
152 		hash ^= addr[i];
153 	for (i = 0; i < sizeof(sid_t) * 8; i += 8)
154 		hash ^= (__force __u32)sid >> i;
155 	for (i = 8; (i >>= 1) >= PPPOE_HASH_BITS;)
156 		hash ^= hash >> i;
157 
158 	return hash & PPPOE_HASH_MASK;
159 }
160 
161 /**********************************************************************
162  *
163  *  Set/get/delete/rehash items  (internal versions)
164  *
165  **********************************************************************/
166 static struct pppox_sock *__get_item(struct pppoe_net *pn, __be16 sid,
167 				unsigned char *addr, int ifindex)
168 {
169 	int hash = hash_item(sid, addr);
170 	struct pppox_sock *ret;
171 
172 	ret = pn->hash_table[hash];
173 	while (ret) {
174 		if (cmp_addr(&ret->pppoe_pa, sid, addr) &&
175 		    ret->pppoe_ifindex == ifindex)
176 			return ret;
177 
178 		ret = ret->next;
179 	}
180 
181 	return NULL;
182 }
183 
184 static int __set_item(struct pppoe_net *pn, struct pppox_sock *po)
185 {
186 	int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
187 	struct pppox_sock *ret;
188 
189 	ret = pn->hash_table[hash];
190 	while (ret) {
191 		if (cmp_2_addr(&ret->pppoe_pa, &po->pppoe_pa) &&
192 		    ret->pppoe_ifindex == po->pppoe_ifindex)
193 			return -EALREADY;
194 
195 		ret = ret->next;
196 	}
197 
198 	po->next = pn->hash_table[hash];
199 	pn->hash_table[hash] = po;
200 
201 	return 0;
202 }
203 
204 static void __delete_item(struct pppoe_net *pn, __be16 sid,
205 					char *addr, int ifindex)
206 {
207 	int hash = hash_item(sid, addr);
208 	struct pppox_sock *ret, **src;
209 
210 	ret = pn->hash_table[hash];
211 	src = &pn->hash_table[hash];
212 
213 	while (ret) {
214 		if (cmp_addr(&ret->pppoe_pa, sid, addr) &&
215 		    ret->pppoe_ifindex == ifindex) {
216 			*src = ret->next;
217 			break;
218 		}
219 
220 		src = &ret->next;
221 		ret = ret->next;
222 	}
223 }
224 
225 /**********************************************************************
226  *
227  *  Set/get/delete/rehash items
228  *
229  **********************************************************************/
230 static inline struct pppox_sock *get_item(struct pppoe_net *pn, __be16 sid,
231 					unsigned char *addr, int ifindex)
232 {
233 	struct pppox_sock *po;
234 
235 	read_lock_bh(&pn->hash_lock);
236 	po = __get_item(pn, sid, addr, ifindex);
237 	if (po)
238 		sock_hold(sk_pppox(po));
239 	read_unlock_bh(&pn->hash_lock);
240 
241 	return po;
242 }
243 
244 static inline struct pppox_sock *get_item_by_addr(struct net *net,
245 						struct sockaddr_pppox *sp)
246 {
247 	struct net_device *dev;
248 	struct pppoe_net *pn;
249 	struct pppox_sock *pppox_sock = NULL;
250 
251 	int ifindex;
252 
253 	rcu_read_lock();
254 	dev = dev_get_by_name_rcu(net, sp->sa_addr.pppoe.dev);
255 	if (dev) {
256 		ifindex = dev->ifindex;
257 		pn = pppoe_pernet(net);
258 		pppox_sock = get_item(pn, sp->sa_addr.pppoe.sid,
259 				sp->sa_addr.pppoe.remote, ifindex);
260 	}
261 	rcu_read_unlock();
262 	return pppox_sock;
263 }
264 
265 static inline void delete_item(struct pppoe_net *pn, __be16 sid,
266 					char *addr, int ifindex)
267 {
268 	write_lock_bh(&pn->hash_lock);
269 	__delete_item(pn, sid, addr, ifindex);
270 	write_unlock_bh(&pn->hash_lock);
271 }
272 
273 /***************************************************************************
274  *
275  *  Handler for device events.
276  *  Certain device events require that sockets be unconnected.
277  *
278  **************************************************************************/
279 
280 static void pppoe_flush_dev(struct net_device *dev)
281 {
282 	struct pppoe_net *pn;
283 	int i;
284 
285 	pn = pppoe_pernet(dev_net(dev));
286 	write_lock_bh(&pn->hash_lock);
287 	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
288 		struct pppox_sock *po = pn->hash_table[i];
289 		struct sock *sk;
290 
291 		while (po) {
292 			while (po && po->pppoe_dev != dev) {
293 				po = po->next;
294 			}
295 
296 			if (!po)
297 				break;
298 
299 			sk = sk_pppox(po);
300 
301 			/* We always grab the socket lock, followed by the
302 			 * hash_lock, in that order.  Since we should hold the
303 			 * sock lock while doing any unbinding, we need to
304 			 * release the lock we're holding.  Hold a reference to
305 			 * the sock so it doesn't disappear as we're jumping
306 			 * between locks.
307 			 */
308 
309 			sock_hold(sk);
310 			write_unlock_bh(&pn->hash_lock);
311 			lock_sock(sk);
312 
313 			if (po->pppoe_dev == dev &&
314 			    sk->sk_state & (PPPOX_CONNECTED | PPPOX_BOUND)) {
315 				pppox_unbind_sock(sk);
316 				sk->sk_state_change(sk);
317 				po->pppoe_dev = NULL;
318 				dev_put(dev);
319 			}
320 
321 			release_sock(sk);
322 			sock_put(sk);
323 
324 			/* Restart the process from the start of the current
325 			 * hash chain. We dropped locks so the world may have
326 			 * change from underneath us.
327 			 */
328 
329 			BUG_ON(pppoe_pernet(dev_net(dev)) == NULL);
330 			write_lock_bh(&pn->hash_lock);
331 			po = pn->hash_table[i];
332 		}
333 	}
334 	write_unlock_bh(&pn->hash_lock);
335 }
336 
337 static int pppoe_device_event(struct notifier_block *this,
338 			      unsigned long event, void *ptr)
339 {
340 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
341 
342 	/* Only look at sockets that are using this specific device. */
343 	switch (event) {
344 	case NETDEV_CHANGEADDR:
345 	case NETDEV_CHANGEMTU:
346 		/* A change in mtu or address is a bad thing, requiring
347 		 * LCP re-negotiation.
348 		 */
349 
350 	case NETDEV_GOING_DOWN:
351 	case NETDEV_DOWN:
352 		/* Find every socket on this device and kill it. */
353 		pppoe_flush_dev(dev);
354 		break;
355 
356 	default:
357 		break;
358 	}
359 
360 	return NOTIFY_DONE;
361 }
362 
363 static struct notifier_block pppoe_notifier = {
364 	.notifier_call = pppoe_device_event,
365 };
366 
367 /************************************************************************
368  *
369  * Do the real work of receiving a PPPoE Session frame.
370  *
371  ***********************************************************************/
372 static int pppoe_rcv_core(struct sock *sk, struct sk_buff *skb)
373 {
374 	struct pppox_sock *po = pppox_sk(sk);
375 	struct pppox_sock *relay_po;
376 
377 	/* Backlog receive. Semantics of backlog rcv preclude any code from
378 	 * executing in lock_sock()/release_sock() bounds; meaning sk->sk_state
379 	 * can't change.
380 	 */
381 
382 	if (skb->pkt_type == PACKET_OTHERHOST)
383 		goto abort_kfree;
384 
385 	if (sk->sk_state & PPPOX_BOUND) {
386 		ppp_input(&po->chan, skb);
387 	} else if (sk->sk_state & PPPOX_RELAY) {
388 		relay_po = get_item_by_addr(sock_net(sk),
389 					    &po->pppoe_relay);
390 		if (relay_po == NULL)
391 			goto abort_kfree;
392 
393 		if ((sk_pppox(relay_po)->sk_state & PPPOX_CONNECTED) == 0)
394 			goto abort_put;
395 
396 		if (!__pppoe_xmit(sk_pppox(relay_po), skb))
397 			goto abort_put;
398 	} else {
399 		if (sock_queue_rcv_skb(sk, skb))
400 			goto abort_kfree;
401 	}
402 
403 	return NET_RX_SUCCESS;
404 
405 abort_put:
406 	sock_put(sk_pppox(relay_po));
407 
408 abort_kfree:
409 	kfree_skb(skb);
410 	return NET_RX_DROP;
411 }
412 
413 /************************************************************************
414  *
415  * Receive wrapper called in BH context.
416  *
417  ***********************************************************************/
418 static int pppoe_rcv(struct sk_buff *skb, struct net_device *dev,
419 		     struct packet_type *pt, struct net_device *orig_dev)
420 {
421 	struct pppoe_hdr *ph;
422 	struct pppox_sock *po;
423 	struct pppoe_net *pn;
424 	int len;
425 
426 	skb = skb_share_check(skb, GFP_ATOMIC);
427 	if (!skb)
428 		goto out;
429 
430 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
431 		goto drop;
432 
433 	ph = pppoe_hdr(skb);
434 	len = ntohs(ph->length);
435 
436 	skb_pull_rcsum(skb, sizeof(*ph));
437 	if (skb->len < len)
438 		goto drop;
439 
440 	if (pskb_trim_rcsum(skb, len))
441 		goto drop;
442 
443 	pn = pppoe_pernet(dev_net(dev));
444 
445 	/* Note that get_item does a sock_hold(), so sk_pppox(po)
446 	 * is known to be safe.
447 	 */
448 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
449 	if (!po)
450 		goto drop;
451 
452 	return sk_receive_skb(sk_pppox(po), skb, 0);
453 
454 drop:
455 	kfree_skb(skb);
456 out:
457 	return NET_RX_DROP;
458 }
459 
460 static void pppoe_unbind_sock_work(struct work_struct *work)
461 {
462 	struct pppox_sock *po = container_of(work, struct pppox_sock,
463 					     proto.pppoe.padt_work);
464 	struct sock *sk = sk_pppox(po);
465 
466 	lock_sock(sk);
467 	if (po->pppoe_dev) {
468 		dev_put(po->pppoe_dev);
469 		po->pppoe_dev = NULL;
470 	}
471 	pppox_unbind_sock(sk);
472 	release_sock(sk);
473 	sock_put(sk);
474 }
475 
476 /************************************************************************
477  *
478  * Receive a PPPoE Discovery frame.
479  * This is solely for detection of PADT frames
480  *
481  ***********************************************************************/
482 static int pppoe_disc_rcv(struct sk_buff *skb, struct net_device *dev,
483 			  struct packet_type *pt, struct net_device *orig_dev)
484 
485 {
486 	struct pppoe_hdr *ph;
487 	struct pppox_sock *po;
488 	struct pppoe_net *pn;
489 
490 	skb = skb_share_check(skb, GFP_ATOMIC);
491 	if (!skb)
492 		goto out;
493 
494 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
495 		goto abort;
496 
497 	ph = pppoe_hdr(skb);
498 	if (ph->code != PADT_CODE)
499 		goto abort;
500 
501 	pn = pppoe_pernet(dev_net(dev));
502 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
503 	if (po)
504 		if (!schedule_work(&po->proto.pppoe.padt_work))
505 			sock_put(sk_pppox(po));
506 
507 abort:
508 	kfree_skb(skb);
509 out:
510 	return NET_RX_SUCCESS; /* Lies... :-) */
511 }
512 
513 static struct packet_type pppoes_ptype __read_mostly = {
514 	.type	= cpu_to_be16(ETH_P_PPP_SES),
515 	.func	= pppoe_rcv,
516 };
517 
518 static struct packet_type pppoed_ptype __read_mostly = {
519 	.type	= cpu_to_be16(ETH_P_PPP_DISC),
520 	.func	= pppoe_disc_rcv,
521 };
522 
523 static struct proto pppoe_sk_proto __read_mostly = {
524 	.name	  = "PPPOE",
525 	.owner	  = THIS_MODULE,
526 	.obj_size = sizeof(struct pppox_sock),
527 };
528 
529 /***********************************************************************
530  *
531  * Initialize a new struct sock.
532  *
533  **********************************************************************/
534 static int pppoe_create(struct net *net, struct socket *sock, int kern)
535 {
536 	struct sock *sk;
537 
538 	sk = sk_alloc(net, PF_PPPOX, GFP_KERNEL, &pppoe_sk_proto, kern);
539 	if (!sk)
540 		return -ENOMEM;
541 
542 	sock_init_data(sock, sk);
543 
544 	sock->state	= SS_UNCONNECTED;
545 	sock->ops	= &pppoe_ops;
546 
547 	sk->sk_backlog_rcv	= pppoe_rcv_core;
548 	sk->sk_state		= PPPOX_NONE;
549 	sk->sk_type		= SOCK_STREAM;
550 	sk->sk_family		= PF_PPPOX;
551 	sk->sk_protocol		= PX_PROTO_OE;
552 
553 	INIT_WORK(&pppox_sk(sk)->proto.pppoe.padt_work,
554 		  pppoe_unbind_sock_work);
555 
556 	return 0;
557 }
558 
559 static int pppoe_release(struct socket *sock)
560 {
561 	struct sock *sk = sock->sk;
562 	struct pppox_sock *po;
563 	struct pppoe_net *pn;
564 	struct net *net = NULL;
565 
566 	if (!sk)
567 		return 0;
568 
569 	lock_sock(sk);
570 	if (sock_flag(sk, SOCK_DEAD)) {
571 		release_sock(sk);
572 		return -EBADF;
573 	}
574 
575 	po = pppox_sk(sk);
576 
577 	if (po->pppoe_dev) {
578 		dev_put(po->pppoe_dev);
579 		po->pppoe_dev = NULL;
580 	}
581 
582 	pppox_unbind_sock(sk);
583 
584 	/* Signal the death of the socket. */
585 	sk->sk_state = PPPOX_DEAD;
586 
587 	net = sock_net(sk);
588 	pn = pppoe_pernet(net);
589 
590 	/*
591 	 * protect "po" from concurrent updates
592 	 * on pppoe_flush_dev
593 	 */
594 	delete_item(pn, po->pppoe_pa.sid, po->pppoe_pa.remote,
595 		    po->pppoe_ifindex);
596 
597 	sock_orphan(sk);
598 	sock->sk = NULL;
599 
600 	skb_queue_purge(&sk->sk_receive_queue);
601 	release_sock(sk);
602 	sock_put(sk);
603 
604 	return 0;
605 }
606 
607 static int pppoe_connect(struct socket *sock, struct sockaddr *uservaddr,
608 		  int sockaddr_len, int flags)
609 {
610 	struct sock *sk = sock->sk;
611 	struct sockaddr_pppox *sp = (struct sockaddr_pppox *)uservaddr;
612 	struct pppox_sock *po = pppox_sk(sk);
613 	struct net_device *dev = NULL;
614 	struct pppoe_net *pn;
615 	struct net *net = NULL;
616 	int error;
617 
618 	lock_sock(sk);
619 
620 	error = -EINVAL;
621 	if (sp->sa_protocol != PX_PROTO_OE)
622 		goto end;
623 
624 	/* Check for already bound sockets */
625 	error = -EBUSY;
626 	if ((sk->sk_state & PPPOX_CONNECTED) &&
627 	     stage_session(sp->sa_addr.pppoe.sid))
628 		goto end;
629 
630 	/* Check for already disconnected sockets, on attempts to disconnect */
631 	error = -EALREADY;
632 	if ((sk->sk_state & PPPOX_DEAD) &&
633 	     !stage_session(sp->sa_addr.pppoe.sid))
634 		goto end;
635 
636 	error = 0;
637 
638 	/* Delete the old binding */
639 	if (stage_session(po->pppoe_pa.sid)) {
640 		pppox_unbind_sock(sk);
641 		pn = pppoe_pernet(sock_net(sk));
642 		delete_item(pn, po->pppoe_pa.sid,
643 			    po->pppoe_pa.remote, po->pppoe_ifindex);
644 		if (po->pppoe_dev) {
645 			dev_put(po->pppoe_dev);
646 			po->pppoe_dev = NULL;
647 		}
648 
649 		po->pppoe_ifindex = 0;
650 		memset(&po->pppoe_pa, 0, sizeof(po->pppoe_pa));
651 		memset(&po->pppoe_relay, 0, sizeof(po->pppoe_relay));
652 		memset(&po->chan, 0, sizeof(po->chan));
653 		po->next = NULL;
654 		po->num = 0;
655 
656 		sk->sk_state = PPPOX_NONE;
657 	}
658 
659 	/* Re-bind in session stage only */
660 	if (stage_session(sp->sa_addr.pppoe.sid)) {
661 		error = -ENODEV;
662 		net = sock_net(sk);
663 		dev = dev_get_by_name(net, sp->sa_addr.pppoe.dev);
664 		if (!dev)
665 			goto err_put;
666 
667 		po->pppoe_dev = dev;
668 		po->pppoe_ifindex = dev->ifindex;
669 		pn = pppoe_pernet(net);
670 		if (!(dev->flags & IFF_UP)) {
671 			goto err_put;
672 		}
673 
674 		memcpy(&po->pppoe_pa,
675 		       &sp->sa_addr.pppoe,
676 		       sizeof(struct pppoe_addr));
677 
678 		write_lock_bh(&pn->hash_lock);
679 		error = __set_item(pn, po);
680 		write_unlock_bh(&pn->hash_lock);
681 		if (error < 0)
682 			goto err_put;
683 
684 		po->chan.hdrlen = (sizeof(struct pppoe_hdr) +
685 				   dev->hard_header_len);
686 
687 		po->chan.mtu = dev->mtu - sizeof(struct pppoe_hdr) - 2;
688 		po->chan.private = sk;
689 		po->chan.ops = &pppoe_chan_ops;
690 
691 		error = ppp_register_net_channel(dev_net(dev), &po->chan);
692 		if (error) {
693 			delete_item(pn, po->pppoe_pa.sid,
694 				    po->pppoe_pa.remote, po->pppoe_ifindex);
695 			goto err_put;
696 		}
697 
698 		sk->sk_state = PPPOX_CONNECTED;
699 	}
700 
701 	po->num = sp->sa_addr.pppoe.sid;
702 
703 end:
704 	release_sock(sk);
705 	return error;
706 err_put:
707 	if (po->pppoe_dev) {
708 		dev_put(po->pppoe_dev);
709 		po->pppoe_dev = NULL;
710 	}
711 	goto end;
712 }
713 
714 static int pppoe_getname(struct socket *sock, struct sockaddr *uaddr,
715 		  int *usockaddr_len, int peer)
716 {
717 	int len = sizeof(struct sockaddr_pppox);
718 	struct sockaddr_pppox sp;
719 
720 	sp.sa_family	= AF_PPPOX;
721 	sp.sa_protocol	= PX_PROTO_OE;
722 	memcpy(&sp.sa_addr.pppoe, &pppox_sk(sock->sk)->pppoe_pa,
723 	       sizeof(struct pppoe_addr));
724 
725 	memcpy(uaddr, &sp, len);
726 
727 	*usockaddr_len = len;
728 
729 	return 0;
730 }
731 
732 static int pppoe_ioctl(struct socket *sock, unsigned int cmd,
733 		unsigned long arg)
734 {
735 	struct sock *sk = sock->sk;
736 	struct pppox_sock *po = pppox_sk(sk);
737 	int val;
738 	int err;
739 
740 	switch (cmd) {
741 	case PPPIOCGMRU:
742 		err = -ENXIO;
743 		if (!(sk->sk_state & PPPOX_CONNECTED))
744 			break;
745 
746 		err = -EFAULT;
747 		if (put_user(po->pppoe_dev->mtu -
748 			     sizeof(struct pppoe_hdr) -
749 			     PPP_HDRLEN,
750 			     (int __user *)arg))
751 			break;
752 		err = 0;
753 		break;
754 
755 	case PPPIOCSMRU:
756 		err = -ENXIO;
757 		if (!(sk->sk_state & PPPOX_CONNECTED))
758 			break;
759 
760 		err = -EFAULT;
761 		if (get_user(val, (int __user *)arg))
762 			break;
763 
764 		if (val < (po->pppoe_dev->mtu
765 			   - sizeof(struct pppoe_hdr)
766 			   - PPP_HDRLEN))
767 			err = 0;
768 		else
769 			err = -EINVAL;
770 		break;
771 
772 	case PPPIOCSFLAGS:
773 		err = -EFAULT;
774 		if (get_user(val, (int __user *)arg))
775 			break;
776 		err = 0;
777 		break;
778 
779 	case PPPOEIOCSFWD:
780 	{
781 		struct pppox_sock *relay_po;
782 
783 		err = -EBUSY;
784 		if (sk->sk_state & (PPPOX_BOUND | PPPOX_DEAD))
785 			break;
786 
787 		err = -ENOTCONN;
788 		if (!(sk->sk_state & PPPOX_CONNECTED))
789 			break;
790 
791 		/* PPPoE address from the user specifies an outbound
792 		   PPPoE address which frames are forwarded to */
793 		err = -EFAULT;
794 		if (copy_from_user(&po->pppoe_relay,
795 				   (void __user *)arg,
796 				   sizeof(struct sockaddr_pppox)))
797 			break;
798 
799 		err = -EINVAL;
800 		if (po->pppoe_relay.sa_family != AF_PPPOX ||
801 		    po->pppoe_relay.sa_protocol != PX_PROTO_OE)
802 			break;
803 
804 		/* Check that the socket referenced by the address
805 		   actually exists. */
806 		relay_po = get_item_by_addr(sock_net(sk), &po->pppoe_relay);
807 		if (!relay_po)
808 			break;
809 
810 		sock_put(sk_pppox(relay_po));
811 		sk->sk_state |= PPPOX_RELAY;
812 		err = 0;
813 		break;
814 	}
815 
816 	case PPPOEIOCDFWD:
817 		err = -EALREADY;
818 		if (!(sk->sk_state & PPPOX_RELAY))
819 			break;
820 
821 		sk->sk_state &= ~PPPOX_RELAY;
822 		err = 0;
823 		break;
824 
825 	default:
826 		err = -ENOTTY;
827 	}
828 
829 	return err;
830 }
831 
832 static int pppoe_sendmsg(struct socket *sock, struct msghdr *m,
833 			 size_t total_len)
834 {
835 	struct sk_buff *skb;
836 	struct sock *sk = sock->sk;
837 	struct pppox_sock *po = pppox_sk(sk);
838 	int error;
839 	struct pppoe_hdr hdr;
840 	struct pppoe_hdr *ph;
841 	struct net_device *dev;
842 	char *start;
843 
844 	lock_sock(sk);
845 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED)) {
846 		error = -ENOTCONN;
847 		goto end;
848 	}
849 
850 	hdr.ver = 1;
851 	hdr.type = 1;
852 	hdr.code = 0;
853 	hdr.sid = po->num;
854 
855 	dev = po->pppoe_dev;
856 
857 	error = -EMSGSIZE;
858 	if (total_len > (dev->mtu + dev->hard_header_len))
859 		goto end;
860 
861 
862 	skb = sock_wmalloc(sk, total_len + dev->hard_header_len + 32,
863 			   0, GFP_KERNEL);
864 	if (!skb) {
865 		error = -ENOMEM;
866 		goto end;
867 	}
868 
869 	/* Reserve space for headers. */
870 	skb_reserve(skb, dev->hard_header_len);
871 	skb_reset_network_header(skb);
872 
873 	skb->dev = dev;
874 
875 	skb->priority = sk->sk_priority;
876 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
877 
878 	ph = (struct pppoe_hdr *)skb_put(skb, total_len + sizeof(struct pppoe_hdr));
879 	start = (char *)&ph->tag[0];
880 
881 	error = memcpy_from_msg(start, m, total_len);
882 	if (error < 0) {
883 		kfree_skb(skb);
884 		goto end;
885 	}
886 
887 	error = total_len;
888 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
889 			po->pppoe_pa.remote, NULL, total_len);
890 
891 	memcpy(ph, &hdr, sizeof(struct pppoe_hdr));
892 
893 	ph->length = htons(total_len);
894 
895 	dev_queue_xmit(skb);
896 
897 end:
898 	release_sock(sk);
899 	return error;
900 }
901 
902 /************************************************************************
903  *
904  * xmit function for internal use.
905  *
906  ***********************************************************************/
907 static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb)
908 {
909 	struct pppox_sock *po = pppox_sk(sk);
910 	struct net_device *dev = po->pppoe_dev;
911 	struct pppoe_hdr *ph;
912 	int data_len = skb->len;
913 
914 	/* The higher-level PPP code (ppp_unregister_channel()) ensures the PPP
915 	 * xmit operations conclude prior to an unregistration call.  Thus
916 	 * sk->sk_state cannot change, so we don't need to do lock_sock().
917 	 * But, we also can't do a lock_sock since that introduces a potential
918 	 * deadlock as we'd reverse the lock ordering used when calling
919 	 * ppp_unregister_channel().
920 	 */
921 
922 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED))
923 		goto abort;
924 
925 	if (!dev)
926 		goto abort;
927 
928 	/* Copy the data if there is no space for the header or if it's
929 	 * read-only.
930 	 */
931 	if (skb_cow_head(skb, sizeof(*ph) + dev->hard_header_len))
932 		goto abort;
933 
934 	__skb_push(skb, sizeof(*ph));
935 	skb_reset_network_header(skb);
936 
937 	ph = pppoe_hdr(skb);
938 	ph->ver	= 1;
939 	ph->type = 1;
940 	ph->code = 0;
941 	ph->sid	= po->num;
942 	ph->length = htons(data_len);
943 
944 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
945 	skb->dev = dev;
946 
947 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
948 			po->pppoe_pa.remote, NULL, data_len);
949 
950 	dev_queue_xmit(skb);
951 	return 1;
952 
953 abort:
954 	kfree_skb(skb);
955 	return 1;
956 }
957 
958 /************************************************************************
959  *
960  * xmit function called by generic PPP driver
961  * sends PPP frame over PPPoE socket
962  *
963  ***********************************************************************/
964 static int pppoe_xmit(struct ppp_channel *chan, struct sk_buff *skb)
965 {
966 	struct sock *sk = (struct sock *)chan->private;
967 	return __pppoe_xmit(sk, skb);
968 }
969 
970 static const struct ppp_channel_ops pppoe_chan_ops = {
971 	.start_xmit = pppoe_xmit,
972 };
973 
974 static int pppoe_recvmsg(struct socket *sock, struct msghdr *m,
975 			 size_t total_len, int flags)
976 {
977 	struct sock *sk = sock->sk;
978 	struct sk_buff *skb;
979 	int error = 0;
980 
981 	if (sk->sk_state & PPPOX_BOUND) {
982 		error = -EIO;
983 		goto end;
984 	}
985 
986 	skb = skb_recv_datagram(sk, flags & ~MSG_DONTWAIT,
987 				flags & MSG_DONTWAIT, &error);
988 	if (error < 0)
989 		goto end;
990 
991 	if (skb) {
992 		total_len = min_t(size_t, total_len, skb->len);
993 		error = skb_copy_datagram_msg(skb, 0, m, total_len);
994 		if (error == 0) {
995 			consume_skb(skb);
996 			return total_len;
997 		}
998 	}
999 
1000 	kfree_skb(skb);
1001 end:
1002 	return error;
1003 }
1004 
1005 #ifdef CONFIG_PROC_FS
1006 static int pppoe_seq_show(struct seq_file *seq, void *v)
1007 {
1008 	struct pppox_sock *po;
1009 	char *dev_name;
1010 
1011 	if (v == SEQ_START_TOKEN) {
1012 		seq_puts(seq, "Id       Address              Device\n");
1013 		goto out;
1014 	}
1015 
1016 	po = v;
1017 	dev_name = po->pppoe_pa.dev;
1018 
1019 	seq_printf(seq, "%08X %pM %8s\n",
1020 		po->pppoe_pa.sid, po->pppoe_pa.remote, dev_name);
1021 out:
1022 	return 0;
1023 }
1024 
1025 static inline struct pppox_sock *pppoe_get_idx(struct pppoe_net *pn, loff_t pos)
1026 {
1027 	struct pppox_sock *po;
1028 	int i;
1029 
1030 	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
1031 		po = pn->hash_table[i];
1032 		while (po) {
1033 			if (!pos--)
1034 				goto out;
1035 			po = po->next;
1036 		}
1037 	}
1038 
1039 out:
1040 	return po;
1041 }
1042 
1043 static void *pppoe_seq_start(struct seq_file *seq, loff_t *pos)
1044 	__acquires(pn->hash_lock)
1045 {
1046 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1047 	loff_t l = *pos;
1048 
1049 	read_lock_bh(&pn->hash_lock);
1050 	return l ? pppoe_get_idx(pn, --l) : SEQ_START_TOKEN;
1051 }
1052 
1053 static void *pppoe_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1054 {
1055 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1056 	struct pppox_sock *po;
1057 
1058 	++*pos;
1059 	if (v == SEQ_START_TOKEN) {
1060 		po = pppoe_get_idx(pn, 0);
1061 		goto out;
1062 	}
1063 	po = v;
1064 	if (po->next)
1065 		po = po->next;
1066 	else {
1067 		int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
1068 
1069 		po = NULL;
1070 		while (++hash < PPPOE_HASH_SIZE) {
1071 			po = pn->hash_table[hash];
1072 			if (po)
1073 				break;
1074 		}
1075 	}
1076 
1077 out:
1078 	return po;
1079 }
1080 
1081 static void pppoe_seq_stop(struct seq_file *seq, void *v)
1082 	__releases(pn->hash_lock)
1083 {
1084 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1085 	read_unlock_bh(&pn->hash_lock);
1086 }
1087 
1088 static const struct seq_operations pppoe_seq_ops = {
1089 	.start		= pppoe_seq_start,
1090 	.next		= pppoe_seq_next,
1091 	.stop		= pppoe_seq_stop,
1092 	.show		= pppoe_seq_show,
1093 };
1094 
1095 static int pppoe_seq_open(struct inode *inode, struct file *file)
1096 {
1097 	return seq_open_net(inode, file, &pppoe_seq_ops,
1098 			sizeof(struct seq_net_private));
1099 }
1100 
1101 static const struct file_operations pppoe_seq_fops = {
1102 	.owner		= THIS_MODULE,
1103 	.open		= pppoe_seq_open,
1104 	.read		= seq_read,
1105 	.llseek		= seq_lseek,
1106 	.release	= seq_release_net,
1107 };
1108 
1109 #endif /* CONFIG_PROC_FS */
1110 
1111 static const struct proto_ops pppoe_ops = {
1112 	.family		= AF_PPPOX,
1113 	.owner		= THIS_MODULE,
1114 	.release	= pppoe_release,
1115 	.bind		= sock_no_bind,
1116 	.connect	= pppoe_connect,
1117 	.socketpair	= sock_no_socketpair,
1118 	.accept		= sock_no_accept,
1119 	.getname	= pppoe_getname,
1120 	.poll		= datagram_poll,
1121 	.listen		= sock_no_listen,
1122 	.shutdown	= sock_no_shutdown,
1123 	.setsockopt	= sock_no_setsockopt,
1124 	.getsockopt	= sock_no_getsockopt,
1125 	.sendmsg	= pppoe_sendmsg,
1126 	.recvmsg	= pppoe_recvmsg,
1127 	.mmap		= sock_no_mmap,
1128 	.ioctl		= pppox_ioctl,
1129 };
1130 
1131 static const struct pppox_proto pppoe_proto = {
1132 	.create	= pppoe_create,
1133 	.ioctl	= pppoe_ioctl,
1134 	.owner	= THIS_MODULE,
1135 };
1136 
1137 static __net_init int pppoe_init_net(struct net *net)
1138 {
1139 	struct pppoe_net *pn = pppoe_pernet(net);
1140 	struct proc_dir_entry *pde;
1141 
1142 	rwlock_init(&pn->hash_lock);
1143 
1144 	pde = proc_create("pppoe", S_IRUGO, net->proc_net, &pppoe_seq_fops);
1145 #ifdef CONFIG_PROC_FS
1146 	if (!pde)
1147 		return -ENOMEM;
1148 #endif
1149 
1150 	return 0;
1151 }
1152 
1153 static __net_exit void pppoe_exit_net(struct net *net)
1154 {
1155 	remove_proc_entry("pppoe", net->proc_net);
1156 }
1157 
1158 static struct pernet_operations pppoe_net_ops = {
1159 	.init = pppoe_init_net,
1160 	.exit = pppoe_exit_net,
1161 	.id   = &pppoe_net_id,
1162 	.size = sizeof(struct pppoe_net),
1163 };
1164 
1165 static int __init pppoe_init(void)
1166 {
1167 	int err;
1168 
1169 	err = register_pernet_device(&pppoe_net_ops);
1170 	if (err)
1171 		goto out;
1172 
1173 	err = proto_register(&pppoe_sk_proto, 0);
1174 	if (err)
1175 		goto out_unregister_net_ops;
1176 
1177 	err = register_pppox_proto(PX_PROTO_OE, &pppoe_proto);
1178 	if (err)
1179 		goto out_unregister_pppoe_proto;
1180 
1181 	dev_add_pack(&pppoes_ptype);
1182 	dev_add_pack(&pppoed_ptype);
1183 	register_netdevice_notifier(&pppoe_notifier);
1184 
1185 	return 0;
1186 
1187 out_unregister_pppoe_proto:
1188 	proto_unregister(&pppoe_sk_proto);
1189 out_unregister_net_ops:
1190 	unregister_pernet_device(&pppoe_net_ops);
1191 out:
1192 	return err;
1193 }
1194 
1195 static void __exit pppoe_exit(void)
1196 {
1197 	unregister_netdevice_notifier(&pppoe_notifier);
1198 	dev_remove_pack(&pppoed_ptype);
1199 	dev_remove_pack(&pppoes_ptype);
1200 	unregister_pppox_proto(PX_PROTO_OE);
1201 	proto_unregister(&pppoe_sk_proto);
1202 	unregister_pernet_device(&pppoe_net_ops);
1203 }
1204 
1205 module_init(pppoe_init);
1206 module_exit(pppoe_exit);
1207 
1208 MODULE_AUTHOR("Michal Ostrowski <mostrows@speakeasy.net>");
1209 MODULE_DESCRIPTION("PPP over Ethernet driver");
1210 MODULE_LICENSE("GPL");
1211 MODULE_ALIAS_NET_PF_PROTO(PF_PPPOX, PX_PROTO_OE);
1212