xref: /linux/drivers/net/ppp/pppoe.c (revision 3ce095c16263630dde46d6051854073edaacf3d7)
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 | PPPOX_ZOMBIE)) {
315 				pppox_unbind_sock(sk);
316 				sk->sk_state = PPPOX_ZOMBIE;
317 				sk->sk_state_change(sk);
318 				po->pppoe_dev = NULL;
319 				dev_put(dev);
320 			}
321 
322 			release_sock(sk);
323 			sock_put(sk);
324 
325 			/* Restart the process from the start of the current
326 			 * hash chain. We dropped locks so the world may have
327 			 * change from underneath us.
328 			 */
329 
330 			BUG_ON(pppoe_pernet(dev_net(dev)) == NULL);
331 			write_lock_bh(&pn->hash_lock);
332 			po = pn->hash_table[i];
333 		}
334 	}
335 	write_unlock_bh(&pn->hash_lock);
336 }
337 
338 static int pppoe_device_event(struct notifier_block *this,
339 			      unsigned long event, void *ptr)
340 {
341 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
342 
343 	/* Only look at sockets that are using this specific device. */
344 	switch (event) {
345 	case NETDEV_CHANGEADDR:
346 	case NETDEV_CHANGEMTU:
347 		/* A change in mtu or address is a bad thing, requiring
348 		 * LCP re-negotiation.
349 		 */
350 
351 	case NETDEV_GOING_DOWN:
352 	case NETDEV_DOWN:
353 		/* Find every socket on this device and kill it. */
354 		pppoe_flush_dev(dev);
355 		break;
356 
357 	default:
358 		break;
359 	}
360 
361 	return NOTIFY_DONE;
362 }
363 
364 static struct notifier_block pppoe_notifier = {
365 	.notifier_call = pppoe_device_event,
366 };
367 
368 /************************************************************************
369  *
370  * Do the real work of receiving a PPPoE Session frame.
371  *
372  ***********************************************************************/
373 static int pppoe_rcv_core(struct sock *sk, struct sk_buff *skb)
374 {
375 	struct pppox_sock *po = pppox_sk(sk);
376 	struct pppox_sock *relay_po;
377 
378 	/* Backlog receive. Semantics of backlog rcv preclude any code from
379 	 * executing in lock_sock()/release_sock() bounds; meaning sk->sk_state
380 	 * can't change.
381 	 */
382 
383 	if (skb->pkt_type == PACKET_OTHERHOST)
384 		goto abort_kfree;
385 
386 	if (sk->sk_state & PPPOX_BOUND) {
387 		ppp_input(&po->chan, skb);
388 	} else if (sk->sk_state & PPPOX_RELAY) {
389 		relay_po = get_item_by_addr(sock_net(sk),
390 					    &po->pppoe_relay);
391 		if (relay_po == NULL)
392 			goto abort_kfree;
393 
394 		if ((sk_pppox(relay_po)->sk_state & PPPOX_CONNECTED) == 0)
395 			goto abort_put;
396 
397 		if (!__pppoe_xmit(sk_pppox(relay_po), skb))
398 			goto abort_put;
399 	} else {
400 		if (sock_queue_rcv_skb(sk, skb))
401 			goto abort_kfree;
402 	}
403 
404 	return NET_RX_SUCCESS;
405 
406 abort_put:
407 	sock_put(sk_pppox(relay_po));
408 
409 abort_kfree:
410 	kfree_skb(skb);
411 	return NET_RX_DROP;
412 }
413 
414 /************************************************************************
415  *
416  * Receive wrapper called in BH context.
417  *
418  ***********************************************************************/
419 static int pppoe_rcv(struct sk_buff *skb, struct net_device *dev,
420 		     struct packet_type *pt, struct net_device *orig_dev)
421 {
422 	struct pppoe_hdr *ph;
423 	struct pppox_sock *po;
424 	struct pppoe_net *pn;
425 	int len;
426 
427 	skb = skb_share_check(skb, GFP_ATOMIC);
428 	if (!skb)
429 		goto out;
430 
431 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
432 		goto drop;
433 
434 	ph = pppoe_hdr(skb);
435 	len = ntohs(ph->length);
436 
437 	skb_pull_rcsum(skb, sizeof(*ph));
438 	if (skb->len < len)
439 		goto drop;
440 
441 	if (pskb_trim_rcsum(skb, len))
442 		goto drop;
443 
444 	pn = pppoe_pernet(dev_net(dev));
445 
446 	/* Note that get_item does a sock_hold(), so sk_pppox(po)
447 	 * is known to be safe.
448 	 */
449 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
450 	if (!po)
451 		goto drop;
452 
453 	return sk_receive_skb(sk_pppox(po), skb, 0);
454 
455 drop:
456 	kfree_skb(skb);
457 out:
458 	return NET_RX_DROP;
459 }
460 
461 static void pppoe_unbind_sock_work(struct work_struct *work)
462 {
463 	struct pppox_sock *po = container_of(work, struct pppox_sock,
464 					     proto.pppoe.padt_work);
465 	struct sock *sk = sk_pppox(po);
466 
467 	lock_sock(sk);
468 	if (po->pppoe_dev) {
469 		dev_put(po->pppoe_dev);
470 		po->pppoe_dev = NULL;
471 	}
472 	pppox_unbind_sock(sk);
473 	release_sock(sk);
474 	sock_put(sk);
475 }
476 
477 /************************************************************************
478  *
479  * Receive a PPPoE Discovery frame.
480  * This is solely for detection of PADT frames
481  *
482  ***********************************************************************/
483 static int pppoe_disc_rcv(struct sk_buff *skb, struct net_device *dev,
484 			  struct packet_type *pt, struct net_device *orig_dev)
485 
486 {
487 	struct pppoe_hdr *ph;
488 	struct pppox_sock *po;
489 	struct pppoe_net *pn;
490 
491 	skb = skb_share_check(skb, GFP_ATOMIC);
492 	if (!skb)
493 		goto out;
494 
495 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
496 		goto abort;
497 
498 	ph = pppoe_hdr(skb);
499 	if (ph->code != PADT_CODE)
500 		goto abort;
501 
502 	pn = pppoe_pernet(dev_net(dev));
503 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
504 	if (po) {
505 		struct sock *sk = sk_pppox(po);
506 
507 		bh_lock_sock(sk);
508 
509 		/* If the user has locked the socket, just ignore
510 		 * the packet.  With the way two rcv protocols hook into
511 		 * one socket family type, we cannot (easily) distinguish
512 		 * what kind of SKB it is during backlog rcv.
513 		 */
514 		if (sock_owned_by_user(sk) == 0) {
515 			/* We're no longer connect at the PPPOE layer,
516 			 * and must wait for ppp channel to disconnect us.
517 			 */
518 			sk->sk_state = PPPOX_ZOMBIE;
519 		}
520 
521 		bh_unlock_sock(sk);
522 		if (!schedule_work(&po->proto.pppoe.padt_work))
523 			sock_put(sk);
524 	}
525 
526 abort:
527 	kfree_skb(skb);
528 out:
529 	return NET_RX_SUCCESS; /* Lies... :-) */
530 }
531 
532 static struct packet_type pppoes_ptype __read_mostly = {
533 	.type	= cpu_to_be16(ETH_P_PPP_SES),
534 	.func	= pppoe_rcv,
535 };
536 
537 static struct packet_type pppoed_ptype __read_mostly = {
538 	.type	= cpu_to_be16(ETH_P_PPP_DISC),
539 	.func	= pppoe_disc_rcv,
540 };
541 
542 static struct proto pppoe_sk_proto __read_mostly = {
543 	.name	  = "PPPOE",
544 	.owner	  = THIS_MODULE,
545 	.obj_size = sizeof(struct pppox_sock),
546 };
547 
548 /***********************************************************************
549  *
550  * Initialize a new struct sock.
551  *
552  **********************************************************************/
553 static int pppoe_create(struct net *net, struct socket *sock)
554 {
555 	struct sock *sk;
556 
557 	sk = sk_alloc(net, PF_PPPOX, GFP_KERNEL, &pppoe_sk_proto);
558 	if (!sk)
559 		return -ENOMEM;
560 
561 	sock_init_data(sock, sk);
562 
563 	sock->state	= SS_UNCONNECTED;
564 	sock->ops	= &pppoe_ops;
565 
566 	sk->sk_backlog_rcv	= pppoe_rcv_core;
567 	sk->sk_state		= PPPOX_NONE;
568 	sk->sk_type		= SOCK_STREAM;
569 	sk->sk_family		= PF_PPPOX;
570 	sk->sk_protocol		= PX_PROTO_OE;
571 
572 	return 0;
573 }
574 
575 static int pppoe_release(struct socket *sock)
576 {
577 	struct sock *sk = sock->sk;
578 	struct pppox_sock *po;
579 	struct pppoe_net *pn;
580 	struct net *net = NULL;
581 
582 	if (!sk)
583 		return 0;
584 
585 	lock_sock(sk);
586 	if (sock_flag(sk, SOCK_DEAD)) {
587 		release_sock(sk);
588 		return -EBADF;
589 	}
590 
591 	po = pppox_sk(sk);
592 
593 	if (sk->sk_state & (PPPOX_CONNECTED | PPPOX_BOUND | PPPOX_ZOMBIE)) {
594 		dev_put(po->pppoe_dev);
595 		po->pppoe_dev = NULL;
596 	}
597 
598 	pppox_unbind_sock(sk);
599 
600 	/* Signal the death of the socket. */
601 	sk->sk_state = PPPOX_DEAD;
602 
603 	net = sock_net(sk);
604 	pn = pppoe_pernet(net);
605 
606 	/*
607 	 * protect "po" from concurrent updates
608 	 * on pppoe_flush_dev
609 	 */
610 	delete_item(pn, po->pppoe_pa.sid, po->pppoe_pa.remote,
611 		    po->pppoe_ifindex);
612 
613 	sock_orphan(sk);
614 	sock->sk = NULL;
615 
616 	skb_queue_purge(&sk->sk_receive_queue);
617 	release_sock(sk);
618 	sock_put(sk);
619 
620 	return 0;
621 }
622 
623 static int pppoe_connect(struct socket *sock, struct sockaddr *uservaddr,
624 		  int sockaddr_len, int flags)
625 {
626 	struct sock *sk = sock->sk;
627 	struct sockaddr_pppox *sp = (struct sockaddr_pppox *)uservaddr;
628 	struct pppox_sock *po = pppox_sk(sk);
629 	struct net_device *dev = NULL;
630 	struct pppoe_net *pn;
631 	struct net *net = NULL;
632 	int error;
633 
634 	lock_sock(sk);
635 
636 	INIT_WORK(&po->proto.pppoe.padt_work, pppoe_unbind_sock_work);
637 
638 	error = -EINVAL;
639 	if (sp->sa_protocol != PX_PROTO_OE)
640 		goto end;
641 
642 	/* Check for already bound sockets */
643 	error = -EBUSY;
644 	if ((sk->sk_state & PPPOX_CONNECTED) &&
645 	     stage_session(sp->sa_addr.pppoe.sid))
646 		goto end;
647 
648 	/* Check for already disconnected sockets, on attempts to disconnect */
649 	error = -EALREADY;
650 	if ((sk->sk_state & PPPOX_DEAD) &&
651 	     !stage_session(sp->sa_addr.pppoe.sid))
652 		goto end;
653 
654 	error = 0;
655 
656 	/* Delete the old binding */
657 	if (stage_session(po->pppoe_pa.sid)) {
658 		pppox_unbind_sock(sk);
659 		pn = pppoe_pernet(sock_net(sk));
660 		delete_item(pn, po->pppoe_pa.sid,
661 			    po->pppoe_pa.remote, po->pppoe_ifindex);
662 		if (po->pppoe_dev) {
663 			dev_put(po->pppoe_dev);
664 			po->pppoe_dev = NULL;
665 		}
666 
667 		memset(sk_pppox(po) + 1, 0,
668 		       sizeof(struct pppox_sock) - sizeof(struct sock));
669 		sk->sk_state = PPPOX_NONE;
670 	}
671 
672 	/* Re-bind in session stage only */
673 	if (stage_session(sp->sa_addr.pppoe.sid)) {
674 		error = -ENODEV;
675 		net = sock_net(sk);
676 		dev = dev_get_by_name(net, sp->sa_addr.pppoe.dev);
677 		if (!dev)
678 			goto err_put;
679 
680 		po->pppoe_dev = dev;
681 		po->pppoe_ifindex = dev->ifindex;
682 		pn = pppoe_pernet(net);
683 		if (!(dev->flags & IFF_UP)) {
684 			goto err_put;
685 		}
686 
687 		memcpy(&po->pppoe_pa,
688 		       &sp->sa_addr.pppoe,
689 		       sizeof(struct pppoe_addr));
690 
691 		write_lock_bh(&pn->hash_lock);
692 		error = __set_item(pn, po);
693 		write_unlock_bh(&pn->hash_lock);
694 		if (error < 0)
695 			goto err_put;
696 
697 		po->chan.hdrlen = (sizeof(struct pppoe_hdr) +
698 				   dev->hard_header_len);
699 
700 		po->chan.mtu = dev->mtu - sizeof(struct pppoe_hdr) - 2;
701 		po->chan.private = sk;
702 		po->chan.ops = &pppoe_chan_ops;
703 
704 		error = ppp_register_net_channel(dev_net(dev), &po->chan);
705 		if (error) {
706 			delete_item(pn, po->pppoe_pa.sid,
707 				    po->pppoe_pa.remote, po->pppoe_ifindex);
708 			goto err_put;
709 		}
710 
711 		sk->sk_state = PPPOX_CONNECTED;
712 	}
713 
714 	po->num = sp->sa_addr.pppoe.sid;
715 
716 end:
717 	release_sock(sk);
718 	return error;
719 err_put:
720 	if (po->pppoe_dev) {
721 		dev_put(po->pppoe_dev);
722 		po->pppoe_dev = NULL;
723 	}
724 	goto end;
725 }
726 
727 static int pppoe_getname(struct socket *sock, struct sockaddr *uaddr,
728 		  int *usockaddr_len, int peer)
729 {
730 	int len = sizeof(struct sockaddr_pppox);
731 	struct sockaddr_pppox sp;
732 
733 	sp.sa_family	= AF_PPPOX;
734 	sp.sa_protocol	= PX_PROTO_OE;
735 	memcpy(&sp.sa_addr.pppoe, &pppox_sk(sock->sk)->pppoe_pa,
736 	       sizeof(struct pppoe_addr));
737 
738 	memcpy(uaddr, &sp, len);
739 
740 	*usockaddr_len = len;
741 
742 	return 0;
743 }
744 
745 static int pppoe_ioctl(struct socket *sock, unsigned int cmd,
746 		unsigned long arg)
747 {
748 	struct sock *sk = sock->sk;
749 	struct pppox_sock *po = pppox_sk(sk);
750 	int val;
751 	int err;
752 
753 	switch (cmd) {
754 	case PPPIOCGMRU:
755 		err = -ENXIO;
756 		if (!(sk->sk_state & PPPOX_CONNECTED))
757 			break;
758 
759 		err = -EFAULT;
760 		if (put_user(po->pppoe_dev->mtu -
761 			     sizeof(struct pppoe_hdr) -
762 			     PPP_HDRLEN,
763 			     (int __user *)arg))
764 			break;
765 		err = 0;
766 		break;
767 
768 	case PPPIOCSMRU:
769 		err = -ENXIO;
770 		if (!(sk->sk_state & PPPOX_CONNECTED))
771 			break;
772 
773 		err = -EFAULT;
774 		if (get_user(val, (int __user *)arg))
775 			break;
776 
777 		if (val < (po->pppoe_dev->mtu
778 			   - sizeof(struct pppoe_hdr)
779 			   - PPP_HDRLEN))
780 			err = 0;
781 		else
782 			err = -EINVAL;
783 		break;
784 
785 	case PPPIOCSFLAGS:
786 		err = -EFAULT;
787 		if (get_user(val, (int __user *)arg))
788 			break;
789 		err = 0;
790 		break;
791 
792 	case PPPOEIOCSFWD:
793 	{
794 		struct pppox_sock *relay_po;
795 
796 		err = -EBUSY;
797 		if (sk->sk_state & (PPPOX_BOUND | PPPOX_ZOMBIE | PPPOX_DEAD))
798 			break;
799 
800 		err = -ENOTCONN;
801 		if (!(sk->sk_state & PPPOX_CONNECTED))
802 			break;
803 
804 		/* PPPoE address from the user specifies an outbound
805 		   PPPoE address which frames are forwarded to */
806 		err = -EFAULT;
807 		if (copy_from_user(&po->pppoe_relay,
808 				   (void __user *)arg,
809 				   sizeof(struct sockaddr_pppox)))
810 			break;
811 
812 		err = -EINVAL;
813 		if (po->pppoe_relay.sa_family != AF_PPPOX ||
814 		    po->pppoe_relay.sa_protocol != PX_PROTO_OE)
815 			break;
816 
817 		/* Check that the socket referenced by the address
818 		   actually exists. */
819 		relay_po = get_item_by_addr(sock_net(sk), &po->pppoe_relay);
820 		if (!relay_po)
821 			break;
822 
823 		sock_put(sk_pppox(relay_po));
824 		sk->sk_state |= PPPOX_RELAY;
825 		err = 0;
826 		break;
827 	}
828 
829 	case PPPOEIOCDFWD:
830 		err = -EALREADY;
831 		if (!(sk->sk_state & PPPOX_RELAY))
832 			break;
833 
834 		sk->sk_state &= ~PPPOX_RELAY;
835 		err = 0;
836 		break;
837 
838 	default:
839 		err = -ENOTTY;
840 	}
841 
842 	return err;
843 }
844 
845 static int pppoe_sendmsg(struct socket *sock, struct msghdr *m,
846 			 size_t total_len)
847 {
848 	struct sk_buff *skb;
849 	struct sock *sk = sock->sk;
850 	struct pppox_sock *po = pppox_sk(sk);
851 	int error;
852 	struct pppoe_hdr hdr;
853 	struct pppoe_hdr *ph;
854 	struct net_device *dev;
855 	char *start;
856 
857 	lock_sock(sk);
858 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED)) {
859 		error = -ENOTCONN;
860 		goto end;
861 	}
862 
863 	hdr.ver = 1;
864 	hdr.type = 1;
865 	hdr.code = 0;
866 	hdr.sid = po->num;
867 
868 	dev = po->pppoe_dev;
869 
870 	error = -EMSGSIZE;
871 	if (total_len > (dev->mtu + dev->hard_header_len))
872 		goto end;
873 
874 
875 	skb = sock_wmalloc(sk, total_len + dev->hard_header_len + 32,
876 			   0, GFP_KERNEL);
877 	if (!skb) {
878 		error = -ENOMEM;
879 		goto end;
880 	}
881 
882 	/* Reserve space for headers. */
883 	skb_reserve(skb, dev->hard_header_len);
884 	skb_reset_network_header(skb);
885 
886 	skb->dev = dev;
887 
888 	skb->priority = sk->sk_priority;
889 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
890 
891 	ph = (struct pppoe_hdr *)skb_put(skb, total_len + sizeof(struct pppoe_hdr));
892 	start = (char *)&ph->tag[0];
893 
894 	error = memcpy_from_msg(start, m, total_len);
895 	if (error < 0) {
896 		kfree_skb(skb);
897 		goto end;
898 	}
899 
900 	error = total_len;
901 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
902 			po->pppoe_pa.remote, NULL, total_len);
903 
904 	memcpy(ph, &hdr, sizeof(struct pppoe_hdr));
905 
906 	ph->length = htons(total_len);
907 
908 	dev_queue_xmit(skb);
909 
910 end:
911 	release_sock(sk);
912 	return error;
913 }
914 
915 /************************************************************************
916  *
917  * xmit function for internal use.
918  *
919  ***********************************************************************/
920 static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb)
921 {
922 	struct pppox_sock *po = pppox_sk(sk);
923 	struct net_device *dev = po->pppoe_dev;
924 	struct pppoe_hdr *ph;
925 	int data_len = skb->len;
926 
927 	/* The higher-level PPP code (ppp_unregister_channel()) ensures the PPP
928 	 * xmit operations conclude prior to an unregistration call.  Thus
929 	 * sk->sk_state cannot change, so we don't need to do lock_sock().
930 	 * But, we also can't do a lock_sock since that introduces a potential
931 	 * deadlock as we'd reverse the lock ordering used when calling
932 	 * ppp_unregister_channel().
933 	 */
934 
935 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED))
936 		goto abort;
937 
938 	if (!dev)
939 		goto abort;
940 
941 	/* Copy the data if there is no space for the header or if it's
942 	 * read-only.
943 	 */
944 	if (skb_cow_head(skb, sizeof(*ph) + dev->hard_header_len))
945 		goto abort;
946 
947 	__skb_push(skb, sizeof(*ph));
948 	skb_reset_network_header(skb);
949 
950 	ph = pppoe_hdr(skb);
951 	ph->ver	= 1;
952 	ph->type = 1;
953 	ph->code = 0;
954 	ph->sid	= po->num;
955 	ph->length = htons(data_len);
956 
957 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
958 	skb->dev = dev;
959 
960 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
961 			po->pppoe_pa.remote, NULL, data_len);
962 
963 	dev_queue_xmit(skb);
964 	return 1;
965 
966 abort:
967 	kfree_skb(skb);
968 	return 1;
969 }
970 
971 /************************************************************************
972  *
973  * xmit function called by generic PPP driver
974  * sends PPP frame over PPPoE socket
975  *
976  ***********************************************************************/
977 static int pppoe_xmit(struct ppp_channel *chan, struct sk_buff *skb)
978 {
979 	struct sock *sk = (struct sock *)chan->private;
980 	return __pppoe_xmit(sk, skb);
981 }
982 
983 static const struct ppp_channel_ops pppoe_chan_ops = {
984 	.start_xmit = pppoe_xmit,
985 };
986 
987 static int pppoe_recvmsg(struct socket *sock, struct msghdr *m,
988 			 size_t total_len, int flags)
989 {
990 	struct sock *sk = sock->sk;
991 	struct sk_buff *skb;
992 	int error = 0;
993 
994 	if (sk->sk_state & PPPOX_BOUND) {
995 		error = -EIO;
996 		goto end;
997 	}
998 
999 	skb = skb_recv_datagram(sk, flags & ~MSG_DONTWAIT,
1000 				flags & MSG_DONTWAIT, &error);
1001 	if (error < 0)
1002 		goto end;
1003 
1004 	if (skb) {
1005 		total_len = min_t(size_t, total_len, skb->len);
1006 		error = skb_copy_datagram_msg(skb, 0, m, total_len);
1007 		if (error == 0) {
1008 			consume_skb(skb);
1009 			return total_len;
1010 		}
1011 	}
1012 
1013 	kfree_skb(skb);
1014 end:
1015 	return error;
1016 }
1017 
1018 #ifdef CONFIG_PROC_FS
1019 static int pppoe_seq_show(struct seq_file *seq, void *v)
1020 {
1021 	struct pppox_sock *po;
1022 	char *dev_name;
1023 
1024 	if (v == SEQ_START_TOKEN) {
1025 		seq_puts(seq, "Id       Address              Device\n");
1026 		goto out;
1027 	}
1028 
1029 	po = v;
1030 	dev_name = po->pppoe_pa.dev;
1031 
1032 	seq_printf(seq, "%08X %pM %8s\n",
1033 		po->pppoe_pa.sid, po->pppoe_pa.remote, dev_name);
1034 out:
1035 	return 0;
1036 }
1037 
1038 static inline struct pppox_sock *pppoe_get_idx(struct pppoe_net *pn, loff_t pos)
1039 {
1040 	struct pppox_sock *po;
1041 	int i;
1042 
1043 	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
1044 		po = pn->hash_table[i];
1045 		while (po) {
1046 			if (!pos--)
1047 				goto out;
1048 			po = po->next;
1049 		}
1050 	}
1051 
1052 out:
1053 	return po;
1054 }
1055 
1056 static void *pppoe_seq_start(struct seq_file *seq, loff_t *pos)
1057 	__acquires(pn->hash_lock)
1058 {
1059 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1060 	loff_t l = *pos;
1061 
1062 	read_lock_bh(&pn->hash_lock);
1063 	return l ? pppoe_get_idx(pn, --l) : SEQ_START_TOKEN;
1064 }
1065 
1066 static void *pppoe_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1067 {
1068 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1069 	struct pppox_sock *po;
1070 
1071 	++*pos;
1072 	if (v == SEQ_START_TOKEN) {
1073 		po = pppoe_get_idx(pn, 0);
1074 		goto out;
1075 	}
1076 	po = v;
1077 	if (po->next)
1078 		po = po->next;
1079 	else {
1080 		int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
1081 
1082 		po = NULL;
1083 		while (++hash < PPPOE_HASH_SIZE) {
1084 			po = pn->hash_table[hash];
1085 			if (po)
1086 				break;
1087 		}
1088 	}
1089 
1090 out:
1091 	return po;
1092 }
1093 
1094 static void pppoe_seq_stop(struct seq_file *seq, void *v)
1095 	__releases(pn->hash_lock)
1096 {
1097 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1098 	read_unlock_bh(&pn->hash_lock);
1099 }
1100 
1101 static const struct seq_operations pppoe_seq_ops = {
1102 	.start		= pppoe_seq_start,
1103 	.next		= pppoe_seq_next,
1104 	.stop		= pppoe_seq_stop,
1105 	.show		= pppoe_seq_show,
1106 };
1107 
1108 static int pppoe_seq_open(struct inode *inode, struct file *file)
1109 {
1110 	return seq_open_net(inode, file, &pppoe_seq_ops,
1111 			sizeof(struct seq_net_private));
1112 }
1113 
1114 static const struct file_operations pppoe_seq_fops = {
1115 	.owner		= THIS_MODULE,
1116 	.open		= pppoe_seq_open,
1117 	.read		= seq_read,
1118 	.llseek		= seq_lseek,
1119 	.release	= seq_release_net,
1120 };
1121 
1122 #endif /* CONFIG_PROC_FS */
1123 
1124 static const struct proto_ops pppoe_ops = {
1125 	.family		= AF_PPPOX,
1126 	.owner		= THIS_MODULE,
1127 	.release	= pppoe_release,
1128 	.bind		= sock_no_bind,
1129 	.connect	= pppoe_connect,
1130 	.socketpair	= sock_no_socketpair,
1131 	.accept		= sock_no_accept,
1132 	.getname	= pppoe_getname,
1133 	.poll		= datagram_poll,
1134 	.listen		= sock_no_listen,
1135 	.shutdown	= sock_no_shutdown,
1136 	.setsockopt	= sock_no_setsockopt,
1137 	.getsockopt	= sock_no_getsockopt,
1138 	.sendmsg	= pppoe_sendmsg,
1139 	.recvmsg	= pppoe_recvmsg,
1140 	.mmap		= sock_no_mmap,
1141 	.ioctl		= pppox_ioctl,
1142 };
1143 
1144 static const struct pppox_proto pppoe_proto = {
1145 	.create	= pppoe_create,
1146 	.ioctl	= pppoe_ioctl,
1147 	.owner	= THIS_MODULE,
1148 };
1149 
1150 static __net_init int pppoe_init_net(struct net *net)
1151 {
1152 	struct pppoe_net *pn = pppoe_pernet(net);
1153 	struct proc_dir_entry *pde;
1154 
1155 	rwlock_init(&pn->hash_lock);
1156 
1157 	pde = proc_create("pppoe", S_IRUGO, net->proc_net, &pppoe_seq_fops);
1158 #ifdef CONFIG_PROC_FS
1159 	if (!pde)
1160 		return -ENOMEM;
1161 #endif
1162 
1163 	return 0;
1164 }
1165 
1166 static __net_exit void pppoe_exit_net(struct net *net)
1167 {
1168 	remove_proc_entry("pppoe", net->proc_net);
1169 }
1170 
1171 static struct pernet_operations pppoe_net_ops = {
1172 	.init = pppoe_init_net,
1173 	.exit = pppoe_exit_net,
1174 	.id   = &pppoe_net_id,
1175 	.size = sizeof(struct pppoe_net),
1176 };
1177 
1178 static int __init pppoe_init(void)
1179 {
1180 	int err;
1181 
1182 	err = register_pernet_device(&pppoe_net_ops);
1183 	if (err)
1184 		goto out;
1185 
1186 	err = proto_register(&pppoe_sk_proto, 0);
1187 	if (err)
1188 		goto out_unregister_net_ops;
1189 
1190 	err = register_pppox_proto(PX_PROTO_OE, &pppoe_proto);
1191 	if (err)
1192 		goto out_unregister_pppoe_proto;
1193 
1194 	dev_add_pack(&pppoes_ptype);
1195 	dev_add_pack(&pppoed_ptype);
1196 	register_netdevice_notifier(&pppoe_notifier);
1197 
1198 	return 0;
1199 
1200 out_unregister_pppoe_proto:
1201 	proto_unregister(&pppoe_sk_proto);
1202 out_unregister_net_ops:
1203 	unregister_pernet_device(&pppoe_net_ops);
1204 out:
1205 	return err;
1206 }
1207 
1208 static void __exit pppoe_exit(void)
1209 {
1210 	unregister_netdevice_notifier(&pppoe_notifier);
1211 	dev_remove_pack(&pppoed_ptype);
1212 	dev_remove_pack(&pppoes_ptype);
1213 	unregister_pppox_proto(PX_PROTO_OE);
1214 	proto_unregister(&pppoe_sk_proto);
1215 	unregister_pernet_device(&pppoe_net_ops);
1216 }
1217 
1218 module_init(pppoe_init);
1219 module_exit(pppoe_exit);
1220 
1221 MODULE_AUTHOR("Michal Ostrowski <mostrows@speakeasy.net>");
1222 MODULE_DESCRIPTION("PPP over Ethernet driver");
1223 MODULE_LICENSE("GPL");
1224 MODULE_ALIAS_NETPROTO(PF_PPPOX);
1225