5fa906881b86a0ed800ff38141ff520e7155fad2
[pandora-kernel.git] / net / l2tp / l2tp_core.c
1 /*
2  * L2TP core.
3  *
4  * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
5  *
6  * This file contains some code of the original L2TPv2 pppol2tp
7  * driver, which has the following copyright:
8  *
9  * Authors:     Martijn van Oosterhout <kleptog@svana.org>
10  *              James Chapman (jchapman@katalix.com)
11  * Contributors:
12  *              Michal Ostrowski <mostrows@speakeasy.net>
13  *              Arnaldo Carvalho de Melo <acme@xconectiva.com.br>
14  *              David S. Miller (davem@redhat.com)
15  *
16  * This program is free software; you can redistribute it and/or modify
17  * it under the terms of the GNU General Public License version 2 as
18  * published by the Free Software Foundation.
19  */
20
21 #include <linux/module.h>
22 #include <linux/string.h>
23 #include <linux/list.h>
24 #include <linux/rculist.h>
25 #include <linux/uaccess.h>
26
27 #include <linux/kernel.h>
28 #include <linux/spinlock.h>
29 #include <linux/kthread.h>
30 #include <linux/sched.h>
31 #include <linux/slab.h>
32 #include <linux/errno.h>
33 #include <linux/jiffies.h>
34
35 #include <linux/netdevice.h>
36 #include <linux/net.h>
37 #include <linux/inetdevice.h>
38 #include <linux/skbuff.h>
39 #include <linux/init.h>
40 #include <linux/in.h>
41 #include <linux/ip.h>
42 #include <linux/udp.h>
43 #include <linux/l2tp.h>
44 #include <linux/hash.h>
45 #include <linux/sort.h>
46 #include <linux/file.h>
47 #include <linux/nsproxy.h>
48 #include <net/net_namespace.h>
49 #include <net/netns/generic.h>
50 #include <net/dst.h>
51 #include <net/ip.h>
52 #include <net/udp.h>
53 #include <net/inet_common.h>
54 #include <net/xfrm.h>
55 #include <net/protocol.h>
56
57 #include <asm/byteorder.h>
58 #include <linux/atomic.h>
59
60 #include "l2tp_core.h"
61
62 #define L2TP_DRV_VERSION        "V2.0"
63
64 /* L2TP header constants */
65 #define L2TP_HDRFLAG_T     0x8000
66 #define L2TP_HDRFLAG_L     0x4000
67 #define L2TP_HDRFLAG_S     0x0800
68 #define L2TP_HDRFLAG_O     0x0200
69 #define L2TP_HDRFLAG_P     0x0100
70
71 #define L2TP_HDR_VER_MASK  0x000F
72 #define L2TP_HDR_VER_2     0x0002
73 #define L2TP_HDR_VER_3     0x0003
74
75 /* L2TPv3 default L2-specific sublayer */
76 #define L2TP_SLFLAG_S      0x40000000
77 #define L2TP_SL_SEQ_MASK   0x00ffffff
78
79 #define L2TP_HDR_SIZE_SEQ               10
80 #define L2TP_HDR_SIZE_NOSEQ             6
81
82 /* Default trace flags */
83 #define L2TP_DEFAULT_DEBUG_FLAGS        0
84
85 #define PRINTK(_mask, _type, _lvl, _fmt, args...)                       \
86         do {                                                            \
87                 if ((_mask) & (_type))                                  \
88                         printk(_lvl "L2TP: " _fmt, ##args);             \
89         } while (0)
90
91 /* Private data stored for received packets in the skb.
92  */
93 struct l2tp_skb_cb {
94         u32                     ns;
95         u16                     has_seq;
96         u16                     length;
97         unsigned long           expires;
98 };
99
100 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
101
102 static atomic_t l2tp_tunnel_count;
103 static atomic_t l2tp_session_count;
104
105 /* per-net private data for this module */
106 static unsigned int l2tp_net_id;
107 struct l2tp_net {
108         struct list_head l2tp_tunnel_list;
109         spinlock_t l2tp_tunnel_list_lock;
110         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
111         spinlock_t l2tp_session_hlist_lock;
112 };
113
114 static void l2tp_session_set_header_len(struct l2tp_session *session, int version);
115 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel);
116
117 static inline struct l2tp_net *l2tp_pernet(const struct net *net)
118 {
119         BUG_ON(!net);
120
121         return net_generic(net, l2tp_net_id);
122 }
123
124
125 /* Session hash global list for L2TPv3.
126  * The session_id SHOULD be random according to RFC3931, but several
127  * L2TP implementations use incrementing session_ids.  So we do a real
128  * hash on the session_id, rather than a simple bitmask.
129  */
130 static inline struct hlist_head *
131 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
132 {
133         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
134
135 }
136
137 /* Lookup a session by id in the global session list
138  */
139 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
140 {
141         struct l2tp_net *pn = l2tp_pernet(net);
142         struct hlist_head *session_list =
143                 l2tp_session_id_hash_2(pn, session_id);
144         struct l2tp_session *session;
145         struct hlist_node *walk;
146
147         rcu_read_lock_bh();
148         hlist_for_each_entry_rcu(session, walk, session_list, global_hlist) {
149                 if (session->session_id == session_id) {
150                         rcu_read_unlock_bh();
151                         return session;
152                 }
153         }
154         rcu_read_unlock_bh();
155
156         return NULL;
157 }
158
159 /* Session hash list.
160  * The session_id SHOULD be random according to RFC2661, but several
161  * L2TP implementations (Cisco and Microsoft) use incrementing
162  * session_ids.  So we do a real hash on the session_id, rather than a
163  * simple bitmask.
164  */
165 static inline struct hlist_head *
166 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
167 {
168         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
169 }
170
171 /* Lookup a session by id
172  */
173 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
174 {
175         struct hlist_head *session_list;
176         struct l2tp_session *session;
177         struct hlist_node *walk;
178
179         /* In L2TPv3, session_ids are unique over all tunnels and we
180          * sometimes need to look them up before we know the
181          * tunnel.
182          */
183         if (tunnel == NULL)
184                 return l2tp_session_find_2(net, session_id);
185
186         session_list = l2tp_session_id_hash(tunnel, session_id);
187         read_lock_bh(&tunnel->hlist_lock);
188         hlist_for_each_entry(session, walk, session_list, hlist) {
189                 if (session->session_id == session_id) {
190                         read_unlock_bh(&tunnel->hlist_lock);
191                         return session;
192                 }
193         }
194         read_unlock_bh(&tunnel->hlist_lock);
195
196         return NULL;
197 }
198 EXPORT_SYMBOL_GPL(l2tp_session_find);
199
200 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
201 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
202 {
203         const struct l2tp_net *pn = l2tp_pernet(net);
204         struct l2tp_tunnel *tunnel;
205
206         rcu_read_lock_bh();
207         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
208                 if (tunnel->tunnel_id == tunnel_id) {
209                         l2tp_tunnel_inc_refcount(tunnel);
210                         rcu_read_unlock_bh();
211
212                         return tunnel;
213                 }
214         }
215         rcu_read_unlock_bh();
216
217         return NULL;
218 }
219 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
220
221 /* Like l2tp_session_find() but takes a reference on the returned session.
222  * Optionally calls session->ref() too if do_ref is true.
223  */
224 struct l2tp_session *l2tp_session_get(const struct net *net,
225                                       struct l2tp_tunnel *tunnel,
226                                       u32 session_id, bool do_ref)
227 {
228         struct hlist_head *session_list;
229         struct l2tp_session *session;
230         struct hlist_node *walk;
231
232         if (!tunnel) {
233                 struct l2tp_net *pn = l2tp_pernet(net);
234
235                 session_list = l2tp_session_id_hash_2(pn, session_id);
236
237                 rcu_read_lock_bh();
238                 hlist_for_each_entry_rcu(session, walk, session_list, global_hlist) {
239                         if (session->session_id == session_id) {
240                                 l2tp_session_inc_refcount(session);
241                                 if (do_ref && session->ref)
242                                         session->ref(session);
243                                 rcu_read_unlock_bh();
244
245                                 return session;
246                         }
247                 }
248                 rcu_read_unlock_bh();
249
250                 return NULL;
251         }
252
253         session_list = l2tp_session_id_hash(tunnel, session_id);
254         read_lock_bh(&tunnel->hlist_lock);
255         hlist_for_each_entry(session, walk, session_list, hlist) {
256                 if (session->session_id == session_id) {
257                         l2tp_session_inc_refcount(session);
258                         if (do_ref && session->ref)
259                                 session->ref(session);
260                         read_unlock_bh(&tunnel->hlist_lock);
261
262                         return session;
263                 }
264         }
265         read_unlock_bh(&tunnel->hlist_lock);
266
267         return NULL;
268 }
269 EXPORT_SYMBOL_GPL(l2tp_session_get);
270
271 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth,
272                                           bool do_ref)
273 {
274         int hash;
275         struct hlist_node *walk;
276         struct l2tp_session *session;
277         int count = 0;
278
279         read_lock_bh(&tunnel->hlist_lock);
280         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
281                 hlist_for_each_entry(session, walk, &tunnel->session_hlist[hash], hlist) {
282                         if (++count > nth) {
283                                 l2tp_session_inc_refcount(session);
284                                 if (do_ref && session->ref)
285                                         session->ref(session);
286                                 read_unlock_bh(&tunnel->hlist_lock);
287                                 return session;
288                         }
289                 }
290         }
291
292         read_unlock_bh(&tunnel->hlist_lock);
293
294         return NULL;
295 }
296 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
297
298 /* Lookup a session by interface name.
299  * This is very inefficient but is only used by management interfaces.
300  */
301 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
302                                                 const char *ifname,
303                                                 bool do_ref)
304 {
305         struct l2tp_net *pn = l2tp_pernet(net);
306         int hash;
307         struct hlist_node *walk;
308         struct l2tp_session *session;
309
310         rcu_read_lock_bh();
311         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
312                 hlist_for_each_entry_rcu(session, walk, &pn->l2tp_session_hlist[hash], global_hlist) {
313                         if (!strcmp(session->ifname, ifname)) {
314                                 l2tp_session_inc_refcount(session);
315                                 if (do_ref && session->ref)
316                                         session->ref(session);
317                                 rcu_read_unlock_bh();
318
319                                 return session;
320                         }
321                 }
322         }
323
324         rcu_read_unlock_bh();
325
326         return NULL;
327 }
328 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
329
330 static int l2tp_session_add_to_tunnel(struct l2tp_tunnel *tunnel,
331                                       struct l2tp_session *session)
332 {
333         struct l2tp_session *session_walk;
334         struct hlist_head *g_head;
335         struct hlist_head *head;
336         struct l2tp_net *pn;
337         struct hlist_node *walk;
338         int err;
339
340         head = l2tp_session_id_hash(tunnel, session->session_id);
341
342         write_lock_bh(&tunnel->hlist_lock);
343         if (!tunnel->acpt_newsess) {
344                 err = -ENODEV;
345                 goto err_tlock;
346         }
347
348         hlist_for_each_entry(session_walk, walk, head, hlist)
349                 if (session_walk->session_id == session->session_id) {
350                         err = -EEXIST;
351                         goto err_tlock;
352                 }
353
354         if (tunnel->version == L2TP_HDR_VER_3) {
355                 pn = l2tp_pernet(tunnel->l2tp_net);
356                 g_head = l2tp_session_id_hash_2(l2tp_pernet(tunnel->l2tp_net),
357                                                 session->session_id);
358
359                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
360
361                 hlist_for_each_entry(session_walk, walk, g_head, global_hlist)
362                         if (session_walk->session_id == session->session_id) {
363                                 err = -EEXIST;
364                                 goto err_tlock_pnlock;
365                         }
366
367                 l2tp_tunnel_inc_refcount(tunnel);
368                 sock_hold(tunnel->sock);
369                 hlist_add_head_rcu(&session->global_hlist, g_head);
370
371                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
372         } else {
373                 l2tp_tunnel_inc_refcount(tunnel);
374                 sock_hold(tunnel->sock);
375         }
376
377         hlist_add_head(&session->hlist, head);
378         write_unlock_bh(&tunnel->hlist_lock);
379
380         return 0;
381
382 err_tlock_pnlock:
383         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
384 err_tlock:
385         write_unlock_bh(&tunnel->hlist_lock);
386
387         return err;
388 }
389
390 /* Lookup a tunnel by id
391  */
392 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
393 {
394         struct l2tp_tunnel *tunnel;
395         struct l2tp_net *pn = l2tp_pernet(net);
396
397         rcu_read_lock_bh();
398         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
399                 if (tunnel->tunnel_id == tunnel_id) {
400                         rcu_read_unlock_bh();
401                         return tunnel;
402                 }
403         }
404         rcu_read_unlock_bh();
405
406         return NULL;
407 }
408 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
409
410 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
411 {
412         struct l2tp_net *pn = l2tp_pernet(net);
413         struct l2tp_tunnel *tunnel;
414         int count = 0;
415
416         rcu_read_lock_bh();
417         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
418                 if (++count > nth) {
419                         rcu_read_unlock_bh();
420                         return tunnel;
421                 }
422         }
423
424         rcu_read_unlock_bh();
425
426         return NULL;
427 }
428 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
429
430 /*****************************************************************************
431  * Receive data handling
432  *****************************************************************************/
433
434 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
435  * number.
436  */
437 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
438 {
439         struct sk_buff *skbp;
440         struct sk_buff *tmp;
441         u32 ns = L2TP_SKB_CB(skb)->ns;
442
443         spin_lock_bh(&session->reorder_q.lock);
444         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
445                 if (L2TP_SKB_CB(skbp)->ns > ns) {
446                         __skb_queue_before(&session->reorder_q, skbp, skb);
447                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
448                                "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
449                                session->name, ns, L2TP_SKB_CB(skbp)->ns,
450                                skb_queue_len(&session->reorder_q));
451                         session->stats.rx_oos_packets++;
452                         goto out;
453                 }
454         }
455
456         __skb_queue_tail(&session->reorder_q, skb);
457
458 out:
459         spin_unlock_bh(&session->reorder_q.lock);
460 }
461
462 /* Dequeue a single skb.
463  */
464 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
465 {
466         struct l2tp_tunnel *tunnel = session->tunnel;
467         int length = L2TP_SKB_CB(skb)->length;
468
469         /* We're about to requeue the skb, so return resources
470          * to its current owner (a socket receive buffer).
471          */
472         skb_orphan(skb);
473
474         tunnel->stats.rx_packets++;
475         tunnel->stats.rx_bytes += length;
476         session->stats.rx_packets++;
477         session->stats.rx_bytes += length;
478
479         if (L2TP_SKB_CB(skb)->has_seq) {
480                 /* Bump our Nr */
481                 session->nr++;
482                 if (tunnel->version == L2TP_HDR_VER_2)
483                         session->nr &= 0xffff;
484                 else
485                         session->nr &= 0xffffff;
486
487                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
488                        "%s: updated nr to %hu\n", session->name, session->nr);
489         }
490
491         /* call private receive handler */
492         if (session->recv_skb != NULL)
493                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
494         else
495                 kfree_skb(skb);
496
497         if (session->deref)
498                 (*session->deref)(session);
499 }
500
501 /* Dequeue skbs from the session's reorder_q, subject to packet order.
502  * Skbs that have been in the queue for too long are simply discarded.
503  */
504 static void l2tp_recv_dequeue(struct l2tp_session *session)
505 {
506         struct sk_buff *skb;
507         struct sk_buff *tmp;
508
509         /* If the pkt at the head of the queue has the nr that we
510          * expect to send up next, dequeue it and any other
511          * in-sequence packets behind it.
512          */
513 start:
514         spin_lock_bh(&session->reorder_q.lock);
515         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
516                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
517                         session->stats.rx_seq_discards++;
518                         session->stats.rx_errors++;
519                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
520                                "%s: oos pkt %u len %d discarded (too old), "
521                                "waiting for %u, reorder_q_len=%d\n",
522                                session->name, L2TP_SKB_CB(skb)->ns,
523                                L2TP_SKB_CB(skb)->length, session->nr,
524                                skb_queue_len(&session->reorder_q));
525                         __skb_unlink(skb, &session->reorder_q);
526                         kfree_skb(skb);
527                         if (session->deref)
528                                 (*session->deref)(session);
529                         continue;
530                 }
531
532                 if (L2TP_SKB_CB(skb)->has_seq) {
533                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
534                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
535                                        "%s: holding oos pkt %u len %d, "
536                                        "waiting for %u, reorder_q_len=%d\n",
537                                        session->name, L2TP_SKB_CB(skb)->ns,
538                                        L2TP_SKB_CB(skb)->length, session->nr,
539                                        skb_queue_len(&session->reorder_q));
540                                 goto out;
541                         }
542                 }
543                 __skb_unlink(skb, &session->reorder_q);
544
545                 /* Process the skb. We release the queue lock while we
546                  * do so to let other contexts process the queue.
547                  */
548                 spin_unlock_bh(&session->reorder_q.lock);
549                 l2tp_recv_dequeue_skb(session, skb);
550                 goto start;
551         }
552
553 out:
554         spin_unlock_bh(&session->reorder_q.lock);
555 }
556
557 static inline int l2tp_verify_udp_checksum(struct sock *sk,
558                                            struct sk_buff *skb)
559 {
560         struct udphdr *uh = udp_hdr(skb);
561         u16 ulen = ntohs(uh->len);
562         struct inet_sock *inet;
563         __wsum psum;
564
565         if (sk->sk_no_check || skb_csum_unnecessary(skb) || !uh->check)
566                 return 0;
567
568         inet = inet_sk(sk);
569         psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr, ulen,
570                                   IPPROTO_UDP, 0);
571
572         if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
573             !csum_fold(csum_add(psum, skb->csum)))
574                 return 0;
575
576         skb->csum = psum;
577
578         return __skb_checksum_complete(skb);
579 }
580
581 /* Do receive processing of L2TP data frames. We handle both L2TPv2
582  * and L2TPv3 data frames here.
583  *
584  * L2TPv2 Data Message Header
585  *
586  *  0                   1                   2                   3
587  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
588  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
589  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
590  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
591  * |           Tunnel ID           |           Session ID          |
592  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
593  * |             Ns (opt)          |             Nr (opt)          |
594  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
595  * |      Offset Size (opt)        |    Offset pad... (opt)
596  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
597  *
598  * Data frames are marked by T=0. All other fields are the same as
599  * those in L2TP control frames.
600  *
601  * L2TPv3 Data Message Header
602  *
603  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
604  * |                      L2TP Session Header                      |
605  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
606  * |                      L2-Specific Sublayer                     |
607  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
608  * |                        Tunnel Payload                      ...
609  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
610  *
611  * L2TPv3 Session Header Over IP
612  *
613  *  0                   1                   2                   3
614  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
615  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
616  * |                           Session ID                          |
617  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
618  * |               Cookie (optional, maximum 64 bits)...
619  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
620  *                                                                 |
621  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
622  *
623  * L2TPv3 L2-Specific Sublayer Format
624  *
625  *  0                   1                   2                   3
626  *  0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
627  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
628  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
629  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
630  *
631  * Cookie value, sublayer format and offset (pad) are negotiated with
632  * the peer when the session is set up. Unlike L2TPv2, we do not need
633  * to parse the packet header to determine if optional fields are
634  * present.
635  *
636  * Caller must already have parsed the frame and determined that it is
637  * a data (not control) frame before coming here. Fields up to the
638  * session-id have already been parsed and ptr points to the data
639  * after the session-id.
640  *
641  * session->ref() must have been called prior to l2tp_recv_common().
642  * session->deref() will be called automatically after skb is processed.
643  */
644 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
645                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
646                       int length, int (*payload_hook)(struct sk_buff *skb))
647 {
648         struct l2tp_tunnel *tunnel = session->tunnel;
649         int offset;
650         u32 ns, nr;
651
652         /* Parse and check optional cookie */
653         if (session->peer_cookie_len > 0) {
654                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
655                         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
656                                "%s: cookie mismatch (%u/%u). Discarding.\n",
657                                tunnel->name, tunnel->tunnel_id, session->session_id);
658                         session->stats.rx_cookie_discards++;
659                         goto discard;
660                 }
661                 ptr += session->peer_cookie_len;
662         }
663
664         /* Handle the optional sequence numbers. Sequence numbers are
665          * in different places for L2TPv2 and L2TPv3.
666          *
667          * If we are the LAC, enable/disable sequence numbers under
668          * the control of the LNS.  If no sequence numbers present but
669          * we were expecting them, discard frame.
670          */
671         ns = nr = 0;
672         L2TP_SKB_CB(skb)->has_seq = 0;
673         if (tunnel->version == L2TP_HDR_VER_2) {
674                 if (hdrflags & L2TP_HDRFLAG_S) {
675                         ns = ntohs(*(__be16 *) ptr);
676                         ptr += 2;
677                         nr = ntohs(*(__be16 *) ptr);
678                         ptr += 2;
679
680                         /* Store L2TP info in the skb */
681                         L2TP_SKB_CB(skb)->ns = ns;
682                         L2TP_SKB_CB(skb)->has_seq = 1;
683
684                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
685                                "%s: recv data ns=%u, nr=%u, session nr=%u\n",
686                                session->name, ns, nr, session->nr);
687                 }
688         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
689                 u32 l2h = ntohl(*(__be32 *) ptr);
690
691                 if (l2h & 0x40000000) {
692                         ns = l2h & 0x00ffffff;
693
694                         /* Store L2TP info in the skb */
695                         L2TP_SKB_CB(skb)->ns = ns;
696                         L2TP_SKB_CB(skb)->has_seq = 1;
697
698                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
699                                "%s: recv data ns=%u, session nr=%u\n",
700                                session->name, ns, session->nr);
701                 }
702         }
703
704         /* Advance past L2-specific header, if present */
705         ptr += session->l2specific_len;
706
707         if (L2TP_SKB_CB(skb)->has_seq) {
708                 /* Received a packet with sequence numbers. If we're the LNS,
709                  * check if we sre sending sequence numbers and if not,
710                  * configure it so.
711                  */
712                 if ((!session->lns_mode) && (!session->send_seq)) {
713                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
714                                "%s: requested to enable seq numbers by LNS\n",
715                                session->name);
716                         session->send_seq = -1;
717                         l2tp_session_set_header_len(session, tunnel->version);
718                 }
719         } else {
720                 /* No sequence numbers.
721                  * If user has configured mandatory sequence numbers, discard.
722                  */
723                 if (session->recv_seq) {
724                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
725                                "%s: recv data has no seq numbers when required. "
726                                "Discarding\n", session->name);
727                         session->stats.rx_seq_discards++;
728                         goto discard;
729                 }
730
731                 /* If we're the LAC and we're sending sequence numbers, the
732                  * LNS has requested that we no longer send sequence numbers.
733                  * If we're the LNS and we're sending sequence numbers, the
734                  * LAC is broken. Discard the frame.
735                  */
736                 if ((!session->lns_mode) && (session->send_seq)) {
737                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
738                                "%s: requested to disable seq numbers by LNS\n",
739                                session->name);
740                         session->send_seq = 0;
741                         l2tp_session_set_header_len(session, tunnel->version);
742                 } else if (session->send_seq) {
743                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
744                                "%s: recv data has no seq numbers when required. "
745                                "Discarding\n", session->name);
746                         session->stats.rx_seq_discards++;
747                         goto discard;
748                 }
749         }
750
751         /* Session data offset is handled differently for L2TPv2 and
752          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
753          * the header. For L2TPv3, the offset is negotiated using AVPs
754          * in the session setup control protocol.
755          */
756         if (tunnel->version == L2TP_HDR_VER_2) {
757                 /* If offset bit set, skip it. */
758                 if (hdrflags & L2TP_HDRFLAG_O) {
759                         offset = ntohs(*(__be16 *)ptr);
760                         ptr += 2 + offset;
761                 }
762         } else
763                 ptr += session->offset;
764
765         offset = ptr - optr;
766         if (!pskb_may_pull(skb, offset))
767                 goto discard;
768
769         __skb_pull(skb, offset);
770
771         /* If caller wants to process the payload before we queue the
772          * packet, do so now.
773          */
774         if (payload_hook)
775                 if ((*payload_hook)(skb))
776                         goto discard;
777
778         /* Prepare skb for adding to the session's reorder_q.  Hold
779          * packets for max reorder_timeout or 1 second if not
780          * reordering.
781          */
782         L2TP_SKB_CB(skb)->length = length;
783         L2TP_SKB_CB(skb)->expires = jiffies +
784                 (session->reorder_timeout ? session->reorder_timeout : HZ);
785
786         /* Add packet to the session's receive queue. Reordering is done here, if
787          * enabled. Saved L2TP protocol info is stored in skb->sb[].
788          */
789         if (L2TP_SKB_CB(skb)->has_seq) {
790                 if (session->reorder_timeout != 0) {
791                         /* Packet reordering enabled. Add skb to session's
792                          * reorder queue, in order of ns.
793                          */
794                         l2tp_recv_queue_skb(session, skb);
795                 } else {
796                         /* Packet reordering disabled. Discard out-of-sequence
797                          * packets
798                          */
799                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
800                                 session->stats.rx_seq_discards++;
801                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
802                                        "%s: oos pkt %u len %d discarded, "
803                                        "waiting for %u, reorder_q_len=%d\n",
804                                        session->name, L2TP_SKB_CB(skb)->ns,
805                                        L2TP_SKB_CB(skb)->length, session->nr,
806                                        skb_queue_len(&session->reorder_q));
807                                 goto discard;
808                         }
809                         skb_queue_tail(&session->reorder_q, skb);
810                 }
811         } else {
812                 /* No sequence numbers. Add the skb to the tail of the
813                  * reorder queue. This ensures that it will be
814                  * delivered after all previous sequenced skbs.
815                  */
816                 skb_queue_tail(&session->reorder_q, skb);
817         }
818
819         /* Try to dequeue as many skbs from reorder_q as we can. */
820         l2tp_recv_dequeue(session);
821
822         return;
823
824 discard:
825         session->stats.rx_errors++;
826         kfree_skb(skb);
827
828         if (session->deref)
829                 (*session->deref)(session);
830 }
831 EXPORT_SYMBOL(l2tp_recv_common);
832
833 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
834  * here. The skb is not on a list when we get here.
835  * Returns 0 if the packet was a data packet and was successfully passed on.
836  * Returns 1 if the packet was not a good data packet and could not be
837  * forwarded.  All such packets are passed up to userspace to deal with.
838  */
839 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
840                               int (*payload_hook)(struct sk_buff *skb))
841 {
842         struct l2tp_session *session = NULL;
843         unsigned char *ptr, *optr;
844         u16 hdrflags;
845         u32 tunnel_id, session_id;
846         int offset;
847         u16 version;
848         int length;
849
850         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
851                 goto discard_bad_csum;
852
853         /* UDP always verifies the packet length. */
854         __skb_pull(skb, sizeof(struct udphdr));
855
856         /* Short packet? */
857         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
858                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
859                        "%s: recv short packet (len=%d)\n", tunnel->name, skb->len);
860                 goto error;
861         }
862
863         /* Trace packet contents, if enabled */
864         if (tunnel->debug & L2TP_MSG_DATA) {
865                 length = min(32u, skb->len);
866                 if (!pskb_may_pull(skb, length))
867                         goto error;
868
869                 printk(KERN_DEBUG "%s: recv: ", tunnel->name);
870
871                 offset = 0;
872                 do {
873                         printk(" %02X", skb->data[offset]);
874                 } while (++offset < length);
875
876                 printk("\n");
877         }
878
879         /* Point to L2TP header */
880         optr = ptr = skb->data;
881
882         /* Get L2TP header flags */
883         hdrflags = ntohs(*(__be16 *) ptr);
884
885         /* Check protocol version */
886         version = hdrflags & L2TP_HDR_VER_MASK;
887         if (version != tunnel->version) {
888                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
889                        "%s: recv protocol version mismatch: got %d expected %d\n",
890                        tunnel->name, version, tunnel->version);
891                 goto error;
892         }
893
894         /* Get length of L2TP packet */
895         length = skb->len;
896
897         /* If type is control packet, it is handled by userspace. */
898         if (hdrflags & L2TP_HDRFLAG_T) {
899                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
900                        "%s: recv control packet, len=%d\n", tunnel->name, length);
901                 goto error;
902         }
903
904         /* Skip flags */
905         ptr += 2;
906
907         if (tunnel->version == L2TP_HDR_VER_2) {
908                 /* If length is present, skip it */
909                 if (hdrflags & L2TP_HDRFLAG_L)
910                         ptr += 2;
911
912                 /* Extract tunnel and session ID */
913                 tunnel_id = ntohs(*(__be16 *) ptr);
914                 ptr += 2;
915                 session_id = ntohs(*(__be16 *) ptr);
916                 ptr += 2;
917         } else {
918                 ptr += 2;       /* skip reserved bits */
919                 tunnel_id = tunnel->tunnel_id;
920                 session_id = ntohl(*(__be32 *) ptr);
921                 ptr += 4;
922         }
923
924         /* Find the session context */
925         session = l2tp_session_get(tunnel->l2tp_net, tunnel, session_id, true);
926         if (!session || !session->recv_skb) {
927                 if (session) {
928                         if (session->deref)
929                                 session->deref(session);
930                         l2tp_session_dec_refcount(session);
931                 }
932
933                 /* Not found? Pass to userspace to deal with */
934                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
935                        "%s: no session found (%u/%u). Passing up.\n",
936                        tunnel->name, tunnel_id, session_id);
937                 goto error;
938         }
939
940         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
941         l2tp_session_dec_refcount(session);
942
943         return 0;
944
945 discard_bad_csum:
946         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
947         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
948         tunnel->stats.rx_errors++;
949         kfree_skb(skb);
950
951         return 0;
952
953 error:
954         /* Put UDP header back */
955         __skb_push(skb, sizeof(struct udphdr));
956
957         return 1;
958 }
959
960 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
961  * Return codes:
962  * 0 : success.
963  * <0: error
964  * >0: skb should be passed up to userspace as UDP.
965  */
966 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
967 {
968         struct l2tp_tunnel *tunnel;
969
970         tunnel = l2tp_sock_to_tunnel(sk);
971         if (tunnel == NULL)
972                 goto pass_up;
973
974         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
975                "%s: received %d bytes\n", tunnel->name, skb->len);
976
977         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
978                 goto pass_up_put;
979
980         sock_put(sk);
981         return 0;
982
983 pass_up_put:
984         sock_put(sk);
985 pass_up:
986         return 1;
987 }
988 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
989
990 /************************************************************************
991  * Transmit handling
992  ***********************************************************************/
993
994 /* Build an L2TP header for the session into the buffer provided.
995  */
996 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
997 {
998         struct l2tp_tunnel *tunnel = session->tunnel;
999         __be16 *bufp = buf;
1000         __be16 *optr = buf;
1001         u16 flags = L2TP_HDR_VER_2;
1002         u32 tunnel_id = tunnel->peer_tunnel_id;
1003         u32 session_id = session->peer_session_id;
1004
1005         if (session->send_seq)
1006                 flags |= L2TP_HDRFLAG_S;
1007
1008         /* Setup L2TP header. */
1009         *bufp++ = htons(flags);
1010         *bufp++ = htons(tunnel_id);
1011         *bufp++ = htons(session_id);
1012         if (session->send_seq) {
1013                 *bufp++ = htons(session->ns);
1014                 *bufp++ = 0;
1015                 session->ns++;
1016                 session->ns &= 0xffff;
1017                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
1018                        "%s: updated ns to %u\n", session->name, session->ns);
1019         }
1020
1021         return bufp - optr;
1022 }
1023
1024 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
1025 {
1026         struct l2tp_tunnel *tunnel = session->tunnel;
1027         char *bufp = buf;
1028         char *optr = bufp;
1029
1030         /* Setup L2TP header. The header differs slightly for UDP and
1031          * IP encapsulations. For UDP, there is 4 bytes of flags.
1032          */
1033         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1034                 u16 flags = L2TP_HDR_VER_3;
1035                 *((__be16 *) bufp) = htons(flags);
1036                 bufp += 2;
1037                 *((__be16 *) bufp) = 0;
1038                 bufp += 2;
1039         }
1040
1041         *((__be32 *) bufp) = htonl(session->peer_session_id);
1042         bufp += 4;
1043         if (session->cookie_len) {
1044                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1045                 bufp += session->cookie_len;
1046         }
1047         if (session->l2specific_len) {
1048                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1049                         u32 l2h = 0;
1050                         if (session->send_seq) {
1051                                 l2h = 0x40000000 | session->ns;
1052                                 session->ns++;
1053                                 session->ns &= 0xffffff;
1054                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
1055                                        "%s: updated ns to %u\n", session->name, session->ns);
1056                         }
1057
1058                         *((__be32 *) bufp) = htonl(l2h);
1059                 }
1060                 bufp += session->l2specific_len;
1061         }
1062         if (session->offset)
1063                 bufp += session->offset;
1064
1065         return bufp - optr;
1066 }
1067
1068 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1069                           struct flowi *fl, size_t data_len)
1070 {
1071         struct l2tp_tunnel *tunnel = session->tunnel;
1072         unsigned int len = skb->len;
1073         int error;
1074
1075         /* Debug */
1076         if (session->send_seq)
1077                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
1078                        "%s: send %Zd bytes, ns=%u\n", session->name,
1079                        data_len, session->ns - 1);
1080         else
1081                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
1082                        "%s: send %Zd bytes\n", session->name, data_len);
1083
1084         if (session->debug & L2TP_MSG_DATA) {
1085                 int i;
1086                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1087                 unsigned char *datap = skb->data + uhlen;
1088
1089                 printk(KERN_DEBUG "%s: xmit:", session->name);
1090                 for (i = 0; i < (len - uhlen); i++) {
1091                         printk(" %02X", *datap++);
1092                         if (i == 31) {
1093                                 printk(" ...");
1094                                 break;
1095                         }
1096                 }
1097                 printk("\n");
1098         }
1099
1100         /* Queue the packet to IP for output */
1101         skb->local_df = 1;
1102         error = ip_queue_xmit(skb, fl);
1103
1104         /* Update stats */
1105         if (error >= 0) {
1106                 tunnel->stats.tx_packets++;
1107                 tunnel->stats.tx_bytes += len;
1108                 session->stats.tx_packets++;
1109                 session->stats.tx_bytes += len;
1110         } else {
1111                 tunnel->stats.tx_errors++;
1112                 session->stats.tx_errors++;
1113         }
1114
1115         return 0;
1116 }
1117
1118 /* Automatically called when the skb is freed.
1119  */
1120 static void l2tp_sock_wfree(struct sk_buff *skb)
1121 {
1122         sock_put(skb->sk);
1123 }
1124
1125 /* For data skbs that we transmit, we associate with the tunnel socket
1126  * but don't do accounting.
1127  */
1128 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1129 {
1130         sock_hold(sk);
1131         skb->sk = sk;
1132         skb->destructor = l2tp_sock_wfree;
1133 }
1134
1135 /* If caller requires the skb to have a ppp header, the header must be
1136  * inserted in the skb data before calling this function.
1137  */
1138 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1139 {
1140         int data_len = skb->len;
1141         struct l2tp_tunnel *tunnel = session->tunnel;
1142         struct sock *sk = tunnel->sock;
1143         struct flowi *fl;
1144         struct udphdr *uh;
1145         struct inet_sock *inet;
1146         __wsum csum;
1147         int old_headroom;
1148         int new_headroom;
1149         int headroom;
1150         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1151         int udp_len;
1152
1153         /* Check that there's enough headroom in the skb to insert IP,
1154          * UDP and L2TP headers. If not enough, expand it to
1155          * make room. Adjust truesize.
1156          */
1157         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1158                 uhlen + hdr_len;
1159         old_headroom = skb_headroom(skb);
1160         if (skb_cow_head(skb, headroom)) {
1161                 dev_kfree_skb(skb);
1162                 goto abort;
1163         }
1164
1165         new_headroom = skb_headroom(skb);
1166         skb_orphan(skb);
1167         skb->truesize += new_headroom - old_headroom;
1168
1169         /* Setup L2TP header */
1170         session->build_header(session, __skb_push(skb, hdr_len));
1171
1172         /* Reset skb netfilter state */
1173         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1174         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1175                               IPSKB_REROUTED);
1176         nf_reset(skb);
1177
1178         bh_lock_sock(sk);
1179         if (sock_owned_by_user(sk)) {
1180                 dev_kfree_skb(skb);
1181                 goto out_unlock;
1182         }
1183
1184         /* Get routing info from the tunnel socket */
1185         skb_dst_drop(skb);
1186         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1187
1188         inet = inet_sk(sk);
1189         fl = &inet->cork.fl;
1190         switch (tunnel->encap) {
1191         case L2TP_ENCAPTYPE_UDP:
1192                 /* Setup UDP header */
1193                 __skb_push(skb, sizeof(*uh));
1194                 skb_reset_transport_header(skb);
1195                 uh = udp_hdr(skb);
1196                 uh->source = inet->inet_sport;
1197                 uh->dest = inet->inet_dport;
1198                 udp_len = uhlen + hdr_len + data_len;
1199                 uh->len = htons(udp_len);
1200                 uh->check = 0;
1201
1202                 /* Calculate UDP checksum if configured to do so */
1203                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1204                         skb->ip_summed = CHECKSUM_NONE;
1205                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1206                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1207                         skb->ip_summed = CHECKSUM_COMPLETE;
1208                         csum = skb_checksum(skb, 0, udp_len, 0);
1209                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1210                                                       inet->inet_daddr,
1211                                                       udp_len, IPPROTO_UDP, csum);
1212                         if (uh->check == 0)
1213                                 uh->check = CSUM_MANGLED_0;
1214                 } else {
1215                         skb->ip_summed = CHECKSUM_PARTIAL;
1216                         skb->csum_start = skb_transport_header(skb) - skb->head;
1217                         skb->csum_offset = offsetof(struct udphdr, check);
1218                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1219                                                        inet->inet_daddr,
1220                                                        udp_len, IPPROTO_UDP, 0);
1221                 }
1222                 break;
1223
1224         case L2TP_ENCAPTYPE_IP:
1225                 break;
1226         }
1227
1228         l2tp_skb_set_owner_w(skb, sk);
1229
1230         l2tp_xmit_core(session, skb, fl, data_len);
1231 out_unlock:
1232         bh_unlock_sock(sk);
1233
1234 abort:
1235         return 0;
1236 }
1237 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1238
1239 /*****************************************************************************
1240  * Tinnel and session create/destroy.
1241  *****************************************************************************/
1242
1243 /* Tunnel socket destruct hook.
1244  * The tunnel context is deleted only when all session sockets have been
1245  * closed.
1246  */
1247 static void l2tp_tunnel_destruct(struct sock *sk)
1248 {
1249         struct l2tp_tunnel *tunnel;
1250
1251         tunnel = sk->sk_user_data;
1252         if (tunnel == NULL)
1253                 goto end;
1254
1255         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1256                "%s: closing...\n", tunnel->name);
1257
1258         /* Close all sessions */
1259         l2tp_tunnel_closeall(tunnel);
1260
1261         switch (tunnel->encap) {
1262         case L2TP_ENCAPTYPE_UDP:
1263                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1264                 (udp_sk(sk))->encap_type = 0;
1265                 (udp_sk(sk))->encap_rcv = NULL;
1266                 break;
1267         case L2TP_ENCAPTYPE_IP:
1268                 break;
1269         }
1270
1271         /* Remove hooks into tunnel socket */
1272         tunnel->sock = NULL;
1273         sk->sk_destruct = tunnel->old_sk_destruct;
1274         sk->sk_user_data = NULL;
1275
1276         /* Call the original destructor */
1277         if (sk->sk_destruct)
1278                 (*sk->sk_destruct)(sk);
1279
1280         /* We're finished with the socket */
1281         l2tp_tunnel_dec_refcount(tunnel);
1282
1283 end:
1284         return;
1285 }
1286
1287 /* When the tunnel is closed, all the attached sessions need to go too.
1288  */
1289 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1290 {
1291         int hash;
1292         struct hlist_node *walk;
1293         struct hlist_node *tmp;
1294         struct l2tp_session *session;
1295
1296         BUG_ON(tunnel == NULL);
1297
1298         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1299                "%s: closing all sessions...\n", tunnel->name);
1300
1301         write_lock_bh(&tunnel->hlist_lock);
1302         tunnel->acpt_newsess = false;
1303         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1304 again:
1305                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1306                         session = hlist_entry(walk, struct l2tp_session, hlist);
1307
1308                         PRINTK(session->debug, L2TP_MSG_CONTROL, KERN_INFO,
1309                                "%s: closing session\n", session->name);
1310
1311                         hlist_del_init(&session->hlist);
1312
1313                         /* Since we should hold the sock lock while
1314                          * doing any unbinding, we need to release the
1315                          * lock we're holding before taking that lock.
1316                          * Hold a reference to the sock so it doesn't
1317                          * disappear as we're jumping between locks.
1318                          */
1319                         if (session->ref != NULL)
1320                                 (*session->ref)(session);
1321
1322                         write_unlock_bh(&tunnel->hlist_lock);
1323
1324                         if (tunnel->version != L2TP_HDR_VER_2) {
1325                                 struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1326
1327                                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
1328                                 hlist_del_init_rcu(&session->global_hlist);
1329                                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1330                                 synchronize_rcu();
1331                         }
1332
1333                         if (session->session_close != NULL)
1334                                 (*session->session_close)(session);
1335
1336                         if (session->deref != NULL)
1337                                 (*session->deref)(session);
1338
1339                         write_lock_bh(&tunnel->hlist_lock);
1340
1341                         /* Now restart from the beginning of this hash
1342                          * chain.  We always remove a session from the
1343                          * list so we are guaranteed to make forward
1344                          * progress.
1345                          */
1346                         goto again;
1347                 }
1348         }
1349         write_unlock_bh(&tunnel->hlist_lock);
1350 }
1351
1352 /* Really kill the tunnel.
1353  * Come here only when all sessions have been cleared from the tunnel.
1354  */
1355 void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1356 {
1357         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1358
1359         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1360         BUG_ON(tunnel->sock != NULL);
1361
1362         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1363                "%s: free...\n", tunnel->name);
1364
1365         /* Remove from tunnel list */
1366         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1367         list_del_rcu(&tunnel->list);
1368         kfree_rcu(tunnel, rcu);
1369         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1370
1371         atomic_dec(&l2tp_tunnel_count);
1372 }
1373 EXPORT_SYMBOL_GPL(l2tp_tunnel_free);
1374
1375 /* Create a socket for the tunnel, if one isn't set up by
1376  * userspace. This is used for static tunnels where there is no
1377  * managing L2TP daemon.
1378  */
1379 static int l2tp_tunnel_sock_create(u32 tunnel_id, u32 peer_tunnel_id, struct l2tp_tunnel_cfg *cfg, struct socket **sockp)
1380 {
1381         int err = -EINVAL;
1382         struct sockaddr_in udp_addr;
1383         struct sockaddr_l2tpip ip_addr;
1384         struct socket *sock = NULL;
1385
1386         switch (cfg->encap) {
1387         case L2TP_ENCAPTYPE_UDP:
1388                 err = sock_create(AF_INET, SOCK_DGRAM, 0, sockp);
1389                 if (err < 0)
1390                         goto out;
1391
1392                 sock = *sockp;
1393
1394                 memset(&udp_addr, 0, sizeof(udp_addr));
1395                 udp_addr.sin_family = AF_INET;
1396                 udp_addr.sin_addr = cfg->local_ip;
1397                 udp_addr.sin_port = htons(cfg->local_udp_port);
1398                 err = kernel_bind(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr));
1399                 if (err < 0)
1400                         goto out;
1401
1402                 udp_addr.sin_family = AF_INET;
1403                 udp_addr.sin_addr = cfg->peer_ip;
1404                 udp_addr.sin_port = htons(cfg->peer_udp_port);
1405                 err = kernel_connect(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr), 0);
1406                 if (err < 0)
1407                         goto out;
1408
1409                 if (!cfg->use_udp_checksums)
1410                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1411
1412                 break;
1413
1414         case L2TP_ENCAPTYPE_IP:
1415                 err = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_L2TP, sockp);
1416                 if (err < 0)
1417                         goto out;
1418
1419                 sock = *sockp;
1420
1421                 memset(&ip_addr, 0, sizeof(ip_addr));
1422                 ip_addr.l2tp_family = AF_INET;
1423                 ip_addr.l2tp_addr = cfg->local_ip;
1424                 ip_addr.l2tp_conn_id = tunnel_id;
1425                 err = kernel_bind(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr));
1426                 if (err < 0)
1427                         goto out;
1428
1429                 ip_addr.l2tp_family = AF_INET;
1430                 ip_addr.l2tp_addr = cfg->peer_ip;
1431                 ip_addr.l2tp_conn_id = peer_tunnel_id;
1432                 err = kernel_connect(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr), 0);
1433                 if (err < 0)
1434                         goto out;
1435
1436                 break;
1437
1438         default:
1439                 goto out;
1440         }
1441
1442 out:
1443         if ((err < 0) && sock) {
1444                 sock_release(sock);
1445                 *sockp = NULL;
1446         }
1447
1448         return err;
1449 }
1450
1451 int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32 peer_tunnel_id, struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1452 {
1453         struct l2tp_tunnel *tunnel = NULL;
1454         int err;
1455         struct socket *sock = NULL;
1456         struct sock *sk = NULL;
1457         struct l2tp_net *pn;
1458         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1459
1460         /* Get the tunnel socket from the fd, which was opened by
1461          * the userspace L2TP daemon. If not specified, create a
1462          * kernel socket.
1463          */
1464         if (fd < 0) {
1465                 err = l2tp_tunnel_sock_create(tunnel_id, peer_tunnel_id, cfg, &sock);
1466                 if (err < 0)
1467                         goto err;
1468         } else {
1469                 err = -EBADF;
1470                 sock = sockfd_lookup(fd, &err);
1471                 if (!sock) {
1472                         printk(KERN_ERR "tunl %hu: sockfd_lookup(fd=%d) returned %d\n",
1473                                tunnel_id, fd, err);
1474                         goto err;
1475                 }
1476         }
1477
1478         sk = sock->sk;
1479
1480         if (cfg != NULL)
1481                 encap = cfg->encap;
1482
1483         /* Quick sanity checks */
1484         switch (encap) {
1485         case L2TP_ENCAPTYPE_UDP:
1486                 err = -EPROTONOSUPPORT;
1487                 if (sk->sk_protocol != IPPROTO_UDP) {
1488                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1489                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1490                         goto err;
1491                 }
1492                 break;
1493         case L2TP_ENCAPTYPE_IP:
1494                 err = -EPROTONOSUPPORT;
1495                 if (sk->sk_protocol != IPPROTO_L2TP) {
1496                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1497                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1498                         goto err;
1499                 }
1500                 break;
1501         }
1502
1503         /* Check if this socket has already been prepped */
1504         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
1505         if (tunnel != NULL) {
1506                 /* This socket has already been prepped */
1507                 err = -EBUSY;
1508                 goto err;
1509         }
1510
1511         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1512         if (tunnel == NULL) {
1513                 err = -ENOMEM;
1514                 goto err;
1515         }
1516
1517         tunnel->version = version;
1518         tunnel->tunnel_id = tunnel_id;
1519         tunnel->peer_tunnel_id = peer_tunnel_id;
1520         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1521
1522         tunnel->magic = L2TP_TUNNEL_MAGIC;
1523         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1524         rwlock_init(&tunnel->hlist_lock);
1525         tunnel->acpt_newsess = true;
1526
1527         /* The net we belong to */
1528         tunnel->l2tp_net = net;
1529         pn = l2tp_pernet(net);
1530
1531         if (cfg != NULL)
1532                 tunnel->debug = cfg->debug;
1533
1534         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1535         tunnel->encap = encap;
1536         if (encap == L2TP_ENCAPTYPE_UDP) {
1537                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1538                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1539                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1540         }
1541
1542         sk->sk_user_data = tunnel;
1543
1544         /* Hook on the tunnel socket destructor so that we can cleanup
1545          * if the tunnel socket goes away.
1546          */
1547         tunnel->old_sk_destruct = sk->sk_destruct;
1548         sk->sk_destruct = &l2tp_tunnel_destruct;
1549         tunnel->sock = sk;
1550         sk->sk_allocation = GFP_ATOMIC;
1551
1552         /* Add tunnel to our list */
1553         INIT_LIST_HEAD(&tunnel->list);
1554         atomic_inc(&l2tp_tunnel_count);
1555
1556         /* Bump the reference count. The tunnel context is deleted
1557          * only when this drops to zero. Must be done before list insertion
1558          */
1559         l2tp_tunnel_inc_refcount(tunnel);
1560         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1561         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1562         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1563
1564         err = 0;
1565 err:
1566         if (tunnelp)
1567                 *tunnelp = tunnel;
1568
1569         /* If tunnel's socket was created by the kernel, it doesn't
1570          *  have a file.
1571          */
1572         if (sock && sock->file)
1573                 sockfd_put(sock);
1574
1575         return err;
1576 }
1577 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1578
1579 /* This function is used by the netlink TUNNEL_DELETE command.
1580  */
1581 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1582 {
1583         int err = 0;
1584         struct socket *sock = tunnel->sock ? tunnel->sock->sk_socket : NULL;
1585
1586         /* Force the tunnel socket to close. This will eventually
1587          * cause the tunnel to be deleted via the normal socket close
1588          * mechanisms when userspace closes the tunnel socket.
1589          */
1590         if (sock != NULL) {
1591                 err = inet_shutdown(sock, 2);
1592
1593                 /* If the tunnel's socket was created by the kernel,
1594                  * close the socket here since the socket was not
1595                  * created by userspace.
1596                  */
1597                 if (sock->file == NULL)
1598                         err = inet_release(sock);
1599         }
1600
1601         return err;
1602 }
1603 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1604
1605 /* Really kill the session.
1606  */
1607 void l2tp_session_free(struct l2tp_session *session)
1608 {
1609         struct l2tp_tunnel *tunnel;
1610
1611         BUG_ON(atomic_read(&session->ref_count) != 0);
1612
1613         tunnel = session->tunnel;
1614         if (tunnel != NULL) {
1615                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1616
1617                 /* Delete the session from the hash */
1618                 write_lock_bh(&tunnel->hlist_lock);
1619                 hlist_del_init(&session->hlist);
1620                 write_unlock_bh(&tunnel->hlist_lock);
1621
1622                 /* Unlink from the global hash if not L2TPv2 */
1623                 if (tunnel->version != L2TP_HDR_VER_2) {
1624                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1625
1626                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1627                         hlist_del_init_rcu(&session->global_hlist);
1628                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1629                         synchronize_rcu();
1630                 }
1631
1632                 if (session->session_id != 0)
1633                         atomic_dec(&l2tp_session_count);
1634
1635                 sock_put(tunnel->sock);
1636
1637                 /* This will delete the tunnel context if this
1638                  * is the last session on the tunnel.
1639                  */
1640                 session->tunnel = NULL;
1641                 l2tp_tunnel_dec_refcount(tunnel);
1642         }
1643
1644         kfree(session);
1645
1646         return;
1647 }
1648 EXPORT_SYMBOL_GPL(l2tp_session_free);
1649
1650 /* This function is used by the netlink SESSION_DELETE command and by
1651    pseudowire modules.
1652  */
1653 int l2tp_session_delete(struct l2tp_session *session)
1654 {
1655         if (session->session_close != NULL)
1656                 (*session->session_close)(session);
1657
1658         l2tp_session_dec_refcount(session);
1659
1660         return 0;
1661 }
1662 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1663
1664
1665 /* We come here whenever a session's send_seq, cookie_len or
1666  * l2specific_len parameters are set.
1667  */
1668 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1669 {
1670         if (version == L2TP_HDR_VER_2) {
1671                 session->hdr_len = 6;
1672                 if (session->send_seq)
1673                         session->hdr_len += 4;
1674         } else {
1675                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1676                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1677                         session->hdr_len += 4;
1678         }
1679
1680 }
1681
1682 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id, u32 peer_session_id, struct l2tp_session_cfg *cfg)
1683 {
1684         struct l2tp_session *session;
1685         int err;
1686
1687         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1688         if (session != NULL) {
1689                 session->magic = L2TP_SESSION_MAGIC;
1690                 session->tunnel = tunnel;
1691
1692                 session->session_id = session_id;
1693                 session->peer_session_id = peer_session_id;
1694                 session->nr = 1;
1695
1696                 sprintf(&session->name[0], "sess %u/%u",
1697                         tunnel->tunnel_id, session->session_id);
1698
1699                 skb_queue_head_init(&session->reorder_q);
1700
1701                 INIT_HLIST_NODE(&session->hlist);
1702                 INIT_HLIST_NODE(&session->global_hlist);
1703
1704                 /* Inherit debug options from tunnel */
1705                 session->debug = tunnel->debug;
1706
1707                 if (cfg) {
1708                         session->pwtype = cfg->pw_type;
1709                         session->debug = cfg->debug;
1710                         session->mtu = cfg->mtu;
1711                         session->mru = cfg->mru;
1712                         session->send_seq = cfg->send_seq;
1713                         session->recv_seq = cfg->recv_seq;
1714                         session->lns_mode = cfg->lns_mode;
1715                         session->reorder_timeout = cfg->reorder_timeout;
1716                         session->offset = cfg->offset;
1717                         session->l2specific_type = cfg->l2specific_type;
1718                         session->l2specific_len = cfg->l2specific_len;
1719                         session->cookie_len = cfg->cookie_len;
1720                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1721                         session->peer_cookie_len = cfg->peer_cookie_len;
1722                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1723                 }
1724
1725                 if (tunnel->version == L2TP_HDR_VER_2)
1726                         session->build_header = l2tp_build_l2tpv2_header;
1727                 else
1728                         session->build_header = l2tp_build_l2tpv3_header;
1729
1730                 l2tp_session_set_header_len(session, tunnel->version);
1731
1732                 l2tp_session_inc_refcount(session);
1733
1734                 err = l2tp_session_add_to_tunnel(tunnel, session);
1735                 if (err) {
1736                         kfree(session);
1737
1738                         return ERR_PTR(err);
1739                 }
1740
1741                 /* Ignore management session in session count value */
1742                 if (session->session_id != 0)
1743                         atomic_inc(&l2tp_session_count);
1744
1745                 return session;
1746         }
1747
1748         return ERR_PTR(-ENOMEM);
1749 }
1750 EXPORT_SYMBOL_GPL(l2tp_session_create);
1751
1752 /*****************************************************************************
1753  * Init and cleanup
1754  *****************************************************************************/
1755
1756 static __net_init int l2tp_init_net(struct net *net)
1757 {
1758         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1759         int hash;
1760
1761         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1762         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1763
1764         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1765                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1766
1767         spin_lock_init(&pn->l2tp_session_hlist_lock);
1768
1769         return 0;
1770 }
1771
1772 static struct pernet_operations l2tp_net_ops = {
1773         .init = l2tp_init_net,
1774         .id   = &l2tp_net_id,
1775         .size = sizeof(struct l2tp_net),
1776 };
1777
1778 static int __init l2tp_init(void)
1779 {
1780         int rc = 0;
1781
1782         rc = register_pernet_device(&l2tp_net_ops);
1783         if (rc)
1784                 goto out;
1785
1786         printk(KERN_INFO "L2TP core driver, %s\n", L2TP_DRV_VERSION);
1787
1788 out:
1789         return rc;
1790 }
1791
1792 static void __exit l2tp_exit(void)
1793 {
1794         unregister_pernet_device(&l2tp_net_ops);
1795 }
1796
1797 module_init(l2tp_init);
1798 module_exit(l2tp_exit);
1799
1800 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1801 MODULE_DESCRIPTION("L2TP core");
1802 MODULE_LICENSE("GPL");
1803 MODULE_VERSION(L2TP_DRV_VERSION);
1804