xref: /openbmc/linux/drivers/net/ppp/pppoe.c (revision dc6a81c3)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /** -*- linux-c -*- ***********************************************************
3  * Linux PPP over Ethernet (PPPoX/PPPoE) Sockets
4  *
5  * PPPoX --- Generic PPP encapsulation socket family
6  * PPPoE --- PPP over Ethernet (RFC 2516)
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  */
54 
55 #include <linux/string.h>
56 #include <linux/module.h>
57 #include <linux/kernel.h>
58 #include <linux/slab.h>
59 #include <linux/errno.h>
60 #include <linux/netdevice.h>
61 #include <linux/net.h>
62 #include <linux/inetdevice.h>
63 #include <linux/etherdevice.h>
64 #include <linux/skbuff.h>
65 #include <linux/init.h>
66 #include <linux/if_ether.h>
67 #include <linux/if_pppox.h>
68 #include <linux/ppp_channel.h>
69 #include <linux/ppp_defs.h>
70 #include <linux/ppp-ioctl.h>
71 #include <linux/notifier.h>
72 #include <linux/file.h>
73 #include <linux/proc_fs.h>
74 #include <linux/seq_file.h>
75 
76 #include <linux/nsproxy.h>
77 #include <net/net_namespace.h>
78 #include <net/netns/generic.h>
79 #include <net/sock.h>
80 
81 #include <linux/uaccess.h>
82 
83 #define PPPOE_HASH_BITS 4
84 #define PPPOE_HASH_SIZE (1 << PPPOE_HASH_BITS)
85 #define PPPOE_HASH_MASK	(PPPOE_HASH_SIZE - 1)
86 
87 static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb);
88 
89 static const struct proto_ops pppoe_ops;
90 static const struct ppp_channel_ops pppoe_chan_ops;
91 
92 /* per-net private data for this module */
93 static unsigned int pppoe_net_id __read_mostly;
94 struct pppoe_net {
95 	/*
96 	 * we could use _single_ hash table for all
97 	 * nets by injecting net id into the hash but
98 	 * it would increase hash chains and add
99 	 * a few additional math comparations messy
100 	 * as well, moreover in case of SMP less locking
101 	 * controversy here
102 	 */
103 	struct pppox_sock *hash_table[PPPOE_HASH_SIZE];
104 	rwlock_t hash_lock;
105 };
106 
107 /*
108  * PPPoE could be in the following stages:
109  * 1) Discovery stage (to obtain remote MAC and Session ID)
110  * 2) Session stage (MAC and SID are known)
111  *
112  * Ethernet frames have a special tag for this but
113  * we use simpler approach based on session id
114  */
115 static inline bool stage_session(__be16 sid)
116 {
117 	return sid != 0;
118 }
119 
120 static inline struct pppoe_net *pppoe_pernet(struct net *net)
121 {
122 	return net_generic(net, pppoe_net_id);
123 }
124 
125 static inline int cmp_2_addr(struct pppoe_addr *a, struct pppoe_addr *b)
126 {
127 	return a->sid == b->sid && ether_addr_equal(a->remote, b->remote);
128 }
129 
130 static inline int cmp_addr(struct pppoe_addr *a, __be16 sid, char *addr)
131 {
132 	return a->sid == sid && ether_addr_equal(a->remote, addr);
133 }
134 
135 #if 8 % PPPOE_HASH_BITS
136 #error 8 must be a multiple of PPPOE_HASH_BITS
137 #endif
138 
139 static int hash_item(__be16 sid, unsigned char *addr)
140 {
141 	unsigned char hash = 0;
142 	unsigned int i;
143 
144 	for (i = 0; i < ETH_ALEN; i++)
145 		hash ^= addr[i];
146 	for (i = 0; i < sizeof(sid_t) * 8; i += 8)
147 		hash ^= (__force __u32)sid >> i;
148 	for (i = 8; (i >>= 1) >= PPPOE_HASH_BITS;)
149 		hash ^= hash >> i;
150 
151 	return hash & PPPOE_HASH_MASK;
152 }
153 
154 /**********************************************************************
155  *
156  *  Set/get/delete/rehash items  (internal versions)
157  *
158  **********************************************************************/
159 static struct pppox_sock *__get_item(struct pppoe_net *pn, __be16 sid,
160 				unsigned char *addr, int ifindex)
161 {
162 	int hash = hash_item(sid, addr);
163 	struct pppox_sock *ret;
164 
165 	ret = pn->hash_table[hash];
166 	while (ret) {
167 		if (cmp_addr(&ret->pppoe_pa, sid, addr) &&
168 		    ret->pppoe_ifindex == ifindex)
169 			return ret;
170 
171 		ret = ret->next;
172 	}
173 
174 	return NULL;
175 }
176 
177 static int __set_item(struct pppoe_net *pn, struct pppox_sock *po)
178 {
179 	int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
180 	struct pppox_sock *ret;
181 
182 	ret = pn->hash_table[hash];
183 	while (ret) {
184 		if (cmp_2_addr(&ret->pppoe_pa, &po->pppoe_pa) &&
185 		    ret->pppoe_ifindex == po->pppoe_ifindex)
186 			return -EALREADY;
187 
188 		ret = ret->next;
189 	}
190 
191 	po->next = pn->hash_table[hash];
192 	pn->hash_table[hash] = po;
193 
194 	return 0;
195 }
196 
197 static void __delete_item(struct pppoe_net *pn, __be16 sid,
198 					char *addr, int ifindex)
199 {
200 	int hash = hash_item(sid, addr);
201 	struct pppox_sock *ret, **src;
202 
203 	ret = pn->hash_table[hash];
204 	src = &pn->hash_table[hash];
205 
206 	while (ret) {
207 		if (cmp_addr(&ret->pppoe_pa, sid, addr) &&
208 		    ret->pppoe_ifindex == ifindex) {
209 			*src = ret->next;
210 			break;
211 		}
212 
213 		src = &ret->next;
214 		ret = ret->next;
215 	}
216 }
217 
218 /**********************************************************************
219  *
220  *  Set/get/delete/rehash items
221  *
222  **********************************************************************/
223 static inline struct pppox_sock *get_item(struct pppoe_net *pn, __be16 sid,
224 					unsigned char *addr, int ifindex)
225 {
226 	struct pppox_sock *po;
227 
228 	read_lock_bh(&pn->hash_lock);
229 	po = __get_item(pn, sid, addr, ifindex);
230 	if (po)
231 		sock_hold(sk_pppox(po));
232 	read_unlock_bh(&pn->hash_lock);
233 
234 	return po;
235 }
236 
237 static inline struct pppox_sock *get_item_by_addr(struct net *net,
238 						struct sockaddr_pppox *sp)
239 {
240 	struct net_device *dev;
241 	struct pppoe_net *pn;
242 	struct pppox_sock *pppox_sock = NULL;
243 
244 	int ifindex;
245 
246 	rcu_read_lock();
247 	dev = dev_get_by_name_rcu(net, sp->sa_addr.pppoe.dev);
248 	if (dev) {
249 		ifindex = dev->ifindex;
250 		pn = pppoe_pernet(net);
251 		pppox_sock = get_item(pn, sp->sa_addr.pppoe.sid,
252 				sp->sa_addr.pppoe.remote, ifindex);
253 	}
254 	rcu_read_unlock();
255 	return pppox_sock;
256 }
257 
258 static inline void delete_item(struct pppoe_net *pn, __be16 sid,
259 					char *addr, int ifindex)
260 {
261 	write_lock_bh(&pn->hash_lock);
262 	__delete_item(pn, sid, addr, ifindex);
263 	write_unlock_bh(&pn->hash_lock);
264 }
265 
266 /***************************************************************************
267  *
268  *  Handler for device events.
269  *  Certain device events require that sockets be unconnected.
270  *
271  **************************************************************************/
272 
273 static void pppoe_flush_dev(struct net_device *dev)
274 {
275 	struct pppoe_net *pn;
276 	int i;
277 
278 	pn = pppoe_pernet(dev_net(dev));
279 	write_lock_bh(&pn->hash_lock);
280 	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
281 		struct pppox_sock *po = pn->hash_table[i];
282 		struct sock *sk;
283 
284 		while (po) {
285 			while (po && po->pppoe_dev != dev) {
286 				po = po->next;
287 			}
288 
289 			if (!po)
290 				break;
291 
292 			sk = sk_pppox(po);
293 
294 			/* We always grab the socket lock, followed by the
295 			 * hash_lock, in that order.  Since we should hold the
296 			 * sock lock while doing any unbinding, we need to
297 			 * release the lock we're holding.  Hold a reference to
298 			 * the sock so it doesn't disappear as we're jumping
299 			 * between locks.
300 			 */
301 
302 			sock_hold(sk);
303 			write_unlock_bh(&pn->hash_lock);
304 			lock_sock(sk);
305 
306 			if (po->pppoe_dev == dev &&
307 			    sk->sk_state & (PPPOX_CONNECTED | PPPOX_BOUND)) {
308 				pppox_unbind_sock(sk);
309 				sk->sk_state_change(sk);
310 				po->pppoe_dev = NULL;
311 				dev_put(dev);
312 			}
313 
314 			release_sock(sk);
315 			sock_put(sk);
316 
317 			/* Restart the process from the start of the current
318 			 * hash chain. We dropped locks so the world may have
319 			 * change from underneath us.
320 			 */
321 
322 			BUG_ON(pppoe_pernet(dev_net(dev)) == NULL);
323 			write_lock_bh(&pn->hash_lock);
324 			po = pn->hash_table[i];
325 		}
326 	}
327 	write_unlock_bh(&pn->hash_lock);
328 }
329 
330 static int pppoe_device_event(struct notifier_block *this,
331 			      unsigned long event, void *ptr)
332 {
333 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
334 
335 	/* Only look at sockets that are using this specific device. */
336 	switch (event) {
337 	case NETDEV_CHANGEADDR:
338 	case NETDEV_CHANGEMTU:
339 		/* A change in mtu or address is a bad thing, requiring
340 		 * LCP re-negotiation.
341 		 */
342 
343 	case NETDEV_GOING_DOWN:
344 	case NETDEV_DOWN:
345 		/* Find every socket on this device and kill it. */
346 		pppoe_flush_dev(dev);
347 		break;
348 
349 	default:
350 		break;
351 	}
352 
353 	return NOTIFY_DONE;
354 }
355 
356 static struct notifier_block pppoe_notifier = {
357 	.notifier_call = pppoe_device_event,
358 };
359 
360 /************************************************************************
361  *
362  * Do the real work of receiving a PPPoE Session frame.
363  *
364  ***********************************************************************/
365 static int pppoe_rcv_core(struct sock *sk, struct sk_buff *skb)
366 {
367 	struct pppox_sock *po = pppox_sk(sk);
368 	struct pppox_sock *relay_po;
369 
370 	/* Backlog receive. Semantics of backlog rcv preclude any code from
371 	 * executing in lock_sock()/release_sock() bounds; meaning sk->sk_state
372 	 * can't change.
373 	 */
374 
375 	if (skb->pkt_type == PACKET_OTHERHOST)
376 		goto abort_kfree;
377 
378 	if (sk->sk_state & PPPOX_BOUND) {
379 		ppp_input(&po->chan, skb);
380 	} else if (sk->sk_state & PPPOX_RELAY) {
381 		relay_po = get_item_by_addr(sock_net(sk),
382 					    &po->pppoe_relay);
383 		if (relay_po == NULL)
384 			goto abort_kfree;
385 
386 		if ((sk_pppox(relay_po)->sk_state & PPPOX_CONNECTED) == 0)
387 			goto abort_put;
388 
389 		if (!__pppoe_xmit(sk_pppox(relay_po), skb))
390 			goto abort_put;
391 
392 		sock_put(sk_pppox(relay_po));
393 	} else {
394 		if (sock_queue_rcv_skb(sk, skb))
395 			goto abort_kfree;
396 	}
397 
398 	return NET_RX_SUCCESS;
399 
400 abort_put:
401 	sock_put(sk_pppox(relay_po));
402 
403 abort_kfree:
404 	kfree_skb(skb);
405 	return NET_RX_DROP;
406 }
407 
408 /************************************************************************
409  *
410  * Receive wrapper called in BH context.
411  *
412  ***********************************************************************/
413 static int pppoe_rcv(struct sk_buff *skb, struct net_device *dev,
414 		     struct packet_type *pt, struct net_device *orig_dev)
415 {
416 	struct pppoe_hdr *ph;
417 	struct pppox_sock *po;
418 	struct pppoe_net *pn;
419 	int len;
420 
421 	skb = skb_share_check(skb, GFP_ATOMIC);
422 	if (!skb)
423 		goto out;
424 
425 	if (skb_mac_header_len(skb) < ETH_HLEN)
426 		goto drop;
427 
428 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
429 		goto drop;
430 
431 	ph = pppoe_hdr(skb);
432 	len = ntohs(ph->length);
433 
434 	skb_pull_rcsum(skb, sizeof(*ph));
435 	if (skb->len < len)
436 		goto drop;
437 
438 	if (pskb_trim_rcsum(skb, len))
439 		goto drop;
440 
441 	ph = pppoe_hdr(skb);
442 	pn = pppoe_pernet(dev_net(dev));
443 
444 	/* Note that get_item does a sock_hold(), so sk_pppox(po)
445 	 * is known to be safe.
446 	 */
447 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
448 	if (!po)
449 		goto drop;
450 
451 	return sk_receive_skb(sk_pppox(po), skb, 0);
452 
453 drop:
454 	kfree_skb(skb);
455 out:
456 	return NET_RX_DROP;
457 }
458 
459 static void pppoe_unbind_sock_work(struct work_struct *work)
460 {
461 	struct pppox_sock *po = container_of(work, struct pppox_sock,
462 					     proto.pppoe.padt_work);
463 	struct sock *sk = sk_pppox(po);
464 
465 	lock_sock(sk);
466 	if (po->pppoe_dev) {
467 		dev_put(po->pppoe_dev);
468 		po->pppoe_dev = NULL;
469 	}
470 	pppox_unbind_sock(sk);
471 	release_sock(sk);
472 	sock_put(sk);
473 }
474 
475 /************************************************************************
476  *
477  * Receive a PPPoE Discovery frame.
478  * This is solely for detection of PADT frames
479  *
480  ***********************************************************************/
481 static int pppoe_disc_rcv(struct sk_buff *skb, struct net_device *dev,
482 			  struct packet_type *pt, struct net_device *orig_dev)
483 
484 {
485 	struct pppoe_hdr *ph;
486 	struct pppox_sock *po;
487 	struct pppoe_net *pn;
488 
489 	skb = skb_share_check(skb, GFP_ATOMIC);
490 	if (!skb)
491 		goto out;
492 
493 	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
494 		goto abort;
495 
496 	ph = pppoe_hdr(skb);
497 	if (ph->code != PADT_CODE)
498 		goto abort;
499 
500 	pn = pppoe_pernet(dev_net(dev));
501 	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
502 	if (po)
503 		if (!schedule_work(&po->proto.pppoe.padt_work))
504 			sock_put(sk_pppox(po));
505 
506 abort:
507 	kfree_skb(skb);
508 out:
509 	return NET_RX_SUCCESS; /* Lies... :-) */
510 }
511 
512 static struct packet_type pppoes_ptype __read_mostly = {
513 	.type	= cpu_to_be16(ETH_P_PPP_SES),
514 	.func	= pppoe_rcv,
515 };
516 
517 static struct packet_type pppoed_ptype __read_mostly = {
518 	.type	= cpu_to_be16(ETH_P_PPP_DISC),
519 	.func	= pppoe_disc_rcv,
520 };
521 
522 static struct proto pppoe_sk_proto __read_mostly = {
523 	.name	  = "PPPOE",
524 	.owner	  = THIS_MODULE,
525 	.obj_size = sizeof(struct pppox_sock),
526 };
527 
528 /***********************************************************************
529  *
530  * Initialize a new struct sock.
531  *
532  **********************************************************************/
533 static int pppoe_create(struct net *net, struct socket *sock, int kern)
534 {
535 	struct sock *sk;
536 
537 	sk = sk_alloc(net, PF_PPPOX, GFP_KERNEL, &pppoe_sk_proto, kern);
538 	if (!sk)
539 		return -ENOMEM;
540 
541 	sock_init_data(sock, sk);
542 
543 	sock->state	= SS_UNCONNECTED;
544 	sock->ops	= &pppoe_ops;
545 
546 	sk->sk_backlog_rcv	= pppoe_rcv_core;
547 	sk->sk_state		= PPPOX_NONE;
548 	sk->sk_type		= SOCK_STREAM;
549 	sk->sk_family		= PF_PPPOX;
550 	sk->sk_protocol		= PX_PROTO_OE;
551 
552 	INIT_WORK(&pppox_sk(sk)->proto.pppoe.padt_work,
553 		  pppoe_unbind_sock_work);
554 
555 	return 0;
556 }
557 
558 static int pppoe_release(struct socket *sock)
559 {
560 	struct sock *sk = sock->sk;
561 	struct pppox_sock *po;
562 	struct pppoe_net *pn;
563 	struct net *net = NULL;
564 
565 	if (!sk)
566 		return 0;
567 
568 	lock_sock(sk);
569 	if (sock_flag(sk, SOCK_DEAD)) {
570 		release_sock(sk);
571 		return -EBADF;
572 	}
573 
574 	po = pppox_sk(sk);
575 
576 	if (po->pppoe_dev) {
577 		dev_put(po->pppoe_dev);
578 		po->pppoe_dev = NULL;
579 	}
580 
581 	pppox_unbind_sock(sk);
582 
583 	/* Signal the death of the socket. */
584 	sk->sk_state = PPPOX_DEAD;
585 
586 	net = sock_net(sk);
587 	pn = pppoe_pernet(net);
588 
589 	/*
590 	 * protect "po" from concurrent updates
591 	 * on pppoe_flush_dev
592 	 */
593 	delete_item(pn, po->pppoe_pa.sid, po->pppoe_pa.remote,
594 		    po->pppoe_ifindex);
595 
596 	sock_orphan(sk);
597 	sock->sk = NULL;
598 
599 	skb_queue_purge(&sk->sk_receive_queue);
600 	release_sock(sk);
601 	sock_put(sk);
602 
603 	return 0;
604 }
605 
606 static int pppoe_connect(struct socket *sock, struct sockaddr *uservaddr,
607 		  int sockaddr_len, int flags)
608 {
609 	struct sock *sk = sock->sk;
610 	struct sockaddr_pppox *sp = (struct sockaddr_pppox *)uservaddr;
611 	struct pppox_sock *po = pppox_sk(sk);
612 	struct net_device *dev = NULL;
613 	struct pppoe_net *pn;
614 	struct net *net = NULL;
615 	int error;
616 
617 	lock_sock(sk);
618 
619 	error = -EINVAL;
620 
621 	if (sockaddr_len != sizeof(struct sockaddr_pppox))
622 		goto end;
623 
624 	if (sp->sa_protocol != PX_PROTO_OE)
625 		goto end;
626 
627 	/* Check for already bound sockets */
628 	error = -EBUSY;
629 	if ((sk->sk_state & PPPOX_CONNECTED) &&
630 	     stage_session(sp->sa_addr.pppoe.sid))
631 		goto end;
632 
633 	/* Check for already disconnected sockets, on attempts to disconnect */
634 	error = -EALREADY;
635 	if ((sk->sk_state & PPPOX_DEAD) &&
636 	     !stage_session(sp->sa_addr.pppoe.sid))
637 		goto end;
638 
639 	error = 0;
640 
641 	/* Delete the old binding */
642 	if (stage_session(po->pppoe_pa.sid)) {
643 		pppox_unbind_sock(sk);
644 		pn = pppoe_pernet(sock_net(sk));
645 		delete_item(pn, po->pppoe_pa.sid,
646 			    po->pppoe_pa.remote, po->pppoe_ifindex);
647 		if (po->pppoe_dev) {
648 			dev_put(po->pppoe_dev);
649 			po->pppoe_dev = NULL;
650 		}
651 
652 		po->pppoe_ifindex = 0;
653 		memset(&po->pppoe_pa, 0, sizeof(po->pppoe_pa));
654 		memset(&po->pppoe_relay, 0, sizeof(po->pppoe_relay));
655 		memset(&po->chan, 0, sizeof(po->chan));
656 		po->next = NULL;
657 		po->num = 0;
658 
659 		sk->sk_state = PPPOX_NONE;
660 	}
661 
662 	/* Re-bind in session stage only */
663 	if (stage_session(sp->sa_addr.pppoe.sid)) {
664 		error = -ENODEV;
665 		net = sock_net(sk);
666 		dev = dev_get_by_name(net, sp->sa_addr.pppoe.dev);
667 		if (!dev)
668 			goto err_put;
669 
670 		po->pppoe_dev = dev;
671 		po->pppoe_ifindex = dev->ifindex;
672 		pn = pppoe_pernet(net);
673 		if (!(dev->flags & IFF_UP)) {
674 			goto err_put;
675 		}
676 
677 		memcpy(&po->pppoe_pa,
678 		       &sp->sa_addr.pppoe,
679 		       sizeof(struct pppoe_addr));
680 
681 		write_lock_bh(&pn->hash_lock);
682 		error = __set_item(pn, po);
683 		write_unlock_bh(&pn->hash_lock);
684 		if (error < 0)
685 			goto err_put;
686 
687 		po->chan.hdrlen = (sizeof(struct pppoe_hdr) +
688 				   dev->hard_header_len);
689 
690 		po->chan.mtu = dev->mtu - sizeof(struct pppoe_hdr) - 2;
691 		po->chan.private = sk;
692 		po->chan.ops = &pppoe_chan_ops;
693 
694 		error = ppp_register_net_channel(dev_net(dev), &po->chan);
695 		if (error) {
696 			delete_item(pn, po->pppoe_pa.sid,
697 				    po->pppoe_pa.remote, po->pppoe_ifindex);
698 			goto err_put;
699 		}
700 
701 		sk->sk_state = PPPOX_CONNECTED;
702 	}
703 
704 	po->num = sp->sa_addr.pppoe.sid;
705 
706 end:
707 	release_sock(sk);
708 	return error;
709 err_put:
710 	if (po->pppoe_dev) {
711 		dev_put(po->pppoe_dev);
712 		po->pppoe_dev = NULL;
713 	}
714 	goto end;
715 }
716 
717 static int pppoe_getname(struct socket *sock, struct sockaddr *uaddr,
718 		  int peer)
719 {
720 	int len = sizeof(struct sockaddr_pppox);
721 	struct sockaddr_pppox sp;
722 
723 	sp.sa_family	= AF_PPPOX;
724 	sp.sa_protocol	= PX_PROTO_OE;
725 	memcpy(&sp.sa_addr.pppoe, &pppox_sk(sock->sk)->pppoe_pa,
726 	       sizeof(struct pppoe_addr));
727 
728 	memcpy(uaddr, &sp, len);
729 
730 	return len;
731 }
732 
733 static int pppoe_ioctl(struct socket *sock, unsigned int cmd,
734 		unsigned long arg)
735 {
736 	struct sock *sk = sock->sk;
737 	struct pppox_sock *po = pppox_sk(sk);
738 	int val;
739 	int err;
740 
741 	switch (cmd) {
742 	case PPPIOCGMRU:
743 		err = -ENXIO;
744 		if (!(sk->sk_state & PPPOX_CONNECTED))
745 			break;
746 
747 		err = -EFAULT;
748 		if (put_user(po->pppoe_dev->mtu -
749 			     sizeof(struct pppoe_hdr) -
750 			     PPP_HDRLEN,
751 			     (int __user *)arg))
752 			break;
753 		err = 0;
754 		break;
755 
756 	case PPPIOCSMRU:
757 		err = -ENXIO;
758 		if (!(sk->sk_state & PPPOX_CONNECTED))
759 			break;
760 
761 		err = -EFAULT;
762 		if (get_user(val, (int __user *)arg))
763 			break;
764 
765 		if (val < (po->pppoe_dev->mtu
766 			   - sizeof(struct pppoe_hdr)
767 			   - PPP_HDRLEN))
768 			err = 0;
769 		else
770 			err = -EINVAL;
771 		break;
772 
773 	case PPPIOCSFLAGS:
774 		err = -EFAULT;
775 		if (get_user(val, (int __user *)arg))
776 			break;
777 		err = 0;
778 		break;
779 
780 	case PPPOEIOCSFWD:
781 	{
782 		struct pppox_sock *relay_po;
783 
784 		err = -EBUSY;
785 		if (sk->sk_state & (PPPOX_BOUND | PPPOX_DEAD))
786 			break;
787 
788 		err = -ENOTCONN;
789 		if (!(sk->sk_state & PPPOX_CONNECTED))
790 			break;
791 
792 		/* PPPoE address from the user specifies an outbound
793 		   PPPoE address which frames are forwarded to */
794 		err = -EFAULT;
795 		if (copy_from_user(&po->pppoe_relay,
796 				   (void __user *)arg,
797 				   sizeof(struct sockaddr_pppox)))
798 			break;
799 
800 		err = -EINVAL;
801 		if (po->pppoe_relay.sa_family != AF_PPPOX ||
802 		    po->pppoe_relay.sa_protocol != PX_PROTO_OE)
803 			break;
804 
805 		/* Check that the socket referenced by the address
806 		   actually exists. */
807 		relay_po = get_item_by_addr(sock_net(sk), &po->pppoe_relay);
808 		if (!relay_po)
809 			break;
810 
811 		sock_put(sk_pppox(relay_po));
812 		sk->sk_state |= PPPOX_RELAY;
813 		err = 0;
814 		break;
815 	}
816 
817 	case PPPOEIOCDFWD:
818 		err = -EALREADY;
819 		if (!(sk->sk_state & PPPOX_RELAY))
820 			break;
821 
822 		sk->sk_state &= ~PPPOX_RELAY;
823 		err = 0;
824 		break;
825 
826 	default:
827 		err = -ENOTTY;
828 	}
829 
830 	return err;
831 }
832 
833 static int pppoe_sendmsg(struct socket *sock, struct msghdr *m,
834 			 size_t total_len)
835 {
836 	struct sk_buff *skb;
837 	struct sock *sk = sock->sk;
838 	struct pppox_sock *po = pppox_sk(sk);
839 	int error;
840 	struct pppoe_hdr hdr;
841 	struct pppoe_hdr *ph;
842 	struct net_device *dev;
843 	char *start;
844 	int hlen;
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 	hlen = LL_RESERVED_SPACE(dev);
864 	skb = sock_wmalloc(sk, hlen + sizeof(*ph) + total_len +
865 			   dev->needed_tailroom, 0, GFP_KERNEL);
866 	if (!skb) {
867 		error = -ENOMEM;
868 		goto end;
869 	}
870 
871 	/* Reserve space for headers. */
872 	skb_reserve(skb, hlen);
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 = 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, LL_RESERVED_SPACE(dev) + sizeof(*ph)))
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 #endif /* CONFIG_PROC_FS */
1097 
1098 static const struct proto_ops pppoe_ops = {
1099 	.family		= AF_PPPOX,
1100 	.owner		= THIS_MODULE,
1101 	.release	= pppoe_release,
1102 	.bind		= sock_no_bind,
1103 	.connect	= pppoe_connect,
1104 	.socketpair	= sock_no_socketpair,
1105 	.accept		= sock_no_accept,
1106 	.getname	= pppoe_getname,
1107 	.poll		= datagram_poll,
1108 	.listen		= sock_no_listen,
1109 	.shutdown	= sock_no_shutdown,
1110 	.setsockopt	= sock_no_setsockopt,
1111 	.getsockopt	= sock_no_getsockopt,
1112 	.sendmsg	= pppoe_sendmsg,
1113 	.recvmsg	= pppoe_recvmsg,
1114 	.mmap		= sock_no_mmap,
1115 	.ioctl		= pppox_ioctl,
1116 #ifdef CONFIG_COMPAT
1117 	.compat_ioctl	= pppox_compat_ioctl,
1118 #endif
1119 };
1120 
1121 static const struct pppox_proto pppoe_proto = {
1122 	.create	= pppoe_create,
1123 	.ioctl	= pppoe_ioctl,
1124 	.owner	= THIS_MODULE,
1125 };
1126 
1127 static __net_init int pppoe_init_net(struct net *net)
1128 {
1129 	struct pppoe_net *pn = pppoe_pernet(net);
1130 	struct proc_dir_entry *pde;
1131 
1132 	rwlock_init(&pn->hash_lock);
1133 
1134 	pde = proc_create_net("pppoe", 0444, net->proc_net,
1135 			&pppoe_seq_ops, sizeof(struct seq_net_private));
1136 #ifdef CONFIG_PROC_FS
1137 	if (!pde)
1138 		return -ENOMEM;
1139 #endif
1140 
1141 	return 0;
1142 }
1143 
1144 static __net_exit void pppoe_exit_net(struct net *net)
1145 {
1146 	remove_proc_entry("pppoe", net->proc_net);
1147 }
1148 
1149 static struct pernet_operations pppoe_net_ops = {
1150 	.init = pppoe_init_net,
1151 	.exit = pppoe_exit_net,
1152 	.id   = &pppoe_net_id,
1153 	.size = sizeof(struct pppoe_net),
1154 };
1155 
1156 static int __init pppoe_init(void)
1157 {
1158 	int err;
1159 
1160 	err = register_pernet_device(&pppoe_net_ops);
1161 	if (err)
1162 		goto out;
1163 
1164 	err = proto_register(&pppoe_sk_proto, 0);
1165 	if (err)
1166 		goto out_unregister_net_ops;
1167 
1168 	err = register_pppox_proto(PX_PROTO_OE, &pppoe_proto);
1169 	if (err)
1170 		goto out_unregister_pppoe_proto;
1171 
1172 	dev_add_pack(&pppoes_ptype);
1173 	dev_add_pack(&pppoed_ptype);
1174 	register_netdevice_notifier(&pppoe_notifier);
1175 
1176 	return 0;
1177 
1178 out_unregister_pppoe_proto:
1179 	proto_unregister(&pppoe_sk_proto);
1180 out_unregister_net_ops:
1181 	unregister_pernet_device(&pppoe_net_ops);
1182 out:
1183 	return err;
1184 }
1185 
1186 static void __exit pppoe_exit(void)
1187 {
1188 	unregister_netdevice_notifier(&pppoe_notifier);
1189 	dev_remove_pack(&pppoed_ptype);
1190 	dev_remove_pack(&pppoes_ptype);
1191 	unregister_pppox_proto(PX_PROTO_OE);
1192 	proto_unregister(&pppoe_sk_proto);
1193 	unregister_pernet_device(&pppoe_net_ops);
1194 }
1195 
1196 module_init(pppoe_init);
1197 module_exit(pppoe_exit);
1198 
1199 MODULE_AUTHOR("Michal Ostrowski <mostrows@speakeasy.net>");
1200 MODULE_DESCRIPTION("PPP over Ethernet driver");
1201 MODULE_LICENSE("GPL");
1202 MODULE_ALIAS_NET_PF_PROTO(PF_PPPOX, PX_PROTO_OE);
1203