l2tp: Split pppol2tp patch into separate l2tp and ppp parts
[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/uaccess.h>
25
26 #include <linux/kernel.h>
27 #include <linux/spinlock.h>
28 #include <linux/kthread.h>
29 #include <linux/sched.h>
30 #include <linux/slab.h>
31 #include <linux/errno.h>
32 #include <linux/jiffies.h>
33
34 #include <linux/netdevice.h>
35 #include <linux/net.h>
36 #include <linux/inetdevice.h>
37 #include <linux/skbuff.h>
38 #include <linux/init.h>
39 #include <linux/ip.h>
40 #include <linux/udp.h>
41 #include <linux/hash.h>
42 #include <linux/sort.h>
43 #include <linux/file.h>
44 #include <linux/nsproxy.h>
45 #include <net/net_namespace.h>
46 #include <net/netns/generic.h>
47 #include <net/dst.h>
48 #include <net/ip.h>
49 #include <net/udp.h>
50 #include <net/xfrm.h>
51
52 #include <asm/byteorder.h>
53 #include <asm/atomic.h>
54
55 #include "l2tp_core.h"
56
57 #define L2TP_DRV_VERSION        "V2.0"
58
59 /* L2TP header constants */
60 #define L2TP_HDRFLAG_T     0x8000
61 #define L2TP_HDRFLAG_L     0x4000
62 #define L2TP_HDRFLAG_S     0x0800
63 #define L2TP_HDRFLAG_O     0x0200
64 #define L2TP_HDRFLAG_P     0x0100
65
66 #define L2TP_HDR_VER_MASK  0x000F
67 #define L2TP_HDR_VER_2     0x0002
68
69 /* L2TPv3 default L2-specific sublayer */
70 #define L2TP_SLFLAG_S      0x40000000
71 #define L2TP_SL_SEQ_MASK   0x00ffffff
72
73 #define L2TP_HDR_SIZE_SEQ               10
74 #define L2TP_HDR_SIZE_NOSEQ             6
75
76 /* Default trace flags */
77 #define L2TP_DEFAULT_DEBUG_FLAGS        0
78
79 #define PRINTK(_mask, _type, _lvl, _fmt, args...)                       \
80         do {                                                            \
81                 if ((_mask) & (_type))                                  \
82                         printk(_lvl "L2TP: " _fmt, ##args);             \
83         } while (0)
84
85 /* Private data stored for received packets in the skb.
86  */
87 struct l2tp_skb_cb {
88         u16                     ns;
89         u16                     has_seq;
90         u16                     length;
91         unsigned long           expires;
92 };
93
94 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
95
96 static atomic_t l2tp_tunnel_count;
97 static atomic_t l2tp_session_count;
98
99 /* per-net private data for this module */
100 static unsigned int l2tp_net_id;
101 struct l2tp_net {
102         struct list_head l2tp_tunnel_list;
103         rwlock_t l2tp_tunnel_list_lock;
104 };
105
106 static inline struct l2tp_net *l2tp_pernet(struct net *net)
107 {
108         BUG_ON(!net);
109
110         return net_generic(net, l2tp_net_id);
111 }
112
113 /* Session hash list.
114  * The session_id SHOULD be random according to RFC2661, but several
115  * L2TP implementations (Cisco and Microsoft) use incrementing
116  * session_ids.  So we do a real hash on the session_id, rather than a
117  * simple bitmask.
118  */
119 static inline struct hlist_head *
120 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
121 {
122         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
123 }
124
125 /* Lookup a session by id
126  */
127 struct l2tp_session *l2tp_session_find(struct l2tp_tunnel *tunnel, u32 session_id)
128 {
129         struct hlist_head *session_list =
130                 l2tp_session_id_hash(tunnel, session_id);
131         struct l2tp_session *session;
132         struct hlist_node *walk;
133
134         read_lock_bh(&tunnel->hlist_lock);
135         hlist_for_each_entry(session, walk, session_list, hlist) {
136                 if (session->session_id == session_id) {
137                         read_unlock_bh(&tunnel->hlist_lock);
138                         return session;
139                 }
140         }
141         read_unlock_bh(&tunnel->hlist_lock);
142
143         return NULL;
144 }
145 EXPORT_SYMBOL_GPL(l2tp_session_find);
146
147 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
148 {
149         int hash;
150         struct hlist_node *walk;
151         struct l2tp_session *session;
152         int count = 0;
153
154         read_lock_bh(&tunnel->hlist_lock);
155         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
156                 hlist_for_each_entry(session, walk, &tunnel->session_hlist[hash], hlist) {
157                         if (++count > nth) {
158                                 read_unlock_bh(&tunnel->hlist_lock);
159                                 return session;
160                         }
161                 }
162         }
163
164         read_unlock_bh(&tunnel->hlist_lock);
165
166         return NULL;
167 }
168 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
169
170 /* Lookup a tunnel by id
171  */
172 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
173 {
174         struct l2tp_tunnel *tunnel;
175         struct l2tp_net *pn = l2tp_pernet(net);
176
177         read_lock_bh(&pn->l2tp_tunnel_list_lock);
178         list_for_each_entry(tunnel, &pn->l2tp_tunnel_list, list) {
179                 if (tunnel->tunnel_id == tunnel_id) {
180                         read_unlock_bh(&pn->l2tp_tunnel_list_lock);
181                         return tunnel;
182                 }
183         }
184         read_unlock_bh(&pn->l2tp_tunnel_list_lock);
185
186         return NULL;
187 }
188 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
189
190 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
191 {
192         struct l2tp_net *pn = l2tp_pernet(net);
193         struct l2tp_tunnel *tunnel;
194         int count = 0;
195
196         read_lock_bh(&pn->l2tp_tunnel_list_lock);
197         list_for_each_entry(tunnel, &pn->l2tp_tunnel_list, list) {
198                 if (++count > nth) {
199                         read_unlock_bh(&pn->l2tp_tunnel_list_lock);
200                         return tunnel;
201                 }
202         }
203
204         read_unlock_bh(&pn->l2tp_tunnel_list_lock);
205
206         return NULL;
207 }
208 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
209
210 /*****************************************************************************
211  * Receive data handling
212  *****************************************************************************/
213
214 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
215  * number.
216  */
217 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
218 {
219         struct sk_buff *skbp;
220         struct sk_buff *tmp;
221         u16 ns = L2TP_SKB_CB(skb)->ns;
222
223         spin_lock_bh(&session->reorder_q.lock);
224         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
225                 if (L2TP_SKB_CB(skbp)->ns > ns) {
226                         __skb_queue_before(&session->reorder_q, skbp, skb);
227                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
228                                "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
229                                session->name, ns, L2TP_SKB_CB(skbp)->ns,
230                                skb_queue_len(&session->reorder_q));
231                         session->stats.rx_oos_packets++;
232                         goto out;
233                 }
234         }
235
236         __skb_queue_tail(&session->reorder_q, skb);
237
238 out:
239         spin_unlock_bh(&session->reorder_q.lock);
240 }
241
242 /* Dequeue a single skb.
243  */
244 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
245 {
246         struct l2tp_tunnel *tunnel = session->tunnel;
247         int length = L2TP_SKB_CB(skb)->length;
248
249         /* We're about to requeue the skb, so return resources
250          * to its current owner (a socket receive buffer).
251          */
252         skb_orphan(skb);
253
254         tunnel->stats.rx_packets++;
255         tunnel->stats.rx_bytes += length;
256         session->stats.rx_packets++;
257         session->stats.rx_bytes += length;
258
259         if (L2TP_SKB_CB(skb)->has_seq) {
260                 /* Bump our Nr */
261                 session->nr++;
262                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
263                        "%s: updated nr to %hu\n", session->name, session->nr);
264         }
265
266         /* call private receive handler */
267         if (session->recv_skb != NULL)
268                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
269         else
270                 kfree_skb(skb);
271
272         if (session->deref)
273                 (*session->deref)(session);
274 }
275
276 /* Dequeue skbs from the session's reorder_q, subject to packet order.
277  * Skbs that have been in the queue for too long are simply discarded.
278  */
279 static void l2tp_recv_dequeue(struct l2tp_session *session)
280 {
281         struct sk_buff *skb;
282         struct sk_buff *tmp;
283
284         /* If the pkt at the head of the queue has the nr that we
285          * expect to send up next, dequeue it and any other
286          * in-sequence packets behind it.
287          */
288         spin_lock_bh(&session->reorder_q.lock);
289         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
290                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
291                         session->stats.rx_seq_discards++;
292                         session->stats.rx_errors++;
293                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
294                                "%s: oos pkt %hu len %d discarded (too old), "
295                                "waiting for %hu, reorder_q_len=%d\n",
296                                session->name, L2TP_SKB_CB(skb)->ns,
297                                L2TP_SKB_CB(skb)->length, session->nr,
298                                skb_queue_len(&session->reorder_q));
299                         __skb_unlink(skb, &session->reorder_q);
300                         kfree_skb(skb);
301                         if (session->deref)
302                                 (*session->deref)(session);
303                         continue;
304                 }
305
306                 if (L2TP_SKB_CB(skb)->has_seq) {
307                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
308                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
309                                        "%s: holding oos pkt %hu len %d, "
310                                        "waiting for %hu, reorder_q_len=%d\n",
311                                        session->name, L2TP_SKB_CB(skb)->ns,
312                                        L2TP_SKB_CB(skb)->length, session->nr,
313                                        skb_queue_len(&session->reorder_q));
314                                 goto out;
315                         }
316                 }
317                 __skb_unlink(skb, &session->reorder_q);
318
319                 /* Process the skb. We release the queue lock while we
320                  * do so to let other contexts process the queue.
321                  */
322                 spin_unlock_bh(&session->reorder_q.lock);
323                 l2tp_recv_dequeue_skb(session, skb);
324                 spin_lock_bh(&session->reorder_q.lock);
325         }
326
327 out:
328         spin_unlock_bh(&session->reorder_q.lock);
329 }
330
331 static inline int l2tp_verify_udp_checksum(struct sock *sk,
332                                            struct sk_buff *skb)
333 {
334         struct udphdr *uh = udp_hdr(skb);
335         u16 ulen = ntohs(uh->len);
336         struct inet_sock *inet;
337         __wsum psum;
338
339         if (sk->sk_no_check || skb_csum_unnecessary(skb) || !uh->check)
340                 return 0;
341
342         inet = inet_sk(sk);
343         psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr, ulen,
344                                   IPPROTO_UDP, 0);
345
346         if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
347             !csum_fold(csum_add(psum, skb->csum)))
348                 return 0;
349
350         skb->csum = psum;
351
352         return __skb_checksum_complete(skb);
353 }
354
355 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
356  * here. The skb is not on a list when we get here.
357  * Returns 0 if the packet was a data packet and was successfully passed on.
358  * Returns 1 if the packet was not a good data packet and could not be
359  * forwarded.  All such packets are passed up to userspace to deal with.
360  */
361 int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
362                        int (*payload_hook)(struct sk_buff *skb))
363 {
364         struct l2tp_session *session = NULL;
365         unsigned char *ptr, *optr;
366         u16 hdrflags;
367         u32 tunnel_id, session_id;
368         int length;
369         int offset;
370         u16 version;
371         u16 ns, nr;
372
373         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
374                 goto discard_bad_csum;
375
376         /* UDP always verifies the packet length. */
377         __skb_pull(skb, sizeof(struct udphdr));
378
379         /* Short packet? */
380         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
381                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
382                        "%s: recv short packet (len=%d)\n", tunnel->name, skb->len);
383                 goto error;
384         }
385
386         /* Point to L2TP header */
387         optr = ptr = skb->data;
388
389         /* Trace packet contents, if enabled */
390         if (tunnel->debug & L2TP_MSG_DATA) {
391                 length = min(32u, skb->len);
392                 if (!pskb_may_pull(skb, length))
393                         goto error;
394
395                 printk(KERN_DEBUG "%s: recv: ", tunnel->name);
396
397                 offset = 0;
398                 do {
399                         printk(" %02X", ptr[offset]);
400                 } while (++offset < length);
401
402                 printk("\n");
403         }
404
405         /* Get L2TP header flags */
406         hdrflags = ntohs(*(__be16 *)ptr);
407
408         /* Check protocol version */
409         version = hdrflags & L2TP_HDR_VER_MASK;
410         if (version != tunnel->version) {
411                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
412                        "%s: recv protocol version mismatch: got %d expected %d\n",
413                        tunnel->name, version, tunnel->version);
414                 goto error;
415         }
416
417         /* Get length of L2TP packet */
418         length = skb->len;
419
420         /* If type is control packet, it is handled by userspace. */
421         if (hdrflags & L2TP_HDRFLAG_T) {
422                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
423                        "%s: recv control packet, len=%d\n", tunnel->name, length);
424                 goto error;
425         }
426
427         /* Skip flags */
428         ptr += 2;
429
430         /* If length is present, skip it */
431         if (hdrflags & L2TP_HDRFLAG_L)
432                 ptr += 2;
433
434         /* Extract tunnel and session ID */
435         tunnel_id = ntohs(*(__be16 *) ptr);
436         ptr += 2;
437         session_id = ntohs(*(__be16 *) ptr);
438         ptr += 2;
439
440         /* Find the session context */
441         session = l2tp_session_find(tunnel, session_id);
442         if (!session) {
443                 /* Not found? Pass to userspace to deal with */
444                 PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_INFO,
445                        "%s: no session found (%hu/%hu). Passing up.\n",
446                        tunnel->name, tunnel_id, session_id);
447                 goto error;
448         }
449
450         /* The ref count is increased since we now hold a pointer to
451          * the session. Take care to decrement the refcnt when exiting
452          * this function from now on...
453          */
454         l2tp_session_inc_refcount(session);
455         if (session->ref)
456                 (*session->ref)(session);
457
458         /* Handle the optional sequence numbers. Sequence numbers are
459          * in different places for L2TPv2 and L2TPv3.
460          *
461          * If we are the LAC, enable/disable sequence numbers under
462          * the control of the LNS.  If no sequence numbers present but
463          * we were expecting them, discard frame.
464          */
465         ns = nr = 0;
466         L2TP_SKB_CB(skb)->has_seq = 0;
467         if (hdrflags & L2TP_HDRFLAG_S) {
468                 ns = (u16) ntohs(*(__be16 *) ptr);
469                 ptr += 2;
470                 nr = ntohs(*(__be16 *) ptr);
471                 ptr += 2;
472
473                 /* Store L2TP info in the skb */
474                 L2TP_SKB_CB(skb)->ns = ns;
475                 L2TP_SKB_CB(skb)->has_seq = 1;
476
477                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
478                        "%s: recv data ns=%hu, nr=%hu, session nr=%hu\n",
479                        session->name, ns, nr, session->nr);
480         }
481
482         if (L2TP_SKB_CB(skb)->has_seq) {
483                 /* Received a packet with sequence numbers. If we're the LNS,
484                  * check if we sre sending sequence numbers and if not,
485                  * configure it so.
486                  */
487                 if ((!session->lns_mode) && (!session->send_seq)) {
488                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
489                                "%s: requested to enable seq numbers by LNS\n",
490                                session->name);
491                         session->send_seq = -1;
492                 }
493         } else {
494                 /* No sequence numbers.
495                  * If user has configured mandatory sequence numbers, discard.
496                  */
497                 if (session->recv_seq) {
498                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
499                                "%s: recv data has no seq numbers when required. "
500                                "Discarding\n", session->name);
501                         session->stats.rx_seq_discards++;
502                         goto discard;
503                 }
504
505                 /* If we're the LAC and we're sending sequence numbers, the
506                  * LNS has requested that we no longer send sequence numbers.
507                  * If we're the LNS and we're sending sequence numbers, the
508                  * LAC is broken. Discard the frame.
509                  */
510                 if ((!session->lns_mode) && (session->send_seq)) {
511                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_INFO,
512                                "%s: requested to disable seq numbers by LNS\n",
513                                session->name);
514                         session->send_seq = 0;
515                 } else if (session->send_seq) {
516                         PRINTK(session->debug, L2TP_MSG_SEQ, KERN_WARNING,
517                                "%s: recv data has no seq numbers when required. "
518                                "Discarding\n", session->name);
519                         session->stats.rx_seq_discards++;
520                         goto discard;
521                 }
522         }
523
524         /* If offset bit set, skip it. */
525         if (hdrflags & L2TP_HDRFLAG_O) {
526                 offset = ntohs(*(__be16 *)ptr);
527                 ptr += 2 + offset;
528         }
529
530         offset = ptr - optr;
531         if (!pskb_may_pull(skb, offset))
532                 goto discard;
533
534         __skb_pull(skb, offset);
535
536         /* If caller wants to process the payload before we queue the
537          * packet, do so now.
538          */
539         if (payload_hook)
540                 if ((*payload_hook)(skb))
541                         goto discard;
542
543         /* Prepare skb for adding to the session's reorder_q.  Hold
544          * packets for max reorder_timeout or 1 second if not
545          * reordering.
546          */
547         L2TP_SKB_CB(skb)->length = length;
548         L2TP_SKB_CB(skb)->expires = jiffies +
549                 (session->reorder_timeout ? session->reorder_timeout : HZ);
550
551         /* Add packet to the session's receive queue. Reordering is done here, if
552          * enabled. Saved L2TP protocol info is stored in skb->sb[].
553          */
554         if (L2TP_SKB_CB(skb)->has_seq) {
555                 if (session->reorder_timeout != 0) {
556                         /* Packet reordering enabled. Add skb to session's
557                          * reorder queue, in order of ns.
558                          */
559                         l2tp_recv_queue_skb(session, skb);
560                 } else {
561                         /* Packet reordering disabled. Discard out-of-sequence
562                          * packets
563                          */
564                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
565                                 session->stats.rx_seq_discards++;
566                                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
567                                        "%s: oos pkt %hu len %d discarded, "
568                                        "waiting for %hu, reorder_q_len=%d\n",
569                                        session->name, L2TP_SKB_CB(skb)->ns,
570                                        L2TP_SKB_CB(skb)->length, session->nr,
571                                        skb_queue_len(&session->reorder_q));
572                                 goto discard;
573                         }
574                         skb_queue_tail(&session->reorder_q, skb);
575                 }
576         } else {
577                 /* No sequence numbers. Add the skb to the tail of the
578                  * reorder queue. This ensures that it will be
579                  * delivered after all previous sequenced skbs.
580                  */
581                 skb_queue_tail(&session->reorder_q, skb);
582         }
583
584         /* Try to dequeue as many skbs from reorder_q as we can. */
585         l2tp_recv_dequeue(session);
586
587         l2tp_session_dec_refcount(session);
588
589         return 0;
590
591 discard:
592         session->stats.rx_errors++;
593         kfree_skb(skb);
594
595         if (session->deref)
596                 (*session->deref)(session);
597
598         l2tp_session_dec_refcount(session);
599
600         return 0;
601
602 discard_bad_csum:
603         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
604         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
605         tunnel->stats.rx_errors++;
606         kfree_skb(skb);
607
608         return 0;
609
610 error:
611         /* Put UDP header back */
612         __skb_push(skb, sizeof(struct udphdr));
613
614         return 1;
615 }
616 EXPORT_SYMBOL_GPL(l2tp_udp_recv_core);
617
618 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
619  * Return codes:
620  * 0 : success.
621  * <0: error
622  * >0: skb should be passed up to userspace as UDP.
623  */
624 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
625 {
626         struct l2tp_tunnel *tunnel;
627
628         tunnel = l2tp_sock_to_tunnel(sk);
629         if (tunnel == NULL)
630                 goto pass_up;
631
632         PRINTK(tunnel->debug, L2TP_MSG_DATA, KERN_DEBUG,
633                "%s: received %d bytes\n", tunnel->name, skb->len);
634
635         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
636                 goto pass_up_put;
637
638         sock_put(sk);
639         return 0;
640
641 pass_up_put:
642         sock_put(sk);
643 pass_up:
644         return 1;
645 }
646 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
647
648 /************************************************************************
649  * Transmit handling
650  ***********************************************************************/
651
652 /* Build an L2TP header for the session into the buffer provided.
653  */
654 static void l2tp_build_l2tpv2_header(struct l2tp_tunnel *tunnel,
655                                      struct l2tp_session *session,
656                                      void *buf)
657 {
658         __be16 *bufp = buf;
659         u16 flags = L2TP_HDR_VER_2;
660         u32 tunnel_id = tunnel->peer_tunnel_id;
661         u32 session_id = session->peer_session_id;
662
663         if (session->send_seq)
664                 flags |= L2TP_HDRFLAG_S;
665
666         /* Setup L2TP header. */
667         *bufp++ = htons(flags);
668         *bufp++ = htons(tunnel_id);
669         *bufp++ = htons(session_id);
670         if (session->send_seq) {
671                 *bufp++ = htons(session->ns);
672                 *bufp++ = 0;
673                 session->ns++;
674                 PRINTK(session->debug, L2TP_MSG_SEQ, KERN_DEBUG,
675                        "%s: updated ns to %hu\n", session->name, session->ns);
676         }
677 }
678
679 void l2tp_build_l2tp_header(struct l2tp_session *session, void *buf)
680 {
681         struct l2tp_tunnel *tunnel = session->tunnel;
682
683         BUG_ON(tunnel->version != L2TP_HDR_VER_2);
684         l2tp_build_l2tpv2_header(tunnel, session, buf);
685 }
686 EXPORT_SYMBOL_GPL(l2tp_build_l2tp_header);
687
688 int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb, size_t data_len)
689 {
690         struct l2tp_tunnel *tunnel = session->tunnel;
691         unsigned int len = skb->len;
692         int error;
693
694         /* Debug */
695         if (session->send_seq)
696                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
697                        "%s: send %Zd bytes, ns=%hu\n", session->name,
698                        data_len, session->ns - 1);
699         else
700                 PRINTK(session->debug, L2TP_MSG_DATA, KERN_DEBUG,
701                        "%s: send %Zd bytes\n", session->name, data_len);
702
703         if (session->debug & L2TP_MSG_DATA) {
704                 int i;
705                 unsigned char *datap = skb->data + sizeof(struct udphdr);
706
707                 printk(KERN_DEBUG "%s: xmit:", session->name);
708                 for (i = 0; i < (len - sizeof(struct udphdr)); i++) {
709                         printk(" %02X", *datap++);
710                         if (i == 31) {
711                                 printk(" ...");
712                                 break;
713                         }
714                 }
715                 printk("\n");
716         }
717
718         /* Queue the packet to IP for output */
719         error = ip_queue_xmit(skb, 1);
720
721         /* Update stats */
722         if (error >= 0) {
723                 tunnel->stats.tx_packets++;
724                 tunnel->stats.tx_bytes += len;
725                 session->stats.tx_packets++;
726                 session->stats.tx_bytes += len;
727         } else {
728                 tunnel->stats.tx_errors++;
729                 session->stats.tx_errors++;
730         }
731
732         return 0;
733 }
734 EXPORT_SYMBOL_GPL(l2tp_xmit_core);
735
736 /* Automatically called when the skb is freed.
737  */
738 static void l2tp_sock_wfree(struct sk_buff *skb)
739 {
740         sock_put(skb->sk);
741 }
742
743 /* For data skbs that we transmit, we associate with the tunnel socket
744  * but don't do accounting.
745  */
746 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
747 {
748         sock_hold(sk);
749         skb->sk = sk;
750         skb->destructor = l2tp_sock_wfree;
751 }
752
753 /* If caller requires the skb to have a ppp header, the header must be
754  * inserted in the skb data before calling this function.
755  */
756 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
757 {
758         int data_len = skb->len;
759         struct sock *sk = session->tunnel->sock;
760         struct udphdr *uh;
761         unsigned int udp_len;
762         struct inet_sock *inet;
763         __wsum csum;
764         int old_headroom;
765         int new_headroom;
766         int headroom;
767
768         /* Check that there's enough headroom in the skb to insert IP,
769          * UDP and L2TP headers. If not enough, expand it to
770          * make room. Adjust truesize.
771          */
772         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
773                 sizeof(struct udphdr) + hdr_len;
774         old_headroom = skb_headroom(skb);
775         if (skb_cow_head(skb, headroom))
776                 goto abort;
777
778         new_headroom = skb_headroom(skb);
779         skb_orphan(skb);
780         skb->truesize += new_headroom - old_headroom;
781
782         /* Setup L2TP header */
783         l2tp_build_l2tp_header(session, __skb_push(skb, hdr_len));
784         udp_len = sizeof(struct udphdr) + hdr_len + data_len;
785
786         /* Setup UDP header */
787         inet = inet_sk(sk);
788         __skb_push(skb, sizeof(*uh));
789         skb_reset_transport_header(skb);
790         uh = udp_hdr(skb);
791         uh->source = inet->inet_sport;
792         uh->dest = inet->inet_dport;
793         uh->len = htons(udp_len);
794
795         uh->check = 0;
796
797         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
798         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
799                               IPSKB_REROUTED);
800         nf_reset(skb);
801
802         /* Get routing info from the tunnel socket */
803         skb_dst_drop(skb);
804         skb_dst_set(skb, dst_clone(__sk_dst_get(sk)));
805         l2tp_skb_set_owner_w(skb, sk);
806
807         /* Calculate UDP checksum if configured to do so */
808         if (sk->sk_no_check == UDP_CSUM_NOXMIT)
809                 skb->ip_summed = CHECKSUM_NONE;
810         else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
811                  (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
812                 skb->ip_summed = CHECKSUM_COMPLETE;
813                 csum = skb_checksum(skb, 0, udp_len, 0);
814                 uh->check = csum_tcpudp_magic(inet->inet_saddr,
815                                               inet->inet_daddr,
816                                               udp_len, IPPROTO_UDP, csum);
817                 if (uh->check == 0)
818                         uh->check = CSUM_MANGLED_0;
819         } else {
820                 skb->ip_summed = CHECKSUM_PARTIAL;
821                 skb->csum_start = skb_transport_header(skb) - skb->head;
822                 skb->csum_offset = offsetof(struct udphdr, check);
823                 uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
824                                                inet->inet_daddr,
825                                                udp_len, IPPROTO_UDP, 0);
826         }
827
828         l2tp_xmit_core(session, skb, data_len);
829
830 abort:
831         return 0;
832 }
833 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
834
835 /*****************************************************************************
836  * Tinnel and session create/destroy.
837  *****************************************************************************/
838
839 /* Tunnel socket destruct hook.
840  * The tunnel context is deleted only when all session sockets have been
841  * closed.
842  */
843 void l2tp_tunnel_destruct(struct sock *sk)
844 {
845         struct l2tp_tunnel *tunnel;
846
847         tunnel = sk->sk_user_data;
848         if (tunnel == NULL)
849                 goto end;
850
851         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
852                "%s: closing...\n", tunnel->name);
853
854         /* Close all sessions */
855         l2tp_tunnel_closeall(tunnel);
856
857         /* No longer an encapsulation socket. See net/ipv4/udp.c */
858         (udp_sk(sk))->encap_type = 0;
859         (udp_sk(sk))->encap_rcv = NULL;
860
861         /* Remove hooks into tunnel socket */
862         tunnel->sock = NULL;
863         sk->sk_destruct = tunnel->old_sk_destruct;
864         sk->sk_user_data = NULL;
865
866         /* Call the original destructor */
867         if (sk->sk_destruct)
868                 (*sk->sk_destruct)(sk);
869
870         /* We're finished with the socket */
871         l2tp_tunnel_dec_refcount(tunnel);
872
873 end:
874         return;
875 }
876 EXPORT_SYMBOL(l2tp_tunnel_destruct);
877
878 /* When the tunnel is closed, all the attached sessions need to go too.
879  */
880 void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
881 {
882         int hash;
883         struct hlist_node *walk;
884         struct hlist_node *tmp;
885         struct l2tp_session *session;
886
887         BUG_ON(tunnel == NULL);
888
889         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
890                "%s: closing all sessions...\n", tunnel->name);
891
892         write_lock_bh(&tunnel->hlist_lock);
893         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
894 again:
895                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
896                         session = hlist_entry(walk, struct l2tp_session, hlist);
897
898                         PRINTK(session->debug, L2TP_MSG_CONTROL, KERN_INFO,
899                                "%s: closing session\n", session->name);
900
901                         hlist_del_init(&session->hlist);
902
903                         /* Since we should hold the sock lock while
904                          * doing any unbinding, we need to release the
905                          * lock we're holding before taking that lock.
906                          * Hold a reference to the sock so it doesn't
907                          * disappear as we're jumping between locks.
908                          */
909                         if (session->ref != NULL)
910                                 (*session->ref)(session);
911
912                         write_unlock_bh(&tunnel->hlist_lock);
913
914                         if (session->session_close != NULL)
915                                 (*session->session_close)(session);
916
917                         if (session->deref != NULL)
918                                 (*session->deref)(session);
919
920                         write_lock_bh(&tunnel->hlist_lock);
921
922                         /* Now restart from the beginning of this hash
923                          * chain.  We always remove a session from the
924                          * list so we are guaranteed to make forward
925                          * progress.
926                          */
927                         goto again;
928                 }
929         }
930         write_unlock_bh(&tunnel->hlist_lock);
931 }
932 EXPORT_SYMBOL_GPL(l2tp_tunnel_closeall);
933
934 /* Really kill the tunnel.
935  * Come here only when all sessions have been cleared from the tunnel.
936  */
937 void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
938 {
939         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
940
941         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
942         BUG_ON(tunnel->sock != NULL);
943
944         PRINTK(tunnel->debug, L2TP_MSG_CONTROL, KERN_INFO,
945                "%s: free...\n", tunnel->name);
946
947         /* Remove from tunnel list */
948         write_lock_bh(&pn->l2tp_tunnel_list_lock);
949         list_del_init(&tunnel->list);
950         write_unlock_bh(&pn->l2tp_tunnel_list_lock);
951
952         atomic_dec(&l2tp_tunnel_count);
953         kfree(tunnel);
954 }
955 EXPORT_SYMBOL_GPL(l2tp_tunnel_free);
956
957 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)
958 {
959         struct l2tp_tunnel *tunnel = NULL;
960         int err;
961         struct socket *sock = NULL;
962         struct sock *sk = NULL;
963         struct l2tp_net *pn;
964
965         /* Get the tunnel socket from the fd, which was opened by
966          * the userspace L2TP daemon.
967          */
968         err = -EBADF;
969         sock = sockfd_lookup(fd, &err);
970         if (!sock) {
971                 printk(KERN_ERR "tunl %hu: sockfd_lookup(fd=%d) returned %d\n",
972                        tunnel_id, fd, err);
973                 goto err;
974         }
975
976         sk = sock->sk;
977
978         /* Quick sanity checks */
979         err = -EPROTONOSUPPORT;
980         if (sk->sk_protocol != IPPROTO_UDP) {
981                 printk(KERN_ERR "tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
982                        tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
983                 goto err;
984         }
985         err = -EAFNOSUPPORT;
986         if (sock->ops->family != AF_INET) {
987                 printk(KERN_ERR "tunl %hu: fd %d wrong family, got %d, expected %d\n",
988                        tunnel_id, fd, sock->ops->family, AF_INET);
989                 goto err;
990         }
991
992         /* Check if this socket has already been prepped */
993         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
994         if (tunnel != NULL) {
995                 /* This socket has already been prepped */
996                 err = -EBUSY;
997                 goto err;
998         }
999
1000         if (version != L2TP_HDR_VER_2)
1001                 goto err;
1002
1003         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1004         if (tunnel == NULL) {
1005                 err = -ENOMEM;
1006                 goto err;
1007         }
1008
1009         tunnel->version = version;
1010         tunnel->tunnel_id = tunnel_id;
1011         tunnel->peer_tunnel_id = peer_tunnel_id;
1012         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1013
1014         tunnel->magic = L2TP_TUNNEL_MAGIC;
1015         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1016         rwlock_init(&tunnel->hlist_lock);
1017
1018         /* The net we belong to */
1019         tunnel->l2tp_net = net;
1020         pn = l2tp_pernet(net);
1021
1022         if (cfg)
1023                 tunnel->debug = cfg->debug;
1024
1025         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1026         udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1027         udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1028
1029         sk->sk_user_data = tunnel;
1030
1031         /* Hook on the tunnel socket destructor so that we can cleanup
1032          * if the tunnel socket goes away.
1033          */
1034         tunnel->old_sk_destruct = sk->sk_destruct;
1035         sk->sk_destruct = &l2tp_tunnel_destruct;
1036         tunnel->sock = sk;
1037         sk->sk_allocation = GFP_ATOMIC;
1038
1039         /* Add tunnel to our list */
1040         INIT_LIST_HEAD(&tunnel->list);
1041         write_lock_bh(&pn->l2tp_tunnel_list_lock);
1042         list_add(&tunnel->list, &pn->l2tp_tunnel_list);
1043         write_unlock_bh(&pn->l2tp_tunnel_list_lock);
1044         atomic_inc(&l2tp_tunnel_count);
1045
1046         /* Bump the reference count. The tunnel context is deleted
1047          * only when this drops to zero.
1048          */
1049         l2tp_tunnel_inc_refcount(tunnel);
1050
1051         err = 0;
1052 err:
1053         if (tunnelp)
1054                 *tunnelp = tunnel;
1055
1056         if (sock)
1057                 sockfd_put(sock);
1058
1059         return err;
1060 }
1061 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1062
1063 /* Really kill the session.
1064  */
1065 void l2tp_session_free(struct l2tp_session *session)
1066 {
1067         struct l2tp_tunnel *tunnel;
1068
1069         BUG_ON(atomic_read(&session->ref_count) != 0);
1070
1071         tunnel = session->tunnel;
1072         if (tunnel != NULL) {
1073                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1074
1075                 /* Delete the session from the hash */
1076                 write_lock_bh(&tunnel->hlist_lock);
1077                 hlist_del_init(&session->hlist);
1078                 write_unlock_bh(&tunnel->hlist_lock);
1079
1080                 if (session->session_id != 0)
1081                         atomic_dec(&l2tp_session_count);
1082
1083                 sock_put(tunnel->sock);
1084
1085                 /* This will delete the tunnel context if this
1086                  * is the last session on the tunnel.
1087                  */
1088                 session->tunnel = NULL;
1089                 l2tp_tunnel_dec_refcount(tunnel);
1090         }
1091
1092         kfree(session);
1093
1094         return;
1095 }
1096 EXPORT_SYMBOL_GPL(l2tp_session_free);
1097
1098 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)
1099 {
1100         struct l2tp_session *session;
1101
1102         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1103         if (session != NULL) {
1104                 session->magic = L2TP_SESSION_MAGIC;
1105                 session->tunnel = tunnel;
1106
1107                 session->session_id = session_id;
1108                 session->peer_session_id = peer_session_id;
1109
1110                 sprintf(&session->name[0], "sess %u/%u",
1111                         tunnel->tunnel_id, session->session_id);
1112
1113                 skb_queue_head_init(&session->reorder_q);
1114
1115                 INIT_HLIST_NODE(&session->hlist);
1116
1117                 /* Inherit debug options from tunnel */
1118                 session->debug = tunnel->debug;
1119
1120                 if (cfg) {
1121                         session->debug = cfg->debug;
1122                         session->hdr_len = cfg->hdr_len;
1123                         session->mtu = cfg->mtu;
1124                         session->mru = cfg->mru;
1125                         session->send_seq = cfg->send_seq;
1126                         session->recv_seq = cfg->recv_seq;
1127                         session->lns_mode = cfg->lns_mode;
1128                 }
1129
1130                 /* Bump the reference count. The session context is deleted
1131                  * only when this drops to zero.
1132                  */
1133                 l2tp_session_inc_refcount(session);
1134                 l2tp_tunnel_inc_refcount(tunnel);
1135
1136                 /* Ensure tunnel socket isn't deleted */
1137                 sock_hold(tunnel->sock);
1138
1139                 /* Add session to the tunnel's hash list */
1140                 write_lock_bh(&tunnel->hlist_lock);
1141                 hlist_add_head(&session->hlist,
1142                                l2tp_session_id_hash(tunnel, session_id));
1143                 write_unlock_bh(&tunnel->hlist_lock);
1144
1145                 /* Ignore management session in session count value */
1146                 if (session->session_id != 0)
1147                         atomic_inc(&l2tp_session_count);
1148         }
1149
1150         return session;
1151 }
1152 EXPORT_SYMBOL_GPL(l2tp_session_create);
1153
1154 /*****************************************************************************
1155  * Init and cleanup
1156  *****************************************************************************/
1157
1158 static __net_init int l2tp_init_net(struct net *net)
1159 {
1160         struct l2tp_net *pn;
1161         int err;
1162
1163         pn = kzalloc(sizeof(*pn), GFP_KERNEL);
1164         if (!pn)
1165                 return -ENOMEM;
1166
1167         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1168         rwlock_init(&pn->l2tp_tunnel_list_lock);
1169
1170         err = net_assign_generic(net, l2tp_net_id, pn);
1171         if (err)
1172                 goto out;
1173
1174         return 0;
1175
1176 out:
1177         kfree(pn);
1178         return err;
1179 }
1180
1181 static __net_exit void l2tp_exit_net(struct net *net)
1182 {
1183         struct l2tp_net *pn;
1184
1185         pn = net_generic(net, l2tp_net_id);
1186         /*
1187          * if someone has cached our net then
1188          * further net_generic call will return NULL
1189          */
1190         net_assign_generic(net, l2tp_net_id, NULL);
1191         kfree(pn);
1192 }
1193
1194 static struct pernet_operations l2tp_net_ops = {
1195         .init = l2tp_init_net,
1196         .exit = l2tp_exit_net,
1197         .id   = &l2tp_net_id,
1198         .size = sizeof(struct l2tp_net),
1199 };
1200
1201 static int __init l2tp_init(void)
1202 {
1203         int rc = 0;
1204
1205         rc = register_pernet_device(&l2tp_net_ops);
1206         if (rc)
1207                 goto out;
1208
1209         printk(KERN_INFO "L2TP core driver, %s\n", L2TP_DRV_VERSION);
1210
1211 out:
1212         return rc;
1213 }
1214
1215 static void __exit l2tp_exit(void)
1216 {
1217         unregister_pernet_device(&l2tp_net_ops);
1218 }
1219
1220 module_init(l2tp_init);
1221 module_exit(l2tp_exit);
1222
1223 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1224 MODULE_DESCRIPTION("L2TP core");
1225 MODULE_LICENSE("GPL");
1226 MODULE_VERSION(L2TP_DRV_VERSION);
1227