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
92static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb);
93
94static const struct proto_ops pppoe_ops;
95static const struct ppp_channel_ops pppoe_chan_ops;
96
97/* per-net private data for this module */
98static int pppoe_net_id __read_mostly;
99struct 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 */
120static inline bool stage_session(__be16 sid)
121{
122	return sid != 0;
123}
124
125static 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
132static inline int cmp_2_addr(struct pppoe_addr *a, struct pppoe_addr *b)
133{
134	return a->sid == b->sid && !memcmp(a->remote, b->remote, ETH_ALEN);
135}
136
137static inline int cmp_addr(struct pppoe_addr *a, __be16 sid, char *addr)
138{
139	return a->sid == sid && !memcmp(a->remote, addr, ETH_ALEN);
140}
141
142#if 8 % PPPOE_HASH_BITS
143#error 8 must be a multiple of PPPOE_HASH_BITS
144#endif
145
146static 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 **********************************************************************/
166static 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
184static 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
204static struct pppox_sock *__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	return ret;
225}
226
227/**********************************************************************
228 *
229 *  Set/get/delete/rehash items
230 *
231 **********************************************************************/
232static inline struct pppox_sock *get_item(struct pppoe_net *pn, __be16 sid,
233					unsigned char *addr, int ifindex)
234{
235	struct pppox_sock *po;
236
237	read_lock_bh(&pn->hash_lock);
238	po = __get_item(pn, sid, addr, ifindex);
239	if (po)
240		sock_hold(sk_pppox(po));
241	read_unlock_bh(&pn->hash_lock);
242
243	return po;
244}
245
246static inline struct pppox_sock *get_item_by_addr(struct net *net,
247						struct sockaddr_pppox *sp)
248{
249	struct net_device *dev;
250	struct pppoe_net *pn;
251	struct pppox_sock *pppox_sock = NULL;
252
253	int ifindex;
254
255	rcu_read_lock();
256	dev = dev_get_by_name_rcu(net, sp->sa_addr.pppoe.dev);
257	if (dev) {
258		ifindex = dev->ifindex;
259		pn = pppoe_pernet(net);
260		pppox_sock = get_item(pn, sp->sa_addr.pppoe.sid,
261				sp->sa_addr.pppoe.remote, ifindex);
262	}
263	rcu_read_unlock();
264	return pppox_sock;
265}
266
267static inline struct pppox_sock *delete_item(struct pppoe_net *pn, __be16 sid,
268					char *addr, int ifindex)
269{
270	struct pppox_sock *ret;
271
272	write_lock_bh(&pn->hash_lock);
273	ret = __delete_item(pn, sid, addr, ifindex);
274	write_unlock_bh(&pn->hash_lock);
275
276	return ret;
277}
278
279/***************************************************************************
280 *
281 *  Handler for device events.
282 *  Certain device events require that sockets be unconnected.
283 *
284 **************************************************************************/
285
286static void pppoe_flush_dev(struct net_device *dev)
287{
288	struct pppoe_net *pn;
289	int i;
290
291	pn = pppoe_pernet(dev_net(dev));
292	write_lock_bh(&pn->hash_lock);
293	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
294		struct pppox_sock *po = pn->hash_table[i];
295		struct sock *sk;
296
297		while (po) {
298			while (po && po->pppoe_dev != dev) {
299				po = po->next;
300			}
301
302			if (!po)
303				break;
304
305			sk = sk_pppox(po);
306
307			/* We always grab the socket lock, followed by the
308			 * hash_lock, in that order.  Since we should hold the
309			 * sock lock while doing any unbinding, we need to
310			 * release the lock we're holding.  Hold a reference to
311			 * the sock so it doesn't disappear as we're jumping
312			 * between locks.
313			 */
314
315			sock_hold(sk);
316			write_unlock_bh(&pn->hash_lock);
317			lock_sock(sk);
318
319			if (po->pppoe_dev == dev &&
320			    sk->sk_state & (PPPOX_CONNECTED | PPPOX_BOUND | PPPOX_ZOMBIE)) {
321				pppox_unbind_sock(sk);
322				sk->sk_state = PPPOX_ZOMBIE;
323				sk->sk_state_change(sk);
324				po->pppoe_dev = NULL;
325				dev_put(dev);
326			}
327
328			release_sock(sk);
329			sock_put(sk);
330
331			/* Restart the process from the start of the current
332			 * hash chain. We dropped locks so the world may have
333			 * change from underneath us.
334			 */
335
336			BUG_ON(pppoe_pernet(dev_net(dev)) == NULL);
337			write_lock_bh(&pn->hash_lock);
338			po = pn->hash_table[i];
339		}
340	}
341	write_unlock_bh(&pn->hash_lock);
342}
343
344static int pppoe_device_event(struct notifier_block *this,
345			      unsigned long event, void *ptr)
346{
347	struct net_device *dev = (struct net_device *)ptr;
348
349	/* Only look at sockets that are using this specific device. */
350	switch (event) {
351	case NETDEV_CHANGEADDR:
352	case NETDEV_CHANGEMTU:
353		/* A change in mtu or address is a bad thing, requiring
354		 * LCP re-negotiation.
355		 */
356
357	case NETDEV_GOING_DOWN:
358	case NETDEV_DOWN:
359		/* Find every socket on this device and kill it. */
360		pppoe_flush_dev(dev);
361		break;
362
363	default:
364		break;
365	}
366
367	return NOTIFY_DONE;
368}
369
370static struct notifier_block pppoe_notifier = {
371	.notifier_call = pppoe_device_event,
372};
373
374/************************************************************************
375 *
376 * Do the real work of receiving a PPPoE Session frame.
377 *
378 ***********************************************************************/
379static int pppoe_rcv_core(struct sock *sk, struct sk_buff *skb)
380{
381	struct pppox_sock *po = pppox_sk(sk);
382	struct pppox_sock *relay_po;
383
384	/* Backlog receive. Semantics of backlog rcv preclude any code from
385	 * executing in lock_sock()/release_sock() bounds; meaning sk->sk_state
386	 * can't change.
387	 */
388
389	if (sk->sk_state & PPPOX_BOUND) {
390		ppp_input(&po->chan, skb);
391	} else if (sk->sk_state & PPPOX_RELAY) {
392		relay_po = get_item_by_addr(sock_net(sk),
393					    &po->pppoe_relay);
394		if (relay_po == NULL)
395			goto abort_kfree;
396
397		if ((sk_pppox(relay_po)->sk_state & PPPOX_CONNECTED) == 0)
398			goto abort_put;
399
400		if (!__pppoe_xmit(sk_pppox(relay_po), skb))
401			goto abort_put;
402	} else {
403		if (sock_queue_rcv_skb(sk, skb))
404			goto abort_kfree;
405	}
406
407	return NET_RX_SUCCESS;
408
409abort_put:
410	sock_put(sk_pppox(relay_po));
411
412abort_kfree:
413	kfree_skb(skb);
414	return NET_RX_DROP;
415}
416
417/************************************************************************
418 *
419 * Receive wrapper called in BH context.
420 *
421 ***********************************************************************/
422static int pppoe_rcv(struct sk_buff *skb, struct net_device *dev,
423		     struct packet_type *pt, struct net_device *orig_dev)
424{
425	struct pppoe_hdr *ph;
426	struct pppox_sock *po;
427	struct pppoe_net *pn;
428	int len;
429
430	skb = skb_share_check(skb, GFP_ATOMIC);
431	if (!skb)
432		goto out;
433
434	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
435		goto drop;
436
437	ph = pppoe_hdr(skb);
438	len = ntohs(ph->length);
439
440	skb_pull_rcsum(skb, sizeof(*ph));
441	if (skb->len < len)
442		goto drop;
443
444	if (pskb_trim_rcsum(skb, len))
445		goto drop;
446
447	pn = pppoe_pernet(dev_net(dev));
448
449	/* Note that get_item does a sock_hold(), so sk_pppox(po)
450	 * is known to be safe.
451	 */
452	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
453	if (!po)
454		goto drop;
455
456	return sk_receive_skb(sk_pppox(po), skb, 0);
457
458drop:
459	kfree_skb(skb);
460out:
461	return NET_RX_DROP;
462}
463
464/************************************************************************
465 *
466 * Receive a PPPoE Discovery frame.
467 * This is solely for detection of PADT frames
468 *
469 ***********************************************************************/
470static int pppoe_disc_rcv(struct sk_buff *skb, struct net_device *dev,
471			  struct packet_type *pt, struct net_device *orig_dev)
472
473{
474	struct pppoe_hdr *ph;
475	struct pppox_sock *po;
476	struct pppoe_net *pn;
477
478	skb = skb_share_check(skb, GFP_ATOMIC);
479	if (!skb)
480		goto out;
481
482	if (!pskb_may_pull(skb, sizeof(struct pppoe_hdr)))
483		goto abort;
484
485	ph = pppoe_hdr(skb);
486	if (ph->code != PADT_CODE)
487		goto abort;
488
489	pn = pppoe_pernet(dev_net(dev));
490	po = get_item(pn, ph->sid, eth_hdr(skb)->h_source, dev->ifindex);
491	if (po) {
492		struct sock *sk = sk_pppox(po);
493
494		bh_lock_sock(sk);
495
496		/* If the user has locked the socket, just ignore
497		 * the packet.  With the way two rcv protocols hook into
498		 * one socket family type, we cannot (easily) distinguish
499		 * what kind of SKB it is during backlog rcv.
500		 */
501		if (sock_owned_by_user(sk) == 0) {
502			/* We're no longer connect at the PPPOE layer,
503			 * and must wait for ppp channel to disconnect us.
504			 */
505			sk->sk_state = PPPOX_ZOMBIE;
506		}
507
508		bh_unlock_sock(sk);
509		sock_put(sk);
510	}
511
512abort:
513	kfree_skb(skb);
514out:
515	return NET_RX_SUCCESS; /* Lies... :-) */
516}
517
518static struct packet_type pppoes_ptype __read_mostly = {
519	.type	= cpu_to_be16(ETH_P_PPP_SES),
520	.func	= pppoe_rcv,
521};
522
523static struct packet_type pppoed_ptype __read_mostly = {
524	.type	= cpu_to_be16(ETH_P_PPP_DISC),
525	.func	= pppoe_disc_rcv,
526};
527
528static struct proto pppoe_sk_proto __read_mostly = {
529	.name	  = "PPPOE",
530	.owner	  = THIS_MODULE,
531	.obj_size = sizeof(struct pppox_sock),
532};
533
534/***********************************************************************
535 *
536 * Initialize a new struct sock.
537 *
538 **********************************************************************/
539static int pppoe_create(struct net *net, struct socket *sock)
540{
541	struct sock *sk;
542
543	sk = sk_alloc(net, PF_PPPOX, GFP_KERNEL, &pppoe_sk_proto);
544	if (!sk)
545		return -ENOMEM;
546
547	sock_init_data(sock, sk);
548
549	sock->state	= SS_UNCONNECTED;
550	sock->ops	= &pppoe_ops;
551
552	sk->sk_backlog_rcv	= pppoe_rcv_core;
553	sk->sk_state		= PPPOX_NONE;
554	sk->sk_type		= SOCK_STREAM;
555	sk->sk_family		= PF_PPPOX;
556	sk->sk_protocol		= PX_PROTO_OE;
557
558	return 0;
559}
560
561static 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 (sk->sk_state & (PPPOX_CONNECTED | PPPOX_BOUND)) {
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
609static 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		memset(sk_pppox(po) + 1, 0,
652		       sizeof(struct pppox_sock) - sizeof(struct sock));
653		sk->sk_state = PPPOX_NONE;
654	}
655
656	/* Re-bind in session stage only */
657	if (stage_session(sp->sa_addr.pppoe.sid)) {
658		error = -ENODEV;
659		net = sock_net(sk);
660		dev = dev_get_by_name(net, sp->sa_addr.pppoe.dev);
661		if (!dev)
662			goto err_put;
663
664		po->pppoe_dev = dev;
665		po->pppoe_ifindex = dev->ifindex;
666		pn = pppoe_pernet(net);
667		if (!(dev->flags & IFF_UP)) {
668			goto err_put;
669		}
670
671		memcpy(&po->pppoe_pa,
672		       &sp->sa_addr.pppoe,
673		       sizeof(struct pppoe_addr));
674
675		write_lock_bh(&pn->hash_lock);
676		error = __set_item(pn, po);
677		write_unlock_bh(&pn->hash_lock);
678		if (error < 0)
679			goto err_put;
680
681		po->chan.hdrlen = (sizeof(struct pppoe_hdr) +
682				   dev->hard_header_len);
683
684		po->chan.mtu = dev->mtu - sizeof(struct pppoe_hdr);
685		po->chan.private = sk;
686		po->chan.ops = &pppoe_chan_ops;
687
688		error = ppp_register_net_channel(dev_net(dev), &po->chan);
689		if (error) {
690			delete_item(pn, po->pppoe_pa.sid,
691				    po->pppoe_pa.remote, po->pppoe_ifindex);
692			goto err_put;
693		}
694
695		sk->sk_state = PPPOX_CONNECTED;
696	}
697
698	po->num = sp->sa_addr.pppoe.sid;
699
700end:
701	release_sock(sk);
702	return error;
703err_put:
704	if (po->pppoe_dev) {
705		dev_put(po->pppoe_dev);
706		po->pppoe_dev = NULL;
707	}
708	goto end;
709}
710
711static int pppoe_getname(struct socket *sock, struct sockaddr *uaddr,
712		  int *usockaddr_len, int peer)
713{
714	int len = sizeof(struct sockaddr_pppox);
715	struct sockaddr_pppox sp;
716
717	sp.sa_family	= AF_PPPOX;
718	sp.sa_protocol	= PX_PROTO_OE;
719	memcpy(&sp.sa_addr.pppoe, &pppox_sk(sock->sk)->pppoe_pa,
720	       sizeof(struct pppoe_addr));
721
722	memcpy(uaddr, &sp, len);
723
724	*usockaddr_len = len;
725
726	return 0;
727}
728
729static int pppoe_ioctl(struct socket *sock, unsigned int cmd,
730		unsigned long arg)
731{
732	struct sock *sk = sock->sk;
733	struct pppox_sock *po = pppox_sk(sk);
734	int val;
735	int err;
736
737	switch (cmd) {
738	case PPPIOCGMRU:
739		err = -ENXIO;
740		if (!(sk->sk_state & PPPOX_CONNECTED))
741			break;
742
743		err = -EFAULT;
744		if (put_user(po->pppoe_dev->mtu -
745			     sizeof(struct pppoe_hdr) -
746			     PPP_HDRLEN,
747			     (int __user *)arg))
748			break;
749		err = 0;
750		break;
751
752	case PPPIOCSMRU:
753		err = -ENXIO;
754		if (!(sk->sk_state & PPPOX_CONNECTED))
755			break;
756
757		err = -EFAULT;
758		if (get_user(val, (int __user *)arg))
759			break;
760
761		if (val < (po->pppoe_dev->mtu
762			   - sizeof(struct pppoe_hdr)
763			   - PPP_HDRLEN))
764			err = 0;
765		else
766			err = -EINVAL;
767		break;
768
769	case PPPIOCSFLAGS:
770		err = -EFAULT;
771		if (get_user(val, (int __user *)arg))
772			break;
773		err = 0;
774		break;
775
776	case PPPOEIOCSFWD:
777	{
778		struct pppox_sock *relay_po;
779
780		err = -EBUSY;
781		if (sk->sk_state & (PPPOX_BOUND | PPPOX_ZOMBIE | PPPOX_DEAD))
782			break;
783
784		err = -ENOTCONN;
785		if (!(sk->sk_state & PPPOX_CONNECTED))
786			break;
787
788		/* PPPoE address from the user specifies an outbound
789		   PPPoE address which frames are forwarded to */
790		err = -EFAULT;
791		if (copy_from_user(&po->pppoe_relay,
792				   (void __user *)arg,
793				   sizeof(struct sockaddr_pppox)))
794			break;
795
796		err = -EINVAL;
797		if (po->pppoe_relay.sa_family != AF_PPPOX ||
798		    po->pppoe_relay.sa_protocol != PX_PROTO_OE)
799			break;
800
801		/* Check that the socket referenced by the address
802		   actually exists. */
803		relay_po = get_item_by_addr(sock_net(sk), &po->pppoe_relay);
804		if (!relay_po)
805			break;
806
807		sock_put(sk_pppox(relay_po));
808		sk->sk_state |= PPPOX_RELAY;
809		err = 0;
810		break;
811	}
812
813	case PPPOEIOCDFWD:
814		err = -EALREADY;
815		if (!(sk->sk_state & PPPOX_RELAY))
816			break;
817
818		sk->sk_state &= ~PPPOX_RELAY;
819		err = 0;
820		break;
821
822	default:
823		err = -ENOTTY;
824	}
825
826	return err;
827}
828
829static int pppoe_sendmsg(struct kiocb *iocb, struct socket *sock,
830		  struct msghdr *m, size_t total_len)
831{
832	struct sk_buff *skb;
833	struct sock *sk = sock->sk;
834	struct pppox_sock *po = pppox_sk(sk);
835	int error;
836	struct pppoe_hdr hdr;
837	struct pppoe_hdr *ph;
838	struct net_device *dev;
839	char *start;
840
841	lock_sock(sk);
842	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED)) {
843		error = -ENOTCONN;
844		goto end;
845	}
846
847	hdr.ver = 1;
848	hdr.type = 1;
849	hdr.code = 0;
850	hdr.sid = po->num;
851
852	dev = po->pppoe_dev;
853
854	error = -EMSGSIZE;
855	if (total_len > (dev->mtu + dev->hard_header_len))
856		goto end;
857
858
859	skb = sock_wmalloc(sk, total_len + dev->hard_header_len + 32,
860			   0, GFP_KERNEL);
861	if (!skb) {
862		error = -ENOMEM;
863		goto end;
864	}
865
866	/* Reserve space for headers. */
867	skb_reserve(skb, dev->hard_header_len);
868	skb_reset_network_header(skb);
869
870	skb->dev = dev;
871
872	skb->priority = sk->sk_priority;
873	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
874
875	ph = (struct pppoe_hdr *)skb_put(skb, total_len + sizeof(struct pppoe_hdr));
876	start = (char *)&ph->tag[0];
877
878	error = memcpy_fromiovec(start, m->msg_iov, total_len);
879	if (error < 0) {
880		kfree_skb(skb);
881		goto end;
882	}
883
884	error = total_len;
885	dev_hard_header(skb, dev, ETH_P_PPP_SES,
886			po->pppoe_pa.remote, NULL, total_len);
887
888	memcpy(ph, &hdr, sizeof(struct pppoe_hdr));
889
890	ph->length = htons(total_len);
891
892	dev_queue_xmit(skb);
893
894end:
895	release_sock(sk);
896	return error;
897}
898
899/************************************************************************
900 *
901 * xmit function for internal use.
902 *
903 ***********************************************************************/
904static int __pppoe_xmit(struct sock *sk, struct sk_buff *skb)
905{
906	struct pppox_sock *po = pppox_sk(sk);
907	struct net_device *dev = po->pppoe_dev;
908	struct pppoe_hdr *ph;
909	int data_len = skb->len;
910
911	/* The higher-level PPP code (ppp_unregister_channel()) ensures the PPP
912	 * xmit operations conclude prior to an unregistration call.  Thus
913	 * sk->sk_state cannot change, so we don't need to do lock_sock().
914	 * But, we also can't do a lock_sock since that introduces a potential
915	 * deadlock as we'd reverse the lock ordering used when calling
916	 * ppp_unregister_channel().
917	 */
918
919	if (sock_flag(sk, SOCK_DEAD) || !(sk->sk_state & PPPOX_CONNECTED))
920		goto abort;
921
922	if (!dev)
923		goto abort;
924
925	/* Copy the data if there is no space for the header or if it's
926	 * read-only.
927	 */
928	if (skb_cow_head(skb, sizeof(*ph) + dev->hard_header_len))
929		goto abort;
930
931	__skb_push(skb, sizeof(*ph));
932	skb_reset_network_header(skb);
933
934	ph = pppoe_hdr(skb);
935	ph->ver	= 1;
936	ph->type = 1;
937	ph->code = 0;
938	ph->sid	= po->num;
939	ph->length = htons(data_len);
940
941	skb->protocol = cpu_to_be16(ETH_P_PPP_SES);
942	skb->dev = dev;
943
944	dev_hard_header(skb, dev, ETH_P_PPP_SES,
945			po->pppoe_pa.remote, NULL, data_len);
946
947	dev_queue_xmit(skb);
948	return 1;
949
950abort:
951	kfree_skb(skb);
952	return 1;
953}
954
955/************************************************************************
956 *
957 * xmit function called by generic PPP driver
958 * sends PPP frame over PPPoE socket
959 *
960 ***********************************************************************/
961static int pppoe_xmit(struct ppp_channel *chan, struct sk_buff *skb)
962{
963	struct sock *sk = (struct sock *)chan->private;
964	return __pppoe_xmit(sk, skb);
965}
966
967static const struct ppp_channel_ops pppoe_chan_ops = {
968	.start_xmit = pppoe_xmit,
969};
970
971static int pppoe_recvmsg(struct kiocb *iocb, struct socket *sock,
972		  struct msghdr *m, size_t total_len, int flags)
973{
974	struct sock *sk = sock->sk;
975	struct sk_buff *skb;
976	int error = 0;
977
978	if (sk->sk_state & PPPOX_BOUND) {
979		error = -EIO;
980		goto end;
981	}
982
983	skb = skb_recv_datagram(sk, flags & ~MSG_DONTWAIT,
984				flags & MSG_DONTWAIT, &error);
985	if (error < 0)
986		goto end;
987
988	m->msg_namelen = 0;
989
990	if (skb) {
991		total_len = min_t(size_t, total_len, skb->len);
992		error = skb_copy_datagram_iovec(skb, 0, m->msg_iov, total_len);
993		if (error == 0)
994			error = total_len;
995	}
996
997	kfree_skb(skb);
998end:
999	return error;
1000}
1001
1002#ifdef CONFIG_PROC_FS
1003static int pppoe_seq_show(struct seq_file *seq, void *v)
1004{
1005	struct pppox_sock *po;
1006	char *dev_name;
1007
1008	if (v == SEQ_START_TOKEN) {
1009		seq_puts(seq, "Id       Address              Device\n");
1010		goto out;
1011	}
1012
1013	po = v;
1014	dev_name = po->pppoe_pa.dev;
1015
1016	seq_printf(seq, "%08X %pM %8s\n",
1017		po->pppoe_pa.sid, po->pppoe_pa.remote, dev_name);
1018out:
1019	return 0;
1020}
1021
1022static inline struct pppox_sock *pppoe_get_idx(struct pppoe_net *pn, loff_t pos)
1023{
1024	struct pppox_sock *po;
1025	int i;
1026
1027	for (i = 0; i < PPPOE_HASH_SIZE; i++) {
1028		po = pn->hash_table[i];
1029		while (po) {
1030			if (!pos--)
1031				goto out;
1032			po = po->next;
1033		}
1034	}
1035
1036out:
1037	return po;
1038}
1039
1040static void *pppoe_seq_start(struct seq_file *seq, loff_t *pos)
1041	__acquires(pn->hash_lock)
1042{
1043	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1044	loff_t l = *pos;
1045
1046	read_lock_bh(&pn->hash_lock);
1047	return l ? pppoe_get_idx(pn, --l) : SEQ_START_TOKEN;
1048}
1049
1050static void *pppoe_seq_next(struct seq_file *seq, void *v, loff_t *pos)
1051{
1052	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1053	struct pppox_sock *po;
1054
1055	++*pos;
1056	if (v == SEQ_START_TOKEN) {
1057		po = pppoe_get_idx(pn, 0);
1058		goto out;
1059	}
1060	po = v;
1061	if (po->next)
1062		po = po->next;
1063	else {
1064		int hash = hash_item(po->pppoe_pa.sid, po->pppoe_pa.remote);
1065
1066		po = NULL;
1067		while (++hash < PPPOE_HASH_SIZE) {
1068			po = pn->hash_table[hash];
1069			if (po)
1070				break;
1071		}
1072	}
1073
1074out:
1075	return po;
1076}
1077
1078static void pppoe_seq_stop(struct seq_file *seq, void *v)
1079	__releases(pn->hash_lock)
1080{
1081	struct pppoe_net *pn = pppoe_pernet(seq_file_net(seq));
1082	read_unlock_bh(&pn->hash_lock);
1083}
1084
1085static const struct seq_operations pppoe_seq_ops = {
1086	.start		= pppoe_seq_start,
1087	.next		= pppoe_seq_next,
1088	.stop		= pppoe_seq_stop,
1089	.show		= pppoe_seq_show,
1090};
1091
1092static int pppoe_seq_open(struct inode *inode, struct file *file)
1093{
1094	return seq_open_net(inode, file, &pppoe_seq_ops,
1095			sizeof(struct seq_net_private));
1096}
1097
1098static const struct file_operations pppoe_seq_fops = {
1099	.owner		= THIS_MODULE,
1100	.open		= pppoe_seq_open,
1101	.read		= seq_read,
1102	.llseek		= seq_lseek,
1103	.release	= seq_release_net,
1104};
1105
1106#endif /* CONFIG_PROC_FS */
1107
1108static const struct proto_ops pppoe_ops = {
1109	.family		= AF_PPPOX,
1110	.owner		= THIS_MODULE,
1111	.release	= pppoe_release,
1112	.bind		= sock_no_bind,
1113	.connect	= pppoe_connect,
1114	.socketpair	= sock_no_socketpair,
1115	.accept		= sock_no_accept,
1116	.getname	= pppoe_getname,
1117	.poll		= datagram_poll,
1118	.listen		= sock_no_listen,
1119	.shutdown	= sock_no_shutdown,
1120	.setsockopt	= sock_no_setsockopt,
1121	.getsockopt	= sock_no_getsockopt,
1122	.sendmsg	= pppoe_sendmsg,
1123	.recvmsg	= pppoe_recvmsg,
1124	.mmap		= sock_no_mmap,
1125	.ioctl		= pppox_ioctl,
1126};
1127
1128static const struct pppox_proto pppoe_proto = {
1129	.create	= pppoe_create,
1130	.ioctl	= pppoe_ioctl,
1131	.owner	= THIS_MODULE,
1132};
1133
1134static __net_init int pppoe_init_net(struct net *net)
1135{
1136	struct pppoe_net *pn = pppoe_pernet(net);
1137	struct proc_dir_entry *pde;
1138
1139	rwlock_init(&pn->hash_lock);
1140
1141	pde = proc_net_fops_create(net, "pppoe", S_IRUGO, &pppoe_seq_fops);
1142#ifdef CONFIG_PROC_FS
1143	if (!pde)
1144		return -ENOMEM;
1145#endif
1146
1147	return 0;
1148}
1149
1150static __net_exit void pppoe_exit_net(struct net *net)
1151{
1152	proc_net_remove(net, "pppoe");
1153}
1154
1155static struct pernet_operations pppoe_net_ops = {
1156	.init = pppoe_init_net,
1157	.exit = pppoe_exit_net,
1158	.id   = &pppoe_net_id,
1159	.size = sizeof(struct pppoe_net),
1160};
1161
1162static int __init pppoe_init(void)
1163{
1164	int err;
1165
1166	err = register_pernet_device(&pppoe_net_ops);
1167	if (err)
1168		goto out;
1169
1170	err = proto_register(&pppoe_sk_proto, 0);
1171	if (err)
1172		goto out_unregister_net_ops;
1173
1174	err = register_pppox_proto(PX_PROTO_OE, &pppoe_proto);
1175	if (err)
1176		goto out_unregister_pppoe_proto;
1177
1178	dev_add_pack(&pppoes_ptype);
1179	dev_add_pack(&pppoed_ptype);
1180	register_netdevice_notifier(&pppoe_notifier);
1181
1182	return 0;
1183
1184out_unregister_pppoe_proto:
1185	proto_unregister(&pppoe_sk_proto);
1186out_unregister_net_ops:
1187	unregister_pernet_device(&pppoe_net_ops);
1188out:
1189	return err;
1190}
1191
1192static void __exit pppoe_exit(void)
1193{
1194	unregister_netdevice_notifier(&pppoe_notifier);
1195	dev_remove_pack(&pppoed_ptype);
1196	dev_remove_pack(&pppoes_ptype);
1197	unregister_pppox_proto(PX_PROTO_OE);
1198	proto_unregister(&pppoe_sk_proto);
1199	unregister_pernet_device(&pppoe_net_ops);
1200}
1201
1202module_init(pppoe_init);
1203module_exit(pppoe_exit);
1204
1205MODULE_AUTHOR("Michal Ostrowski <mostrows@speakeasy.net>");
1206MODULE_DESCRIPTION("PPP over Ethernet driver");
1207MODULE_LICENSE("GPL");
1208MODULE_ALIAS_NETPROTO(PF_PPPOX);
1209