xref: /openbmc/linux/drivers/net/ppp/pppoe.c (revision 2891f2d5)
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 <linux/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 unsigned 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 
399 		sock_put(sk_pppox(relay_po));
400 	} else {
401 		if (sock_queue_rcv_skb(sk, skb))
402 			goto abort_kfree;
403 	}
404 
405 	return NET_RX_SUCCESS;
406 
407 abort_put:
408 	sock_put(sk_pppox(relay_po));
409 
410 abort_kfree:
411 	kfree_skb(skb);
412 	return NET_RX_DROP;
413 }
414 
415 /************************************************************************
416  *
417  * Receive wrapper called in BH context.
418  *
419  ***********************************************************************/
420 static int pppoe_rcv(struct sk_buff *skb, struct net_device *dev,
421 		     struct packet_type *pt, struct net_device *orig_dev)
422 {
423 	struct pppoe_hdr *ph;
424 	struct pppox_sock *po;
425 	struct pppoe_net *pn;
426 	int len;
427 
428 	skb = skb_share_check(skb, GFP_ATOMIC);
429 	if (!skb)
430 		goto out;
431 
432 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
433 		goto drop;
434 
435 	ph = pppoe_hdr(skb);
436 	len = ntohs(ph->length);
437 
438 	skb_pull_rcsum(skb, sizeof(*ph));
439 	if (skb->len < len)
440 		goto drop;
441 
442 	if (pskb_trim_rcsum(skb, len))
443 		goto drop;
444 
445 	pn = pppoe_pernet(dev_net(dev));
446 
447 	/* Note that get_item does a sock_hold(), so sk_pppox(po)
448 	 * is known to be safe.
449 	 */
450 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
451 	if (!po)
452 		goto drop;
453 
454 	return sk_receive_skb(sk_pppox(po), skb, 0);
455 
456 drop:
457 	kfree_skb(skb);
458 out:
459 	return NET_RX_DROP;
460 }
461 
462 static void pppoe_unbind_sock_work(struct work_struct *work)
463 {
464 	struct pppox_sock *po = container_of(work, struct pppox_sock,
465 					     proto.pppoe.padt_work);
466 	struct sock *sk = sk_pppox(po);
467 
468 	lock_sock(sk);
469 	if (po->pppoe_dev) {
470 		dev_put(po->pppoe_dev);
471 		po->pppoe_dev = NULL;
472 	}
473 	pppox_unbind_sock(sk);
474 	release_sock(sk);
475 	sock_put(sk);
476 }
477 
478 /************************************************************************
479  *
480  * Receive a PPPoE Discovery frame.
481  * This is solely for detection of PADT frames
482  *
483  ***********************************************************************/
484 static int pppoe_disc_rcv(struct sk_buff *skb, struct net_device *dev,
485 			  struct packet_type *pt, struct net_device *orig_dev)
486 
487 {
488 	struct pppoe_hdr *ph;
489 	struct pppox_sock *po;
490 	struct pppoe_net *pn;
491 
492 	skb = skb_share_check(skb, GFP_ATOMIC);
493 	if (!skb)
494 		goto out;
495 
496 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
497 		goto abort;
498 
499 	ph = pppoe_hdr(skb);
500 	if (ph->code != PADT_CODE)
501 		goto abort;
502 
503 	pn = pppoe_pernet(dev_net(dev));
504 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
505 	if (po)
506 		if (!schedule_work(&po->proto.pppoe.padt_work))
507 			sock_put(sk_pppox(po));
508 
509 abort:
510 	kfree_skb(skb);
511 out:
512 	return NET_RX_SUCCESS; /* Lies... :-) */
513 }
514 
515 static struct packet_type pppoes_ptype __read_mostly = {
516 	.type	= cpu_to_be16(ETH_P_PPP_SES),
517 	.func	= pppoe_rcv,
518 };
519 
520 static struct packet_type pppoed_ptype __read_mostly = {
521 	.type	= cpu_to_be16(ETH_P_PPP_DISC),
522 	.func	= pppoe_disc_rcv,
523 };
524 
525 static struct proto pppoe_sk_proto __read_mostly = {
526 	.name	  = "PPPOE",
527 	.owner	  = THIS_MODULE,
528 	.obj_size = sizeof(struct pppox_sock),
529 };
530 
531 /***********************************************************************
532  *
533  * Initialize a new struct sock.
534  *
535  **********************************************************************/
536 static int pppoe_create(struct net *net, struct socket *sock, int kern)
537 {
538 	struct sock *sk;
539 
540 	sk = sk_alloc(net, PF_PPPOX, GFP_KERNEL, &pppoe_sk_proto, kern);
541 	if (!sk)
542 		return -ENOMEM;
543 
544 	sock_init_data(sock, sk);
545 
546 	sock->state	= SS_UNCONNECTED;
547 	sock->ops	= &pppoe_ops;
548 
549 	sk->sk_backlog_rcv	= pppoe_rcv_core;
550 	sk->sk_state		= PPPOX_NONE;
551 	sk->sk_type		= SOCK_STREAM;
552 	sk->sk_family		= PF_PPPOX;
553 	sk->sk_protocol		= PX_PROTO_OE;
554 
555 	INIT_WORK(&pppox_sk(sk)->proto.pppoe.padt_work,
556 		  pppoe_unbind_sock_work);
557 
558 	return 0;
559 }
560 
561 static int pppoe_release(struct socket *sock)
562 {
563 	struct sock *sk = sock->sk;
564 	struct pppox_sock *po;
565 	struct pppoe_net *pn;
566 	struct net *net = NULL;
567 
568 	if (!sk)
569 		return 0;
570 
571 	lock_sock(sk);
572 	if (sock_flag(sk, SOCK_DEAD)) {
573 		release_sock(sk);
574 		return -EBADF;
575 	}
576 
577 	po = pppox_sk(sk);
578 
579 	if (po->pppoe_dev) {
580 		dev_put(po->pppoe_dev);
581 		po->pppoe_dev = NULL;
582 	}
583 
584 	pppox_unbind_sock(sk);
585 
586 	/* Signal the death of the socket. */
587 	sk->sk_state = PPPOX_DEAD;
588 
589 	net = sock_net(sk);
590 	pn = pppoe_pernet(net);
591 
592 	/*
593 	 * protect "po" from concurrent updates
594 	 * on pppoe_flush_dev
595 	 */
596 	delete_item(pn, po->pppoe_pa.sid, po->pppoe_pa.remote,
597 		    po->pppoe_ifindex);
598 
599 	sock_orphan(sk);
600 	sock->sk = NULL;
601 
602 	skb_queue_purge(&sk->sk_receive_queue);
603 	release_sock(sk);
604 	sock_put(sk);
605 
606 	return 0;
607 }
608 
609 static int pppoe_connect(struct socket *sock, struct sockaddr *uservaddr,
610 		  int sockaddr_len, int flags)
611 {
612 	struct sock *sk = sock->sk;
613 	struct sockaddr_pppox *sp = (struct sockaddr_pppox *)uservaddr;
614 	struct pppox_sock *po = pppox_sk(sk);
615 	struct net_device *dev = NULL;
616 	struct pppoe_net *pn;
617 	struct net *net = NULL;
618 	int error;
619 
620 	lock_sock(sk);
621 
622 	error = -EINVAL;
623 	if (sp->sa_protocol != PX_PROTO_OE)
624 		goto end;
625 
626 	/* Check for already bound sockets */
627 	error = -EBUSY;
628 	if ((sk->sk_state & PPPOX_CONNECTED) &&
629 	     stage_session(sp->sa_addr.pppoe.sid))
630 		goto end;
631 
632 	/* Check for already disconnected sockets, on attempts to disconnect */
633 	error = -EALREADY;
634 	if ((sk->sk_state & PPPOX_DEAD) &&
635 	     !stage_session(sp->sa_addr.pppoe.sid))
636 		goto end;
637 
638 	error = 0;
639 
640 	/* Delete the old binding */
641 	if (stage_session(po->pppoe_pa.sid)) {
642 		pppox_unbind_sock(sk);
643 		pn = pppoe_pernet(sock_net(sk));
644 		delete_item(pn, po->pppoe_pa.sid,
645 			    po->pppoe_pa.remote, po->pppoe_ifindex);
646 		if (po->pppoe_dev) {
647 			dev_put(po->pppoe_dev);
648 			po->pppoe_dev = NULL;
649 		}
650 
651 		po->pppoe_ifindex = 0;
652 		memset(&po->pppoe_pa, 0, sizeof(po->pppoe_pa));
653 		memset(&po->pppoe_relay, 0, sizeof(po->pppoe_relay));
654 		memset(&po->chan, 0, sizeof(po->chan));
655 		po->next = NULL;
656 		po->num = 0;
657 
658 		sk->sk_state = PPPOX_NONE;
659 	}
660 
661 	/* Re-bind in session stage only */
662 	if (stage_session(sp->sa_addr.pppoe.sid)) {
663 		error = -ENODEV;
664 		net = sock_net(sk);
665 		dev = dev_get_by_name(net, sp->sa_addr.pppoe.dev);
666 		if (!dev)
667 			goto err_put;
668 
669 		po->pppoe_dev = dev;
670 		po->pppoe_ifindex = dev->ifindex;
671 		pn = pppoe_pernet(net);
672 		if (!(dev->flags & IFF_UP)) {
673 			goto err_put;
674 		}
675 
676 		memcpy(&po->pppoe_pa,
677 		       &sp->sa_addr.pppoe,
678 		       sizeof(struct pppoe_addr));
679 
680 		write_lock_bh(&pn->hash_lock);
681 		error = __set_item(pn, po);
682 		write_unlock_bh(&pn->hash_lock);
683 		if (error < 0)
684 			goto err_put;
685 
686 		po->chan.hdrlen = (sizeof(struct pppoe_hdr) +
687 				   dev->hard_header_len);
688 
689 		po->chan.mtu = dev->mtu - sizeof(struct pppoe_hdr) - 2;
690 		po->chan.private = sk;
691 		po->chan.ops = &pppoe_chan_ops;
692 
693 		error = ppp_register_net_channel(dev_net(dev), &po->chan);
694 		if (error) {
695 			delete_item(pn, po->pppoe_pa.sid,
696 				    po->pppoe_pa.remote, po->pppoe_ifindex);
697 			goto err_put;
698 		}
699 
700 		sk->sk_state = PPPOX_CONNECTED;
701 	}
702 
703 	po->num = sp->sa_addr.pppoe.sid;
704 
705 end:
706 	release_sock(sk);
707 	return error;
708 err_put:
709 	if (po->pppoe_dev) {
710 		dev_put(po->pppoe_dev);
711 		po->pppoe_dev = NULL;
712 	}
713 	goto end;
714 }
715 
716 static int pppoe_getname(struct socket *sock, struct sockaddr *uaddr,
717 		  int *usockaddr_len, int peer)
718 {
719 	int len = sizeof(struct sockaddr_pppox);
720 	struct sockaddr_pppox sp;
721 
722 	sp.sa_family	= AF_PPPOX;
723 	sp.sa_protocol	= PX_PROTO_OE;
724 	memcpy(&sp.sa_addr.pppoe, &pppox_sk(sock->sk)->pppoe_pa,
725 	       sizeof(struct pppoe_addr));
726 
727 	memcpy(uaddr, &sp, len);
728 
729 	*usockaddr_len = len;
730 
731 	return 0;
732 }
733 
734 static int pppoe_ioctl(struct socket *sock, unsigned int cmd,
735 		unsigned long arg)
736 {
737 	struct sock *sk = sock->sk;
738 	struct pppox_sock *po = pppox_sk(sk);
739 	int val;
740 	int err;
741 
742 	switch (cmd) {
743 	case PPPIOCGMRU:
744 		err = -ENXIO;
745 		if (!(sk->sk_state & PPPOX_CONNECTED))
746 			break;
747 
748 		err = -EFAULT;
749 		if (put_user(po->pppoe_dev->mtu -
750 			     sizeof(struct pppoe_hdr) -
751 			     PPP_HDRLEN,
752 			     (int __user *)arg))
753 			break;
754 		err = 0;
755 		break;
756 
757 	case PPPIOCSMRU:
758 		err = -ENXIO;
759 		if (!(sk->sk_state & PPPOX_CONNECTED))
760 			break;
761 
762 		err = -EFAULT;
763 		if (get_user(val, (int __user *)arg))
764 			break;
765 
766 		if (val < (po->pppoe_dev->mtu
767 			   - sizeof(struct pppoe_hdr)
768 			   - PPP_HDRLEN))
769 			err = 0;
770 		else
771 			err = -EINVAL;
772 		break;
773 
774 	case PPPIOCSFLAGS:
775 		err = -EFAULT;
776 		if (get_user(val, (int __user *)arg))
777 			break;
778 		err = 0;
779 		break;
780 
781 	case PPPOEIOCSFWD:
782 	{
783 		struct pppox_sock *relay_po;
784 
785 		err = -EBUSY;
786 		if (sk->sk_state & (PPPOX_BOUND | PPPOX_DEAD))
787 			break;
788 
789 		err = -ENOTCONN;
790 		if (!(sk->sk_state & PPPOX_CONNECTED))
791 			break;
792 
793 		/* PPPoE address from the user specifies an outbound
794 		   PPPoE address which frames are forwarded to */
795 		err = -EFAULT;
796 		if (copy_from_user(&po->pppoe_relay,
797 				   (void __user *)arg,
798 				   sizeof(struct sockaddr_pppox)))
799 			break;
800 
801 		err = -EINVAL;
802 		if (po->pppoe_relay.sa_family != AF_PPPOX ||
803 		    po->pppoe_relay.sa_protocol != PX_PROTO_OE)
804 			break;
805 
806 		/* Check that the socket referenced by the address
807 		   actually exists. */
808 		relay_po = get_item_by_addr(sock_net(sk), &po->pppoe_relay);
809 		if (!relay_po)
810 			break;
811 
812 		sock_put(sk_pppox(relay_po));
813 		sk->sk_state |= PPPOX_RELAY;
814 		err = 0;
815 		break;
816 	}
817 
818 	case PPPOEIOCDFWD:
819 		err = -EALREADY;
820 		if (!(sk->sk_state & PPPOX_RELAY))
821 			break;
822 
823 		sk->sk_state &= ~PPPOX_RELAY;
824 		err = 0;
825 		break;
826 
827 	default:
828 		err = -ENOTTY;
829 	}
830 
831 	return err;
832 }
833 
834 static int pppoe_sendmsg(struct socket *sock, struct msghdr *m,
835 			 size_t total_len)
836 {
837 	struct sk_buff *skb;
838 	struct sock *sk = sock->sk;
839 	struct pppox_sock *po = pppox_sk(sk);
840 	int error;
841 	struct pppoe_hdr hdr;
842 	struct pppoe_hdr *ph;
843 	struct net_device *dev;
844 	char *start;
845 
846 	lock_sock(sk);
847 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED)) {
848 		error = -ENOTCONN;
849 		goto end;
850 	}
851 
852 	hdr.ver = 1;
853 	hdr.type = 1;
854 	hdr.code = 0;
855 	hdr.sid = po->num;
856 
857 	dev = po->pppoe_dev;
858 
859 	error = -EMSGSIZE;
860 	if (total_len > (dev->mtu + dev->hard_header_len))
861 		goto end;
862 
863 
864 	skb = sock_wmalloc(sk, total_len + dev->hard_header_len + 32,
865 			   0, GFP_KERNEL);
866 	if (!skb) {
867 		error = -ENOMEM;
868 		goto end;
869 	}
870 
871 	/* Reserve space for headers. */
872 	skb_reserve(skb, dev->hard_header_len);
873 	skb_reset_network_header(skb);
874 
875 	skb->dev = dev;
876 
877 	skb->priority = sk->sk_priority;
878 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
879 
880 	ph = (struct pppoe_hdr *)skb_put(skb, total_len + sizeof(struct pppoe_hdr));
881 	start = (char *)&ph->tag[0];
882 
883 	error = memcpy_from_msg(start, m, total_len);
884 	if (error < 0) {
885 		kfree_skb(skb);
886 		goto end;
887 	}
888 
889 	error = total_len;
890 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
891 			po->pppoe_pa.remote, NULL, total_len);
892 
893 	memcpy(ph, &hdr, sizeof(struct pppoe_hdr));
894 
895 	ph->length = htons(total_len);
896 
897 	dev_queue_xmit(skb);
898 
899 end:
900 	release_sock(sk);
901 	return error;
902 }
903 
904 /************************************************************************
905  *
906  * xmit function for internal use.
907  *
908  ***********************************************************************/
909 static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb)
910 {
911 	struct pppox_sock *po = pppox_sk(sk);
912 	struct net_device *dev = po->pppoe_dev;
913 	struct pppoe_hdr *ph;
914 	int data_len = skb->len;
915 
916 	/* The higher-level PPP code (ppp_unregister_channel()) ensures the PPP
917 	 * xmit operations conclude prior to an unregistration call.  Thus
918 	 * sk->sk_state cannot change, so we don't need to do lock_sock().
919 	 * But, we also can't do a lock_sock since that introduces a potential
920 	 * deadlock as we'd reverse the lock ordering used when calling
921 	 * ppp_unregister_channel().
922 	 */
923 
924 	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED))
925 		goto abort;
926 
927 	if (!dev)
928 		goto abort;
929 
930 	/* Copy the data if there is no space for the header or if it's
931 	 * read-only.
932 	 */
933 	if (skb_cow_head(skb, sizeof(*ph) + dev->hard_header_len))
934 		goto abort;
935 
936 	__skb_push(skb, sizeof(*ph));
937 	skb_reset_network_header(skb);
938 
939 	ph = pppoe_hdr(skb);
940 	ph->ver	= 1;
941 	ph->type = 1;
942 	ph->code = 0;
943 	ph->sid	= po->num;
944 	ph->length = htons(data_len);
945 
946 	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
947 	skb->dev = dev;
948 
949 	dev_hard_header(skb, dev, ETH_P_PPP_SES,
950 			po->pppoe_pa.remote, NULL, data_len);
951 
952 	dev_queue_xmit(skb);
953 	return 1;
954 
955 abort:
956 	kfree_skb(skb);
957 	return 1;
958 }
959 
960 /************************************************************************
961  *
962  * xmit function called by generic PPP driver
963  * sends PPP frame over PPPoE socket
964  *
965  ***********************************************************************/
966 static int pppoe_xmit(struct ppp_channel *chan, struct sk_buff *skb)
967 {
968 	struct sock *sk = (struct sock *)chan->private;
969 	return __pppoe_xmit(sk, skb);
970 }
971 
972 static const struct ppp_channel_ops pppoe_chan_ops = {
973 	.start_xmit = pppoe_xmit,
974 };
975 
976 static int pppoe_recvmsg(struct socket *sock, struct msghdr *m,
977 			 size_t total_len, int flags)
978 {
979 	struct sock *sk = sock->sk;
980 	struct sk_buff *skb;
981 	int error = 0;
982 
983 	if (sk->sk_state & PPPOX_BOUND) {
984 		error = -EIO;
985 		goto end;
986 	}
987 
988 	skb = skb_recv_datagram(sk, flags & ~MSG_DONTWAIT,
989 				flags & MSG_DONTWAIT, &error);
990 	if (error < 0)
991 		goto end;
992 
993 	if (skb) {
994 		total_len = min_t(size_t, total_len, skb->len);
995 		error = skb_copy_datagram_msg(skb, 0, m, total_len);
996 		if (error == 0) {
997 			consume_skb(skb);
998 			return total_len;
999 		}
1000 	}
1001 
1002 	kfree_skb(skb);
1003 end:
1004 	return error;
1005 }
1006 
1007 #ifdef CONFIG_PROC_FS
1008 static int pppoe_seq_show(struct seq_file *seq, void *v)
1009 {
1010 	struct pppox_sock *po;
1011 	char *dev_name;
1012 
1013 	if (v == SEQ_START_TOKEN) {
1014 		seq_puts(seq, "Id       Address              Device\n");
1015 		goto out;
1016 	}
1017 
1018 	po = v;
1019 	dev_name = po->pppoe_pa.dev;
1020 
1021 	seq_printf(seq, "%08X %pM %8s\n",
1022 		po->pppoe_pa.sid, po->pppoe_pa.remote, dev_name);
1023 out:
1024 	return 0;
1025 }
1026 
1027 static inline struct pppox_sock *pppoe_get_idx(struct pppoe_net *pn, loff_t pos)
1028 {
1029 	struct pppox_sock *po;
1030 	int i;
1031 
1032 	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
1033 		po = pn->hash_table[i];
1034 		while (po) {
1035 			if (!pos--)
1036 				goto out;
1037 			po = po->next;
1038 		}
1039 	}
1040 
1041 out:
1042 	return po;
1043 }
1044 
1045 static void *pppoe_seq_start(struct seq_file *seq, loff_t *pos)
1046 	__acquires(pn->hash_lock)
1047 {
1048 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1049 	loff_t l = *pos;
1050 
1051 	read_lock_bh(&pn->hash_lock);
1052 	return l ? pppoe_get_idx(pn, --l) : SEQ_START_TOKEN;
1053 }
1054 
1055 static void *pppoe_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1056 {
1057 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1058 	struct pppox_sock *po;
1059 
1060 	++*pos;
1061 	if (v == SEQ_START_TOKEN) {
1062 		po = pppoe_get_idx(pn, 0);
1063 		goto out;
1064 	}
1065 	po = v;
1066 	if (po->next)
1067 		po = po->next;
1068 	else {
1069 		int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
1070 
1071 		po = NULL;
1072 		while (++hash < PPPOE_HASH_SIZE) {
1073 			po = pn->hash_table[hash];
1074 			if (po)
1075 				break;
1076 		}
1077 	}
1078 
1079 out:
1080 	return po;
1081 }
1082 
1083 static void pppoe_seq_stop(struct seq_file *seq, void *v)
1084 	__releases(pn->hash_lock)
1085 {
1086 	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1087 	read_unlock_bh(&pn->hash_lock);
1088 }
1089 
1090 static const struct seq_operations pppoe_seq_ops = {
1091 	.start		= pppoe_seq_start,
1092 	.next		= pppoe_seq_next,
1093 	.stop		= pppoe_seq_stop,
1094 	.show		= pppoe_seq_show,
1095 };
1096 
1097 static int pppoe_seq_open(struct inode *inode, struct file *file)
1098 {
1099 	return seq_open_net(inode, file, &pppoe_seq_ops,
1100 			sizeof(struct seq_net_private));
1101 }
1102 
1103 static const struct file_operations pppoe_seq_fops = {
1104 	.owner		= THIS_MODULE,
1105 	.open		= pppoe_seq_open,
1106 	.read		= seq_read,
1107 	.llseek		= seq_lseek,
1108 	.release	= seq_release_net,
1109 };
1110 
1111 #endif /* CONFIG_PROC_FS */
1112 
1113 static const struct proto_ops pppoe_ops = {
1114 	.family		= AF_PPPOX,
1115 	.owner		= THIS_MODULE,
1116 	.release	= pppoe_release,
1117 	.bind		= sock_no_bind,
1118 	.connect	= pppoe_connect,
1119 	.socketpair	= sock_no_socketpair,
1120 	.accept		= sock_no_accept,
1121 	.getname	= pppoe_getname,
1122 	.poll		= datagram_poll,
1123 	.listen		= sock_no_listen,
1124 	.shutdown	= sock_no_shutdown,
1125 	.setsockopt	= sock_no_setsockopt,
1126 	.getsockopt	= sock_no_getsockopt,
1127 	.sendmsg	= pppoe_sendmsg,
1128 	.recvmsg	= pppoe_recvmsg,
1129 	.mmap		= sock_no_mmap,
1130 	.ioctl		= pppox_ioctl,
1131 };
1132 
1133 static const struct pppox_proto pppoe_proto = {
1134 	.create	= pppoe_create,
1135 	.ioctl	= pppoe_ioctl,
1136 	.owner	= THIS_MODULE,
1137 };
1138 
1139 static __net_init int pppoe_init_net(struct net *net)
1140 {
1141 	struct pppoe_net *pn = pppoe_pernet(net);
1142 	struct proc_dir_entry *pde;
1143 
1144 	rwlock_init(&pn->hash_lock);
1145 
1146 	pde = proc_create("pppoe", S_IRUGO, net->proc_net, &pppoe_seq_fops);
1147 #ifdef CONFIG_PROC_FS
1148 	if (!pde)
1149 		return -ENOMEM;
1150 #endif
1151 
1152 	return 0;
1153 }
1154 
1155 static __net_exit void pppoe_exit_net(struct net *net)
1156 {
1157 	remove_proc_entry("pppoe", net->proc_net);
1158 }
1159 
1160 static struct pernet_operations pppoe_net_ops = {
1161 	.init = pppoe_init_net,
1162 	.exit = pppoe_exit_net,
1163 	.id   = &pppoe_net_id,
1164 	.size = sizeof(struct pppoe_net),
1165 };
1166 
1167 static int __init pppoe_init(void)
1168 {
1169 	int err;
1170 
1171 	err = register_pernet_device(&pppoe_net_ops);
1172 	if (err)
1173 		goto out;
1174 
1175 	err = proto_register(&pppoe_sk_proto, 0);
1176 	if (err)
1177 		goto out_unregister_net_ops;
1178 
1179 	err = register_pppox_proto(PX_PROTO_OE, &pppoe_proto);
1180 	if (err)
1181 		goto out_unregister_pppoe_proto;
1182 
1183 	dev_add_pack(&pppoes_ptype);
1184 	dev_add_pack(&pppoed_ptype);
1185 	register_netdevice_notifier(&pppoe_notifier);
1186 
1187 	return 0;
1188 
1189 out_unregister_pppoe_proto:
1190 	proto_unregister(&pppoe_sk_proto);
1191 out_unregister_net_ops:
1192 	unregister_pernet_device(&pppoe_net_ops);
1193 out:
1194 	return err;
1195 }
1196 
1197 static void __exit pppoe_exit(void)
1198 {
1199 	unregister_netdevice_notifier(&pppoe_notifier);
1200 	dev_remove_pack(&pppoed_ptype);
1201 	dev_remove_pack(&pppoes_ptype);
1202 	unregister_pppox_proto(PX_PROTO_OE);
1203 	proto_unregister(&pppoe_sk_proto);
1204 	unregister_pernet_device(&pppoe_net_ops);
1205 }
1206 
1207 module_init(pppoe_init);
1208 module_exit(pppoe_exit);
1209 
1210 MODULE_AUTHOR("Michal Ostrowski <mostrows@speakeasy.net>");
1211 MODULE_DESCRIPTION("PPP over Ethernet driver");
1212 MODULE_LICENSE("GPL");
1213 MODULE_ALIAS_NET_PF_PROTO(PF_PPPOX, PX_PROTO_OE);
1214