34b2ddeacb673b3b8d4f3e92bf726fc567c49a81
[linux-2.6.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_free(struct l2tp_tunnel *tunnel);
116 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel);
117
118 static inline struct l2tp_net *l2tp_pernet(struct net *net)
119 {
120         BUG_ON(!net);
121
122         return net_generic(net, l2tp_net_id);
123 }
124
125
126 /* Tunnel reference counts. Incremented per session that is added to
127  * the tunnel.
128  */
129 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
130 {
131         atomic_inc(&tunnel->ref_count);
132 }
133
134 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
135 {
136         if (atomic_dec_and_test(&tunnel->ref_count))
137                 l2tp_tunnel_free(tunnel);
138 }
139 #ifdef L2TP_REFCNT_DEBUG
140 #define l2tp_tunnel_inc_refcount(_t) do { \
141                 printk(KERN_DEBUG "l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n", __func__, __LINE__, (_t)->name, atomic_read(&_t->ref_count)); \
142                 l2tp_tunnel_inc_refcount_1(_t);                         \
143         } while (0)
144 #define l2tp_tunnel_dec_refcount(_t) do { \
145                 printk(KERN_DEBUG "l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n", __func__, __LINE__, (_t)->name, atomic_read(&_t->ref_count)); \
146                 l2tp_tunnel_dec_refcount_1(_t);                         \
147         } while (0)
148 #else
149 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
150 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
151 #endif
152
153 /* Session hash global list for L2TPv3.
154  * The session_id SHOULD be random according to RFC3931, but several
155  * L2TP implementations use incrementing session_ids.  So we do a real
156  * hash on the session_id, rather than a simple bitmask.
157  */
158 static inline struct hlist_head *
159 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
160 {
161         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
162
163 }
164
165 /* Lookup a session by id in the global session list
166  */
167 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
168 {
169         struct l2tp_net *pn = l2tp_pernet(net);
170         struct hlist_head *session_list =
171                 l2tp_session_id_hash_2(pn, session_id);
172         struct l2tp_session *session;
173         struct hlist_node *walk;
174
175         rcu_read_lock_bh();
176         hlist_for_each_entry_rcu(session, walk, session_list, global_hlist) {
177                 if (session->session_id == session_id) {
178                         rcu_read_unlock_bh();
179                         return session;
180                 }
181         }
182         rcu_read_unlock_bh();
183
184         return NULL;
185 }
186
187 /* Session hash list.
188  * The session_id SHOULD be random according to RFC2661, but several
189  * L2TP implementations (Cisco and Microsoft) use incrementing
190  * session_ids.  So we do a real hash on the session_id, rather than a
191  * simple bitmask.
192  */
193 static inline struct hlist_head *
194 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
195 {
196         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
197 }
198
199 /* Lookup a session by id
200  */
201 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
202 {
203         struct hlist_head *session_list;
204         struct l2tp_session *session;
205         struct hlist_node *walk;
206
207         /* In L2TPv3, session_ids are unique over all tunnels and we
208          * sometimes need to look them up before we know the
209          * tunnel.
210          */
211         if (tunnel == NULL)
212                 return l2tp_session_find_2(net, session_id);
213
214         session_list = l2tp_session_id_hash(tunnel, session_id);
215         read_lock_bh(&tunnel->hlist_lock);
216         hlist_for_each_entry(session, walk, session_list, hlist) {
217                 if (session->session_id == session_id) {
218                         read_unlock_bh(&tunnel->hlist_lock);
219                         return session;
220                 }
221         }
222         read_unlock_bh(&tunnel->hlist_lock);
223
224         return NULL;
225 }
226 EXPORT_SYMBOL_GPL(l2tp_session_find);
227
228 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
229 {
230         int hash;
231         struct hlist_node *walk;
232         struct l2tp_session *session;
233         int count = 0;
234
235         read_lock_bh(&tunnel->hlist_lock);
236         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
237                 hlist_for_each_entry(session, walk, &tunnel->session_hlist[hash], hlist) {
238                         if (++count > nth) {
239                                 read_unlock_bh(&tunnel->hlist_lock);
240                                 return session;
241                         }
242                 }
243         }
244
245         read_unlock_bh(&tunnel->hlist_lock);
246
247         return NULL;
248 }
249 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
250
251 /* Lookup a session by interface name.
252  * This is very inefficient but is only used by management interfaces.
253  */
254 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
255 {
256         struct l2tp_net *pn = l2tp_pernet(net);
257         int hash;
258         struct hlist_node *walk;
259         struct l2tp_session *session;
260
261         rcu_read_lock_bh();
262         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
263                 hlist_for_each_entry_rcu(session, walk, &pn->l2tp_session_hlist[hash], global_hlist) {
264                         if (!strcmp(session->ifname, ifname)) {
265                                 rcu_read_unlock_bh();
266                                 return session;
267                         }
268                 }
269         }
270
271         rcu_read_unlock_bh();
272
273         return NULL;
274 }
275 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
276
277 /* Lookup a tunnel by id
278  */
279 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
280 {
281         struct l2tp_tunnel *tunnel;
282         struct l2tp_net *pn = l2tp_pernet(net);
283
284         rcu_read_lock_bh();
285         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
286                 if (tunnel->tunnel_id == tunnel_id) {
287                         rcu_read_unlock_bh();
288                         return tunnel;
289                 }
290         }
291         rcu_read_unlock_bh();
292
293         return NULL;
294 }
295 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
296
297 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
298 {
299         struct l2tp_net *pn = l2tp_pernet(net);
300         struct l2tp_tunnel *tunnel;
301         int count = 0;
302
303         rcu_read_lock_bh();
304         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
305                 if (++count > nth) {
306                         rcu_read_unlock_bh();
307                         return tunnel;
308                 }
309         }
310
311         rcu_read_unlock_bh();
312
313         return NULL;
314 }
315 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
316
317 /*****************************************************************************
318  * Receive data handling
319  *****************************************************************************/
320
321 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
322  * number.
323  */
324 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
325 {
326         struct sk_buff *skbp;
327         struct sk_buff *tmp;
328         u32 ns = L2TP_SKB_CB(skb)->ns;
329
330         spin_lock_bh(&session->reorder_q.lock);
331         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
332                 if (L2TP_SKB_CB(skbp)->ns > ns) {
333                         __skb_queue_before(&session->reorder_q, skbp, skb);
334                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
335                                "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
336                                session->name, ns, L2TP_SKB_CB(skbp)->ns,
337                                skb_queue_len(&session->reorder_q));
338                         session->stats.rx_oos_packets++;
339                         goto out;
340                 }
341         }
342
343         __skb_queue_tail(&session->reorder_q, skb);
344
345 out:
346         spin_unlock_bh(&session->reorder_q.lock);
347 }
348
349 /* Dequeue a single skb.
350  */
351 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
352 {
353         struct l2tp_tunnel *tunnel = session->tunnel;
354         int length = L2TP_SKB_CB(skb)->length;
355
356         /* We're about to requeue the skb, so return resources
357          * to its current owner (a socket receive buffer).
358          */
359         skb_orphan(skb);
360
361         tunnel->stats.rx_packets++;
362         tunnel->stats.rx_bytes += length;
363         session->stats.rx_packets++;
364         session->stats.rx_bytes += length;
365
366         if (L2TP_SKB_CB(skb)->has_seq) {
367                 /* Bump our Nr */
368                 session->nr++;
369                 if (tunnel->version == L2TP_HDR_VER_2)
370                         session->nr &= 0xffff;
371                 else
372                         session->nr &= 0xffffff;
373
374                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
375                        "%s: updated nr to %hu\n", session->name, session->nr);
376         }
377
378         /* call private receive handler */
379         if (session->recv_skb != NULL)
380                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
381         else
382                 kfree_skb(skb);
383
384         if (session->deref)
385                 (*session->deref)(session);
386 }
387
388 /* Dequeue skbs from the session's reorder_q, subject to packet order.
389  * Skbs that have been in the queue for too long are simply discarded.
390  */
391 static void l2tp_recv_dequeue(struct l2tp_session *session)
392 {
393         struct sk_buff *skb;
394         struct sk_buff *tmp;
395
396         /* If the pkt at the head of the queue has the nr that we
397          * expect to send up next, dequeue it and any other
398          * in-sequence packets behind it.
399          */
400         spin_lock_bh(&session->reorder_q.lock);
401         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
402                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
403                         session->stats.rx_seq_discards++;
404                         session->stats.rx_errors++;
405                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
406                                "%s: oos pkt %u len %d discarded (too old), "
407                                "waiting for %u, reorder_q_len=%d\n",
408                                session->name, L2TP_SKB_CB(skb)->ns,
409                                L2TP_SKB_CB(skb)->length, session->nr,
410                                skb_queue_len(&session->reorder_q));
411                         __skb_unlink(skb, &session->reorder_q);
412                         kfree_skb(skb);
413                         if (session->deref)
414                                 (*session->deref)(session);
415                         continue;
416                 }
417
418                 if (L2TP_SKB_CB(skb)->has_seq) {
419                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
420                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
421                                        "%s: holding oos pkt %u len %d, "
422                                        "waiting for %u, reorder_q_len=%d\n",
423                                        session->name, L2TP_SKB_CB(skb)->ns,
424                                        L2TP_SKB_CB(skb)->length, session->nr,
425                                        skb_queue_len(&session->reorder_q));
426                                 goto out;
427                         }
428                 }
429                 __skb_unlink(skb, &session->reorder_q);
430
431                 /* Process the skb. We release the queue lock while we
432                  * do so to let other contexts process the queue.
433                  */
434                 spin_unlock_bh(&session->reorder_q.lock);
435                 l2tp_recv_dequeue_skb(session, skb);
436                 spin_lock_bh(&session->reorder_q.lock);
437         }
438
439 out:
440         spin_unlock_bh(&session->reorder_q.lock);
441 }
442
443 static inline int l2tp_verify_udp_checksum(struct sock *sk,
444                                            struct sk_buff *skb)
445 {
446         struct udphdr *uh = udp_hdr(skb);
447         u16 ulen = ntohs(uh->len);
448         struct inet_sock *inet;
449         __wsum psum;
450
451         if (sk->sk_no_check || skb_csum_unnecessary(skb) || !uh->check)
452                 return 0;
453
454         inet = inet_sk(sk);
455         psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr, ulen,
456                                   IPPROTO_UDP, 0);
457
458         if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
459             !csum_fold(csum_add(psum, skb->csum)))
460                 return 0;
461
462         skb->csum = psum;
463
464         return __skb_checksum_complete(skb);
465 }
466
467 /* Do receive processing of L2TP data frames. We handle both L2TPv2
468  * and L2TPv3 data frames here.
469  *
470  * L2TPv2 Data Message Header
471  *
472  *  0                   1                   2                   3
473  *  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
474  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
475  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
476  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
477  * |           Tunnel ID           |           Session ID          |
478  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
479  * |             Ns (opt)          |             Nr (opt)          |
480  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
481  * |      Offset Size (opt)        |    Offset pad... (opt)
482  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
483  *
484  * Data frames are marked by T=0. All other fields are the same as
485  * those in L2TP control frames.
486  *
487  * L2TPv3 Data Message Header
488  *
489  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
490  * |                      L2TP Session Header                      |
491  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
492  * |                      L2-Specific Sublayer                     |
493  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
494  * |                        Tunnel Payload                      ...
495  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
496  *
497  * L2TPv3 Session Header Over IP
498  *
499  *  0                   1                   2                   3
500  *  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
501  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
502  * |                           Session ID                          |
503  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
504  * |               Cookie (optional, maximum 64 bits)...
505  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
506  *                                                                 |
507  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
508  *
509  * L2TPv3 L2-Specific Sublayer Format
510  *
511  *  0                   1                   2                   3
512  *  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
513  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
514  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
515  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
516  *
517  * Cookie value, sublayer format and offset (pad) are negotiated with
518  * the peer when the session is set up. Unlike L2TPv2, we do not need
519  * to parse the packet header to determine if optional fields are
520  * present.
521  *
522  * Caller must already have parsed the frame and determined that it is
523  * a data (not control) frame before coming here. Fields up to the
524  * session-id have already been parsed and ptr points to the data
525  * after the session-id.
526  */
527 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
528                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
529                       int length, int (*payload_hook)(struct sk_buff *skb))
530 {
531         struct l2tp_tunnel *tunnel = session->tunnel;
532         int offset;
533         u32 ns, nr;
534
535         /* The ref count is increased since we now hold a pointer to
536          * the session. Take care to decrement the refcnt when exiting
537          * this function from now on...
538          */
539         l2tp_session_inc_refcount(session);
540         if (session->ref)
541                 (*session->ref)(session);
542
543         /* Parse and check optional cookie */
544         if (session->peer_cookie_len > 0) {
545                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
546                         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
547                                "%s: cookie mismatch (%u/%u). Discarding.\n",
548                                tunnel->name, tunnel->tunnel_id, session->session_id);
549                         session->stats.rx_cookie_discards++;
550                         goto discard;
551                 }
552                 ptr += session->peer_cookie_len;
553         }
554
555         /* Handle the optional sequence numbers. Sequence numbers are
556          * in different places for L2TPv2 and L2TPv3.
557          *
558          * If we are the LAC, enable/disable sequence numbers under
559          * the control of the LNS.  If no sequence numbers present but
560          * we were expecting them, discard frame.
561          */
562         ns = nr = 0;
563         L2TP_SKB_CB(skb)->has_seq = 0;
564         if (tunnel->version == L2TP_HDR_VER_2) {
565                 if (hdrflags & L2TP_HDRFLAG_S) {
566                         ns = ntohs(*(__be16 *) ptr);
567                         ptr += 2;
568                         nr = ntohs(*(__be16 *) ptr);
569                         ptr += 2;
570
571                         /* Store L2TP info in the skb */
572                         L2TP_SKB_CB(skb)->ns = ns;
573                         L2TP_SKB_CB(skb)->has_seq = 1;
574
575                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
576                                "%s: recv data ns=%u, nr=%u, session nr=%u\n",
577                                session->name, ns, nr, session->nr);
578                 }
579         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
580                 u32 l2h = ntohl(*(__be32 *) ptr);
581
582                 if (l2h & 0x40000000) {
583                         ns = l2h & 0x00ffffff;
584
585                         /* Store L2TP info in the skb */
586                         L2TP_SKB_CB(skb)->ns = ns;
587                         L2TP_SKB_CB(skb)->has_seq = 1;
588
589                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
590                                "%s: recv data ns=%u, session nr=%u\n",
591                                session->name, ns, session->nr);
592                 }
593         }
594
595         /* Advance past L2-specific header, if present */
596         ptr += session->l2specific_len;
597
598         if (L2TP_SKB_CB(skb)->has_seq) {
599                 /* Received a packet with sequence numbers. If we're the LNS,
600                  * check if we sre sending sequence numbers and if not,
601                  * configure it so.
602                  */
603                 if ((!session->lns_mode) && (!session->send_seq)) {
604                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
605                                "%s: requested to enable seq numbers by LNS\n",
606                                session->name);
607                         session->send_seq = -1;
608                         l2tp_session_set_header_len(session, tunnel->version);
609                 }
610         } else {
611                 /* No sequence numbers.
612                  * If user has configured mandatory sequence numbers, discard.
613                  */
614                 if (session->recv_seq) {
615                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
616                                "%s: recv data has no seq numbers when required. "
617                                "Discarding\n", session->name);
618                         session->stats.rx_seq_discards++;
619                         goto discard;
620                 }
621
622                 /* If we're the LAC and we're sending sequence numbers, the
623                  * LNS has requested that we no longer send sequence numbers.
624                  * If we're the LNS and we're sending sequence numbers, the
625                  * LAC is broken. Discard the frame.
626                  */
627                 if ((!session->lns_mode) && (session->send_seq)) {
628                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
629                                "%s: requested to disable seq numbers by LNS\n",
630                                session->name);
631                         session->send_seq = 0;
632                         l2tp_session_set_header_len(session, tunnel->version);
633                 } else if (session->send_seq) {
634                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
635                                "%s: recv data has no seq numbers when required. "
636                                "Discarding\n", session->name);
637                         session->stats.rx_seq_discards++;
638                         goto discard;
639                 }
640         }
641
642         /* Session data offset is handled differently for L2TPv2 and
643          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
644          * the header. For L2TPv3, the offset is negotiated using AVPs
645          * in the session setup control protocol.
646          */
647         if (tunnel->version == L2TP_HDR_VER_2) {
648                 /* If offset bit set, skip it. */
649                 if (hdrflags & L2TP_HDRFLAG_O) {
650                         offset = ntohs(*(__be16 *)ptr);
651                         ptr += 2 + offset;
652                 }
653         } else
654                 ptr += session->offset;
655
656         offset = ptr - optr;
657         if (!pskb_may_pull(skb, offset))
658                 goto discard;
659
660         __skb_pull(skb, offset);
661
662         /* If caller wants to process the payload before we queue the
663          * packet, do so now.
664          */
665         if (payload_hook)
666                 if ((*payload_hook)(skb))
667                         goto discard;
668
669         /* Prepare skb for adding to the session's reorder_q.  Hold
670          * packets for max reorder_timeout or 1 second if not
671          * reordering.
672          */
673         L2TP_SKB_CB(skb)->length = length;
674         L2TP_SKB_CB(skb)->expires = jiffies +
675                 (session->reorder_timeout ? session->reorder_timeout : HZ);
676
677         /* Add packet to the session's receive queue. Reordering is done here, if
678          * enabled. Saved L2TP protocol info is stored in skb->sb[].
679          */
680         if (L2TP_SKB_CB(skb)->has_seq) {
681                 if (session->reorder_timeout != 0) {
682                         /* Packet reordering enabled. Add skb to session's
683                          * reorder queue, in order of ns.
684                          */
685                         l2tp_recv_queue_skb(session, skb);
686                 } else {
687                         /* Packet reordering disabled. Discard out-of-sequence
688                          * packets
689                          */
690                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
691                                 session->stats.rx_seq_discards++;
692                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
693                                        "%s: oos pkt %u len %d discarded, "
694                                        "waiting for %u, reorder_q_len=%d\n",
695                                        session->name, L2TP_SKB_CB(skb)->ns,
696                                        L2TP_SKB_CB(skb)->length, session->nr,
697                                        skb_queue_len(&session->reorder_q));
698                                 goto discard;
699                         }
700                         skb_queue_tail(&session->reorder_q, skb);
701                 }
702         } else {
703                 /* No sequence numbers. Add the skb to the tail of the
704                  * reorder queue. This ensures that it will be
705                  * delivered after all previous sequenced skbs.
706                  */
707                 skb_queue_tail(&session->reorder_q, skb);
708         }
709
710         /* Try to dequeue as many skbs from reorder_q as we can. */
711         l2tp_recv_dequeue(session);
712
713         l2tp_session_dec_refcount(session);
714
715         return;
716
717 discard:
718         session->stats.rx_errors++;
719         kfree_skb(skb);
720
721         if (session->deref)
722                 (*session->deref)(session);
723
724         l2tp_session_dec_refcount(session);
725 }
726 EXPORT_SYMBOL(l2tp_recv_common);
727
728 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
729  * here. The skb is not on a list when we get here.
730  * Returns 0 if the packet was a data packet and was successfully passed on.
731  * Returns 1 if the packet was not a good data packet and could not be
732  * forwarded.  All such packets are passed up to userspace to deal with.
733  */
734 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
735                               int (*payload_hook)(struct sk_buff *skb))
736 {
737         struct l2tp_session *session = NULL;
738         unsigned char *ptr, *optr;
739         u16 hdrflags;
740         u32 tunnel_id, session_id;
741         int offset;
742         u16 version;
743         int length;
744
745         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
746                 goto discard_bad_csum;
747
748         /* UDP always verifies the packet length. */
749         __skb_pull(skb, sizeof(struct udphdr));
750
751         /* Short packet? */
752         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
753                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
754                        "%s: recv short packet (len=%d)\n", tunnel->name, skb->len);
755                 goto error;
756         }
757
758         /* Point to L2TP header */
759         optr = ptr = skb->data;
760
761         /* Trace packet contents, if enabled */
762         if (tunnel->debug & L2TP_MSG_DATA) {
763                 length = min(32u, skb->len);
764                 if (!pskb_may_pull(skb, length))
765                         goto error;
766
767                 printk(KERN_DEBUG "%s: recv: ", tunnel->name);
768
769                 offset = 0;
770                 do {
771                         printk(" %02X", ptr[offset]);
772                 } while (++offset < length);
773
774                 printk("\n");
775         }
776
777         /* Get L2TP header flags */
778         hdrflags = ntohs(*(__be16 *) ptr);
779
780         /* Check protocol version */
781         version = hdrflags & L2TP_HDR_VER_MASK;
782         if (version != tunnel->version) {
783                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
784                        "%s: recv protocol version mismatch: got %d expected %d\n",
785                        tunnel->name, version, tunnel->version);
786                 goto error;
787         }
788
789         /* Get length of L2TP packet */
790         length = skb->len;
791
792         /* If type is control packet, it is handled by userspace. */
793         if (hdrflags & L2TP_HDRFLAG_T) {
794                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
795                        "%s: recv control packet, len=%d\n", tunnel->name, length);
796                 goto error;
797         }
798
799         /* Skip flags */
800         ptr += 2;
801
802         if (tunnel->version == L2TP_HDR_VER_2) {
803                 /* If length is present, skip it */
804                 if (hdrflags & L2TP_HDRFLAG_L)
805                         ptr += 2;
806
807                 /* Extract tunnel and session ID */
808                 tunnel_id = ntohs(*(__be16 *) ptr);
809                 ptr += 2;
810                 session_id = ntohs(*(__be16 *) ptr);
811                 ptr += 2;
812         } else {
813                 ptr += 2;       /* skip reserved bits */
814                 tunnel_id = tunnel->tunnel_id;
815                 session_id = ntohl(*(__be32 *) ptr);
816                 ptr += 4;
817         }
818
819         /* Find the session context */
820         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
821         if (!session || !session->recv_skb) {
822                 /* Not found? Pass to userspace to deal with */
823                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
824                        "%s: no session found (%u/%u). Passing up.\n",
825                        tunnel->name, tunnel_id, session_id);
826                 goto error;
827         }
828
829         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
830
831         return 0;
832
833 discard_bad_csum:
834         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
835         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
836         tunnel->stats.rx_errors++;
837         kfree_skb(skb);
838
839         return 0;
840
841 error:
842         /* Put UDP header back */
843         __skb_push(skb, sizeof(struct udphdr));
844
845         return 1;
846 }
847
848 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
849  * Return codes:
850  * 0 : success.
851  * <0: error
852  * >0: skb should be passed up to userspace as UDP.
853  */
854 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
855 {
856         struct l2tp_tunnel *tunnel;
857
858         tunnel = l2tp_sock_to_tunnel(sk);
859         if (tunnel == NULL)
860                 goto pass_up;
861
862         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
863                "%s: received %d bytes\n", tunnel->name, skb->len);
864
865         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
866                 goto pass_up_put;
867
868         sock_put(sk);
869         return 0;
870
871 pass_up_put:
872         sock_put(sk);
873 pass_up:
874         return 1;
875 }
876 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
877
878 /************************************************************************
879  * Transmit handling
880  ***********************************************************************/
881
882 /* Build an L2TP header for the session into the buffer provided.
883  */
884 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
885 {
886         struct l2tp_tunnel *tunnel = session->tunnel;
887         __be16 *bufp = buf;
888         __be16 *optr = buf;
889         u16 flags = L2TP_HDR_VER_2;
890         u32 tunnel_id = tunnel->peer_tunnel_id;
891         u32 session_id = session->peer_session_id;
892
893         if (session->send_seq)
894                 flags |= L2TP_HDRFLAG_S;
895
896         /* Setup L2TP header. */
897         *bufp++ = htons(flags);
898         *bufp++ = htons(tunnel_id);
899         *bufp++ = htons(session_id);
900         if (session->send_seq) {
901                 *bufp++ = htons(session->ns);
902                 *bufp++ = 0;
903                 session->ns++;
904                 session->ns &= 0xffff;
905                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
906                        "%s: updated ns to %u\n", session->name, session->ns);
907         }
908
909         return bufp - optr;
910 }
911
912 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
913 {
914         struct l2tp_tunnel *tunnel = session->tunnel;
915         char *bufp = buf;
916         char *optr = bufp;
917
918         /* Setup L2TP header. The header differs slightly for UDP and
919          * IP encapsulations. For UDP, there is 4 bytes of flags.
920          */
921         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
922                 u16 flags = L2TP_HDR_VER_3;
923                 *((__be16 *) bufp) = htons(flags);
924                 bufp += 2;
925                 *((__be16 *) bufp) = 0;
926                 bufp += 2;
927         }
928
929         *((__be32 *) bufp) = htonl(session->peer_session_id);
930         bufp += 4;
931         if (session->cookie_len) {
932                 memcpy(bufp, &session->cookie[0], session->cookie_len);
933                 bufp += session->cookie_len;
934         }
935         if (session->l2specific_len) {
936                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
937                         u32 l2h = 0;
938                         if (session->send_seq) {
939                                 l2h = 0x40000000 | session->ns;
940                                 session->ns++;
941                                 session->ns &= 0xffffff;
942                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
943                                        "%s: updated ns to %u\n", session->name, session->ns);
944                         }
945
946                         *((__be32 *) bufp) = htonl(l2h);
947                 }
948                 bufp += session->l2specific_len;
949         }
950         if (session->offset)
951                 bufp += session->offset;
952
953         return bufp - optr;
954 }
955
956 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
957                           struct flowi *fl, size_t data_len)
958 {
959         struct l2tp_tunnel *tunnel = session->tunnel;
960         unsigned int len = skb->len;
961         int error;
962
963         /* Debug */
964         if (session->send_seq)
965                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
966                        "%s: send %Zd bytes, ns=%u\n", session->name,
967                        data_len, session->ns - 1);
968         else
969                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
970                        "%s: send %Zd bytes\n", session->name, data_len);
971
972         if (session->debug & L2TP_MSG_DATA) {
973                 int i;
974                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
975                 unsigned char *datap = skb->data + uhlen;
976
977                 printk(KERN_DEBUG "%s: xmit:", session->name);
978                 for (i = 0; i < (len - uhlen); i++) {
979                         printk(" %02X", *datap++);
980                         if (i == 31) {
981                                 printk(" ...");
982                                 break;
983                         }
984                 }
985                 printk("\n");
986         }
987
988         /* Queue the packet to IP for output */
989         skb->local_df = 1;
990         error = ip_queue_xmit(skb, fl);
991
992         /* Update stats */
993         if (error >= 0) {
994                 tunnel->stats.tx_packets++;
995                 tunnel->stats.tx_bytes += len;
996                 session->stats.tx_packets++;
997                 session->stats.tx_bytes += len;
998         } else {
999                 tunnel->stats.tx_errors++;
1000                 session->stats.tx_errors++;
1001         }
1002
1003         return 0;
1004 }
1005
1006 /* Automatically called when the skb is freed.
1007  */
1008 static void l2tp_sock_wfree(struct sk_buff *skb)
1009 {
1010         sock_put(skb->sk);
1011 }
1012
1013 /* For data skbs that we transmit, we associate with the tunnel socket
1014  * but don't do accounting.
1015  */
1016 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1017 {
1018         sock_hold(sk);
1019         skb->sk = sk;
1020         skb->destructor = l2tp_sock_wfree;
1021 }
1022
1023 /* If caller requires the skb to have a ppp header, the header must be
1024  * inserted in the skb data before calling this function.
1025  */
1026 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1027 {
1028         int data_len = skb->len;
1029         struct l2tp_tunnel *tunnel = session->tunnel;
1030         struct sock *sk = tunnel->sock;
1031         struct flowi *fl;
1032         struct udphdr *uh;
1033         struct inet_sock *inet;
1034         __wsum csum;
1035         int old_headroom;
1036         int new_headroom;
1037         int headroom;
1038         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1039         int udp_len;
1040
1041         /* Check that there's enough headroom in the skb to insert IP,
1042          * UDP and L2TP headers. If not enough, expand it to
1043          * make room. Adjust truesize.
1044          */
1045         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1046                 uhlen + hdr_len;
1047         old_headroom = skb_headroom(skb);
1048         if (skb_cow_head(skb, headroom)) {
1049                 dev_kfree_skb(skb);
1050                 goto abort;
1051         }
1052
1053         new_headroom = skb_headroom(skb);
1054         skb_orphan(skb);
1055         skb->truesize += new_headroom - old_headroom;
1056
1057         /* Setup L2TP header */
1058         session->build_header(session, __skb_push(skb, hdr_len));
1059
1060         /* Reset skb netfilter state */
1061         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1062         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1063                               IPSKB_REROUTED);
1064         nf_reset(skb);
1065
1066         bh_lock_sock(sk);
1067         if (sock_owned_by_user(sk)) {
1068                 dev_kfree_skb(skb);
1069                 goto out_unlock;
1070         }
1071
1072         /* Get routing info from the tunnel socket */
1073         skb_dst_drop(skb);
1074         skb_dst_set(skb, dst_clone(__sk_dst_get(sk)));
1075
1076         inet = inet_sk(sk);
1077         fl = &inet->cork.fl;
1078         switch (tunnel->encap) {
1079         case L2TP_ENCAPTYPE_UDP:
1080                 /* Setup UDP header */
1081                 __skb_push(skb, sizeof(*uh));
1082                 skb_reset_transport_header(skb);
1083                 uh = udp_hdr(skb);
1084                 uh->source = inet->inet_sport;
1085                 uh->dest = inet->inet_dport;
1086                 udp_len = uhlen + hdr_len + data_len;
1087                 uh->len = htons(udp_len);
1088                 uh->check = 0;
1089
1090                 /* Calculate UDP checksum if configured to do so */
1091                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1092                         skb->ip_summed = CHECKSUM_NONE;
1093                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1094                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1095                         skb->ip_summed = CHECKSUM_COMPLETE;
1096                         csum = skb_checksum(skb, 0, udp_len, 0);
1097                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1098                                                       inet->inet_daddr,
1099                                                       udp_len, IPPROTO_UDP, csum);
1100                         if (uh->check == 0)
1101                                 uh->check = CSUM_MANGLED_0;
1102                 } else {
1103                         skb->ip_summed = CHECKSUM_PARTIAL;
1104                         skb->csum_start = skb_transport_header(skb) - skb->head;
1105                         skb->csum_offset = offsetof(struct udphdr, check);
1106                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1107                                                        inet->inet_daddr,
1108                                                        udp_len, IPPROTO_UDP, 0);
1109                 }
1110                 break;
1111
1112         case L2TP_ENCAPTYPE_IP:
1113                 break;
1114         }
1115
1116         l2tp_skb_set_owner_w(skb, sk);
1117
1118         l2tp_xmit_core(session, skb, fl, data_len);
1119 out_unlock:
1120         bh_unlock_sock(sk);
1121
1122 abort:
1123         return 0;
1124 }
1125 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1126
1127 /*****************************************************************************
1128  * Tinnel and session create/destroy.
1129  *****************************************************************************/
1130
1131 /* Tunnel socket destruct hook.
1132  * The tunnel context is deleted only when all session sockets have been
1133  * closed.
1134  */
1135 static void l2tp_tunnel_destruct(struct sock *sk)
1136 {
1137         struct l2tp_tunnel *tunnel;
1138
1139         tunnel = sk->sk_user_data;
1140         if (tunnel == NULL)
1141                 goto end;
1142
1143         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1144                "%s: closing...\n", tunnel->name);
1145
1146         /* Close all sessions */
1147         l2tp_tunnel_closeall(tunnel);
1148
1149         switch (tunnel->encap) {
1150         case L2TP_ENCAPTYPE_UDP:
1151                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1152                 (udp_sk(sk))->encap_type = 0;
1153                 (udp_sk(sk))->encap_rcv = NULL;
1154                 break;
1155         case L2TP_ENCAPTYPE_IP:
1156                 break;
1157         }
1158
1159         /* Remove hooks into tunnel socket */
1160         tunnel->sock = NULL;
1161         sk->sk_destruct = tunnel->old_sk_destruct;
1162         sk->sk_user_data = NULL;
1163
1164         /* Call the original destructor */
1165         if (sk->sk_destruct)
1166                 (*sk->sk_destruct)(sk);
1167
1168         /* We're finished with the socket */
1169         l2tp_tunnel_dec_refcount(tunnel);
1170
1171 end:
1172         return;
1173 }
1174
1175 /* When the tunnel is closed, all the attached sessions need to go too.
1176  */
1177 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1178 {
1179         int hash;
1180         struct hlist_node *walk;
1181         struct hlist_node *tmp;
1182         struct l2tp_session *session;
1183
1184         BUG_ON(tunnel == NULL);
1185
1186         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1187                "%s: closing all sessions...\n", tunnel->name);
1188
1189         write_lock_bh(&tunnel->hlist_lock);
1190         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1191 again:
1192                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1193                         session = hlist_entry(walk, struct l2tp_session, hlist);
1194
1195                         PRINTK(session->debug, L2TP_MSG_CONTROL, KERN_INFO,
1196                                "%s: closing session\n", session->name);
1197
1198                         hlist_del_init(&session->hlist);
1199
1200                         /* Since we should hold the sock lock while
1201                          * doing any unbinding, we need to release the
1202                          * lock we're holding before taking that lock.
1203                          * Hold a reference to the sock so it doesn't
1204                          * disappear as we're jumping between locks.
1205                          */
1206                         if (session->ref != NULL)
1207                                 (*session->ref)(session);
1208
1209                         write_unlock_bh(&tunnel->hlist_lock);
1210
1211                         if (tunnel->version != L2TP_HDR_VER_2) {
1212                                 struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1213
1214                                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
1215                                 hlist_del_init_rcu(&session->global_hlist);
1216                                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1217                                 synchronize_rcu();
1218                         }
1219
1220                         if (session->session_close != NULL)
1221                                 (*session->session_close)(session);
1222
1223                         if (session->deref != NULL)
1224                                 (*session->deref)(session);
1225
1226                         write_lock_bh(&tunnel->hlist_lock);
1227
1228                         /* Now restart from the beginning of this hash
1229                          * chain.  We always remove a session from the
1230                          * list so we are guaranteed to make forward
1231                          * progress.
1232                          */
1233                         goto again;
1234                 }
1235         }
1236         write_unlock_bh(&tunnel->hlist_lock);
1237 }
1238
1239 /* Really kill the tunnel.
1240  * Come here only when all sessions have been cleared from the tunnel.
1241  */
1242 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1243 {
1244         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1245
1246         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1247         BUG_ON(tunnel->sock != NULL);
1248
1249         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1250                "%s: free...\n", tunnel->name);
1251
1252         /* Remove from tunnel list */
1253         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1254         list_del_rcu(&tunnel->list);
1255         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1256         synchronize_rcu();
1257
1258         atomic_dec(&l2tp_tunnel_count);
1259         kfree(tunnel);
1260 }
1261
1262 /* Create a socket for the tunnel, if one isn't set up by
1263  * userspace. This is used for static tunnels where there is no
1264  * managing L2TP daemon.
1265  */
1266 static int l2tp_tunnel_sock_create(u32 tunnel_id, u32 peer_tunnel_id, struct l2tp_tunnel_cfg *cfg, struct socket **sockp)
1267 {
1268         int err = -EINVAL;
1269         struct sockaddr_in udp_addr;
1270         struct sockaddr_l2tpip ip_addr;
1271         struct socket *sock = NULL;
1272
1273         switch (cfg->encap) {
1274         case L2TP_ENCAPTYPE_UDP:
1275                 err = sock_create(AF_INET, SOCK_DGRAM, 0, sockp);
1276                 if (err < 0)
1277                         goto out;
1278
1279                 sock = *sockp;
1280
1281                 memset(&udp_addr, 0, sizeof(udp_addr));
1282                 udp_addr.sin_family = AF_INET;
1283                 udp_addr.sin_addr = cfg->local_ip;
1284                 udp_addr.sin_port = htons(cfg->local_udp_port);
1285                 err = kernel_bind(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr));
1286                 if (err < 0)
1287                         goto out;
1288
1289                 udp_addr.sin_family = AF_INET;
1290                 udp_addr.sin_addr = cfg->peer_ip;
1291                 udp_addr.sin_port = htons(cfg->peer_udp_port);
1292                 err = kernel_connect(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr), 0);
1293                 if (err < 0)
1294                         goto out;
1295
1296                 if (!cfg->use_udp_checksums)
1297                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1298
1299                 break;
1300
1301         case L2TP_ENCAPTYPE_IP:
1302                 err = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_L2TP, sockp);
1303                 if (err < 0)
1304                         goto out;
1305
1306                 sock = *sockp;
1307
1308                 memset(&ip_addr, 0, sizeof(ip_addr));
1309                 ip_addr.l2tp_family = AF_INET;
1310                 ip_addr.l2tp_addr = cfg->local_ip;
1311                 ip_addr.l2tp_conn_id = tunnel_id;
1312                 err = kernel_bind(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr));
1313                 if (err < 0)
1314                         goto out;
1315
1316                 ip_addr.l2tp_family = AF_INET;
1317                 ip_addr.l2tp_addr = cfg->peer_ip;
1318                 ip_addr.l2tp_conn_id = peer_tunnel_id;
1319                 err = kernel_connect(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr), 0);
1320                 if (err < 0)
1321                         goto out;
1322
1323                 break;
1324
1325         default:
1326                 goto out;
1327         }
1328
1329 out:
1330         if ((err < 0) && sock) {
1331                 sock_release(sock);
1332                 *sockp = NULL;
1333         }
1334
1335         return err;
1336 }
1337
1338 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)
1339 {
1340         struct l2tp_tunnel *tunnel = NULL;
1341         int err;
1342         struct socket *sock = NULL;
1343         struct sock *sk = NULL;
1344         struct l2tp_net *pn;
1345         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1346
1347         /* Get the tunnel socket from the fd, which was opened by
1348          * the userspace L2TP daemon. If not specified, create a
1349          * kernel socket.
1350          */
1351         if (fd < 0) {
1352                 err = l2tp_tunnel_sock_create(tunnel_id, peer_tunnel_id, cfg, &sock);
1353                 if (err < 0)
1354                         goto err;
1355         } else {
1356                 err = -EBADF;
1357                 sock = sockfd_lookup(fd, &err);
1358                 if (!sock) {
1359                         printk(KERN_ERR "tunl %hu: sockfd_lookup(fd=%d) returned %d\n",
1360                                tunnel_id, fd, err);
1361                         goto err;
1362                 }
1363         }
1364
1365         sk = sock->sk;
1366
1367         if (cfg != NULL)
1368                 encap = cfg->encap;
1369
1370         /* Quick sanity checks */
1371         switch (encap) {
1372         case L2TP_ENCAPTYPE_UDP:
1373                 err = -EPROTONOSUPPORT;
1374                 if (sk->sk_protocol != IPPROTO_UDP) {
1375                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1376                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1377                         goto err;
1378                 }
1379                 break;
1380         case L2TP_ENCAPTYPE_IP:
1381                 err = -EPROTONOSUPPORT;
1382                 if (sk->sk_protocol != IPPROTO_L2TP) {
1383                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1384                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1385                         goto err;
1386                 }
1387                 break;
1388         }
1389
1390         /* Check if this socket has already been prepped */
1391         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
1392         if (tunnel != NULL) {
1393                 /* This socket has already been prepped */
1394                 err = -EBUSY;
1395                 goto err;
1396         }
1397
1398         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1399         if (tunnel == NULL) {
1400                 err = -ENOMEM;
1401                 goto err;
1402         }
1403
1404         tunnel->version = version;
1405         tunnel->tunnel_id = tunnel_id;
1406         tunnel->peer_tunnel_id = peer_tunnel_id;
1407         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1408
1409         tunnel->magic = L2TP_TUNNEL_MAGIC;
1410         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1411         rwlock_init(&tunnel->hlist_lock);
1412
1413         /* The net we belong to */
1414         tunnel->l2tp_net = net;
1415         pn = l2tp_pernet(net);
1416
1417         if (cfg != NULL)
1418                 tunnel->debug = cfg->debug;
1419
1420         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1421         tunnel->encap = encap;
1422         if (encap == L2TP_ENCAPTYPE_UDP) {
1423                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1424                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1425                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1426         }
1427
1428         sk->sk_user_data = tunnel;
1429
1430         /* Hook on the tunnel socket destructor so that we can cleanup
1431          * if the tunnel socket goes away.
1432          */
1433         tunnel->old_sk_destruct = sk->sk_destruct;
1434         sk->sk_destruct = &l2tp_tunnel_destruct;
1435         tunnel->sock = sk;
1436         sk->sk_allocation = GFP_ATOMIC;
1437
1438         /* Add tunnel to our list */
1439         INIT_LIST_HEAD(&tunnel->list);
1440         atomic_inc(&l2tp_tunnel_count);
1441
1442         /* Bump the reference count. The tunnel context is deleted
1443          * only when this drops to zero. Must be done before list insertion
1444          */
1445         l2tp_tunnel_inc_refcount(tunnel);
1446         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1447         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1448         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1449
1450         err = 0;
1451 err:
1452         if (tunnelp)
1453                 *tunnelp = tunnel;
1454
1455         /* If tunnel's socket was created by the kernel, it doesn't
1456          *  have a file.
1457          */
1458         if (sock && sock->file)
1459                 sockfd_put(sock);
1460
1461         return err;
1462 }
1463 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1464
1465 /* This function is used by the netlink TUNNEL_DELETE command.
1466  */
1467 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1468 {
1469         int err = 0;
1470         struct socket *sock = tunnel->sock ? tunnel->sock->sk_socket : NULL;
1471
1472         /* Force the tunnel socket to close. This will eventually
1473          * cause the tunnel to be deleted via the normal socket close
1474          * mechanisms when userspace closes the tunnel socket.
1475          */
1476         if (sock != NULL) {
1477                 err = inet_shutdown(sock, 2);
1478
1479                 /* If the tunnel's socket was created by the kernel,
1480                  * close the socket here since the socket was not
1481                  * created by userspace.
1482                  */
1483                 if (sock->file == NULL)
1484                         err = inet_release(sock);
1485         }
1486
1487         return err;
1488 }
1489 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1490
1491 /* Really kill the session.
1492  */
1493 void l2tp_session_free(struct l2tp_session *session)
1494 {
1495         struct l2tp_tunnel *tunnel;
1496
1497         BUG_ON(atomic_read(&session->ref_count) != 0);
1498
1499         tunnel = session->tunnel;
1500         if (tunnel != NULL) {
1501                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1502
1503                 /* Delete the session from the hash */
1504                 write_lock_bh(&tunnel->hlist_lock);
1505                 hlist_del_init(&session->hlist);
1506                 write_unlock_bh(&tunnel->hlist_lock);
1507
1508                 /* Unlink from the global hash if not L2TPv2 */
1509                 if (tunnel->version != L2TP_HDR_VER_2) {
1510                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1511
1512                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1513                         hlist_del_init_rcu(&session->global_hlist);
1514                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1515                         synchronize_rcu();
1516                 }
1517
1518                 if (session->session_id != 0)
1519                         atomic_dec(&l2tp_session_count);
1520
1521                 sock_put(tunnel->sock);
1522
1523                 /* This will delete the tunnel context if this
1524                  * is the last session on the tunnel.
1525                  */
1526                 session->tunnel = NULL;
1527                 l2tp_tunnel_dec_refcount(tunnel);
1528         }
1529
1530         kfree(session);
1531
1532         return;
1533 }
1534 EXPORT_SYMBOL_GPL(l2tp_session_free);
1535
1536 /* This function is used by the netlink SESSION_DELETE command and by
1537    pseudowire modules.
1538  */
1539 int l2tp_session_delete(struct l2tp_session *session)
1540 {
1541         if (session->session_close != NULL)
1542                 (*session->session_close)(session);
1543
1544         l2tp_session_dec_refcount(session);
1545
1546         return 0;
1547 }
1548 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1549
1550
1551 /* We come here whenever a session's send_seq, cookie_len or
1552  * l2specific_len parameters are set.
1553  */
1554 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1555 {
1556         if (version == L2TP_HDR_VER_2) {
1557                 session->hdr_len = 6;
1558                 if (session->send_seq)
1559                         session->hdr_len += 4;
1560         } else {
1561                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1562                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1563                         session->hdr_len += 4;
1564         }
1565
1566 }
1567
1568 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)
1569 {
1570         struct l2tp_session *session;
1571
1572         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1573         if (session != NULL) {
1574                 session->magic = L2TP_SESSION_MAGIC;
1575                 session->tunnel = tunnel;
1576
1577                 session->session_id = session_id;
1578                 session->peer_session_id = peer_session_id;
1579                 session->nr = 1;
1580
1581                 sprintf(&session->name[0], "sess %u/%u",
1582                         tunnel->tunnel_id, session->session_id);
1583
1584                 skb_queue_head_init(&session->reorder_q);
1585
1586                 INIT_HLIST_NODE(&session->hlist);
1587                 INIT_HLIST_NODE(&session->global_hlist);
1588
1589                 /* Inherit debug options from tunnel */
1590                 session->debug = tunnel->debug;
1591
1592                 if (cfg) {
1593                         session->pwtype = cfg->pw_type;
1594                         session->debug = cfg->debug;
1595                         session->mtu = cfg->mtu;
1596                         session->mru = cfg->mru;
1597                         session->send_seq = cfg->send_seq;
1598                         session->recv_seq = cfg->recv_seq;
1599                         session->lns_mode = cfg->lns_mode;
1600                         session->reorder_timeout = cfg->reorder_timeout;
1601                         session->offset = cfg->offset;
1602                         session->l2specific_type = cfg->l2specific_type;
1603                         session->l2specific_len = cfg->l2specific_len;
1604                         session->cookie_len = cfg->cookie_len;
1605                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1606                         session->peer_cookie_len = cfg->peer_cookie_len;
1607                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1608                 }
1609
1610                 if (tunnel->version == L2TP_HDR_VER_2)
1611                         session->build_header = l2tp_build_l2tpv2_header;
1612                 else
1613                         session->build_header = l2tp_build_l2tpv3_header;
1614
1615                 l2tp_session_set_header_len(session, tunnel->version);
1616
1617                 /* Bump the reference count. The session context is deleted
1618                  * only when this drops to zero.
1619                  */
1620                 l2tp_session_inc_refcount(session);
1621                 l2tp_tunnel_inc_refcount(tunnel);
1622
1623                 /* Ensure tunnel socket isn't deleted */
1624                 sock_hold(tunnel->sock);
1625
1626                 /* Add session to the tunnel's hash list */
1627                 write_lock_bh(&tunnel->hlist_lock);
1628                 hlist_add_head(&session->hlist,
1629                                l2tp_session_id_hash(tunnel, session_id));
1630                 write_unlock_bh(&tunnel->hlist_lock);
1631
1632                 /* And to the global session list if L2TPv3 */
1633                 if (tunnel->version != L2TP_HDR_VER_2) {
1634                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1635
1636                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1637                         hlist_add_head_rcu(&session->global_hlist,
1638                                            l2tp_session_id_hash_2(pn, session_id));
1639                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1640                 }
1641
1642                 /* Ignore management session in session count value */
1643                 if (session->session_id != 0)
1644                         atomic_inc(&l2tp_session_count);
1645         }
1646
1647         return session;
1648 }
1649 EXPORT_SYMBOL_GPL(l2tp_session_create);
1650
1651 /*****************************************************************************
1652  * Init and cleanup
1653  *****************************************************************************/
1654
1655 static __net_init int l2tp_init_net(struct net *net)
1656 {
1657         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1658         int hash;
1659
1660         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1661         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1662
1663         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1664                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1665
1666         spin_lock_init(&pn->l2tp_session_hlist_lock);
1667
1668         return 0;
1669 }
1670
1671 static struct pernet_operations l2tp_net_ops = {
1672         .init = l2tp_init_net,
1673         .id   = &l2tp_net_id,
1674         .size = sizeof(struct l2tp_net),
1675 };
1676
1677 static int __init l2tp_init(void)
1678 {
1679         int rc = 0;
1680
1681         rc = register_pernet_device(&l2tp_net_ops);
1682         if (rc)
1683                 goto out;
1684
1685         printk(KERN_INFO "L2TP core driver, %s\n", L2TP_DRV_VERSION);
1686
1687 out:
1688         return rc;
1689 }
1690
1691 static void __exit l2tp_exit(void)
1692 {
1693         unregister_pernet_device(&l2tp_net_ops);
1694 }
1695
1696 module_init(l2tp_init);
1697 module_exit(l2tp_exit);
1698
1699 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1700 MODULE_DESCRIPTION("L2TP core");
1701 MODULE_LICENSE("GPL");
1702 MODULE_VERSION(L2TP_DRV_VERSION);
1703