Merge tag 'chrome-platform-4.16-rc4-fixes' of git://git.kernel.org/pub/scm/linux...
[linux-2.6-block.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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
22
23 #include <linux/module.h>
24 #include <linux/string.h>
25 #include <linux/list.h>
26 #include <linux/rculist.h>
27 #include <linux/uaccess.h>
28
29 #include <linux/kernel.h>
30 #include <linux/spinlock.h>
31 #include <linux/kthread.h>
32 #include <linux/sched.h>
33 #include <linux/slab.h>
34 #include <linux/errno.h>
35 #include <linux/jiffies.h>
36
37 #include <linux/netdevice.h>
38 #include <linux/net.h>
39 #include <linux/inetdevice.h>
40 #include <linux/skbuff.h>
41 #include <linux/init.h>
42 #include <linux/in.h>
43 #include <linux/ip.h>
44 #include <linux/udp.h>
45 #include <linux/l2tp.h>
46 #include <linux/hash.h>
47 #include <linux/sort.h>
48 #include <linux/file.h>
49 #include <linux/nsproxy.h>
50 #include <net/net_namespace.h>
51 #include <net/netns/generic.h>
52 #include <net/dst.h>
53 #include <net/ip.h>
54 #include <net/udp.h>
55 #include <net/udp_tunnel.h>
56 #include <net/inet_common.h>
57 #include <net/xfrm.h>
58 #include <net/protocol.h>
59 #include <net/inet6_connection_sock.h>
60 #include <net/inet_ecn.h>
61 #include <net/ip6_route.h>
62 #include <net/ip6_checksum.h>
63
64 #include <asm/byteorder.h>
65 #include <linux/atomic.h>
66
67 #include "l2tp_core.h"
68
69 #define L2TP_DRV_VERSION        "V2.0"
70
71 /* L2TP header constants */
72 #define L2TP_HDRFLAG_T     0x8000
73 #define L2TP_HDRFLAG_L     0x4000
74 #define L2TP_HDRFLAG_S     0x0800
75 #define L2TP_HDRFLAG_O     0x0200
76 #define L2TP_HDRFLAG_P     0x0100
77
78 #define L2TP_HDR_VER_MASK  0x000F
79 #define L2TP_HDR_VER_2     0x0002
80 #define L2TP_HDR_VER_3     0x0003
81
82 /* L2TPv3 default L2-specific sublayer */
83 #define L2TP_SLFLAG_S      0x40000000
84 #define L2TP_SL_SEQ_MASK   0x00ffffff
85
86 #define L2TP_HDR_SIZE_SEQ               10
87 #define L2TP_HDR_SIZE_NOSEQ             6
88
89 /* Default trace flags */
90 #define L2TP_DEFAULT_DEBUG_FLAGS        0
91
92 /* Private data stored for received packets in the skb.
93  */
94 struct l2tp_skb_cb {
95         u32                     ns;
96         u16                     has_seq;
97         u16                     length;
98         unsigned long           expires;
99 };
100
101 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
102
103 static struct workqueue_struct *l2tp_wq;
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
115 static inline struct l2tp_tunnel *l2tp_tunnel(struct sock *sk)
116 {
117         return sk->sk_user_data;
118 }
119
120 static inline struct l2tp_net *l2tp_pernet(const struct net *net)
121 {
122         BUG_ON(!net);
123
124         return net_generic(net, l2tp_net_id);
125 }
126
127 /* Session hash global list for L2TPv3.
128  * The session_id SHOULD be random according to RFC3931, but several
129  * L2TP implementations use incrementing session_ids.  So we do a real
130  * hash on the session_id, rather than a simple bitmask.
131  */
132 static inline struct hlist_head *
133 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
134 {
135         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
136
137 }
138
139 /* Session hash list.
140  * The session_id SHOULD be random according to RFC2661, but several
141  * L2TP implementations (Cisco and Microsoft) use incrementing
142  * session_ids.  So we do a real hash on the session_id, rather than a
143  * simple bitmask.
144  */
145 static inline struct hlist_head *
146 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
147 {
148         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
149 }
150
151 void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
152 {
153         sock_put(tunnel->sock);
154         /* the tunnel is freed in the socket destructor */
155 }
156 EXPORT_SYMBOL(l2tp_tunnel_free);
157
158 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
159 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
160 {
161         const struct l2tp_net *pn = l2tp_pernet(net);
162         struct l2tp_tunnel *tunnel;
163
164         rcu_read_lock_bh();
165         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
166                 if (tunnel->tunnel_id == tunnel_id) {
167                         l2tp_tunnel_inc_refcount(tunnel);
168                         rcu_read_unlock_bh();
169
170                         return tunnel;
171                 }
172         }
173         rcu_read_unlock_bh();
174
175         return NULL;
176 }
177 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
178
179 /* Lookup a session. A new reference is held on the returned session. */
180 struct l2tp_session *l2tp_session_get(const struct net *net,
181                                       struct l2tp_tunnel *tunnel,
182                                       u32 session_id)
183 {
184         struct hlist_head *session_list;
185         struct l2tp_session *session;
186
187         if (!tunnel) {
188                 struct l2tp_net *pn = l2tp_pernet(net);
189
190                 session_list = l2tp_session_id_hash_2(pn, session_id);
191
192                 rcu_read_lock_bh();
193                 hlist_for_each_entry_rcu(session, session_list, global_hlist) {
194                         if (session->session_id == session_id) {
195                                 l2tp_session_inc_refcount(session);
196                                 rcu_read_unlock_bh();
197
198                                 return session;
199                         }
200                 }
201                 rcu_read_unlock_bh();
202
203                 return NULL;
204         }
205
206         session_list = l2tp_session_id_hash(tunnel, session_id);
207         read_lock_bh(&tunnel->hlist_lock);
208         hlist_for_each_entry(session, session_list, hlist) {
209                 if (session->session_id == session_id) {
210                         l2tp_session_inc_refcount(session);
211                         read_unlock_bh(&tunnel->hlist_lock);
212
213                         return session;
214                 }
215         }
216         read_unlock_bh(&tunnel->hlist_lock);
217
218         return NULL;
219 }
220 EXPORT_SYMBOL_GPL(l2tp_session_get);
221
222 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth)
223 {
224         int hash;
225         struct l2tp_session *session;
226         int count = 0;
227
228         read_lock_bh(&tunnel->hlist_lock);
229         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
230                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
231                         if (++count > nth) {
232                                 l2tp_session_inc_refcount(session);
233                                 read_unlock_bh(&tunnel->hlist_lock);
234                                 return session;
235                         }
236                 }
237         }
238
239         read_unlock_bh(&tunnel->hlist_lock);
240
241         return NULL;
242 }
243 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
244
245 /* Lookup a session by interface name.
246  * This is very inefficient but is only used by management interfaces.
247  */
248 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
249                                                 const char *ifname)
250 {
251         struct l2tp_net *pn = l2tp_pernet(net);
252         int hash;
253         struct l2tp_session *session;
254
255         rcu_read_lock_bh();
256         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
257                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
258                         if (!strcmp(session->ifname, ifname)) {
259                                 l2tp_session_inc_refcount(session);
260                                 rcu_read_unlock_bh();
261
262                                 return session;
263                         }
264                 }
265         }
266
267         rcu_read_unlock_bh();
268
269         return NULL;
270 }
271 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
272
273 int l2tp_session_register(struct l2tp_session *session,
274                           struct l2tp_tunnel *tunnel)
275 {
276         struct l2tp_session *session_walk;
277         struct hlist_head *g_head;
278         struct hlist_head *head;
279         struct l2tp_net *pn;
280         int err;
281
282         head = l2tp_session_id_hash(tunnel, session->session_id);
283
284         write_lock_bh(&tunnel->hlist_lock);
285         if (!tunnel->acpt_newsess) {
286                 err = -ENODEV;
287                 goto err_tlock;
288         }
289
290         hlist_for_each_entry(session_walk, head, hlist)
291                 if (session_walk->session_id == session->session_id) {
292                         err = -EEXIST;
293                         goto err_tlock;
294                 }
295
296         if (tunnel->version == L2TP_HDR_VER_3) {
297                 pn = l2tp_pernet(tunnel->l2tp_net);
298                 g_head = l2tp_session_id_hash_2(l2tp_pernet(tunnel->l2tp_net),
299                                                 session->session_id);
300
301                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
302
303                 hlist_for_each_entry(session_walk, g_head, global_hlist)
304                         if (session_walk->session_id == session->session_id) {
305                                 err = -EEXIST;
306                                 goto err_tlock_pnlock;
307                         }
308
309                 l2tp_tunnel_inc_refcount(tunnel);
310                 hlist_add_head_rcu(&session->global_hlist, g_head);
311
312                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
313         } else {
314                 l2tp_tunnel_inc_refcount(tunnel);
315         }
316
317         hlist_add_head(&session->hlist, head);
318         write_unlock_bh(&tunnel->hlist_lock);
319
320         return 0;
321
322 err_tlock_pnlock:
323         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
324 err_tlock:
325         write_unlock_bh(&tunnel->hlist_lock);
326
327         return err;
328 }
329 EXPORT_SYMBOL_GPL(l2tp_session_register);
330
331 /* Lookup a tunnel by id
332  */
333 struct l2tp_tunnel *l2tp_tunnel_find(const struct net *net, u32 tunnel_id)
334 {
335         struct l2tp_tunnel *tunnel;
336         struct l2tp_net *pn = l2tp_pernet(net);
337
338         rcu_read_lock_bh();
339         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
340                 if (tunnel->tunnel_id == tunnel_id) {
341                         rcu_read_unlock_bh();
342                         return tunnel;
343                 }
344         }
345         rcu_read_unlock_bh();
346
347         return NULL;
348 }
349 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
350
351 struct l2tp_tunnel *l2tp_tunnel_find_nth(const struct net *net, int nth)
352 {
353         struct l2tp_net *pn = l2tp_pernet(net);
354         struct l2tp_tunnel *tunnel;
355         int count = 0;
356
357         rcu_read_lock_bh();
358         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
359                 if (++count > nth) {
360                         rcu_read_unlock_bh();
361                         return tunnel;
362                 }
363         }
364
365         rcu_read_unlock_bh();
366
367         return NULL;
368 }
369 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
370
371 /*****************************************************************************
372  * Receive data handling
373  *****************************************************************************/
374
375 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
376  * number.
377  */
378 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
379 {
380         struct sk_buff *skbp;
381         struct sk_buff *tmp;
382         u32 ns = L2TP_SKB_CB(skb)->ns;
383
384         spin_lock_bh(&session->reorder_q.lock);
385         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
386                 if (L2TP_SKB_CB(skbp)->ns > ns) {
387                         __skb_queue_before(&session->reorder_q, skbp, skb);
388                         l2tp_dbg(session, L2TP_MSG_SEQ,
389                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
390                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
391                                  skb_queue_len(&session->reorder_q));
392                         atomic_long_inc(&session->stats.rx_oos_packets);
393                         goto out;
394                 }
395         }
396
397         __skb_queue_tail(&session->reorder_q, skb);
398
399 out:
400         spin_unlock_bh(&session->reorder_q.lock);
401 }
402
403 /* Dequeue a single skb.
404  */
405 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
406 {
407         struct l2tp_tunnel *tunnel = session->tunnel;
408         int length = L2TP_SKB_CB(skb)->length;
409
410         /* We're about to requeue the skb, so return resources
411          * to its current owner (a socket receive buffer).
412          */
413         skb_orphan(skb);
414
415         atomic_long_inc(&tunnel->stats.rx_packets);
416         atomic_long_add(length, &tunnel->stats.rx_bytes);
417         atomic_long_inc(&session->stats.rx_packets);
418         atomic_long_add(length, &session->stats.rx_bytes);
419
420         if (L2TP_SKB_CB(skb)->has_seq) {
421                 /* Bump our Nr */
422                 session->nr++;
423                 session->nr &= session->nr_max;
424
425                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
426                          session->name, session->nr);
427         }
428
429         /* call private receive handler */
430         if (session->recv_skb != NULL)
431                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
432         else
433                 kfree_skb(skb);
434 }
435
436 /* Dequeue skbs from the session's reorder_q, subject to packet order.
437  * Skbs that have been in the queue for too long are simply discarded.
438  */
439 static void l2tp_recv_dequeue(struct l2tp_session *session)
440 {
441         struct sk_buff *skb;
442         struct sk_buff *tmp;
443
444         /* If the pkt at the head of the queue has the nr that we
445          * expect to send up next, dequeue it and any other
446          * in-sequence packets behind it.
447          */
448 start:
449         spin_lock_bh(&session->reorder_q.lock);
450         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
451                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
452                         atomic_long_inc(&session->stats.rx_seq_discards);
453                         atomic_long_inc(&session->stats.rx_errors);
454                         l2tp_dbg(session, L2TP_MSG_SEQ,
455                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
456                                  session->name, L2TP_SKB_CB(skb)->ns,
457                                  L2TP_SKB_CB(skb)->length, session->nr,
458                                  skb_queue_len(&session->reorder_q));
459                         session->reorder_skip = 1;
460                         __skb_unlink(skb, &session->reorder_q);
461                         kfree_skb(skb);
462                         continue;
463                 }
464
465                 if (L2TP_SKB_CB(skb)->has_seq) {
466                         if (session->reorder_skip) {
467                                 l2tp_dbg(session, L2TP_MSG_SEQ,
468                                          "%s: advancing nr to next pkt: %u -> %u",
469                                          session->name, session->nr,
470                                          L2TP_SKB_CB(skb)->ns);
471                                 session->reorder_skip = 0;
472                                 session->nr = L2TP_SKB_CB(skb)->ns;
473                         }
474                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
475                                 l2tp_dbg(session, L2TP_MSG_SEQ,
476                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
477                                          session->name, L2TP_SKB_CB(skb)->ns,
478                                          L2TP_SKB_CB(skb)->length, session->nr,
479                                          skb_queue_len(&session->reorder_q));
480                                 goto out;
481                         }
482                 }
483                 __skb_unlink(skb, &session->reorder_q);
484
485                 /* Process the skb. We release the queue lock while we
486                  * do so to let other contexts process the queue.
487                  */
488                 spin_unlock_bh(&session->reorder_q.lock);
489                 l2tp_recv_dequeue_skb(session, skb);
490                 goto start;
491         }
492
493 out:
494         spin_unlock_bh(&session->reorder_q.lock);
495 }
496
497 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
498 {
499         u32 nws;
500
501         if (nr >= session->nr)
502                 nws = nr - session->nr;
503         else
504                 nws = (session->nr_max + 1) - (session->nr - nr);
505
506         return nws < session->nr_window_size;
507 }
508
509 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
510  * acceptable, else non-zero.
511  */
512 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
513 {
514         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
515                 /* Packet sequence number is outside allowed window.
516                  * Discard it.
517                  */
518                 l2tp_dbg(session, L2TP_MSG_SEQ,
519                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
520                          session->name, L2TP_SKB_CB(skb)->ns,
521                          L2TP_SKB_CB(skb)->length, session->nr);
522                 goto discard;
523         }
524
525         if (session->reorder_timeout != 0) {
526                 /* Packet reordering enabled. Add skb to session's
527                  * reorder queue, in order of ns.
528                  */
529                 l2tp_recv_queue_skb(session, skb);
530                 goto out;
531         }
532
533         /* Packet reordering disabled. Discard out-of-sequence packets, while
534          * tracking the number if in-sequence packets after the first OOS packet
535          * is seen. After nr_oos_count_max in-sequence packets, reset the
536          * sequence number to re-enable packet reception.
537          */
538         if (L2TP_SKB_CB(skb)->ns == session->nr) {
539                 skb_queue_tail(&session->reorder_q, skb);
540         } else {
541                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
542                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
543
544                 if (nr_oos == nr_next)
545                         session->nr_oos_count++;
546                 else
547                         session->nr_oos_count = 0;
548
549                 session->nr_oos = nr_oos;
550                 if (session->nr_oos_count > session->nr_oos_count_max) {
551                         session->reorder_skip = 1;
552                         l2tp_dbg(session, L2TP_MSG_SEQ,
553                                  "%s: %d oos packets received. Resetting sequence numbers\n",
554                                  session->name, session->nr_oos_count);
555                 }
556                 if (!session->reorder_skip) {
557                         atomic_long_inc(&session->stats.rx_seq_discards);
558                         l2tp_dbg(session, L2TP_MSG_SEQ,
559                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
560                                  session->name, L2TP_SKB_CB(skb)->ns,
561                                  L2TP_SKB_CB(skb)->length, session->nr,
562                                  skb_queue_len(&session->reorder_q));
563                         goto discard;
564                 }
565                 skb_queue_tail(&session->reorder_q, skb);
566         }
567
568 out:
569         return 0;
570
571 discard:
572         return 1;
573 }
574
575 /* Do receive processing of L2TP data frames. We handle both L2TPv2
576  * and L2TPv3 data frames here.
577  *
578  * L2TPv2 Data Message Header
579  *
580  *  0                   1                   2                   3
581  *  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
582  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
583  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
584  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
585  * |           Tunnel ID           |           Session ID          |
586  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
587  * |             Ns (opt)          |             Nr (opt)          |
588  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
589  * |      Offset Size (opt)        |    Offset pad... (opt)
590  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
591  *
592  * Data frames are marked by T=0. All other fields are the same as
593  * those in L2TP control frames.
594  *
595  * L2TPv3 Data Message Header
596  *
597  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
598  * |                      L2TP Session Header                      |
599  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
600  * |                      L2-Specific Sublayer                     |
601  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
602  * |                        Tunnel Payload                      ...
603  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
604  *
605  * L2TPv3 Session Header Over IP
606  *
607  *  0                   1                   2                   3
608  *  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
609  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
610  * |                           Session ID                          |
611  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
612  * |               Cookie (optional, maximum 64 bits)...
613  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
614  *                                                                 |
615  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
616  *
617  * L2TPv3 L2-Specific Sublayer Format
618  *
619  *  0                   1                   2                   3
620  *  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
621  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
622  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
623  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
624  *
625  * Cookie value and sublayer format are negotiated with the peer when
626  * the session is set up. Unlike L2TPv2, we do not need to parse the
627  * packet header to determine if optional fields are present.
628  *
629  * Caller must already have parsed the frame and determined that it is
630  * a data (not control) frame before coming here. Fields up to the
631  * session-id have already been parsed and ptr points to the data
632  * after the session-id.
633  */
634 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
635                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
636                       int length, int (*payload_hook)(struct sk_buff *skb))
637 {
638         struct l2tp_tunnel *tunnel = session->tunnel;
639         int offset;
640         u32 ns, nr;
641
642         /* Parse and check optional cookie */
643         if (session->peer_cookie_len > 0) {
644                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
645                         l2tp_info(tunnel, L2TP_MSG_DATA,
646                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
647                                   tunnel->name, tunnel->tunnel_id,
648                                   session->session_id);
649                         atomic_long_inc(&session->stats.rx_cookie_discards);
650                         goto discard;
651                 }
652                 ptr += session->peer_cookie_len;
653         }
654
655         /* Handle the optional sequence numbers. Sequence numbers are
656          * in different places for L2TPv2 and L2TPv3.
657          *
658          * If we are the LAC, enable/disable sequence numbers under
659          * the control of the LNS.  If no sequence numbers present but
660          * we were expecting them, discard frame.
661          */
662         ns = nr = 0;
663         L2TP_SKB_CB(skb)->has_seq = 0;
664         if (tunnel->version == L2TP_HDR_VER_2) {
665                 if (hdrflags & L2TP_HDRFLAG_S) {
666                         ns = ntohs(*(__be16 *) ptr);
667                         ptr += 2;
668                         nr = ntohs(*(__be16 *) ptr);
669                         ptr += 2;
670
671                         /* Store L2TP info in the skb */
672                         L2TP_SKB_CB(skb)->ns = ns;
673                         L2TP_SKB_CB(skb)->has_seq = 1;
674
675                         l2tp_dbg(session, L2TP_MSG_SEQ,
676                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
677                                  session->name, ns, nr, session->nr);
678                 }
679         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
680                 u32 l2h = ntohl(*(__be32 *) ptr);
681
682                 if (l2h & 0x40000000) {
683                         ns = l2h & 0x00ffffff;
684
685                         /* Store L2TP info in the skb */
686                         L2TP_SKB_CB(skb)->ns = ns;
687                         L2TP_SKB_CB(skb)->has_seq = 1;
688
689                         l2tp_dbg(session, L2TP_MSG_SEQ,
690                                  "%s: recv data ns=%u, session nr=%u\n",
691                                  session->name, ns, session->nr);
692                 }
693                 ptr += 4;
694         }
695
696         if (L2TP_SKB_CB(skb)->has_seq) {
697                 /* Received a packet with sequence numbers. If we're the LNS,
698                  * check if we sre sending sequence numbers and if not,
699                  * configure it so.
700                  */
701                 if ((!session->lns_mode) && (!session->send_seq)) {
702                         l2tp_info(session, L2TP_MSG_SEQ,
703                                   "%s: requested to enable seq numbers by LNS\n",
704                                   session->name);
705                         session->send_seq = 1;
706                         l2tp_session_set_header_len(session, tunnel->version);
707                 }
708         } else {
709                 /* No sequence numbers.
710                  * If user has configured mandatory sequence numbers, discard.
711                  */
712                 if (session->recv_seq) {
713                         l2tp_warn(session, L2TP_MSG_SEQ,
714                                   "%s: recv data has no seq numbers when required. Discarding.\n",
715                                   session->name);
716                         atomic_long_inc(&session->stats.rx_seq_discards);
717                         goto discard;
718                 }
719
720                 /* If we're the LAC and we're sending sequence numbers, the
721                  * LNS has requested that we no longer send sequence numbers.
722                  * If we're the LNS and we're sending sequence numbers, the
723                  * LAC is broken. Discard the frame.
724                  */
725                 if ((!session->lns_mode) && (session->send_seq)) {
726                         l2tp_info(session, L2TP_MSG_SEQ,
727                                   "%s: requested to disable seq numbers by LNS\n",
728                                   session->name);
729                         session->send_seq = 0;
730                         l2tp_session_set_header_len(session, tunnel->version);
731                 } else if (session->send_seq) {
732                         l2tp_warn(session, L2TP_MSG_SEQ,
733                                   "%s: recv data has no seq numbers when required. Discarding.\n",
734                                   session->name);
735                         atomic_long_inc(&session->stats.rx_seq_discards);
736                         goto discard;
737                 }
738         }
739
740         /* Session data offset is defined only for L2TPv2 and is
741          * indicated by an optional 16-bit value in the header.
742          */
743         if (tunnel->version == L2TP_HDR_VER_2) {
744                 /* If offset bit set, skip it. */
745                 if (hdrflags & L2TP_HDRFLAG_O) {
746                         offset = ntohs(*(__be16 *)ptr);
747                         ptr += 2 + offset;
748                 }
749         }
750
751         offset = ptr - optr;
752         if (!pskb_may_pull(skb, offset))
753                 goto discard;
754
755         __skb_pull(skb, offset);
756
757         /* If caller wants to process the payload before we queue the
758          * packet, do so now.
759          */
760         if (payload_hook)
761                 if ((*payload_hook)(skb))
762                         goto discard;
763
764         /* Prepare skb for adding to the session's reorder_q.  Hold
765          * packets for max reorder_timeout or 1 second if not
766          * reordering.
767          */
768         L2TP_SKB_CB(skb)->length = length;
769         L2TP_SKB_CB(skb)->expires = jiffies +
770                 (session->reorder_timeout ? session->reorder_timeout : HZ);
771
772         /* Add packet to the session's receive queue. Reordering is done here, if
773          * enabled. Saved L2TP protocol info is stored in skb->sb[].
774          */
775         if (L2TP_SKB_CB(skb)->has_seq) {
776                 if (l2tp_recv_data_seq(session, skb))
777                         goto discard;
778         } else {
779                 /* No sequence numbers. Add the skb to the tail of the
780                  * reorder queue. This ensures that it will be
781                  * delivered after all previous sequenced skbs.
782                  */
783                 skb_queue_tail(&session->reorder_q, skb);
784         }
785
786         /* Try to dequeue as many skbs from reorder_q as we can. */
787         l2tp_recv_dequeue(session);
788
789         return;
790
791 discard:
792         atomic_long_inc(&session->stats.rx_errors);
793         kfree_skb(skb);
794 }
795 EXPORT_SYMBOL(l2tp_recv_common);
796
797 /* Drop skbs from the session's reorder_q
798  */
799 int l2tp_session_queue_purge(struct l2tp_session *session)
800 {
801         struct sk_buff *skb = NULL;
802         BUG_ON(!session);
803         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
804         while ((skb = skb_dequeue(&session->reorder_q))) {
805                 atomic_long_inc(&session->stats.rx_errors);
806                 kfree_skb(skb);
807         }
808         return 0;
809 }
810 EXPORT_SYMBOL_GPL(l2tp_session_queue_purge);
811
812 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
813  * here. The skb is not on a list when we get here.
814  * Returns 0 if the packet was a data packet and was successfully passed on.
815  * Returns 1 if the packet was not a good data packet and could not be
816  * forwarded.  All such packets are passed up to userspace to deal with.
817  */
818 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
819                               int (*payload_hook)(struct sk_buff *skb))
820 {
821         struct l2tp_session *session = NULL;
822         unsigned char *ptr, *optr;
823         u16 hdrflags;
824         u32 tunnel_id, session_id;
825         u16 version;
826         int length;
827
828         /* UDP has verifed checksum */
829
830         /* UDP always verifies the packet length. */
831         __skb_pull(skb, sizeof(struct udphdr));
832
833         /* Short packet? */
834         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
835                 l2tp_info(tunnel, L2TP_MSG_DATA,
836                           "%s: recv short packet (len=%d)\n",
837                           tunnel->name, skb->len);
838                 goto error;
839         }
840
841         /* Trace packet contents, if enabled */
842         if (tunnel->debug & L2TP_MSG_DATA) {
843                 length = min(32u, skb->len);
844                 if (!pskb_may_pull(skb, length))
845                         goto error;
846
847                 pr_debug("%s: recv\n", tunnel->name);
848                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
849         }
850
851         /* Point to L2TP header */
852         optr = ptr = skb->data;
853
854         /* Get L2TP header flags */
855         hdrflags = ntohs(*(__be16 *) ptr);
856
857         /* Check protocol version */
858         version = hdrflags & L2TP_HDR_VER_MASK;
859         if (version != tunnel->version) {
860                 l2tp_info(tunnel, L2TP_MSG_DATA,
861                           "%s: recv protocol version mismatch: got %d expected %d\n",
862                           tunnel->name, version, tunnel->version);
863                 goto error;
864         }
865
866         /* Get length of L2TP packet */
867         length = skb->len;
868
869         /* If type is control packet, it is handled by userspace. */
870         if (hdrflags & L2TP_HDRFLAG_T) {
871                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
872                          "%s: recv control packet, len=%d\n",
873                          tunnel->name, length);
874                 goto error;
875         }
876
877         /* Skip flags */
878         ptr += 2;
879
880         if (tunnel->version == L2TP_HDR_VER_2) {
881                 /* If length is present, skip it */
882                 if (hdrflags & L2TP_HDRFLAG_L)
883                         ptr += 2;
884
885                 /* Extract tunnel and session ID */
886                 tunnel_id = ntohs(*(__be16 *) ptr);
887                 ptr += 2;
888                 session_id = ntohs(*(__be16 *) ptr);
889                 ptr += 2;
890         } else {
891                 ptr += 2;       /* skip reserved bits */
892                 tunnel_id = tunnel->tunnel_id;
893                 session_id = ntohl(*(__be32 *) ptr);
894                 ptr += 4;
895         }
896
897         /* Find the session context */
898         session = l2tp_session_get(tunnel->l2tp_net, tunnel, session_id);
899         if (!session || !session->recv_skb) {
900                 if (session)
901                         l2tp_session_dec_refcount(session);
902
903                 /* Not found? Pass to userspace to deal with */
904                 l2tp_info(tunnel, L2TP_MSG_DATA,
905                           "%s: no session found (%u/%u). Passing up.\n",
906                           tunnel->name, tunnel_id, session_id);
907                 goto error;
908         }
909
910         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
911         l2tp_session_dec_refcount(session);
912
913         return 0;
914
915 error:
916         /* Put UDP header back */
917         __skb_push(skb, sizeof(struct udphdr));
918
919         return 1;
920 }
921
922 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
923  * Return codes:
924  * 0 : success.
925  * <0: error
926  * >0: skb should be passed up to userspace as UDP.
927  */
928 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
929 {
930         struct l2tp_tunnel *tunnel;
931
932         tunnel = l2tp_tunnel(sk);
933         if (tunnel == NULL)
934                 goto pass_up;
935
936         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
937                  tunnel->name, skb->len);
938
939         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
940                 goto pass_up;
941
942         return 0;
943
944 pass_up:
945         return 1;
946 }
947 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
948
949 /************************************************************************
950  * Transmit handling
951  ***********************************************************************/
952
953 /* Build an L2TP header for the session into the buffer provided.
954  */
955 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
956 {
957         struct l2tp_tunnel *tunnel = session->tunnel;
958         __be16 *bufp = buf;
959         __be16 *optr = buf;
960         u16 flags = L2TP_HDR_VER_2;
961         u32 tunnel_id = tunnel->peer_tunnel_id;
962         u32 session_id = session->peer_session_id;
963
964         if (session->send_seq)
965                 flags |= L2TP_HDRFLAG_S;
966
967         /* Setup L2TP header. */
968         *bufp++ = htons(flags);
969         *bufp++ = htons(tunnel_id);
970         *bufp++ = htons(session_id);
971         if (session->send_seq) {
972                 *bufp++ = htons(session->ns);
973                 *bufp++ = 0;
974                 session->ns++;
975                 session->ns &= 0xffff;
976                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
977                          session->name, session->ns);
978         }
979
980         return bufp - optr;
981 }
982
983 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
984 {
985         struct l2tp_tunnel *tunnel = session->tunnel;
986         char *bufp = buf;
987         char *optr = bufp;
988
989         /* Setup L2TP header. The header differs slightly for UDP and
990          * IP encapsulations. For UDP, there is 4 bytes of flags.
991          */
992         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
993                 u16 flags = L2TP_HDR_VER_3;
994                 *((__be16 *) bufp) = htons(flags);
995                 bufp += 2;
996                 *((__be16 *) bufp) = 0;
997                 bufp += 2;
998         }
999
1000         *((__be32 *) bufp) = htonl(session->peer_session_id);
1001         bufp += 4;
1002         if (session->cookie_len) {
1003                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1004                 bufp += session->cookie_len;
1005         }
1006         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1007                 u32 l2h = 0;
1008
1009                 if (session->send_seq) {
1010                         l2h = 0x40000000 | session->ns;
1011                         session->ns++;
1012                         session->ns &= 0xffffff;
1013                         l2tp_dbg(session, L2TP_MSG_SEQ,
1014                                  "%s: updated ns to %u\n",
1015                                  session->name, session->ns);
1016                 }
1017
1018                 *((__be32 *)bufp) = htonl(l2h);
1019                 bufp += 4;
1020         }
1021
1022         return bufp - optr;
1023 }
1024
1025 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1026                           struct flowi *fl, size_t data_len)
1027 {
1028         struct l2tp_tunnel *tunnel = session->tunnel;
1029         unsigned int len = skb->len;
1030         int error;
1031
1032         /* Debug */
1033         if (session->send_seq)
1034                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes, ns=%u\n",
1035                          session->name, data_len, session->ns - 1);
1036         else
1037                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes\n",
1038                          session->name, data_len);
1039
1040         if (session->debug & L2TP_MSG_DATA) {
1041                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1042                 unsigned char *datap = skb->data + uhlen;
1043
1044                 pr_debug("%s: xmit\n", session->name);
1045                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1046                                      datap, min_t(size_t, 32, len - uhlen));
1047         }
1048
1049         /* Queue the packet to IP for output */
1050         skb->ignore_df = 1;
1051 #if IS_ENABLED(CONFIG_IPV6)
1052         if (tunnel->sock->sk_family == PF_INET6 && !tunnel->v4mapped)
1053                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1054         else
1055 #endif
1056                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1057
1058         /* Update stats */
1059         if (error >= 0) {
1060                 atomic_long_inc(&tunnel->stats.tx_packets);
1061                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1062                 atomic_long_inc(&session->stats.tx_packets);
1063                 atomic_long_add(len, &session->stats.tx_bytes);
1064         } else {
1065                 atomic_long_inc(&tunnel->stats.tx_errors);
1066                 atomic_long_inc(&session->stats.tx_errors);
1067         }
1068
1069         return 0;
1070 }
1071
1072 /* If caller requires the skb to have a ppp header, the header must be
1073  * inserted in the skb data before calling this function.
1074  */
1075 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1076 {
1077         int data_len = skb->len;
1078         struct l2tp_tunnel *tunnel = session->tunnel;
1079         struct sock *sk = tunnel->sock;
1080         struct flowi *fl;
1081         struct udphdr *uh;
1082         struct inet_sock *inet;
1083         int headroom;
1084         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1085         int udp_len;
1086         int ret = NET_XMIT_SUCCESS;
1087
1088         /* Check that there's enough headroom in the skb to insert IP,
1089          * UDP and L2TP headers. If not enough, expand it to
1090          * make room. Adjust truesize.
1091          */
1092         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1093                 uhlen + hdr_len;
1094         if (skb_cow_head(skb, headroom)) {
1095                 kfree_skb(skb);
1096                 return NET_XMIT_DROP;
1097         }
1098
1099         /* Setup L2TP header */
1100         session->build_header(session, __skb_push(skb, hdr_len));
1101
1102         /* Reset skb netfilter state */
1103         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1104         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1105                               IPSKB_REROUTED);
1106         nf_reset(skb);
1107
1108         bh_lock_sock(sk);
1109         if (sock_owned_by_user(sk)) {
1110                 kfree_skb(skb);
1111                 ret = NET_XMIT_DROP;
1112                 goto out_unlock;
1113         }
1114
1115         /* Get routing info from the tunnel socket */
1116         skb_dst_drop(skb);
1117         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1118
1119         inet = inet_sk(sk);
1120         fl = &inet->cork.fl;
1121         switch (tunnel->encap) {
1122         case L2TP_ENCAPTYPE_UDP:
1123                 /* Setup UDP header */
1124                 __skb_push(skb, sizeof(*uh));
1125                 skb_reset_transport_header(skb);
1126                 uh = udp_hdr(skb);
1127                 uh->source = inet->inet_sport;
1128                 uh->dest = inet->inet_dport;
1129                 udp_len = uhlen + hdr_len + data_len;
1130                 uh->len = htons(udp_len);
1131
1132                 /* Calculate UDP checksum if configured to do so */
1133 #if IS_ENABLED(CONFIG_IPV6)
1134                 if (sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1135                         udp6_set_csum(udp_get_no_check6_tx(sk),
1136                                       skb, &inet6_sk(sk)->saddr,
1137                                       &sk->sk_v6_daddr, udp_len);
1138                 else
1139 #endif
1140                 udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1141                              inet->inet_daddr, udp_len);
1142                 break;
1143
1144         case L2TP_ENCAPTYPE_IP:
1145                 break;
1146         }
1147
1148         l2tp_xmit_core(session, skb, fl, data_len);
1149 out_unlock:
1150         bh_unlock_sock(sk);
1151
1152         return ret;
1153 }
1154 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1155
1156 /*****************************************************************************
1157  * Tinnel and session create/destroy.
1158  *****************************************************************************/
1159
1160 /* Tunnel socket destruct hook.
1161  * The tunnel context is deleted only when all session sockets have been
1162  * closed.
1163  */
1164 static void l2tp_tunnel_destruct(struct sock *sk)
1165 {
1166         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1167
1168         if (tunnel == NULL)
1169                 goto end;
1170
1171         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1172
1173         /* Disable udp encapsulation */
1174         switch (tunnel->encap) {
1175         case L2TP_ENCAPTYPE_UDP:
1176                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1177                 (udp_sk(sk))->encap_type = 0;
1178                 (udp_sk(sk))->encap_rcv = NULL;
1179                 (udp_sk(sk))->encap_destroy = NULL;
1180                 break;
1181         case L2TP_ENCAPTYPE_IP:
1182                 break;
1183         }
1184
1185         /* Remove hooks into tunnel socket */
1186         sk->sk_destruct = tunnel->old_sk_destruct;
1187         sk->sk_user_data = NULL;
1188
1189         /* Call the original destructor */
1190         if (sk->sk_destruct)
1191                 (*sk->sk_destruct)(sk);
1192
1193         kfree_rcu(tunnel, rcu);
1194 end:
1195         return;
1196 }
1197
1198 /* When the tunnel is closed, all the attached sessions need to go too.
1199  */
1200 void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1201 {
1202         int hash;
1203         struct hlist_node *walk;
1204         struct hlist_node *tmp;
1205         struct l2tp_session *session;
1206
1207         BUG_ON(tunnel == NULL);
1208
1209         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1210                   tunnel->name);
1211
1212         write_lock_bh(&tunnel->hlist_lock);
1213         tunnel->acpt_newsess = false;
1214         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1215 again:
1216                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1217                         session = hlist_entry(walk, struct l2tp_session, hlist);
1218
1219                         l2tp_info(session, L2TP_MSG_CONTROL,
1220                                   "%s: closing session\n", session->name);
1221
1222                         hlist_del_init(&session->hlist);
1223
1224                         if (test_and_set_bit(0, &session->dead))
1225                                 goto again;
1226
1227                         write_unlock_bh(&tunnel->hlist_lock);
1228
1229                         __l2tp_session_unhash(session);
1230                         l2tp_session_queue_purge(session);
1231
1232                         if (session->session_close != NULL)
1233                                 (*session->session_close)(session);
1234
1235                         l2tp_session_dec_refcount(session);
1236
1237                         write_lock_bh(&tunnel->hlist_lock);
1238
1239                         /* Now restart from the beginning of this hash
1240                          * chain.  We always remove a session from the
1241                          * list so we are guaranteed to make forward
1242                          * progress.
1243                          */
1244                         goto again;
1245                 }
1246         }
1247         write_unlock_bh(&tunnel->hlist_lock);
1248 }
1249 EXPORT_SYMBOL_GPL(l2tp_tunnel_closeall);
1250
1251 /* Tunnel socket destroy hook for UDP encapsulation */
1252 static void l2tp_udp_encap_destroy(struct sock *sk)
1253 {
1254         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1255
1256         if (tunnel)
1257                 l2tp_tunnel_delete(tunnel);
1258 }
1259
1260 /* Workqueue tunnel deletion function */
1261 static void l2tp_tunnel_del_work(struct work_struct *work)
1262 {
1263         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1264                                                   del_work);
1265         struct sock *sk = tunnel->sock;
1266         struct socket *sock = sk->sk_socket;
1267         struct l2tp_net *pn;
1268
1269         l2tp_tunnel_closeall(tunnel);
1270
1271         /* If the tunnel socket was created within the kernel, use
1272          * the sk API to release it here.
1273          */
1274         if (tunnel->fd < 0) {
1275                 if (sock) {
1276                         kernel_sock_shutdown(sock, SHUT_RDWR);
1277                         sock_release(sock);
1278                 }
1279         }
1280
1281         /* Remove the tunnel struct from the tunnel list */
1282         pn = l2tp_pernet(tunnel->l2tp_net);
1283         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1284         list_del_rcu(&tunnel->list);
1285         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1286
1287         /* drop initial ref */
1288         l2tp_tunnel_dec_refcount(tunnel);
1289
1290         /* drop workqueue ref */
1291         l2tp_tunnel_dec_refcount(tunnel);
1292 }
1293
1294 /* Create a socket for the tunnel, if one isn't set up by
1295  * userspace. This is used for static tunnels where there is no
1296  * managing L2TP daemon.
1297  *
1298  * Since we don't want these sockets to keep a namespace alive by
1299  * themselves, we drop the socket's namespace refcount after creation.
1300  * These sockets are freed when the namespace exits using the pernet
1301  * exit hook.
1302  */
1303 static int l2tp_tunnel_sock_create(struct net *net,
1304                                 u32 tunnel_id,
1305                                 u32 peer_tunnel_id,
1306                                 struct l2tp_tunnel_cfg *cfg,
1307                                 struct socket **sockp)
1308 {
1309         int err = -EINVAL;
1310         struct socket *sock = NULL;
1311         struct udp_port_cfg udp_conf;
1312
1313         switch (cfg->encap) {
1314         case L2TP_ENCAPTYPE_UDP:
1315                 memset(&udp_conf, 0, sizeof(udp_conf));
1316
1317 #if IS_ENABLED(CONFIG_IPV6)
1318                 if (cfg->local_ip6 && cfg->peer_ip6) {
1319                         udp_conf.family = AF_INET6;
1320                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1321                                sizeof(udp_conf.local_ip6));
1322                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1323                                sizeof(udp_conf.peer_ip6));
1324                         udp_conf.use_udp6_tx_checksums =
1325                           ! cfg->udp6_zero_tx_checksums;
1326                         udp_conf.use_udp6_rx_checksums =
1327                           ! cfg->udp6_zero_rx_checksums;
1328                 } else
1329 #endif
1330                 {
1331                         udp_conf.family = AF_INET;
1332                         udp_conf.local_ip = cfg->local_ip;
1333                         udp_conf.peer_ip = cfg->peer_ip;
1334                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1335                 }
1336
1337                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1338                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1339
1340                 err = udp_sock_create(net, &udp_conf, &sock);
1341                 if (err < 0)
1342                         goto out;
1343
1344                 break;
1345
1346         case L2TP_ENCAPTYPE_IP:
1347 #if IS_ENABLED(CONFIG_IPV6)
1348                 if (cfg->local_ip6 && cfg->peer_ip6) {
1349                         struct sockaddr_l2tpip6 ip6_addr = {0};
1350
1351                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1352                                           IPPROTO_L2TP, &sock);
1353                         if (err < 0)
1354                                 goto out;
1355
1356                         ip6_addr.l2tp_family = AF_INET6;
1357                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1358                                sizeof(ip6_addr.l2tp_addr));
1359                         ip6_addr.l2tp_conn_id = tunnel_id;
1360                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1361                                           sizeof(ip6_addr));
1362                         if (err < 0)
1363                                 goto out;
1364
1365                         ip6_addr.l2tp_family = AF_INET6;
1366                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1367                                sizeof(ip6_addr.l2tp_addr));
1368                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1369                         err = kernel_connect(sock,
1370                                              (struct sockaddr *) &ip6_addr,
1371                                              sizeof(ip6_addr), 0);
1372                         if (err < 0)
1373                                 goto out;
1374                 } else
1375 #endif
1376                 {
1377                         struct sockaddr_l2tpip ip_addr = {0};
1378
1379                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1380                                           IPPROTO_L2TP, &sock);
1381                         if (err < 0)
1382                                 goto out;
1383
1384                         ip_addr.l2tp_family = AF_INET;
1385                         ip_addr.l2tp_addr = cfg->local_ip;
1386                         ip_addr.l2tp_conn_id = tunnel_id;
1387                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1388                                           sizeof(ip_addr));
1389                         if (err < 0)
1390                                 goto out;
1391
1392                         ip_addr.l2tp_family = AF_INET;
1393                         ip_addr.l2tp_addr = cfg->peer_ip;
1394                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1395                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1396                                              sizeof(ip_addr), 0);
1397                         if (err < 0)
1398                                 goto out;
1399                 }
1400                 break;
1401
1402         default:
1403                 goto out;
1404         }
1405
1406 out:
1407         *sockp = sock;
1408         if ((err < 0) && sock) {
1409                 kernel_sock_shutdown(sock, SHUT_RDWR);
1410                 sock_release(sock);
1411                 *sockp = NULL;
1412         }
1413
1414         return err;
1415 }
1416
1417 static struct lock_class_key l2tp_socket_class;
1418
1419 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)
1420 {
1421         struct l2tp_tunnel *tunnel = NULL;
1422         int err;
1423         struct socket *sock = NULL;
1424         struct sock *sk = NULL;
1425         struct l2tp_net *pn;
1426         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1427
1428         /* Get the tunnel socket from the fd, which was opened by
1429          * the userspace L2TP daemon. If not specified, create a
1430          * kernel socket.
1431          */
1432         if (fd < 0) {
1433                 err = l2tp_tunnel_sock_create(net, tunnel_id, peer_tunnel_id,
1434                                 cfg, &sock);
1435                 if (err < 0)
1436                         goto err;
1437         } else {
1438                 sock = sockfd_lookup(fd, &err);
1439                 if (!sock) {
1440                         pr_err("tunl %u: sockfd_lookup(fd=%d) returned %d\n",
1441                                tunnel_id, fd, err);
1442                         err = -EBADF;
1443                         goto err;
1444                 }
1445
1446                 /* Reject namespace mismatches */
1447                 if (!net_eq(sock_net(sock->sk), net)) {
1448                         pr_err("tunl %u: netns mismatch\n", tunnel_id);
1449                         err = -EINVAL;
1450                         goto err;
1451                 }
1452         }
1453
1454         sk = sock->sk;
1455
1456         if (cfg != NULL)
1457                 encap = cfg->encap;
1458
1459         /* Quick sanity checks */
1460         switch (encap) {
1461         case L2TP_ENCAPTYPE_UDP:
1462                 err = -EPROTONOSUPPORT;
1463                 if (sk->sk_protocol != IPPROTO_UDP) {
1464                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1465                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1466                         goto err;
1467                 }
1468                 break;
1469         case L2TP_ENCAPTYPE_IP:
1470                 err = -EPROTONOSUPPORT;
1471                 if (sk->sk_protocol != IPPROTO_L2TP) {
1472                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1473                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1474                         goto err;
1475                 }
1476                 break;
1477         }
1478
1479         /* Check if this socket has already been prepped */
1480         tunnel = l2tp_tunnel(sk);
1481         if (tunnel != NULL) {
1482                 /* This socket has already been prepped */
1483                 err = -EBUSY;
1484                 goto err;
1485         }
1486
1487         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1488         if (tunnel == NULL) {
1489                 err = -ENOMEM;
1490                 goto err;
1491         }
1492
1493         tunnel->version = version;
1494         tunnel->tunnel_id = tunnel_id;
1495         tunnel->peer_tunnel_id = peer_tunnel_id;
1496         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1497
1498         tunnel->magic = L2TP_TUNNEL_MAGIC;
1499         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1500         rwlock_init(&tunnel->hlist_lock);
1501         tunnel->acpt_newsess = true;
1502
1503         /* The net we belong to */
1504         tunnel->l2tp_net = net;
1505         pn = l2tp_pernet(net);
1506
1507         if (cfg != NULL)
1508                 tunnel->debug = cfg->debug;
1509
1510 #if IS_ENABLED(CONFIG_IPV6)
1511         if (sk->sk_family == PF_INET6) {
1512                 struct ipv6_pinfo *np = inet6_sk(sk);
1513
1514                 if (ipv6_addr_v4mapped(&np->saddr) &&
1515                     ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
1516                         struct inet_sock *inet = inet_sk(sk);
1517
1518                         tunnel->v4mapped = true;
1519                         inet->inet_saddr = np->saddr.s6_addr32[3];
1520                         inet->inet_rcv_saddr = sk->sk_v6_rcv_saddr.s6_addr32[3];
1521                         inet->inet_daddr = sk->sk_v6_daddr.s6_addr32[3];
1522                 } else {
1523                         tunnel->v4mapped = false;
1524                 }
1525         }
1526 #endif
1527
1528         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1529         tunnel->encap = encap;
1530         if (encap == L2TP_ENCAPTYPE_UDP) {
1531                 struct udp_tunnel_sock_cfg udp_cfg = { };
1532
1533                 udp_cfg.sk_user_data = tunnel;
1534                 udp_cfg.encap_type = UDP_ENCAP_L2TPINUDP;
1535                 udp_cfg.encap_rcv = l2tp_udp_encap_recv;
1536                 udp_cfg.encap_destroy = l2tp_udp_encap_destroy;
1537
1538                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1539         } else {
1540                 sk->sk_user_data = tunnel;
1541         }
1542
1543         /* Bump the reference count. The tunnel context is deleted
1544          * only when this drops to zero. A reference is also held on
1545          * the tunnel socket to ensure that it is not released while
1546          * the tunnel is extant. Must be done before sk_destruct is
1547          * set.
1548          */
1549         refcount_set(&tunnel->ref_count, 1);
1550         sock_hold(sk);
1551         tunnel->sock = sk;
1552         tunnel->fd = fd;
1553
1554         /* Hook on the tunnel socket destructor so that we can cleanup
1555          * if the tunnel socket goes away.
1556          */
1557         tunnel->old_sk_destruct = sk->sk_destruct;
1558         sk->sk_destruct = &l2tp_tunnel_destruct;
1559         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class, "l2tp_sock");
1560
1561         sk->sk_allocation = GFP_ATOMIC;
1562
1563         /* Init delete workqueue struct */
1564         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1565
1566         /* Add tunnel to our list */
1567         INIT_LIST_HEAD(&tunnel->list);
1568         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1569         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1570         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1571
1572         err = 0;
1573 err:
1574         if (tunnelp)
1575                 *tunnelp = tunnel;
1576
1577         /* If tunnel's socket was created by the kernel, it doesn't
1578          *  have a file.
1579          */
1580         if (sock && sock->file)
1581                 sockfd_put(sock);
1582
1583         return err;
1584 }
1585 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1586
1587 /* This function is used by the netlink TUNNEL_DELETE command.
1588  */
1589 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1590 {
1591         if (!test_and_set_bit(0, &tunnel->dead)) {
1592                 l2tp_tunnel_inc_refcount(tunnel);
1593                 queue_work(l2tp_wq, &tunnel->del_work);
1594         }
1595 }
1596 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1597
1598 /* Really kill the session.
1599  */
1600 void l2tp_session_free(struct l2tp_session *session)
1601 {
1602         struct l2tp_tunnel *tunnel = session->tunnel;
1603
1604         BUG_ON(refcount_read(&session->ref_count) != 0);
1605
1606         if (tunnel) {
1607                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1608                 l2tp_tunnel_dec_refcount(tunnel);
1609         }
1610
1611         kfree(session);
1612 }
1613 EXPORT_SYMBOL_GPL(l2tp_session_free);
1614
1615 /* Remove an l2tp session from l2tp_core's hash lists.
1616  * Provides a tidyup interface for pseudowire code which can't just route all
1617  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1618  * callback.
1619  */
1620 void __l2tp_session_unhash(struct l2tp_session *session)
1621 {
1622         struct l2tp_tunnel *tunnel = session->tunnel;
1623
1624         /* Remove the session from core hashes */
1625         if (tunnel) {
1626                 /* Remove from the per-tunnel hash */
1627                 write_lock_bh(&tunnel->hlist_lock);
1628                 hlist_del_init(&session->hlist);
1629                 write_unlock_bh(&tunnel->hlist_lock);
1630
1631                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1632                 if (tunnel->version != L2TP_HDR_VER_2) {
1633                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1634                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1635                         hlist_del_init_rcu(&session->global_hlist);
1636                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1637                         synchronize_rcu();
1638                 }
1639         }
1640 }
1641 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1642
1643 /* This function is used by the netlink SESSION_DELETE command and by
1644    pseudowire modules.
1645  */
1646 int l2tp_session_delete(struct l2tp_session *session)
1647 {
1648         if (test_and_set_bit(0, &session->dead))
1649                 return 0;
1650
1651         __l2tp_session_unhash(session);
1652         l2tp_session_queue_purge(session);
1653         if (session->session_close != NULL)
1654                 (*session->session_close)(session);
1655
1656         l2tp_session_dec_refcount(session);
1657
1658         return 0;
1659 }
1660 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1661
1662 /* We come here whenever a session's send_seq, cookie_len or
1663  * l2specific_type parameters are set.
1664  */
1665 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1666 {
1667         if (version == L2TP_HDR_VER_2) {
1668                 session->hdr_len = 6;
1669                 if (session->send_seq)
1670                         session->hdr_len += 4;
1671         } else {
1672                 session->hdr_len = 4 + session->cookie_len;
1673                 session->hdr_len += l2tp_get_l2specific_len(session);
1674                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1675                         session->hdr_len += 4;
1676         }
1677
1678 }
1679 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1680
1681 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)
1682 {
1683         struct l2tp_session *session;
1684
1685         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1686         if (session != NULL) {
1687                 session->magic = L2TP_SESSION_MAGIC;
1688                 session->tunnel = tunnel;
1689
1690                 session->session_id = session_id;
1691                 session->peer_session_id = peer_session_id;
1692                 session->nr = 0;
1693                 if (tunnel->version == L2TP_HDR_VER_2)
1694                         session->nr_max = 0xffff;
1695                 else
1696                         session->nr_max = 0xffffff;
1697                 session->nr_window_size = session->nr_max / 2;
1698                 session->nr_oos_count_max = 4;
1699
1700                 /* Use NR of first received packet */
1701                 session->reorder_skip = 1;
1702
1703                 sprintf(&session->name[0], "sess %u/%u",
1704                         tunnel->tunnel_id, session->session_id);
1705
1706                 skb_queue_head_init(&session->reorder_q);
1707
1708                 INIT_HLIST_NODE(&session->hlist);
1709                 INIT_HLIST_NODE(&session->global_hlist);
1710
1711                 /* Inherit debug options from tunnel */
1712                 session->debug = tunnel->debug;
1713
1714                 if (cfg) {
1715                         session->pwtype = cfg->pw_type;
1716                         session->debug = cfg->debug;
1717                         session->mtu = cfg->mtu;
1718                         session->mru = cfg->mru;
1719                         session->send_seq = cfg->send_seq;
1720                         session->recv_seq = cfg->recv_seq;
1721                         session->lns_mode = cfg->lns_mode;
1722                         session->reorder_timeout = cfg->reorder_timeout;
1723                         session->l2specific_type = cfg->l2specific_type;
1724                         session->cookie_len = cfg->cookie_len;
1725                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1726                         session->peer_cookie_len = cfg->peer_cookie_len;
1727                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1728                 }
1729
1730                 if (tunnel->version == L2TP_HDR_VER_2)
1731                         session->build_header = l2tp_build_l2tpv2_header;
1732                 else
1733                         session->build_header = l2tp_build_l2tpv3_header;
1734
1735                 l2tp_session_set_header_len(session, tunnel->version);
1736
1737                 refcount_set(&session->ref_count, 1);
1738
1739                 return session;
1740         }
1741
1742         return ERR_PTR(-ENOMEM);
1743 }
1744 EXPORT_SYMBOL_GPL(l2tp_session_create);
1745
1746 /*****************************************************************************
1747  * Init and cleanup
1748  *****************************************************************************/
1749
1750 static __net_init int l2tp_init_net(struct net *net)
1751 {
1752         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1753         int hash;
1754
1755         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1756         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1757
1758         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1759                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1760
1761         spin_lock_init(&pn->l2tp_session_hlist_lock);
1762
1763         return 0;
1764 }
1765
1766 static __net_exit void l2tp_exit_net(struct net *net)
1767 {
1768         struct l2tp_net *pn = l2tp_pernet(net);
1769         struct l2tp_tunnel *tunnel = NULL;
1770         int hash;
1771
1772         rcu_read_lock_bh();
1773         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1774                 l2tp_tunnel_delete(tunnel);
1775         }
1776         rcu_read_unlock_bh();
1777
1778         flush_workqueue(l2tp_wq);
1779         rcu_barrier();
1780
1781         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1782                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1783 }
1784
1785 static struct pernet_operations l2tp_net_ops = {
1786         .init = l2tp_init_net,
1787         .exit = l2tp_exit_net,
1788         .id   = &l2tp_net_id,
1789         .size = sizeof(struct l2tp_net),
1790 };
1791
1792 static int __init l2tp_init(void)
1793 {
1794         int rc = 0;
1795
1796         rc = register_pernet_device(&l2tp_net_ops);
1797         if (rc)
1798                 goto out;
1799
1800         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1801         if (!l2tp_wq) {
1802                 pr_err("alloc_workqueue failed\n");
1803                 unregister_pernet_device(&l2tp_net_ops);
1804                 rc = -ENOMEM;
1805                 goto out;
1806         }
1807
1808         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1809
1810 out:
1811         return rc;
1812 }
1813
1814 static void __exit l2tp_exit(void)
1815 {
1816         unregister_pernet_device(&l2tp_net_ops);
1817         if (l2tp_wq) {
1818                 destroy_workqueue(l2tp_wq);
1819                 l2tp_wq = NULL;
1820         }
1821 }
1822
1823 module_init(l2tp_init);
1824 module_exit(l2tp_exit);
1825
1826 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1827 MODULE_DESCRIPTION("L2TP core");
1828 MODULE_LICENSE("GPL");
1829 MODULE_VERSION(L2TP_DRV_VERSION);
1830