l2tp: fix locking of 64-bit counters for smp
[linux-3.10.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 #include <net/inet6_connection_sock.h>
57 #include <net/inet_ecn.h>
58 #include <net/ip6_route.h>
59 #include <net/ip6_checksum.h>
60
61 #include <asm/byteorder.h>
62 #include <linux/atomic.h>
63
64 #include "l2tp_core.h"
65
66 #define L2TP_DRV_VERSION        "V2.0"
67
68 /* L2TP header constants */
69 #define L2TP_HDRFLAG_T     0x8000
70 #define L2TP_HDRFLAG_L     0x4000
71 #define L2TP_HDRFLAG_S     0x0800
72 #define L2TP_HDRFLAG_O     0x0200
73 #define L2TP_HDRFLAG_P     0x0100
74
75 #define L2TP_HDR_VER_MASK  0x000F
76 #define L2TP_HDR_VER_2     0x0002
77 #define L2TP_HDR_VER_3     0x0003
78
79 /* L2TPv3 default L2-specific sublayer */
80 #define L2TP_SLFLAG_S      0x40000000
81 #define L2TP_SL_SEQ_MASK   0x00ffffff
82
83 #define L2TP_HDR_SIZE_SEQ               10
84 #define L2TP_HDR_SIZE_NOSEQ             6
85
86 /* Default trace flags */
87 #define L2TP_DEFAULT_DEBUG_FLAGS        0
88
89 #define PRINTK(_mask, _type, _lvl, _fmt, args...)                       \
90         do {                                                            \
91                 if ((_mask) & (_type))                                  \
92                         printk(_lvl "L2TP: " _fmt, ##args);             \
93         } while (0)
94
95 /* Private data stored for received packets in the skb.
96  */
97 struct l2tp_skb_cb {
98         u32                     ns;
99         u16                     has_seq;
100         u16                     length;
101         unsigned long           expires;
102 };
103
104 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
105
106 static atomic_t l2tp_tunnel_count;
107 static atomic_t l2tp_session_count;
108
109 /* per-net private data for this module */
110 static unsigned int l2tp_net_id;
111 struct l2tp_net {
112         struct list_head l2tp_tunnel_list;
113         spinlock_t l2tp_tunnel_list_lock;
114         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
115         spinlock_t l2tp_session_hlist_lock;
116 };
117
118 static void l2tp_session_set_header_len(struct l2tp_session *session, int version);
119 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel);
120 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel);
121
122 static inline struct l2tp_net *l2tp_pernet(struct net *net)
123 {
124         BUG_ON(!net);
125
126         return net_generic(net, l2tp_net_id);
127 }
128
129
130 /* Tunnel reference counts. Incremented per session that is added to
131  * the tunnel.
132  */
133 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
134 {
135         atomic_inc(&tunnel->ref_count);
136 }
137
138 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
139 {
140         if (atomic_dec_and_test(&tunnel->ref_count))
141                 l2tp_tunnel_free(tunnel);
142 }
143 #ifdef L2TP_REFCNT_DEBUG
144 #define l2tp_tunnel_inc_refcount(_t) do { \
145                 printk(KERN_DEBUG "l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n", __func__, __LINE__, (_t)->name, atomic_read(&_t->ref_count)); \
146                 l2tp_tunnel_inc_refcount_1(_t);                         \
147         } while (0)
148 #define l2tp_tunnel_dec_refcount(_t) do { \
149                 printk(KERN_DEBUG "l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n", __func__, __LINE__, (_t)->name, atomic_read(&_t->ref_count)); \
150                 l2tp_tunnel_dec_refcount_1(_t);                         \
151         } while (0)
152 #else
153 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
154 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
155 #endif
156
157 /* Session hash global list for L2TPv3.
158  * The session_id SHOULD be random according to RFC3931, but several
159  * L2TP implementations use incrementing session_ids.  So we do a real
160  * hash on the session_id, rather than a simple bitmask.
161  */
162 static inline struct hlist_head *
163 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
164 {
165         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
166
167 }
168
169 /* Lookup a session by id in the global session list
170  */
171 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
172 {
173         struct l2tp_net *pn = l2tp_pernet(net);
174         struct hlist_head *session_list =
175                 l2tp_session_id_hash_2(pn, session_id);
176         struct l2tp_session *session;
177         struct hlist_node *walk;
178
179         rcu_read_lock_bh();
180         hlist_for_each_entry_rcu(session, walk, session_list, global_hlist) {
181                 if (session->session_id == session_id) {
182                         rcu_read_unlock_bh();
183                         return session;
184                 }
185         }
186         rcu_read_unlock_bh();
187
188         return NULL;
189 }
190
191 /* Session hash list.
192  * The session_id SHOULD be random according to RFC2661, but several
193  * L2TP implementations (Cisco and Microsoft) use incrementing
194  * session_ids.  So we do a real hash on the session_id, rather than a
195  * simple bitmask.
196  */
197 static inline struct hlist_head *
198 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
199 {
200         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
201 }
202
203 /* Lookup a session by id
204  */
205 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
206 {
207         struct hlist_head *session_list;
208         struct l2tp_session *session;
209         struct hlist_node *walk;
210
211         /* In L2TPv3, session_ids are unique over all tunnels and we
212          * sometimes need to look them up before we know the
213          * tunnel.
214          */
215         if (tunnel == NULL)
216                 return l2tp_session_find_2(net, session_id);
217
218         session_list = l2tp_session_id_hash(tunnel, session_id);
219         read_lock_bh(&tunnel->hlist_lock);
220         hlist_for_each_entry(session, walk, session_list, hlist) {
221                 if (session->session_id == session_id) {
222                         read_unlock_bh(&tunnel->hlist_lock);
223                         return session;
224                 }
225         }
226         read_unlock_bh(&tunnel->hlist_lock);
227
228         return NULL;
229 }
230 EXPORT_SYMBOL_GPL(l2tp_session_find);
231
232 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
233 {
234         int hash;
235         struct hlist_node *walk;
236         struct l2tp_session *session;
237         int count = 0;
238
239         read_lock_bh(&tunnel->hlist_lock);
240         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
241                 hlist_for_each_entry(session, walk, &tunnel->session_hlist[hash], hlist) {
242                         if (++count > nth) {
243                                 read_unlock_bh(&tunnel->hlist_lock);
244                                 return session;
245                         }
246                 }
247         }
248
249         read_unlock_bh(&tunnel->hlist_lock);
250
251         return NULL;
252 }
253 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
254
255 /* Lookup a session by interface name.
256  * This is very inefficient but is only used by management interfaces.
257  */
258 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
259 {
260         struct l2tp_net *pn = l2tp_pernet(net);
261         int hash;
262         struct hlist_node *walk;
263         struct l2tp_session *session;
264
265         rcu_read_lock_bh();
266         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
267                 hlist_for_each_entry_rcu(session, walk, &pn->l2tp_session_hlist[hash], global_hlist) {
268                         if (!strcmp(session->ifname, ifname)) {
269                                 rcu_read_unlock_bh();
270                                 return session;
271                         }
272                 }
273         }
274
275         rcu_read_unlock_bh();
276
277         return NULL;
278 }
279 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
280
281 /* Lookup a tunnel by id
282  */
283 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
284 {
285         struct l2tp_tunnel *tunnel;
286         struct l2tp_net *pn = l2tp_pernet(net);
287
288         rcu_read_lock_bh();
289         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
290                 if (tunnel->tunnel_id == tunnel_id) {
291                         rcu_read_unlock_bh();
292                         return tunnel;
293                 }
294         }
295         rcu_read_unlock_bh();
296
297         return NULL;
298 }
299 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
300
301 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
302 {
303         struct l2tp_net *pn = l2tp_pernet(net);
304         struct l2tp_tunnel *tunnel;
305         int count = 0;
306
307         rcu_read_lock_bh();
308         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
309                 if (++count > nth) {
310                         rcu_read_unlock_bh();
311                         return tunnel;
312                 }
313         }
314
315         rcu_read_unlock_bh();
316
317         return NULL;
318 }
319 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
320
321 /*****************************************************************************
322  * Receive data handling
323  *****************************************************************************/
324
325 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
326  * number.
327  */
328 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
329 {
330         struct sk_buff *skbp;
331         struct sk_buff *tmp;
332         u32 ns = L2TP_SKB_CB(skb)->ns;
333         struct l2tp_stats *sstats;
334
335         spin_lock_bh(&session->reorder_q.lock);
336         sstats = &session->stats;
337         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
338                 if (L2TP_SKB_CB(skbp)->ns > ns) {
339                         __skb_queue_before(&session->reorder_q, skbp, skb);
340                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
341                                "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
342                                session->name, ns, L2TP_SKB_CB(skbp)->ns,
343                                skb_queue_len(&session->reorder_q));
344                         u64_stats_update_begin(&sstats->syncp);
345                         sstats->rx_oos_packets++;
346                         u64_stats_update_end(&sstats->syncp);
347                         goto out;
348                 }
349         }
350
351         __skb_queue_tail(&session->reorder_q, skb);
352
353 out:
354         spin_unlock_bh(&session->reorder_q.lock);
355 }
356
357 /* Dequeue a single skb.
358  */
359 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
360 {
361         struct l2tp_tunnel *tunnel = session->tunnel;
362         int length = L2TP_SKB_CB(skb)->length;
363         struct l2tp_stats *tstats, *sstats;
364
365         /* We're about to requeue the skb, so return resources
366          * to its current owner (a socket receive buffer).
367          */
368         skb_orphan(skb);
369
370         tstats = &tunnel->stats;
371         u64_stats_update_begin(&tstats->syncp);
372         sstats = &session->stats;
373         u64_stats_update_begin(&sstats->syncp);
374         tstats->rx_packets++;
375         tstats->rx_bytes += length;
376         sstats->rx_packets++;
377         sstats->rx_bytes += length;
378         u64_stats_update_end(&tstats->syncp);
379         u64_stats_update_end(&sstats->syncp);
380
381         if (L2TP_SKB_CB(skb)->has_seq) {
382                 /* Bump our Nr */
383                 session->nr++;
384                 if (tunnel->version == L2TP_HDR_VER_2)
385                         session->nr &= 0xffff;
386                 else
387                         session->nr &= 0xffffff;
388
389                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
390                        "%s: updated nr to %hu\n", session->name, session->nr);
391         }
392
393         /* call private receive handler */
394         if (session->recv_skb != NULL)
395                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
396         else
397                 kfree_skb(skb);
398
399         if (session->deref)
400                 (*session->deref)(session);
401 }
402
403 /* Dequeue skbs from the session's reorder_q, subject to packet order.
404  * Skbs that have been in the queue for too long are simply discarded.
405  */
406 static void l2tp_recv_dequeue(struct l2tp_session *session)
407 {
408         struct sk_buff *skb;
409         struct sk_buff *tmp;
410         struct l2tp_stats *sstats;
411
412         /* If the pkt at the head of the queue has the nr that we
413          * expect to send up next, dequeue it and any other
414          * in-sequence packets behind it.
415          */
416 start:
417         spin_lock_bh(&session->reorder_q.lock);
418         sstats = &session->stats;
419         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
420                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
421                         u64_stats_update_begin(&sstats->syncp);
422                         sstats->rx_seq_discards++;
423                         sstats->rx_errors++;
424                         u64_stats_update_end(&sstats->syncp);
425                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
426                                "%s: oos pkt %u len %d discarded (too old), "
427                                "waiting for %u, reorder_q_len=%d\n",
428                                session->name, L2TP_SKB_CB(skb)->ns,
429                                L2TP_SKB_CB(skb)->length, session->nr,
430                                skb_queue_len(&session->reorder_q));
431                         __skb_unlink(skb, &session->reorder_q);
432                         kfree_skb(skb);
433                         if (session->deref)
434                                 (*session->deref)(session);
435                         continue;
436                 }
437
438                 if (L2TP_SKB_CB(skb)->has_seq) {
439                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
440                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
441                                        "%s: holding oos pkt %u len %d, "
442                                        "waiting for %u, reorder_q_len=%d\n",
443                                        session->name, L2TP_SKB_CB(skb)->ns,
444                                        L2TP_SKB_CB(skb)->length, session->nr,
445                                        skb_queue_len(&session->reorder_q));
446                                 goto out;
447                         }
448                 }
449                 __skb_unlink(skb, &session->reorder_q);
450
451                 /* Process the skb. We release the queue lock while we
452                  * do so to let other contexts process the queue.
453                  */
454                 spin_unlock_bh(&session->reorder_q.lock);
455                 l2tp_recv_dequeue_skb(session, skb);
456                 goto start;
457         }
458
459 out:
460         spin_unlock_bh(&session->reorder_q.lock);
461 }
462
463 static inline int l2tp_verify_udp_checksum(struct sock *sk,
464                                            struct sk_buff *skb)
465 {
466         struct udphdr *uh = udp_hdr(skb);
467         u16 ulen = ntohs(uh->len);
468         __wsum psum;
469
470         if (sk->sk_no_check || skb_csum_unnecessary(skb))
471                 return 0;
472
473 #if IS_ENABLED(CONFIG_IPV6)
474         if (sk->sk_family == PF_INET6) {
475                 if (!uh->check) {
476                         LIMIT_NETDEBUG(KERN_INFO "L2TP: IPv6: checksum is 0\n");
477                         return 1;
478                 }
479                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
480                     !csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
481                                      &ipv6_hdr(skb)->daddr, ulen,
482                                      IPPROTO_UDP, skb->csum)) {
483                         skb->ip_summed = CHECKSUM_UNNECESSARY;
484                         return 0;
485                 }
486                 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
487                                                          &ipv6_hdr(skb)->daddr,
488                                                          skb->len, IPPROTO_UDP,
489                                                          0));
490         } else
491 #endif
492         {
493                 struct inet_sock *inet;
494                 if (!uh->check)
495                         return 0;
496                 inet = inet_sk(sk);
497                 psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr,
498                                           ulen, IPPROTO_UDP, 0);
499
500                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
501                     !csum_fold(csum_add(psum, skb->csum)))
502                         return 0;
503                 skb->csum = psum;
504         }
505
506         return __skb_checksum_complete(skb);
507 }
508
509 /* Do receive processing of L2TP data frames. We handle both L2TPv2
510  * and L2TPv3 data frames here.
511  *
512  * L2TPv2 Data Message Header
513  *
514  *  0                   1                   2                   3
515  *  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
516  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
517  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
518  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
519  * |           Tunnel ID           |           Session ID          |
520  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
521  * |             Ns (opt)          |             Nr (opt)          |
522  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
523  * |      Offset Size (opt)        |    Offset pad... (opt)
524  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
525  *
526  * Data frames are marked by T=0. All other fields are the same as
527  * those in L2TP control frames.
528  *
529  * L2TPv3 Data Message Header
530  *
531  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
532  * |                      L2TP Session Header                      |
533  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
534  * |                      L2-Specific Sublayer                     |
535  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
536  * |                        Tunnel Payload                      ...
537  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
538  *
539  * L2TPv3 Session Header Over IP
540  *
541  *  0                   1                   2                   3
542  *  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
543  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
544  * |                           Session ID                          |
545  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
546  * |               Cookie (optional, maximum 64 bits)...
547  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
548  *                                                                 |
549  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
550  *
551  * L2TPv3 L2-Specific Sublayer Format
552  *
553  *  0                   1                   2                   3
554  *  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
555  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
556  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
557  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
558  *
559  * Cookie value, sublayer format and offset (pad) are negotiated with
560  * the peer when the session is set up. Unlike L2TPv2, we do not need
561  * to parse the packet header to determine if optional fields are
562  * present.
563  *
564  * Caller must already have parsed the frame and determined that it is
565  * a data (not control) frame before coming here. Fields up to the
566  * session-id have already been parsed and ptr points to the data
567  * after the session-id.
568  */
569 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
570                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
571                       int length, int (*payload_hook)(struct sk_buff *skb))
572 {
573         struct l2tp_tunnel *tunnel = session->tunnel;
574         int offset;
575         u32 ns, nr;
576         struct l2tp_stats *sstats = &session->stats;
577
578         /* The ref count is increased since we now hold a pointer to
579          * the session. Take care to decrement the refcnt when exiting
580          * this function from now on...
581          */
582         l2tp_session_inc_refcount(session);
583         if (session->ref)
584                 (*session->ref)(session);
585
586         /* Parse and check optional cookie */
587         if (session->peer_cookie_len > 0) {
588                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
589                         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
590                                "%s: cookie mismatch (%u/%u). Discarding.\n",
591                                tunnel->name, tunnel->tunnel_id, session->session_id);
592                         u64_stats_update_begin(&sstats->syncp);
593                         sstats->rx_cookie_discards++;
594                         u64_stats_update_end(&sstats->syncp);
595                         goto discard;
596                 }
597                 ptr += session->peer_cookie_len;
598         }
599
600         /* Handle the optional sequence numbers. Sequence numbers are
601          * in different places for L2TPv2 and L2TPv3.
602          *
603          * If we are the LAC, enable/disable sequence numbers under
604          * the control of the LNS.  If no sequence numbers present but
605          * we were expecting them, discard frame.
606          */
607         ns = nr = 0;
608         L2TP_SKB_CB(skb)->has_seq = 0;
609         if (tunnel->version == L2TP_HDR_VER_2) {
610                 if (hdrflags & L2TP_HDRFLAG_S) {
611                         ns = ntohs(*(__be16 *) ptr);
612                         ptr += 2;
613                         nr = ntohs(*(__be16 *) ptr);
614                         ptr += 2;
615
616                         /* Store L2TP info in the skb */
617                         L2TP_SKB_CB(skb)->ns = ns;
618                         L2TP_SKB_CB(skb)->has_seq = 1;
619
620                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
621                                "%s: recv data ns=%u, nr=%u, session nr=%u\n",
622                                session->name, ns, nr, session->nr);
623                 }
624         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
625                 u32 l2h = ntohl(*(__be32 *) ptr);
626
627                 if (l2h & 0x40000000) {
628                         ns = l2h & 0x00ffffff;
629
630                         /* Store L2TP info in the skb */
631                         L2TP_SKB_CB(skb)->ns = ns;
632                         L2TP_SKB_CB(skb)->has_seq = 1;
633
634                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
635                                "%s: recv data ns=%u, session nr=%u\n",
636                                session->name, ns, session->nr);
637                 }
638         }
639
640         /* Advance past L2-specific header, if present */
641         ptr += session->l2specific_len;
642
643         if (L2TP_SKB_CB(skb)->has_seq) {
644                 /* Received a packet with sequence numbers. If we're the LNS,
645                  * check if we sre sending sequence numbers and if not,
646                  * configure it so.
647                  */
648                 if ((!session->lns_mode) && (!session->send_seq)) {
649                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
650                                "%s: requested to enable seq numbers by LNS\n",
651                                session->name);
652                         session->send_seq = -1;
653                         l2tp_session_set_header_len(session, tunnel->version);
654                 }
655         } else {
656                 /* No sequence numbers.
657                  * If user has configured mandatory sequence numbers, discard.
658                  */
659                 if (session->recv_seq) {
660                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
661                                "%s: recv data has no seq numbers when required. "
662                                "Discarding\n", session->name);
663                         u64_stats_update_begin(&sstats->syncp);
664                         sstats->rx_seq_discards++;
665                         u64_stats_update_end(&sstats->syncp);
666                         goto discard;
667                 }
668
669                 /* If we're the LAC and we're sending sequence numbers, the
670                  * LNS has requested that we no longer send sequence numbers.
671                  * If we're the LNS and we're sending sequence numbers, the
672                  * LAC is broken. Discard the frame.
673                  */
674                 if ((!session->lns_mode) && (session->send_seq)) {
675                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
676                                "%s: requested to disable seq numbers by LNS\n",
677                                session->name);
678                         session->send_seq = 0;
679                         l2tp_session_set_header_len(session, tunnel->version);
680                 } else if (session->send_seq) {
681                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
682                                "%s: recv data has no seq numbers when required. "
683                                "Discarding\n", session->name);
684                         u64_stats_update_begin(&sstats->syncp);
685                         sstats->rx_seq_discards++;
686                         u64_stats_update_end(&sstats->syncp);
687                         goto discard;
688                 }
689         }
690
691         /* Session data offset is handled differently for L2TPv2 and
692          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
693          * the header. For L2TPv3, the offset is negotiated using AVPs
694          * in the session setup control protocol.
695          */
696         if (tunnel->version == L2TP_HDR_VER_2) {
697                 /* If offset bit set, skip it. */
698                 if (hdrflags & L2TP_HDRFLAG_O) {
699                         offset = ntohs(*(__be16 *)ptr);
700                         ptr += 2 + offset;
701                 }
702         } else
703                 ptr += session->offset;
704
705         offset = ptr - optr;
706         if (!pskb_may_pull(skb, offset))
707                 goto discard;
708
709         __skb_pull(skb, offset);
710
711         /* If caller wants to process the payload before we queue the
712          * packet, do so now.
713          */
714         if (payload_hook)
715                 if ((*payload_hook)(skb))
716                         goto discard;
717
718         /* Prepare skb for adding to the session's reorder_q.  Hold
719          * packets for max reorder_timeout or 1 second if not
720          * reordering.
721          */
722         L2TP_SKB_CB(skb)->length = length;
723         L2TP_SKB_CB(skb)->expires = jiffies +
724                 (session->reorder_timeout ? session->reorder_timeout : HZ);
725
726         /* Add packet to the session's receive queue. Reordering is done here, if
727          * enabled. Saved L2TP protocol info is stored in skb->sb[].
728          */
729         if (L2TP_SKB_CB(skb)->has_seq) {
730                 if (session->reorder_timeout != 0) {
731                         /* Packet reordering enabled. Add skb to session's
732                          * reorder queue, in order of ns.
733                          */
734                         l2tp_recv_queue_skb(session, skb);
735                 } else {
736                         /* Packet reordering disabled. Discard out-of-sequence
737                          * packets
738                          */
739                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
740                                 u64_stats_update_begin(&sstats->syncp);
741                                 sstats->rx_seq_discards++;
742                                 u64_stats_update_end(&sstats->syncp);
743                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
744                                        "%s: oos pkt %u len %d discarded, "
745                                        "waiting for %u, reorder_q_len=%d\n",
746                                        session->name, L2TP_SKB_CB(skb)->ns,
747                                        L2TP_SKB_CB(skb)->length, session->nr,
748                                        skb_queue_len(&session->reorder_q));
749                                 goto discard;
750                         }
751                         skb_queue_tail(&session->reorder_q, skb);
752                 }
753         } else {
754                 /* No sequence numbers. Add the skb to the tail of the
755                  * reorder queue. This ensures that it will be
756                  * delivered after all previous sequenced skbs.
757                  */
758                 skb_queue_tail(&session->reorder_q, skb);
759         }
760
761         /* Try to dequeue as many skbs from reorder_q as we can. */
762         l2tp_recv_dequeue(session);
763
764         l2tp_session_dec_refcount(session);
765
766         return;
767
768 discard:
769         u64_stats_update_begin(&sstats->syncp);
770         sstats->rx_errors++;
771         u64_stats_update_end(&sstats->syncp);
772         kfree_skb(skb);
773
774         if (session->deref)
775                 (*session->deref)(session);
776
777         l2tp_session_dec_refcount(session);
778 }
779 EXPORT_SYMBOL(l2tp_recv_common);
780
781 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
782  * here. The skb is not on a list when we get here.
783  * Returns 0 if the packet was a data packet and was successfully passed on.
784  * Returns 1 if the packet was not a good data packet and could not be
785  * forwarded.  All such packets are passed up to userspace to deal with.
786  */
787 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
788                               int (*payload_hook)(struct sk_buff *skb))
789 {
790         struct l2tp_session *session = NULL;
791         unsigned char *ptr, *optr;
792         u16 hdrflags;
793         u32 tunnel_id, session_id;
794         int offset;
795         u16 version;
796         int length;
797         struct l2tp_stats *tstats;
798
799         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
800                 goto discard_bad_csum;
801
802         /* UDP always verifies the packet length. */
803         __skb_pull(skb, sizeof(struct udphdr));
804
805         /* Short packet? */
806         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
807                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
808                        "%s: recv short packet (len=%d)\n", tunnel->name, skb->len);
809                 goto error;
810         }
811
812         /* Trace packet contents, if enabled */
813         if (tunnel->debug & L2TP_MSG_DATA) {
814                 length = min(32u, skb->len);
815                 if (!pskb_may_pull(skb, length))
816                         goto error;
817
818                 printk(KERN_DEBUG "%s: recv: ", tunnel->name);
819
820                 offset = 0;
821                 do {
822                         printk(" %02X", skb->data[offset]);
823                 } while (++offset < length);
824
825                 printk("\n");
826         }
827
828         /* Point to L2TP header */
829         optr = ptr = skb->data;
830
831         /* Get L2TP header flags */
832         hdrflags = ntohs(*(__be16 *) ptr);
833
834         /* Check protocol version */
835         version = hdrflags & L2TP_HDR_VER_MASK;
836         if (version != tunnel->version) {
837                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
838                        "%s: recv protocol version mismatch: got %d expected %d\n",
839                        tunnel->name, version, tunnel->version);
840                 goto error;
841         }
842
843         /* Get length of L2TP packet */
844         length = skb->len;
845
846         /* If type is control packet, it is handled by userspace. */
847         if (hdrflags & L2TP_HDRFLAG_T) {
848                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
849                        "%s: recv control packet, len=%d\n", tunnel->name, length);
850                 goto error;
851         }
852
853         /* Skip flags */
854         ptr += 2;
855
856         if (tunnel->version == L2TP_HDR_VER_2) {
857                 /* If length is present, skip it */
858                 if (hdrflags & L2TP_HDRFLAG_L)
859                         ptr += 2;
860
861                 /* Extract tunnel and session ID */
862                 tunnel_id = ntohs(*(__be16 *) ptr);
863                 ptr += 2;
864                 session_id = ntohs(*(__be16 *) ptr);
865                 ptr += 2;
866         } else {
867                 ptr += 2;       /* skip reserved bits */
868                 tunnel_id = tunnel->tunnel_id;
869                 session_id = ntohl(*(__be32 *) ptr);
870                 ptr += 4;
871         }
872
873         /* Find the session context */
874         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
875         if (!session || !session->recv_skb) {
876                 /* Not found? Pass to userspace to deal with */
877                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
878                        "%s: no session found (%u/%u). Passing up.\n",
879                        tunnel->name, tunnel_id, session_id);
880                 goto error;
881         }
882
883         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
884
885         return 0;
886
887 discard_bad_csum:
888         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
889         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
890         tstats = &tunnel->stats;
891         u64_stats_update_begin(&tstats->syncp);
892         tstats->rx_errors++;
893         u64_stats_update_end(&tstats->syncp);
894         kfree_skb(skb);
895
896         return 0;
897
898 error:
899         /* Put UDP header back */
900         __skb_push(skb, sizeof(struct udphdr));
901
902         return 1;
903 }
904
905 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
906  * Return codes:
907  * 0 : success.
908  * <0: error
909  * >0: skb should be passed up to userspace as UDP.
910  */
911 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
912 {
913         struct l2tp_tunnel *tunnel;
914
915         tunnel = l2tp_sock_to_tunnel(sk);
916         if (tunnel == NULL)
917                 goto pass_up;
918
919         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
920                "%s: received %d bytes\n", tunnel->name, skb->len);
921
922         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
923                 goto pass_up_put;
924
925         sock_put(sk);
926         return 0;
927
928 pass_up_put:
929         sock_put(sk);
930 pass_up:
931         return 1;
932 }
933 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
934
935 /************************************************************************
936  * Transmit handling
937  ***********************************************************************/
938
939 /* Build an L2TP header for the session into the buffer provided.
940  */
941 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
942 {
943         struct l2tp_tunnel *tunnel = session->tunnel;
944         __be16 *bufp = buf;
945         __be16 *optr = buf;
946         u16 flags = L2TP_HDR_VER_2;
947         u32 tunnel_id = tunnel->peer_tunnel_id;
948         u32 session_id = session->peer_session_id;
949
950         if (session->send_seq)
951                 flags |= L2TP_HDRFLAG_S;
952
953         /* Setup L2TP header. */
954         *bufp++ = htons(flags);
955         *bufp++ = htons(tunnel_id);
956         *bufp++ = htons(session_id);
957         if (session->send_seq) {
958                 *bufp++ = htons(session->ns);
959                 *bufp++ = 0;
960                 session->ns++;
961                 session->ns &= 0xffff;
962                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
963                        "%s: updated ns to %u\n", session->name, session->ns);
964         }
965
966         return bufp - optr;
967 }
968
969 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
970 {
971         struct l2tp_tunnel *tunnel = session->tunnel;
972         char *bufp = buf;
973         char *optr = bufp;
974
975         /* Setup L2TP header. The header differs slightly for UDP and
976          * IP encapsulations. For UDP, there is 4 bytes of flags.
977          */
978         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
979                 u16 flags = L2TP_HDR_VER_3;
980                 *((__be16 *) bufp) = htons(flags);
981                 bufp += 2;
982                 *((__be16 *) bufp) = 0;
983                 bufp += 2;
984         }
985
986         *((__be32 *) bufp) = htonl(session->peer_session_id);
987         bufp += 4;
988         if (session->cookie_len) {
989                 memcpy(bufp, &session->cookie[0], session->cookie_len);
990                 bufp += session->cookie_len;
991         }
992         if (session->l2specific_len) {
993                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
994                         u32 l2h = 0;
995                         if (session->send_seq) {
996                                 l2h = 0x40000000 | session->ns;
997                                 session->ns++;
998                                 session->ns &= 0xffffff;
999                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
1000                                        "%s: updated ns to %u\n", session->name, session->ns);
1001                         }
1002
1003                         *((__be32 *) bufp) = htonl(l2h);
1004                 }
1005                 bufp += session->l2specific_len;
1006         }
1007         if (session->offset)
1008                 bufp += session->offset;
1009
1010         return bufp - optr;
1011 }
1012
1013 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1014                           struct flowi *fl, size_t data_len)
1015 {
1016         struct l2tp_tunnel *tunnel = session->tunnel;
1017         unsigned int len = skb->len;
1018         int error;
1019         struct l2tp_stats *tstats, *sstats;
1020
1021         /* Debug */
1022         if (session->send_seq)
1023                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
1024                        "%s: send %Zd bytes, ns=%u\n", session->name,
1025                        data_len, session->ns - 1);
1026         else
1027                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
1028                        "%s: send %Zd bytes\n", session->name, data_len);
1029
1030         if (session->debug & L2TP_MSG_DATA) {
1031                 int i;
1032                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1033                 unsigned char *datap = skb->data + uhlen;
1034
1035                 printk(KERN_DEBUG "%s: xmit:", session->name);
1036                 for (i = 0; i < (len - uhlen); i++) {
1037                         printk(" %02X", *datap++);
1038                         if (i == 31) {
1039                                 printk(" ...");
1040                                 break;
1041                         }
1042                 }
1043                 printk("\n");
1044         }
1045
1046         /* Queue the packet to IP for output */
1047         skb->local_df = 1;
1048 #if IS_ENABLED(CONFIG_IPV6)
1049         if (skb->sk->sk_family == PF_INET6)
1050                 error = inet6_csk_xmit(skb, NULL);
1051         else
1052 #endif
1053                 error = ip_queue_xmit(skb, fl);
1054
1055         /* Update stats */
1056         tstats = &tunnel->stats;
1057         u64_stats_update_begin(&tstats->syncp);
1058         sstats = &session->stats;
1059         u64_stats_update_begin(&sstats->syncp);
1060         if (error >= 0) {
1061                 tstats->tx_packets++;
1062                 tstats->tx_bytes += len;
1063                 sstats->tx_packets++;
1064                 sstats->tx_bytes += len;
1065         } else {
1066                 tstats->tx_errors++;
1067                 sstats->tx_errors++;
1068         }
1069         u64_stats_update_end(&tstats->syncp);
1070         u64_stats_update_end(&sstats->syncp);
1071
1072         return 0;
1073 }
1074
1075 /* Automatically called when the skb is freed.
1076  */
1077 static void l2tp_sock_wfree(struct sk_buff *skb)
1078 {
1079         sock_put(skb->sk);
1080 }
1081
1082 /* For data skbs that we transmit, we associate with the tunnel socket
1083  * but don't do accounting.
1084  */
1085 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1086 {
1087         sock_hold(sk);
1088         skb->sk = sk;
1089         skb->destructor = l2tp_sock_wfree;
1090 }
1091
1092 #if IS_ENABLED(CONFIG_IPV6)
1093 static void l2tp_xmit_ipv6_csum(struct sock *sk, struct sk_buff *skb,
1094                                 int udp_len)
1095 {
1096         struct ipv6_pinfo *np = inet6_sk(sk);
1097         struct udphdr *uh = udp_hdr(skb);
1098
1099         if (!skb_dst(skb) || !skb_dst(skb)->dev ||
1100             !(skb_dst(skb)->dev->features & NETIF_F_IPV6_CSUM)) {
1101                 __wsum csum = skb_checksum(skb, 0, udp_len, 0);
1102                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1103                 uh->check = csum_ipv6_magic(&np->saddr, &np->daddr, udp_len,
1104                                             IPPROTO_UDP, csum);
1105                 if (uh->check == 0)
1106                         uh->check = CSUM_MANGLED_0;
1107         } else {
1108                 skb->ip_summed = CHECKSUM_PARTIAL;
1109                 skb->csum_start = skb_transport_header(skb) - skb->head;
1110                 skb->csum_offset = offsetof(struct udphdr, check);
1111                 uh->check = ~csum_ipv6_magic(&np->saddr, &np->daddr,
1112                                              udp_len, IPPROTO_UDP, 0);
1113         }
1114 }
1115 #endif
1116
1117 /* If caller requires the skb to have a ppp header, the header must be
1118  * inserted in the skb data before calling this function.
1119  */
1120 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1121 {
1122         int data_len = skb->len;
1123         struct l2tp_tunnel *tunnel = session->tunnel;
1124         struct sock *sk = tunnel->sock;
1125         struct flowi *fl;
1126         struct udphdr *uh;
1127         struct inet_sock *inet;
1128         __wsum csum;
1129         int old_headroom;
1130         int new_headroom;
1131         int headroom;
1132         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1133         int udp_len;
1134
1135         /* Check that there's enough headroom in the skb to insert IP,
1136          * UDP and L2TP headers. If not enough, expand it to
1137          * make room. Adjust truesize.
1138          */
1139         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1140                 uhlen + hdr_len;
1141         old_headroom = skb_headroom(skb);
1142         if (skb_cow_head(skb, headroom)) {
1143                 dev_kfree_skb(skb);
1144                 goto abort;
1145         }
1146
1147         new_headroom = skb_headroom(skb);
1148         skb_orphan(skb);
1149         skb->truesize += new_headroom - old_headroom;
1150
1151         /* Setup L2TP header */
1152         session->build_header(session, __skb_push(skb, hdr_len));
1153
1154         /* Reset skb netfilter state */
1155         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1156         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1157                               IPSKB_REROUTED);
1158         nf_reset(skb);
1159
1160         bh_lock_sock(sk);
1161         if (sock_owned_by_user(sk)) {
1162                 dev_kfree_skb(skb);
1163                 goto out_unlock;
1164         }
1165
1166         /* Get routing info from the tunnel socket */
1167         skb_dst_drop(skb);
1168         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1169
1170         inet = inet_sk(sk);
1171         fl = &inet->cork.fl;
1172         switch (tunnel->encap) {
1173         case L2TP_ENCAPTYPE_UDP:
1174                 /* Setup UDP header */
1175                 __skb_push(skb, sizeof(*uh));
1176                 skb_reset_transport_header(skb);
1177                 uh = udp_hdr(skb);
1178                 uh->source = inet->inet_sport;
1179                 uh->dest = inet->inet_dport;
1180                 udp_len = uhlen + hdr_len + data_len;
1181                 uh->len = htons(udp_len);
1182                 uh->check = 0;
1183
1184                 /* Calculate UDP checksum if configured to do so */
1185 #if IS_ENABLED(CONFIG_IPV6)
1186                 if (sk->sk_family == PF_INET6)
1187                         l2tp_xmit_ipv6_csum(sk, skb, udp_len);
1188                 else
1189 #endif
1190                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1191                         skb->ip_summed = CHECKSUM_NONE;
1192                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1193                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1194                         skb->ip_summed = CHECKSUM_COMPLETE;
1195                         csum = skb_checksum(skb, 0, udp_len, 0);
1196                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1197                                                       inet->inet_daddr,
1198                                                       udp_len, IPPROTO_UDP, csum);
1199                         if (uh->check == 0)
1200                                 uh->check = CSUM_MANGLED_0;
1201                 } else {
1202                         skb->ip_summed = CHECKSUM_PARTIAL;
1203                         skb->csum_start = skb_transport_header(skb) - skb->head;
1204                         skb->csum_offset = offsetof(struct udphdr, check);
1205                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1206                                                        inet->inet_daddr,
1207                                                        udp_len, IPPROTO_UDP, 0);
1208                 }
1209                 break;
1210
1211         case L2TP_ENCAPTYPE_IP:
1212                 break;
1213         }
1214
1215         l2tp_skb_set_owner_w(skb, sk);
1216
1217         l2tp_xmit_core(session, skb, fl, data_len);
1218 out_unlock:
1219         bh_unlock_sock(sk);
1220
1221 abort:
1222         return 0;
1223 }
1224 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1225
1226 /*****************************************************************************
1227  * Tinnel and session create/destroy.
1228  *****************************************************************************/
1229
1230 /* Tunnel socket destruct hook.
1231  * The tunnel context is deleted only when all session sockets have been
1232  * closed.
1233  */
1234 static void l2tp_tunnel_destruct(struct sock *sk)
1235 {
1236         struct l2tp_tunnel *tunnel;
1237
1238         tunnel = sk->sk_user_data;
1239         if (tunnel == NULL)
1240                 goto end;
1241
1242         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1243                "%s: closing...\n", tunnel->name);
1244
1245         /* Close all sessions */
1246         l2tp_tunnel_closeall(tunnel);
1247
1248         switch (tunnel->encap) {
1249         case L2TP_ENCAPTYPE_UDP:
1250                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1251                 (udp_sk(sk))->encap_type = 0;
1252                 (udp_sk(sk))->encap_rcv = NULL;
1253                 break;
1254         case L2TP_ENCAPTYPE_IP:
1255                 break;
1256         }
1257
1258         /* Remove hooks into tunnel socket */
1259         tunnel->sock = NULL;
1260         sk->sk_destruct = tunnel->old_sk_destruct;
1261         sk->sk_user_data = NULL;
1262
1263         /* Call the original destructor */
1264         if (sk->sk_destruct)
1265                 (*sk->sk_destruct)(sk);
1266
1267         /* We're finished with the socket */
1268         l2tp_tunnel_dec_refcount(tunnel);
1269
1270 end:
1271         return;
1272 }
1273
1274 /* When the tunnel is closed, all the attached sessions need to go too.
1275  */
1276 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1277 {
1278         int hash;
1279         struct hlist_node *walk;
1280         struct hlist_node *tmp;
1281         struct l2tp_session *session;
1282
1283         BUG_ON(tunnel == NULL);
1284
1285         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1286                "%s: closing all sessions...\n", tunnel->name);
1287
1288         write_lock_bh(&tunnel->hlist_lock);
1289         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1290 again:
1291                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1292                         session = hlist_entry(walk, struct l2tp_session, hlist);
1293
1294                         PRINTK(session->debug, L2TP_MSG_CONTROL, KERN_INFO,
1295                                "%s: closing session\n", session->name);
1296
1297                         hlist_del_init(&session->hlist);
1298
1299                         /* Since we should hold the sock lock while
1300                          * doing any unbinding, we need to release the
1301                          * lock we're holding before taking that lock.
1302                          * Hold a reference to the sock so it doesn't
1303                          * disappear as we're jumping between locks.
1304                          */
1305                         if (session->ref != NULL)
1306                                 (*session->ref)(session);
1307
1308                         write_unlock_bh(&tunnel->hlist_lock);
1309
1310                         if (tunnel->version != L2TP_HDR_VER_2) {
1311                                 struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1312
1313                                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
1314                                 hlist_del_init_rcu(&session->global_hlist);
1315                                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1316                                 synchronize_rcu();
1317                         }
1318
1319                         if (session->session_close != NULL)
1320                                 (*session->session_close)(session);
1321
1322                         if (session->deref != NULL)
1323                                 (*session->deref)(session);
1324
1325                         write_lock_bh(&tunnel->hlist_lock);
1326
1327                         /* Now restart from the beginning of this hash
1328                          * chain.  We always remove a session from the
1329                          * list so we are guaranteed to make forward
1330                          * progress.
1331                          */
1332                         goto again;
1333                 }
1334         }
1335         write_unlock_bh(&tunnel->hlist_lock);
1336 }
1337
1338 /* Really kill the tunnel.
1339  * Come here only when all sessions have been cleared from the tunnel.
1340  */
1341 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1342 {
1343         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1344
1345         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1346         BUG_ON(tunnel->sock != NULL);
1347
1348         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
1349                "%s: free...\n", tunnel->name);
1350
1351         /* Remove from tunnel list */
1352         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1353         list_del_rcu(&tunnel->list);
1354         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1355         synchronize_rcu();
1356
1357         atomic_dec(&l2tp_tunnel_count);
1358         kfree(tunnel);
1359 }
1360
1361 /* Create a socket for the tunnel, if one isn't set up by
1362  * userspace. This is used for static tunnels where there is no
1363  * managing L2TP daemon.
1364  */
1365 static int l2tp_tunnel_sock_create(u32 tunnel_id, u32 peer_tunnel_id, struct l2tp_tunnel_cfg *cfg, struct socket **sockp)
1366 {
1367         int err = -EINVAL;
1368         struct sockaddr_in udp_addr;
1369         struct sockaddr_l2tpip ip_addr;
1370         struct socket *sock = NULL;
1371
1372         switch (cfg->encap) {
1373         case L2TP_ENCAPTYPE_UDP:
1374                 err = sock_create(AF_INET, SOCK_DGRAM, 0, sockp);
1375                 if (err < 0)
1376                         goto out;
1377
1378                 sock = *sockp;
1379
1380                 memset(&udp_addr, 0, sizeof(udp_addr));
1381                 udp_addr.sin_family = AF_INET;
1382                 udp_addr.sin_addr = cfg->local_ip;
1383                 udp_addr.sin_port = htons(cfg->local_udp_port);
1384                 err = kernel_bind(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr));
1385                 if (err < 0)
1386                         goto out;
1387
1388                 udp_addr.sin_family = AF_INET;
1389                 udp_addr.sin_addr = cfg->peer_ip;
1390                 udp_addr.sin_port = htons(cfg->peer_udp_port);
1391                 err = kernel_connect(sock, (struct sockaddr *) &udp_addr, sizeof(udp_addr), 0);
1392                 if (err < 0)
1393                         goto out;
1394
1395                 if (!cfg->use_udp_checksums)
1396                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1397
1398                 break;
1399
1400         case L2TP_ENCAPTYPE_IP:
1401                 err = sock_create(AF_INET, SOCK_DGRAM, IPPROTO_L2TP, sockp);
1402                 if (err < 0)
1403                         goto out;
1404
1405                 sock = *sockp;
1406
1407                 memset(&ip_addr, 0, sizeof(ip_addr));
1408                 ip_addr.l2tp_family = AF_INET;
1409                 ip_addr.l2tp_addr = cfg->local_ip;
1410                 ip_addr.l2tp_conn_id = tunnel_id;
1411                 err = kernel_bind(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr));
1412                 if (err < 0)
1413                         goto out;
1414
1415                 ip_addr.l2tp_family = AF_INET;
1416                 ip_addr.l2tp_addr = cfg->peer_ip;
1417                 ip_addr.l2tp_conn_id = peer_tunnel_id;
1418                 err = kernel_connect(sock, (struct sockaddr *) &ip_addr, sizeof(ip_addr), 0);
1419                 if (err < 0)
1420                         goto out;
1421
1422                 break;
1423
1424         default:
1425                 goto out;
1426         }
1427
1428 out:
1429         if ((err < 0) && sock) {
1430                 sock_release(sock);
1431                 *sockp = NULL;
1432         }
1433
1434         return err;
1435 }
1436
1437 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)
1438 {
1439         struct l2tp_tunnel *tunnel = NULL;
1440         int err;
1441         struct socket *sock = NULL;
1442         struct sock *sk = NULL;
1443         struct l2tp_net *pn;
1444         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1445
1446         /* Get the tunnel socket from the fd, which was opened by
1447          * the userspace L2TP daemon. If not specified, create a
1448          * kernel socket.
1449          */
1450         if (fd < 0) {
1451                 err = l2tp_tunnel_sock_create(tunnel_id, peer_tunnel_id, cfg, &sock);
1452                 if (err < 0)
1453                         goto err;
1454         } else {
1455                 err = -EBADF;
1456                 sock = sockfd_lookup(fd, &err);
1457                 if (!sock) {
1458                         printk(KERN_ERR "tunl %hu: sockfd_lookup(fd=%d) returned %d\n",
1459                                tunnel_id, fd, err);
1460                         goto err;
1461                 }
1462         }
1463
1464         sk = sock->sk;
1465
1466         if (cfg != NULL)
1467                 encap = cfg->encap;
1468
1469         /* Quick sanity checks */
1470         switch (encap) {
1471         case L2TP_ENCAPTYPE_UDP:
1472                 err = -EPROTONOSUPPORT;
1473                 if (sk->sk_protocol != IPPROTO_UDP) {
1474                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1475                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1476                         goto err;
1477                 }
1478                 break;
1479         case L2TP_ENCAPTYPE_IP:
1480                 err = -EPROTONOSUPPORT;
1481                 if (sk->sk_protocol != IPPROTO_L2TP) {
1482                         printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1483                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1484                         goto err;
1485                 }
1486                 break;
1487         }
1488
1489         /* Check if this socket has already been prepped */
1490         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
1491         if (tunnel != NULL) {
1492                 /* This socket has already been prepped */
1493                 err = -EBUSY;
1494                 goto err;
1495         }
1496
1497         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1498         if (tunnel == NULL) {
1499                 err = -ENOMEM;
1500                 goto err;
1501         }
1502
1503         tunnel->version = version;
1504         tunnel->tunnel_id = tunnel_id;
1505         tunnel->peer_tunnel_id = peer_tunnel_id;
1506         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1507
1508         tunnel->magic = L2TP_TUNNEL_MAGIC;
1509         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1510         rwlock_init(&tunnel->hlist_lock);
1511
1512         /* The net we belong to */
1513         tunnel->l2tp_net = net;
1514         pn = l2tp_pernet(net);
1515
1516         if (cfg != NULL)
1517                 tunnel->debug = cfg->debug;
1518
1519         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1520         tunnel->encap = encap;
1521         if (encap == L2TP_ENCAPTYPE_UDP) {
1522                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1523                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1524                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1525 #if IS_ENABLED(CONFIG_IPV6)
1526                 if (sk->sk_family == PF_INET6)
1527                         udpv6_encap_enable();
1528                 else
1529 #endif
1530                 udp_encap_enable();
1531         }
1532
1533         sk->sk_user_data = tunnel;
1534
1535         /* Hook on the tunnel socket destructor so that we can cleanup
1536          * if the tunnel socket goes away.
1537          */
1538         tunnel->old_sk_destruct = sk->sk_destruct;
1539         sk->sk_destruct = &l2tp_tunnel_destruct;
1540         tunnel->sock = sk;
1541         sk->sk_allocation = GFP_ATOMIC;
1542
1543         /* Add tunnel to our list */
1544         INIT_LIST_HEAD(&tunnel->list);
1545         atomic_inc(&l2tp_tunnel_count);
1546
1547         /* Bump the reference count. The tunnel context is deleted
1548          * only when this drops to zero. Must be done before list insertion
1549          */
1550         l2tp_tunnel_inc_refcount(tunnel);
1551         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1552         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1553         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1554
1555         err = 0;
1556 err:
1557         if (tunnelp)
1558                 *tunnelp = tunnel;
1559
1560         /* If tunnel's socket was created by the kernel, it doesn't
1561          *  have a file.
1562          */
1563         if (sock && sock->file)
1564                 sockfd_put(sock);
1565
1566         return err;
1567 }
1568 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1569
1570 /* This function is used by the netlink TUNNEL_DELETE command.
1571  */
1572 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1573 {
1574         int err = 0;
1575         struct socket *sock = tunnel->sock ? tunnel->sock->sk_socket : NULL;
1576
1577         /* Force the tunnel socket to close. This will eventually
1578          * cause the tunnel to be deleted via the normal socket close
1579          * mechanisms when userspace closes the tunnel socket.
1580          */
1581         if (sock != NULL) {
1582                 err = inet_shutdown(sock, 2);
1583
1584                 /* If the tunnel's socket was created by the kernel,
1585                  * close the socket here since the socket was not
1586                  * created by userspace.
1587                  */
1588                 if (sock->file == NULL)
1589                         err = inet_release(sock);
1590         }
1591
1592         return err;
1593 }
1594 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1595
1596 /* Really kill the session.
1597  */
1598 void l2tp_session_free(struct l2tp_session *session)
1599 {
1600         struct l2tp_tunnel *tunnel;
1601
1602         BUG_ON(atomic_read(&session->ref_count) != 0);
1603
1604         tunnel = session->tunnel;
1605         if (tunnel != NULL) {
1606                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1607
1608                 /* Delete the session from the hash */
1609                 write_lock_bh(&tunnel->hlist_lock);
1610                 hlist_del_init(&session->hlist);
1611                 write_unlock_bh(&tunnel->hlist_lock);
1612
1613                 /* Unlink from the global hash if not L2TPv2 */
1614                 if (tunnel->version != L2TP_HDR_VER_2) {
1615                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1616
1617                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1618                         hlist_del_init_rcu(&session->global_hlist);
1619                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1620                         synchronize_rcu();
1621                 }
1622
1623                 if (session->session_id != 0)
1624                         atomic_dec(&l2tp_session_count);
1625
1626                 sock_put(tunnel->sock);
1627
1628                 /* This will delete the tunnel context if this
1629                  * is the last session on the tunnel.
1630                  */
1631                 session->tunnel = NULL;
1632                 l2tp_tunnel_dec_refcount(tunnel);
1633         }
1634
1635         kfree(session);
1636
1637         return;
1638 }
1639 EXPORT_SYMBOL_GPL(l2tp_session_free);
1640
1641 /* This function is used by the netlink SESSION_DELETE command and by
1642    pseudowire modules.
1643  */
1644 int l2tp_session_delete(struct l2tp_session *session)
1645 {
1646         if (session->session_close != NULL)
1647                 (*session->session_close)(session);
1648
1649         l2tp_session_dec_refcount(session);
1650
1651         return 0;
1652 }
1653 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1654
1655
1656 /* We come here whenever a session's send_seq, cookie_len or
1657  * l2specific_len parameters are set.
1658  */
1659 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1660 {
1661         if (version == L2TP_HDR_VER_2) {
1662                 session->hdr_len = 6;
1663                 if (session->send_seq)
1664                         session->hdr_len += 4;
1665         } else {
1666                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1667                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1668                         session->hdr_len += 4;
1669         }
1670
1671 }
1672
1673 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)
1674 {
1675         struct l2tp_session *session;
1676
1677         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1678         if (session != NULL) {
1679                 session->magic = L2TP_SESSION_MAGIC;
1680                 session->tunnel = tunnel;
1681
1682                 session->session_id = session_id;
1683                 session->peer_session_id = peer_session_id;
1684                 session->nr = 1;
1685
1686                 sprintf(&session->name[0], "sess %u/%u",
1687                         tunnel->tunnel_id, session->session_id);
1688
1689                 skb_queue_head_init(&session->reorder_q);
1690
1691                 INIT_HLIST_NODE(&session->hlist);
1692                 INIT_HLIST_NODE(&session->global_hlist);
1693
1694                 /* Inherit debug options from tunnel */
1695                 session->debug = tunnel->debug;
1696
1697                 if (cfg) {
1698                         session->pwtype = cfg->pw_type;
1699                         session->debug = cfg->debug;
1700                         session->mtu = cfg->mtu;
1701                         session->mru = cfg->mru;
1702                         session->send_seq = cfg->send_seq;
1703                         session->recv_seq = cfg->recv_seq;
1704                         session->lns_mode = cfg->lns_mode;
1705                         session->reorder_timeout = cfg->reorder_timeout;
1706                         session->offset = cfg->offset;
1707                         session->l2specific_type = cfg->l2specific_type;
1708                         session->l2specific_len = cfg->l2specific_len;
1709                         session->cookie_len = cfg->cookie_len;
1710                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1711                         session->peer_cookie_len = cfg->peer_cookie_len;
1712                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1713                 }
1714
1715                 if (tunnel->version == L2TP_HDR_VER_2)
1716                         session->build_header = l2tp_build_l2tpv2_header;
1717                 else
1718                         session->build_header = l2tp_build_l2tpv3_header;
1719
1720                 l2tp_session_set_header_len(session, tunnel->version);
1721
1722                 /* Bump the reference count. The session context is deleted
1723                  * only when this drops to zero.
1724                  */
1725                 l2tp_session_inc_refcount(session);
1726                 l2tp_tunnel_inc_refcount(tunnel);
1727
1728                 /* Ensure tunnel socket isn't deleted */
1729                 sock_hold(tunnel->sock);
1730
1731                 /* Add session to the tunnel's hash list */
1732                 write_lock_bh(&tunnel->hlist_lock);
1733                 hlist_add_head(&session->hlist,
1734                                l2tp_session_id_hash(tunnel, session_id));
1735                 write_unlock_bh(&tunnel->hlist_lock);
1736
1737                 /* And to the global session list if L2TPv3 */
1738                 if (tunnel->version != L2TP_HDR_VER_2) {
1739                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1740
1741                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1742                         hlist_add_head_rcu(&session->global_hlist,
1743                                            l2tp_session_id_hash_2(pn, session_id));
1744                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1745                 }
1746
1747                 /* Ignore management session in session count value */
1748                 if (session->session_id != 0)
1749                         atomic_inc(&l2tp_session_count);
1750         }
1751
1752         return session;
1753 }
1754 EXPORT_SYMBOL_GPL(l2tp_session_create);
1755
1756 /*****************************************************************************
1757  * Init and cleanup
1758  *****************************************************************************/
1759
1760 static __net_init int l2tp_init_net(struct net *net)
1761 {
1762         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1763         int hash;
1764
1765         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1766         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1767
1768         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1769                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1770
1771         spin_lock_init(&pn->l2tp_session_hlist_lock);
1772
1773         return 0;
1774 }
1775
1776 static struct pernet_operations l2tp_net_ops = {
1777         .init = l2tp_init_net,
1778         .id   = &l2tp_net_id,
1779         .size = sizeof(struct l2tp_net),
1780 };
1781
1782 static int __init l2tp_init(void)
1783 {
1784         int rc = 0;
1785
1786         rc = register_pernet_device(&l2tp_net_ops);
1787         if (rc)
1788                 goto out;
1789
1790         printk(KERN_INFO "L2TP core driver, %s\n", L2TP_DRV_VERSION);
1791
1792 out:
1793         return rc;
1794 }
1795
1796 static void __exit l2tp_exit(void)
1797 {
1798         unregister_pernet_device(&l2tp_net_ops);
1799 }
1800
1801 module_init(l2tp_init);
1802 module_exit(l2tp_exit);
1803
1804 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1805 MODULE_DESCRIPTION("L2TP core");
1806 MODULE_LICENSE("GPL");
1807 MODULE_VERSION(L2TP_DRV_VERSION);
1808