net/l2tp/l2tp_core.c

   1 /*
   2  * L2TP core.
   3  *
   4  * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
   5  *
   6  * This file contains some code of the original L2TPv2 pppol2tp
   7  * driver, which has the following copyright:
   8  *
   9  * Authors:     Martijn van Oosterhout <kleptog@svana.org>
  10  *              James Chapman (jchapman@katalix.com)
  11  * Contributors:
  12  *              Michal Ostrowski <mostrows@speakeasy.net>
  13  *              Arnaldo Carvalho de Melo <acme@xconectiva.com.br>
  14  *              David S. Miller (davem@redhat.com)
  15  *
  16  * This program is free software; you can redistribute it and/or modify
  17  * it under the terms of the GNU General Public License version 2 as
  18  * published by the Free Software Foundation.
  19  */
  20
  21 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  22
  23 #include <linux/module.h>
  24 #include <linux/string.h>
  25 #include <linux/list.h>
  26 #include <linux/rculist.h>
  27 #include <linux/uaccess.h>
  28
  29 #include <linux/kernel.h>
  30 #include <linux/spinlock.h>
  31 #include <linux/kthread.h>
  32 #include <linux/sched.h>
  33 #include <linux/slab.h>
  34 #include <linux/errno.h>
  35 #include <linux/jiffies.h>
  36
  37 #include <linux/netdevice.h>
  38 #include <linux/net.h>
  39 #include <linux/inetdevice.h>
  40 #include <linux/skbuff.h>
  41 #include <linux/init.h>
  42 #include <linux/in.h>
  43 #include <linux/ip.h>
  44 #include <linux/udp.h>
  45 #include <linux/l2tp.h>
  46 #include <linux/hash.h>
  47 #include <linux/sort.h>
  48 #include <linux/file.h>
  49 #include <linux/nsproxy.h>
  50 #include <net/net_namespace.h>
  51 #include <net/netns/generic.h>
  52 #include <net/dst.h>
  53 #include <net/ip.h>
  54 #include <net/udp.h>
  55 #include <net/inet_common.h>
  56 #include <net/xfrm.h>
  57 #include <net/protocol.h>
  58 #include <net/inet6_connection_sock.h>
  59 #include <net/inet_ecn.h>
  60 #include <net/ip6_route.h>
  61 #include <net/ip6_checksum.h>
  62
  63 #include <asm/byteorder.h>
  64 #include <linux/atomic.h>
  65
  66 #include "l2tp_core.h"
  67
  68 #define L2TP_DRV_VERSION        "V2.0"
  69
  70 /* L2TP header constants */
  71 #define L2TP_HDRFLAG_T     0x8000
  72 #define L2TP_HDRFLAG_L     0x4000
  73 #define L2TP_HDRFLAG_S     0x0800
  74 #define L2TP_HDRFLAG_O     0x0200
  75 #define L2TP_HDRFLAG_P     0x0100
  76
  77 #define L2TP_HDR_VER_MASK  0x000F
  78 #define L2TP_HDR_VER_2     0x0002
  79 #define L2TP_HDR_VER_3     0x0003
  80
  81 /* L2TPv3 default L2-specific sublayer */
  82 #define L2TP_SLFLAG_S      0x40000000
  83 #define L2TP_SL_SEQ_MASK   0x00ffffff
  84
  85 #define L2TP_HDR_SIZE_SEQ               10
  86 #define L2TP_HDR_SIZE_NOSEQ             6
  87
  88 /* Default trace flags */
  89 #define L2TP_DEFAULT_DEBUG_FLAGS        0
  90
  91 /* Private data stored for received packets in the skb.
  92  */
  93 struct l2tp_skb_cb {
  94         u32                     ns;
  95         u16                     has_seq;
  96         u16                     length;
  97         unsigned long           expires;
  98 };
  99
 100 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
 101
 102 static atomic_t l2tp_tunnel_count;
 103 static atomic_t l2tp_session_count;
 104 static struct workqueue_struct *l2tp_wq;
 105
 106 /* per-net private data for this module */
 107 static unsigned int l2tp_net_id;
 108 struct l2tp_net {
 109         struct list_head l2tp_tunnel_list;
 110         spinlock_t l2tp_tunnel_list_lock;
 111         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
 112         spinlock_t l2tp_session_hlist_lock;
 113 };
 114
 115 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel);
 116
 117 static inline struct l2tp_tunnel *l2tp_tunnel(struct sock *sk)
 118 {
 119         return sk->sk_user_data;
 120 }
 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 /* Tunnel reference counts. Incremented per session that is added to
 130  * the tunnel.
 131  */
 132 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
 133 {
 134         atomic_inc(&tunnel->ref_count);
 135 }
 136
 137 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
 138 {
 139         if (atomic_dec_and_test(&tunnel->ref_count))
 140                 l2tp_tunnel_free(tunnel);
 141 }
 142 #ifdef L2TP_REFCNT_DEBUG
 143 #define l2tp_tunnel_inc_refcount(_t)                                    \
 144 do {                                                                    \
 145         pr_debug("l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n",        \
 146                  __func__, __LINE__, (_t)->name,                        \
 147                  atomic_read(&_t->ref_count));                          \
 148         l2tp_tunnel_inc_refcount_1(_t);                                 \
 149 } while (0)
 150 #define l2tp_tunnel_dec_refcount(_t)
 151 do {                                                                    \
 152         pr_debug("l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n",        \
 153                  __func__, __LINE__, (_t)->name,                        \
 154                  atomic_read(&_t->ref_count));                          \
 155         l2tp_tunnel_dec_refcount_1(_t);                                 \
 156 } while (0)
 157 #else
 158 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
 159 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
 160 #endif
 161
 162 /* Session hash global list for L2TPv3.
 163  * The session_id SHOULD be random according to RFC3931, but several
 164  * L2TP implementations use incrementing session_ids.  So we do a real
 165  * hash on the session_id, rather than a simple bitmask.
 166  */
 167 static inline struct hlist_head *
 168 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 169 {
 170         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 171
 172 }
 173
 174 /* Lookup the tunnel socket, possibly involving the fs code if the socket is
 175  * owned by userspace.  A struct sock returned from this function must be
 176  * released using l2tp_tunnel_sock_put once you're done with it.
 177  */
 178 static struct sock *l2tp_tunnel_sock_lookup(struct l2tp_tunnel *tunnel)
 179 {
 180         int err = 0;
 181         struct socket *sock = NULL;
 182         struct sock *sk = NULL;
 183
 184         if (!tunnel)
 185                 goto out;
 186
 187         if (tunnel->fd >= 0) {
 188                 /* Socket is owned by userspace, who might be in the process
 189                  * of closing it.  Look the socket up using the fd to ensure
 190                  * consistency.
 191                  */
 192                 sock = sockfd_lookup(tunnel->fd, &err);
 193                 if (sock)
 194                         sk = sock->sk;
 195         } else {
 196                 /* Socket is owned by kernelspace */
 197                 sk = tunnel->sock;
 198                 sock_hold(sk);
 199         }
 200
 201 out:
 202         return sk;
 203 }
 204
 205 /* Drop a reference to a tunnel socket obtained via. l2tp_tunnel_sock_put */
 206 static void l2tp_tunnel_sock_put(struct sock *sk)
 207 {
 208         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
 209         if (tunnel) {
 210                 if (tunnel->fd >= 0) {
 211                         /* Socket is owned by userspace */
 212                         sockfd_put(sk->sk_socket);
 213                 }
 214                 sock_put(sk);
 215         }
 216         sock_put(sk);
 217 }
 218
 219 /* Lookup a session by id in the global session list
 220  */
 221 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
 222 {
 223         struct l2tp_net *pn = l2tp_pernet(net);
 224         struct hlist_head *session_list =
 225                 l2tp_session_id_hash_2(pn, session_id);
 226         struct l2tp_session *session;
 227
 228         rcu_read_lock_bh();
 229         hlist_for_each_entry_rcu(session, session_list, global_hlist) {
 230                 if (session->session_id == session_id) {
 231                         rcu_read_unlock_bh();
 232                         return session;
 233                 }
 234         }
 235         rcu_read_unlock_bh();
 236
 237         return NULL;
 238 }
 239
 240 /* Session hash list.
 241  * The session_id SHOULD be random according to RFC2661, but several
 242  * L2TP implementations (Cisco and Microsoft) use incrementing
 243  * session_ids.  So we do a real hash on the session_id, rather than a
 244  * simple bitmask.
 245  */
 246 static inline struct hlist_head *
 247 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 248 {
 249         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 250 }
 251
 252 /* Lookup a session by id
 253  */
 254 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
 255 {
 256         struct hlist_head *session_list;
 257         struct l2tp_session *session;
 258
 259         /* In L2TPv3, session_ids are unique over all tunnels and we
 260          * sometimes need to look them up before we know the
 261          * tunnel.
 262          */
 263         if (tunnel == NULL)
 264                 return l2tp_session_find_2(net, session_id);
 265
 266         session_list = l2tp_session_id_hash(tunnel, session_id);
 267         read_lock_bh(&tunnel->hlist_lock);
 268         hlist_for_each_entry(session, session_list, hlist) {
 269                 if (session->session_id == session_id) {
 270                         read_unlock_bh(&tunnel->hlist_lock);
 271                         return session;
 272                 }
 273         }
 274         read_unlock_bh(&tunnel->hlist_lock);
 275
 276         return NULL;
 277 }
 278 EXPORT_SYMBOL_GPL(l2tp_session_find);
 279
 280 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
 281 {
 282         int hash;
 283         struct l2tp_session *session;
 284         int count = 0;
 285
 286         read_lock_bh(&tunnel->hlist_lock);
 287         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 288                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 289                         if (++count > nth) {
 290                                 read_unlock_bh(&tunnel->hlist_lock);
 291                                 return session;
 292                         }
 293                 }
 294         }
 295
 296         read_unlock_bh(&tunnel->hlist_lock);
 297
 298         return NULL;
 299 }
 300 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
 301
 302 /* Lookup a session by interface name.
 303  * This is very inefficient but is only used by management interfaces.
 304  */
 305 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
 306 {
 307         struct l2tp_net *pn = l2tp_pernet(net);
 308         int hash;
 309         struct l2tp_session *session;
 310
 311         rcu_read_lock_bh();
 312         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 313                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 314                         if (!strcmp(session->ifname, ifname)) {
 315                                 rcu_read_unlock_bh();
 316                                 return session;
 317                         }
 318                 }
 319         }
 320
 321         rcu_read_unlock_bh();
 322
 323         return NULL;
 324 }
 325 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
 326
 327 /* Lookup a tunnel by id
 328  */
 329 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
 330 {
 331         struct l2tp_tunnel *tunnel;
 332         struct l2tp_net *pn = l2tp_pernet(net);
 333
 334         rcu_read_lock_bh();
 335         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 336                 if (tunnel->tunnel_id == tunnel_id) {
 337                         rcu_read_unlock_bh();
 338                         return tunnel;
 339                 }
 340         }
 341         rcu_read_unlock_bh();
 342
 343         return NULL;
 344 }
 345 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
 346
 347 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
 348 {
 349         struct l2tp_net *pn = l2tp_pernet(net);
 350         struct l2tp_tunnel *tunnel;
 351         int count = 0;
 352
 353         rcu_read_lock_bh();
 354         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 355                 if (++count > nth) {
 356                         rcu_read_unlock_bh();
 357                         return tunnel;
 358                 }
 359         }
 360
 361         rcu_read_unlock_bh();
 362
 363         return NULL;
 364 }
 365 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
 366
 367 /*****************************************************************************
 368  * Receive data handling
 369  *****************************************************************************/
 370
 371 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 372  * number.
 373  */
 374 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 375 {
 376         struct sk_buff *skbp;
 377         struct sk_buff *tmp;
 378         u32 ns = L2TP_SKB_CB(skb)->ns;
 379
 380         spin_lock_bh(&session->reorder_q.lock);
 381         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 382                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 383                         __skb_queue_before(&session->reorder_q, skbp, skb);
 384                         l2tp_dbg(session, L2TP_MSG_SEQ,
 385                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 386                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 387                                  skb_queue_len(&session->reorder_q));
 388                         atomic_long_inc(&session->stats.rx_oos_packets);
 389                         goto out;
 390                 }
 391         }
 392
 393         __skb_queue_tail(&session->reorder_q, skb);
 394
 395 out:
 396         spin_unlock_bh(&session->reorder_q.lock);
 397 }
 398
 399 /* Dequeue a single skb.
 400  */
 401 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 402 {
 403         struct l2tp_tunnel *tunnel = session->tunnel;
 404         int length = L2TP_SKB_CB(skb)->length;
 405
 406         /* We're about to requeue the skb, so return resources
 407          * to its current owner (a socket receive buffer).
 408          */
 409         skb_orphan(skb);
 410
 411         atomic_long_inc(&tunnel->stats.rx_packets);
 412         atomic_long_add(length, &tunnel->stats.rx_bytes);
 413         atomic_long_inc(&session->stats.rx_packets);
 414         atomic_long_add(length, &session->stats.rx_bytes);
 415
 416         if (L2TP_SKB_CB(skb)->has_seq) {
 417                 /* Bump our Nr */
 418                 session->nr++;
 419                 session->nr &= session->nr_max;
 420
 421                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 422                          session->name, session->nr);
 423         }
 424
 425         /* call private receive handler */
 426         if (session->recv_skb != NULL)
 427                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 428         else
 429                 kfree_skb(skb);
 430
 431         if (session->deref)
 432                 (*session->deref)(session);
 433 }
 434
 435 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 436  * Skbs that have been in the queue for too long are simply discarded.
 437  */
 438 static void l2tp_recv_dequeue(struct l2tp_session *session)
 439 {
 440         struct sk_buff *skb;
 441         struct sk_buff *tmp;
 442
 443         /* If the pkt at the head of the queue has the nr that we
 444          * expect to send up next, dequeue it and any other
 445          * in-sequence packets behind it.
 446          */
 447 start:
 448         spin_lock_bh(&session->reorder_q.lock);
 449         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 450                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 451                         atomic_long_inc(&session->stats.rx_seq_discards);
 452                         atomic_long_inc(&session->stats.rx_errors);
 453                         l2tp_dbg(session, L2TP_MSG_SEQ,
 454                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 455                                  session->name, L2TP_SKB_CB(skb)->ns,
 456                                  L2TP_SKB_CB(skb)->length, session->nr,
 457                                  skb_queue_len(&session->reorder_q));
 458                         session->reorder_skip = 1;
 459                         __skb_unlink(skb, &session->reorder_q);
 460                         kfree_skb(skb);
 461                         if (session->deref)
 462                                 (*session->deref)(session);
 463                         continue;
 464                 }
 465
 466                 if (L2TP_SKB_CB(skb)->has_seq) {
 467                         if (session->reorder_skip) {
 468                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 469                                          "%s: advancing nr to next pkt: %u -> %u",
 470                                          session->name, session->nr,
 471                                          L2TP_SKB_CB(skb)->ns);
 472                                 session->reorder_skip = 0;
 473                                 session->nr = L2TP_SKB_CB(skb)->ns;
 474                         }
 475                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 476                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 477                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 478                                          session->name, L2TP_SKB_CB(skb)->ns,
 479                                          L2TP_SKB_CB(skb)->length, session->nr,
 480                                          skb_queue_len(&session->reorder_q));
 481                                 goto out;
 482                         }
 483                 }
 484                 __skb_unlink(skb, &session->reorder_q);
 485
 486                 /* Process the skb. We release the queue lock while we
 487                  * do so to let other contexts process the queue.
 488                  */
 489                 spin_unlock_bh(&session->reorder_q.lock);
 490                 l2tp_recv_dequeue_skb(session, skb);
 491                 goto start;
 492         }
 493
 494 out:
 495         spin_unlock_bh(&session->reorder_q.lock);
 496 }
 497
 498 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 499 {
 500         u32 nws;
 501
 502         if (nr >= session->nr)
 503                 nws = nr - session->nr;
 504         else
 505                 nws = (session->nr_max + 1) - (session->nr - nr);
 506
 507         return nws < session->nr_window_size;
 508 }
 509
 510 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 511  * acceptable, else non-zero.
 512  */
 513 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 514 {
 515         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
 516                 /* Packet sequence number is outside allowed window.
 517                  * Discard it.
 518                  */
 519                 l2tp_dbg(session, L2TP_MSG_SEQ,
 520                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
 521                          session->name, L2TP_SKB_CB(skb)->ns,
 522                          L2TP_SKB_CB(skb)->length, session->nr);
 523                 goto discard;
 524         }
 525
 526         if (session->reorder_timeout != 0) {
 527                 /* Packet reordering enabled. Add skb to session's
 528                  * reorder queue, in order of ns.
 529                  */
 530                 l2tp_recv_queue_skb(session, skb);
 531                 goto out;
 532         }
 533
 534         /* Packet reordering disabled. Discard out-of-sequence packets, while
 535          * tracking the number if in-sequence packets after the first OOS packet
 536          * is seen. After nr_oos_count_max in-sequence packets, reset the
 537          * sequence number to re-enable packet reception.
 538          */
 539         if (L2TP_SKB_CB(skb)->ns == session->nr) {
 540                 skb_queue_tail(&session->reorder_q, skb);
 541         } else {
 542                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
 543                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 544
 545                 if (nr_oos == nr_next)
 546                         session->nr_oos_count++;
 547                 else
 548                         session->nr_oos_count = 0;
 549
 550                 session->nr_oos = nr_oos;
 551                 if (session->nr_oos_count > session->nr_oos_count_max) {
 552                         session->reorder_skip = 1;
 553                         l2tp_dbg(session, L2TP_MSG_SEQ,
 554                                  "%s: %d oos packets received. Resetting sequence numbers\n",
 555                                  session->name, session->nr_oos_count);
 556                 }
 557                 if (!session->reorder_skip) {
 558                         atomic_long_inc(&session->stats.rx_seq_discards);
 559                         l2tp_dbg(session, L2TP_MSG_SEQ,
 560                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 561                                  session->name, L2TP_SKB_CB(skb)->ns,
 562                                  L2TP_SKB_CB(skb)->length, session->nr,
 563                                  skb_queue_len(&session->reorder_q));
 564                         goto discard;
 565                 }
 566                 skb_queue_tail(&session->reorder_q, skb);
 567         }
 568
 569 out:
 570         return 0;
 571
 572 discard:
 573         return 1;
 574 }
 575
 576 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 577  * and L2TPv3 data frames here.
 578  *
 579  * L2TPv2 Data Message Header
 580  *
 581  *  0                   1                   2                   3
 582  *  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
 583  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 584  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 585  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 586  * |           Tunnel ID           |           Session ID          |
 587  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 588  * |             Ns (opt)          |             Nr (opt)          |
 589  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 590  * |      Offset Size (opt)        |    Offset pad... (opt)
 591  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 592  *
 593  * Data frames are marked by T=0. All other fields are the same as
 594  * those in L2TP control frames.
 595  *
 596  * L2TPv3 Data Message Header
 597  *
 598  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 599  * |                      L2TP Session Header                      |
 600  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 601  * |                      L2-Specific Sublayer                     |
 602  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 603  * |                        Tunnel Payload                      ...
 604  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 605  *
 606  * L2TPv3 Session Header Over IP
 607  *
 608  *  0                   1                   2                   3
 609  *  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
 610  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 611  * |                           Session ID                          |
 612  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 613  * |               Cookie (optional, maximum 64 bits)...
 614  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 615  *                                                                 |
 616  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 617  *
 618  * L2TPv3 L2-Specific Sublayer Format
 619  *
 620  *  0                   1                   2                   3
 621  *  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
 622  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 623  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 624  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 625  *
 626  * Cookie value, sublayer format and offset (pad) are negotiated with
 627  * the peer when the session is set up. Unlike L2TPv2, we do not need
 628  * to parse the packet header to determine if optional fields are
 629  * present.
 630  *
 631  * Caller must already have parsed the frame and determined that it is
 632  * a data (not control) frame before coming here. Fields up to the
 633  * session-id have already been parsed and ptr points to the data
 634  * after the session-id.
 635  */
 636 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 637                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 638                       int length, int (*payload_hook)(struct sk_buff *skb))
 639 {
 640         struct l2tp_tunnel *tunnel = session->tunnel;
 641         int offset;
 642         u32 ns, nr;
 643
 644         /* The ref count is increased since we now hold a pointer to
 645          * the session. Take care to decrement the refcnt when exiting
 646          * this function from now on...
 647          */
 648         l2tp_session_inc_refcount(session);
 649         if (session->ref)
 650                 (*session->ref)(session);
 651
 652         /* Parse and check optional cookie */
 653         if (session->peer_cookie_len > 0) {
 654                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 655                         l2tp_info(tunnel, L2TP_MSG_DATA,
 656                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 657                                   tunnel->name, tunnel->tunnel_id,
 658                                   session->session_id);
 659                         atomic_long_inc(&session->stats.rx_cookie_discards);
 660                         goto discard;
 661                 }
 662                 ptr += session->peer_cookie_len;
 663         }
 664
 665         /* Handle the optional sequence numbers. Sequence numbers are
 666          * in different places for L2TPv2 and L2TPv3.
 667          *
 668          * If we are the LAC, enable/disable sequence numbers under
 669          * the control of the LNS.  If no sequence numbers present but
 670          * we were expecting them, discard frame.
 671          */
 672         ns = nr = 0;
 673         L2TP_SKB_CB(skb)->has_seq = 0;
 674         if (tunnel->version == L2TP_HDR_VER_2) {
 675                 if (hdrflags & L2TP_HDRFLAG_S) {
 676                         ns = ntohs(*(__be16 *) ptr);
 677                         ptr += 2;
 678                         nr = ntohs(*(__be16 *) ptr);
 679                         ptr += 2;
 680
 681                         /* Store L2TP info in the skb */
 682                         L2TP_SKB_CB(skb)->ns = ns;
 683                         L2TP_SKB_CB(skb)->has_seq = 1;
 684
 685                         l2tp_dbg(session, L2TP_MSG_SEQ,
 686                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 687                                  session->name, ns, nr, session->nr);
 688                 }
 689         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 690                 u32 l2h = ntohl(*(__be32 *) ptr);
 691
 692                 if (l2h & 0x40000000) {
 693                         ns = l2h & 0x00ffffff;
 694
 695                         /* Store L2TP info in the skb */
 696                         L2TP_SKB_CB(skb)->ns = ns;
 697                         L2TP_SKB_CB(skb)->has_seq = 1;
 698
 699                         l2tp_dbg(session, L2TP_MSG_SEQ,
 700                                  "%s: recv data ns=%u, session nr=%u\n",
 701                                  session->name, ns, session->nr);
 702                 }
 703         }
 704
 705         /* Advance past L2-specific header, if present */
 706         ptr += session->l2specific_len;
 707
 708         if (L2TP_SKB_CB(skb)->has_seq) {
 709                 /* Received a packet with sequence numbers. If we're the LNS,
 710                  * check if we sre sending sequence numbers and if not,
 711                  * configure it so.
 712                  */
 713                 if ((!session->lns_mode) && (!session->send_seq)) {
 714                         l2tp_info(session, L2TP_MSG_SEQ,
 715                                   "%s: requested to enable seq numbers by LNS\n",
 716                                   session->name);
 717                         session->send_seq = -1;
 718                         l2tp_session_set_header_len(session, tunnel->version);
 719                 }
 720         } else {
 721                 /* No sequence numbers.
 722                  * If user has configured mandatory sequence numbers, discard.
 723                  */
 724                 if (session->recv_seq) {
 725                         l2tp_warn(session, L2TP_MSG_SEQ,
 726                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 727                                   session->name);
 728                         atomic_long_inc(&session->stats.rx_seq_discards);
 729                         goto discard;
 730                 }
 731
 732                 /* If we're the LAC and we're sending sequence numbers, the
 733                  * LNS has requested that we no longer send sequence numbers.
 734                  * If we're the LNS and we're sending sequence numbers, the
 735                  * LAC is broken. Discard the frame.
 736                  */
 737                 if ((!session->lns_mode) && (session->send_seq)) {
 738                         l2tp_info(session, L2TP_MSG_SEQ,
 739                                   "%s: requested to disable seq numbers by LNS\n",
 740                                   session->name);
 741                         session->send_seq = 0;
 742                         l2tp_session_set_header_len(session, tunnel->version);
 743                 } else if (session->send_seq) {
 744                         l2tp_warn(session, L2TP_MSG_SEQ,
 745                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 746                                   session->name);
 747                         atomic_long_inc(&session->stats.rx_seq_discards);
 748                         goto discard;
 749                 }
 750         }
 751
 752         /* Session data offset is handled differently for L2TPv2 and
 753          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
 754          * the header. For L2TPv3, the offset is negotiated using AVPs
 755          * in the session setup control protocol.
 756          */
 757         if (tunnel->version == L2TP_HDR_VER_2) {
 758                 /* If offset bit set, skip it. */
 759                 if (hdrflags & L2TP_HDRFLAG_O) {
 760                         offset = ntohs(*(__be16 *)ptr);
 761                         ptr += 2 + offset;
 762                 }
 763         } else
 764                 ptr += session->offset;
 765
 766         offset = ptr - optr;
 767         if (!pskb_may_pull(skb, offset))
 768                 goto discard;
 769
 770         __skb_pull(skb, offset);
 771
 772         /* If caller wants to process the payload before we queue the
 773          * packet, do so now.
 774          */
 775         if (payload_hook)
 776                 if ((*payload_hook)(skb))
 777                         goto discard;
 778
 779         /* Prepare skb for adding to the session's reorder_q.  Hold
 780          * packets for max reorder_timeout or 1 second if not
 781          * reordering.
 782          */
 783         L2TP_SKB_CB(skb)->length = length;
 784         L2TP_SKB_CB(skb)->expires = jiffies +
 785                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 786
 787         /* Add packet to the session's receive queue. Reordering is done here, if
 788          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 789          */
 790         if (L2TP_SKB_CB(skb)->has_seq) {
 791                 if (l2tp_recv_data_seq(session, skb))
 792                         goto discard;
 793         } else {
 794                 /* No sequence numbers. Add the skb to the tail of the
 795                  * reorder queue. This ensures that it will be
 796                  * delivered after all previous sequenced skbs.
 797                  */
 798                 skb_queue_tail(&session->reorder_q, skb);
 799         }
 800
 801         /* Try to dequeue as many skbs from reorder_q as we can. */
 802         l2tp_recv_dequeue(session);
 803
 804         l2tp_session_dec_refcount(session);
 805
 806         return;
 807
 808 discard:
 809         atomic_long_inc(&session->stats.rx_errors);
 810         kfree_skb(skb);
 811
 812         if (session->deref)
 813                 (*session->deref)(session);
 814
 815         l2tp_session_dec_refcount(session);
 816 }
 817 EXPORT_SYMBOL(l2tp_recv_common);
 818
 819 /* Drop skbs from the session's reorder_q
 820  */
 821 int l2tp_session_queue_purge(struct l2tp_session *session)
 822 {
 823         struct sk_buff *skb = NULL;
 824         BUG_ON(!session);
 825         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
 826         while ((skb = skb_dequeue(&session->reorder_q))) {
 827                 atomic_long_inc(&session->stats.rx_errors);
 828                 kfree_skb(skb);
 829                 if (session->deref)
 830                         (*session->deref)(session);
 831         }
 832         return 0;
 833 }
 834 EXPORT_SYMBOL_GPL(l2tp_session_queue_purge);
 835
 836 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 837  * here. The skb is not on a list when we get here.
 838  * Returns 0 if the packet was a data packet and was successfully passed on.
 839  * Returns 1 if the packet was not a good data packet and could not be
 840  * forwarded.  All such packets are passed up to userspace to deal with.
 841  */
 842 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
 843                               int (*payload_hook)(struct sk_buff *skb))
 844 {
 845         struct l2tp_session *session = NULL;
 846         unsigned char *ptr, *optr;
 847         u16 hdrflags;
 848         u32 tunnel_id, session_id;
 849         u16 version;
 850         int length;
 851
 852         /* UDP has verifed checksum */
 853
 854         /* UDP always verifies the packet length. */
 855         __skb_pull(skb, sizeof(struct udphdr));
 856
 857         /* Short packet? */
 858         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
 859                 l2tp_info(tunnel, L2TP_MSG_DATA,
 860                           "%s: recv short packet (len=%d)\n",
 861                           tunnel->name, skb->len);
 862                 goto error;
 863         }
 864
 865         /* Trace packet contents, if enabled */
 866         if (tunnel->debug & L2TP_MSG_DATA) {
 867                 length = min(32u, skb->len);
 868                 if (!pskb_may_pull(skb, length))
 869                         goto error;
 870
 871                 pr_debug("%s: recv\n", tunnel->name);
 872                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 873         }
 874
 875         /* Point to L2TP header */
 876         optr = ptr = skb->data;
 877
 878         /* Get L2TP header flags */
 879         hdrflags = ntohs(*(__be16 *) ptr);
 880
 881         /* Check protocol version */
 882         version = hdrflags & L2TP_HDR_VER_MASK;
 883         if (version != tunnel->version) {
 884                 l2tp_info(tunnel, L2TP_MSG_DATA,
 885                           "%s: recv protocol version mismatch: got %d expected %d\n",
 886                           tunnel->name, version, tunnel->version);
 887                 goto error;
 888         }
 889
 890         /* Get length of L2TP packet */
 891         length = skb->len;
 892
 893         /* If type is control packet, it is handled by userspace. */
 894         if (hdrflags & L2TP_HDRFLAG_T) {
 895                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 896                          "%s: recv control packet, len=%d\n",
 897                          tunnel->name, length);
 898                 goto error;
 899         }
 900
 901         /* Skip flags */
 902         ptr += 2;
 903
 904         if (tunnel->version == L2TP_HDR_VER_2) {
 905                 /* If length is present, skip it */
 906                 if (hdrflags & L2TP_HDRFLAG_L)
 907                         ptr += 2;
 908
 909                 /* Extract tunnel and session ID */
 910                 tunnel_id = ntohs(*(__be16 *) ptr);
 911                 ptr += 2;
 912                 session_id = ntohs(*(__be16 *) ptr);
 913                 ptr += 2;
 914         } else {
 915                 ptr += 2;       /* skip reserved bits */
 916                 tunnel_id = tunnel->tunnel_id;
 917                 session_id = ntohl(*(__be32 *) ptr);
 918                 ptr += 4;
 919         }
 920
 921         /* Find the session context */
 922         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
 923         if (!session || !session->recv_skb) {
 924                 /* Not found? Pass to userspace to deal with */
 925                 l2tp_info(tunnel, L2TP_MSG_DATA,
 926                           "%s: no session found (%u/%u). Passing up.\n",
 927                           tunnel->name, tunnel_id, session_id);
 928                 goto error;
 929         }
 930
 931         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
 932
 933         return 0;
 934
 935 error:
 936         /* Put UDP header back */
 937         __skb_push(skb, sizeof(struct udphdr));
 938
 939         return 1;
 940 }
 941
 942 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 943  * Return codes:
 944  * 0 : success.
 945  * <0: error
 946  * >0: skb should be passed up to userspace as UDP.
 947  */
 948 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 949 {
 950         struct l2tp_tunnel *tunnel;
 951
 952         tunnel = l2tp_sock_to_tunnel(sk);
 953         if (tunnel == NULL)
 954                 goto pass_up;
 955
 956         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 957                  tunnel->name, skb->len);
 958
 959         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
 960                 goto pass_up_put;
 961
 962         sock_put(sk);
 963         return 0;
 964
 965 pass_up_put:
 966         sock_put(sk);
 967 pass_up:
 968         return 1;
 969 }
 970 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 971
 972 /************************************************************************
 973  * Transmit handling
 974  ***********************************************************************/
 975
 976 /* Build an L2TP header for the session into the buffer provided.
 977  */
 978 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 979 {
 980         struct l2tp_tunnel *tunnel = session->tunnel;
 981         __be16 *bufp = buf;
 982         __be16 *optr = buf;
 983         u16 flags = L2TP_HDR_VER_2;
 984         u32 tunnel_id = tunnel->peer_tunnel_id;
 985         u32 session_id = session->peer_session_id;
 986
 987         if (session->send_seq)
 988                 flags |= L2TP_HDRFLAG_S;
 989
 990         /* Setup L2TP header. */
 991         *bufp++ = htons(flags);
 992         *bufp++ = htons(tunnel_id);
 993         *bufp++ = htons(session_id);
 994         if (session->send_seq) {
 995                 *bufp++ = htons(session->ns);
 996                 *bufp++ = 0;
 997                 session->ns++;
 998                 session->ns &= 0xffff;
 999                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
1000                          session->name, session->ns);
1001         }
1002
1003         return bufp - optr;
1004 }
1005
1006 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
1007 {
1008         struct l2tp_tunnel *tunnel = session->tunnel;
1009         char *bufp = buf;
1010         char *optr = bufp;
1011
1012         /* Setup L2TP header. The header differs slightly for UDP and
1013          * IP encapsulations. For UDP, there is 4 bytes of flags.
1014          */
1015         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1016                 u16 flags = L2TP_HDR_VER_3;
1017                 *((__be16 *) bufp) = htons(flags);
1018                 bufp += 2;
1019                 *((__be16 *) bufp) = 0;
1020                 bufp += 2;
1021         }
1022
1023         *((__be32 *) bufp) = htonl(session->peer_session_id);
1024         bufp += 4;
1025         if (session->cookie_len) {
1026                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1027                 bufp += session->cookie_len;
1028         }
1029         if (session->l2specific_len) {
1030                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1031                         u32 l2h = 0;
1032                         if (session->send_seq) {
1033                                 l2h = 0x40000000 | session->ns;
1034                                 session->ns++;
1035                                 session->ns &= 0xffffff;
1036                                 l2tp_dbg(session, L2TP_MSG_SEQ,
1037                                          "%s: updated ns to %u\n",
1038                                          session->name, session->ns);
1039                         }
1040
1041                         *((__be32 *) bufp) = htonl(l2h);
1042                 }
1043                 bufp += session->l2specific_len;
1044         }
1045         if (session->offset)
1046                 bufp += session->offset;
1047
1048         return bufp - optr;
1049 }
1050
1051 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1052                           struct flowi *fl, size_t data_len)
1053 {
1054         struct l2tp_tunnel *tunnel = session->tunnel;
1055         unsigned int len = skb->len;
1056         int error;
1057
1058         /* Debug */
1059         if (session->send_seq)
1060                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes, ns=%u\n",
1061                          session->name, data_len, session->ns - 1);
1062         else
1063                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes\n",
1064                          session->name, data_len);
1065
1066         if (session->debug & L2TP_MSG_DATA) {
1067                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1068                 unsigned char *datap = skb->data + uhlen;
1069
1070                 pr_debug("%s: xmit\n", session->name);
1071                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1072                                      datap, min_t(size_t, 32, len - uhlen));
1073         }
1074
1075         /* Queue the packet to IP for output */
1076         skb->ignore_df = 1;
1077 #if IS_ENABLED(CONFIG_IPV6)
1078         if (tunnel->sock->sk_family == PF_INET6 && !tunnel->v4mapped)
1079                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1080         else
1081 #endif
1082                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1083
1084         /* Update stats */
1085         if (error >= 0) {
1086                 atomic_long_inc(&tunnel->stats.tx_packets);
1087                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1088                 atomic_long_inc(&session->stats.tx_packets);
1089                 atomic_long_add(len, &session->stats.tx_bytes);
1090         } else {
1091                 atomic_long_inc(&tunnel->stats.tx_errors);
1092                 atomic_long_inc(&session->stats.tx_errors);
1093         }
1094
1095         return 0;
1096 }
1097
1098 #if IS_ENABLED(CONFIG_IPV6)
1099 static void l2tp_xmit_ipv6_csum(struct sock *sk, struct sk_buff *skb,
1100                                 int udp_len)
1101 {
1102         struct ipv6_pinfo *np = inet6_sk(sk);
1103         struct udphdr *uh = udp_hdr(skb);
1104
1105         if (udp_get_no_check6_tx(sk))
1106                 skb->ip_summed = CHECKSUM_NONE;
1107         else if (!skb_dst(skb) || !skb_dst(skb)->dev ||
1108             !(skb_dst(skb)->dev->features & NETIF_F_IPV6_CSUM)) {
1109                 __wsum csum = skb_checksum(skb, 0, udp_len, 0);
1110                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1111                 uh->check = csum_ipv6_magic(&np->saddr, &sk->sk_v6_daddr, udp_len,
1112                                             IPPROTO_UDP, csum);
1113                 if (uh->check == 0)
1114                         uh->check = CSUM_MANGLED_0;
1115         } else {
1116                 skb->ip_summed = CHECKSUM_PARTIAL;
1117                 skb->csum_start = skb_transport_header(skb) - skb->head;
1118                 skb->csum_offset = offsetof(struct udphdr, check);
1119                 uh->check = ~csum_ipv6_magic(&np->saddr, &sk->sk_v6_daddr,
1120                                              udp_len, IPPROTO_UDP, 0);
1121         }
1122 }
1123 #endif
1124
1125 /* If caller requires the skb to have a ppp header, the header must be
1126  * inserted in the skb data before calling this function.
1127  */
1128 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1129 {
1130         int data_len = skb->len;
1131         struct l2tp_tunnel *tunnel = session->tunnel;
1132         struct sock *sk = tunnel->sock;
1133         struct flowi *fl;
1134         struct udphdr *uh;
1135         struct inet_sock *inet;
1136         __wsum csum;
1137         int headroom;
1138         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1139         int udp_len;
1140         int ret = NET_XMIT_SUCCESS;
1141
1142         /* Check that there's enough headroom in the skb to insert IP,
1143          * UDP and L2TP headers. If not enough, expand it to
1144          * make room. Adjust truesize.
1145          */
1146         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1147                 uhlen + hdr_len;
1148         if (skb_cow_head(skb, headroom)) {
1149                 kfree_skb(skb);
1150                 return NET_XMIT_DROP;
1151         }
1152
1153         /* Setup L2TP header */
1154         session->build_header(session, __skb_push(skb, hdr_len));
1155
1156         /* Reset skb netfilter state */
1157         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1158         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1159                               IPSKB_REROUTED);
1160         nf_reset(skb);
1161
1162         bh_lock_sock(sk);
1163         if (sock_owned_by_user(sk)) {
1164                 kfree_skb(skb);
1165                 ret = NET_XMIT_DROP;
1166                 goto out_unlock;
1167         }
1168
1169         /* Get routing info from the tunnel socket */
1170         skb_dst_drop(skb);
1171         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1172
1173         inet = inet_sk(sk);
1174         fl = &inet->cork.fl;
1175         switch (tunnel->encap) {
1176         case L2TP_ENCAPTYPE_UDP:
1177                 /* Setup UDP header */
1178                 __skb_push(skb, sizeof(*uh));
1179                 skb_reset_transport_header(skb);
1180                 uh = udp_hdr(skb);
1181                 uh->source = inet->inet_sport;
1182                 uh->dest = inet->inet_dport;
1183                 udp_len = uhlen + hdr_len + data_len;
1184                 uh->len = htons(udp_len);
1185                 uh->check = 0;
1186
1187                 /* Calculate UDP checksum if configured to do so */
1188 #if IS_ENABLED(CONFIG_IPV6)
1189                 if (sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1190                         l2tp_xmit_ipv6_csum(sk, skb, udp_len);
1191                 else
1192 #endif
1193                 if (sk->sk_no_check_tx)
1194                         skb->ip_summed = CHECKSUM_NONE;
1195                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1196                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1197                         skb->ip_summed = CHECKSUM_COMPLETE;
1198                         csum = skb_checksum(skb, 0, udp_len, 0);
1199                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1200                                                       inet->inet_daddr,
1201                                                       udp_len, IPPROTO_UDP, csum);
1202                         if (uh->check == 0)
1203                                 uh->check = CSUM_MANGLED_0;
1204                 } else {
1205                         skb->ip_summed = CHECKSUM_PARTIAL;
1206                         skb->csum_start = skb_transport_header(skb) - skb->head;
1207                         skb->csum_offset = offsetof(struct udphdr, check);
1208                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1209                                                        inet->inet_daddr,
1210                                                        udp_len, IPPROTO_UDP, 0);
1211                 }
1212                 break;
1213
1214         case L2TP_ENCAPTYPE_IP:
1215                 break;
1216         }
1217
1218         l2tp_xmit_core(session, skb, fl, data_len);
1219 out_unlock:
1220         bh_unlock_sock(sk);
1221
1222         return ret;
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 = l2tp_tunnel(sk);
1237         struct l2tp_net *pn;
1238
1239         if (tunnel == NULL)
1240                 goto end;
1241
1242         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1243
1244
1245         /* Disable udp encapsulation */
1246         switch (tunnel->encap) {
1247         case L2TP_ENCAPTYPE_UDP:
1248                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1249                 (udp_sk(sk))->encap_type = 0;
1250                 (udp_sk(sk))->encap_rcv = NULL;
1251                 (udp_sk(sk))->encap_destroy = NULL;
1252                 break;
1253         case L2TP_ENCAPTYPE_IP:
1254                 break;
1255         }
1256
1257         /* Remove hooks into tunnel socket */
1258         sk->sk_destruct = tunnel->old_sk_destruct;
1259         sk->sk_user_data = NULL;
1260         tunnel->sock = NULL;
1261
1262         /* Remove the tunnel struct from the tunnel list */
1263         pn = l2tp_pernet(tunnel->l2tp_net);
1264         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1265         list_del_rcu(&tunnel->list);
1266         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1267         atomic_dec(&l2tp_tunnel_count);
1268
1269         l2tp_tunnel_closeall(tunnel);
1270         l2tp_tunnel_dec_refcount(tunnel);
1271
1272         /* Call the original destructor */
1273         if (sk->sk_destruct)
1274                 (*sk->sk_destruct)(sk);
1275 end:
1276         return;
1277 }
1278
1279 /* When the tunnel is closed, all the attached sessions need to go too.
1280  */
1281 void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1282 {
1283         int hash;
1284         struct hlist_node *walk;
1285         struct hlist_node *tmp;
1286         struct l2tp_session *session;
1287
1288         BUG_ON(tunnel == NULL);
1289
1290         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1291                   tunnel->name);
1292
1293         write_lock_bh(&tunnel->hlist_lock);
1294         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1295 again:
1296                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1297                         session = hlist_entry(walk, struct l2tp_session, hlist);
1298
1299                         l2tp_info(session, L2TP_MSG_CONTROL,
1300                                   "%s: closing session\n", session->name);
1301
1302                         hlist_del_init(&session->hlist);
1303
1304                         if (session->ref != NULL)
1305                                 (*session->ref)(session);
1306
1307                         write_unlock_bh(&tunnel->hlist_lock);
1308
1309                         __l2tp_session_unhash(session);
1310                         l2tp_session_queue_purge(session);
1311
1312                         if (session->session_close != NULL)
1313                                 (*session->session_close)(session);
1314
1315                         if (session->deref != NULL)
1316                                 (*session->deref)(session);
1317
1318                         l2tp_session_dec_refcount(session);
1319
1320                         write_lock_bh(&tunnel->hlist_lock);
1321
1322                         /* Now restart from the beginning of this hash
1323                          * chain.  We always remove a session from the
1324                          * list so we are guaranteed to make forward
1325                          * progress.
1326                          */
1327                         goto again;
1328                 }
1329         }
1330         write_unlock_bh(&tunnel->hlist_lock);
1331 }
1332 EXPORT_SYMBOL_GPL(l2tp_tunnel_closeall);
1333
1334 /* Tunnel socket destroy hook for UDP encapsulation */
1335 static void l2tp_udp_encap_destroy(struct sock *sk)
1336 {
1337         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
1338         if (tunnel) {
1339                 l2tp_tunnel_closeall(tunnel);
1340                 sock_put(sk);
1341         }
1342 }
1343
1344 /* Really kill the tunnel.
1345  * Come here only when all sessions have been cleared from the tunnel.
1346  */
1347 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1348 {
1349         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1350         BUG_ON(tunnel->sock != NULL);
1351         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: free...\n", tunnel->name);
1352         kfree_rcu(tunnel, rcu);
1353 }
1354
1355 /* Workqueue tunnel deletion function */
1356 static void l2tp_tunnel_del_work(struct work_struct *work)
1357 {
1358         struct l2tp_tunnel *tunnel = NULL;
1359         struct socket *sock = NULL;
1360         struct sock *sk = NULL;
1361
1362         tunnel = container_of(work, struct l2tp_tunnel, del_work);
1363         sk = l2tp_tunnel_sock_lookup(tunnel);
1364         if (!sk)
1365                 return;
1366
1367         sock = sk->sk_socket;
1368
1369         /* If the tunnel socket was created by userspace, then go through the
1370          * inet layer to shut the socket down, and let userspace close it.
1371          * Otherwise, if we created the socket directly within the kernel, use
1372          * the sk API to release it here.
1373          * In either case the tunnel resources are freed in the socket
1374          * destructor when the tunnel socket goes away.
1375          */
1376         if (tunnel->fd >= 0) {
1377                 if (sock)
1378                         inet_shutdown(sock, 2);
1379         } else {
1380                 if (sock)
1381                         kernel_sock_shutdown(sock, SHUT_RDWR);
1382                 sk_release_kernel(sk);
1383         }
1384
1385         l2tp_tunnel_sock_put(sk);
1386 }
1387
1388 /* Create a socket for the tunnel, if one isn't set up by
1389  * userspace. This is used for static tunnels where there is no
1390  * managing L2TP daemon.
1391  *
1392  * Since we don't want these sockets to keep a namespace alive by
1393  * themselves, we drop the socket's namespace refcount after creation.
1394  * These sockets are freed when the namespace exits using the pernet
1395  * exit hook.
1396  */
1397 static int l2tp_tunnel_sock_create(struct net *net,
1398                                 u32 tunnel_id,
1399                                 u32 peer_tunnel_id,
1400                                 struct l2tp_tunnel_cfg *cfg,
1401                                 struct socket **sockp)
1402 {
1403         int err = -EINVAL;
1404         struct socket *sock = NULL;
1405         struct sockaddr_in udp_addr = {0};
1406         struct sockaddr_l2tpip ip_addr = {0};
1407 #if IS_ENABLED(CONFIG_IPV6)
1408         struct sockaddr_in6 udp6_addr = {0};
1409         struct sockaddr_l2tpip6 ip6_addr = {0};
1410 #endif
1411
1412         switch (cfg->encap) {
1413         case L2TP_ENCAPTYPE_UDP:
1414 #if IS_ENABLED(CONFIG_IPV6)
1415                 if (cfg->local_ip6 && cfg->peer_ip6) {
1416                         err = sock_create_kern(AF_INET6, SOCK_DGRAM, 0, &sock);
1417                         if (err < 0)
1418                                 goto out;
1419
1420                         sk_change_net(sock->sk, net);
1421
1422                         udp6_addr.sin6_family = AF_INET6;
1423                         memcpy(&udp6_addr.sin6_addr, cfg->local_ip6,
1424                                sizeof(udp6_addr.sin6_addr));
1425                         udp6_addr.sin6_port = htons(cfg->local_udp_port);
1426                         err = kernel_bind(sock, (struct sockaddr *) &udp6_addr,
1427                                           sizeof(udp6_addr));
1428                         if (err < 0)
1429                                 goto out;
1430
1431                         udp6_addr.sin6_family = AF_INET6;
1432                         memcpy(&udp6_addr.sin6_addr, cfg->peer_ip6,
1433                                sizeof(udp6_addr.sin6_addr));
1434                         udp6_addr.sin6_port = htons(cfg->peer_udp_port);
1435                         err = kernel_connect(sock,
1436                                              (struct sockaddr *) &udp6_addr,
1437                                              sizeof(udp6_addr), 0);
1438                         if (err < 0)
1439                                 goto out;
1440
1441                         if (cfg->udp6_zero_tx_checksums)
1442                                 udp_set_no_check6_tx(sock->sk, true);
1443                         if (cfg->udp6_zero_rx_checksums)
1444                                 udp_set_no_check6_rx(sock->sk, true);
1445                 } else
1446 #endif
1447                 {
1448                         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0, &sock);
1449                         if (err < 0)
1450                                 goto out;
1451
1452                         sk_change_net(sock->sk, net);
1453
1454                         udp_addr.sin_family = AF_INET;
1455                         udp_addr.sin_addr = cfg->local_ip;
1456                         udp_addr.sin_port = htons(cfg->local_udp_port);
1457                         err = kernel_bind(sock, (struct sockaddr *) &udp_addr,
1458                                           sizeof(udp_addr));
1459                         if (err < 0)
1460                                 goto out;
1461
1462                         udp_addr.sin_family = AF_INET;
1463                         udp_addr.sin_addr = cfg->peer_ip;
1464                         udp_addr.sin_port = htons(cfg->peer_udp_port);
1465                         err = kernel_connect(sock,
1466                                              (struct sockaddr *) &udp_addr,
1467                                              sizeof(udp_addr), 0);
1468                         if (err < 0)
1469                                 goto out;
1470                 }
1471
1472                 if (!cfg->use_udp_checksums)
1473                         sock->sk->sk_no_check_tx = 1;
1474
1475                 break;
1476
1477         case L2TP_ENCAPTYPE_IP:
1478 #if IS_ENABLED(CONFIG_IPV6)
1479                 if (cfg->local_ip6 && cfg->peer_ip6) {
1480                         err = sock_create_kern(AF_INET6, SOCK_DGRAM,
1481                                           IPPROTO_L2TP, &sock);
1482                         if (err < 0)
1483                                 goto out;
1484
1485                         sk_change_net(sock->sk, net);
1486
1487                         ip6_addr.l2tp_family = AF_INET6;
1488                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1489                                sizeof(ip6_addr.l2tp_addr));
1490                         ip6_addr.l2tp_conn_id = tunnel_id;
1491                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1492                                           sizeof(ip6_addr));
1493                         if (err < 0)
1494                                 goto out;
1495
1496                         ip6_addr.l2tp_family = AF_INET6;
1497                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1498                                sizeof(ip6_addr.l2tp_addr));
1499                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1500                         err = kernel_connect(sock,
1501                                              (struct sockaddr *) &ip6_addr,
1502                                              sizeof(ip6_addr), 0);
1503                         if (err < 0)
1504                                 goto out;
1505                 } else
1506 #endif
1507                 {
1508                         err = sock_create_kern(AF_INET, SOCK_DGRAM,
1509                                           IPPROTO_L2TP, &sock);
1510                         if (err < 0)
1511                                 goto out;
1512
1513                         sk_change_net(sock->sk, net);
1514
1515                         ip_addr.l2tp_family = AF_INET;
1516                         ip_addr.l2tp_addr = cfg->local_ip;
1517                         ip_addr.l2tp_conn_id = tunnel_id;
1518                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1519                                           sizeof(ip_addr));
1520                         if (err < 0)
1521                                 goto out;
1522
1523                         ip_addr.l2tp_family = AF_INET;
1524                         ip_addr.l2tp_addr = cfg->peer_ip;
1525                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1526                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1527                                              sizeof(ip_addr), 0);
1528                         if (err < 0)
1529                                 goto out;
1530                 }
1531                 break;
1532
1533         default:
1534                 goto out;
1535         }
1536
1537 out:
1538         *sockp = sock;
1539         if ((err < 0) && sock) {
1540                 kernel_sock_shutdown(sock, SHUT_RDWR);
1541                 sk_release_kernel(sock->sk);
1542                 *sockp = NULL;
1543         }
1544
1545         return err;
1546 }
1547
1548 static struct lock_class_key l2tp_socket_class;
1549
1550 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)
1551 {
1552         struct l2tp_tunnel *tunnel = NULL;
1553         int err;
1554         struct socket *sock = NULL;
1555         struct sock *sk = NULL;
1556         struct l2tp_net *pn;
1557         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1558
1559         /* Get the tunnel socket from the fd, which was opened by
1560          * the userspace L2TP daemon. If not specified, create a
1561          * kernel socket.
1562          */
1563         if (fd < 0) {
1564                 err = l2tp_tunnel_sock_create(net, tunnel_id, peer_tunnel_id,
1565                                 cfg, &sock);
1566                 if (err < 0)
1567                         goto err;
1568         } else {
1569                 sock = sockfd_lookup(fd, &err);
1570                 if (!sock) {
1571                         pr_err("tunl %u: sockfd_lookup(fd=%d) returned %d\n",
1572                                tunnel_id, fd, err);
1573                         err = -EBADF;
1574                         goto err;
1575                 }
1576
1577                 /* Reject namespace mismatches */
1578                 if (!net_eq(sock_net(sock->sk), net)) {
1579                         pr_err("tunl %u: netns mismatch\n", tunnel_id);
1580                         err = -EINVAL;
1581                         goto err;
1582                 }
1583         }
1584
1585         sk = sock->sk;
1586
1587         if (cfg != NULL)
1588                 encap = cfg->encap;
1589
1590         /* Quick sanity checks */
1591         switch (encap) {
1592         case L2TP_ENCAPTYPE_UDP:
1593                 err = -EPROTONOSUPPORT;
1594                 if (sk->sk_protocol != IPPROTO_UDP) {
1595                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1596                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1597                         goto err;
1598                 }
1599                 break;
1600         case L2TP_ENCAPTYPE_IP:
1601                 err = -EPROTONOSUPPORT;
1602                 if (sk->sk_protocol != IPPROTO_L2TP) {
1603                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1604                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1605                         goto err;
1606                 }
1607                 break;
1608         }
1609
1610         /* Check if this socket has already been prepped */
1611         tunnel = l2tp_tunnel(sk);
1612         if (tunnel != NULL) {
1613                 /* This socket has already been prepped */
1614                 err = -EBUSY;
1615                 goto err;
1616         }
1617
1618         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1619         if (tunnel == NULL) {
1620                 err = -ENOMEM;
1621                 goto err;
1622         }
1623
1624         tunnel->version = version;
1625         tunnel->tunnel_id = tunnel_id;
1626         tunnel->peer_tunnel_id = peer_tunnel_id;
1627         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1628
1629         tunnel->magic = L2TP_TUNNEL_MAGIC;
1630         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1631         rwlock_init(&tunnel->hlist_lock);
1632
1633         /* The net we belong to */
1634         tunnel->l2tp_net = net;
1635         pn = l2tp_pernet(net);
1636
1637         if (cfg != NULL)
1638                 tunnel->debug = cfg->debug;
1639
1640 #if IS_ENABLED(CONFIG_IPV6)
1641         if (sk->sk_family == PF_INET6) {
1642                 struct ipv6_pinfo *np = inet6_sk(sk);
1643
1644                 if (ipv6_addr_v4mapped(&np->saddr) &&
1645                     ipv6_addr_v4mapped(&sk->sk_v6_daddr)) {
1646                         struct inet_sock *inet = inet_sk(sk);
1647
1648                         tunnel->v4mapped = true;
1649                         inet->inet_saddr = np->saddr.s6_addr32[3];
1650                         inet->inet_rcv_saddr = sk->sk_v6_rcv_saddr.s6_addr32[3];
1651                         inet->inet_daddr = sk->sk_v6_daddr.s6_addr32[3];
1652                 } else {
1653                         tunnel->v4mapped = false;
1654                 }
1655         }
1656 #endif
1657
1658         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1659         tunnel->encap = encap;
1660         if (encap == L2TP_ENCAPTYPE_UDP) {
1661                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1662                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1663                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1664                 udp_sk(sk)->encap_destroy = l2tp_udp_encap_destroy;
1665 #if IS_ENABLED(CONFIG_IPV6)
1666                 if (sk->sk_family == PF_INET6 && !tunnel->v4mapped)
1667                         udpv6_encap_enable();
1668                 else
1669 #endif
1670                 udp_encap_enable();
1671         }
1672
1673         sk->sk_user_data = tunnel;
1674
1675         /* Hook on the tunnel socket destructor so that we can cleanup
1676          * if the tunnel socket goes away.
1677          */
1678         tunnel->old_sk_destruct = sk->sk_destruct;
1679         sk->sk_destruct = &l2tp_tunnel_destruct;
1680         tunnel->sock = sk;
1681         tunnel->fd = fd;
1682         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class, "l2tp_sock");
1683
1684         sk->sk_allocation = GFP_ATOMIC;
1685
1686         /* Init delete workqueue struct */
1687         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1688
1689         /* Add tunnel to our list */
1690         INIT_LIST_HEAD(&tunnel->list);
1691         atomic_inc(&l2tp_tunnel_count);
1692
1693         /* Bump the reference count. The tunnel context is deleted
1694          * only when this drops to zero. Must be done before list insertion
1695          */
1696         l2tp_tunnel_inc_refcount(tunnel);
1697         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1698         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1699         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1700
1701         err = 0;
1702 err:
1703         if (tunnelp)
1704                 *tunnelp = tunnel;
1705
1706         /* If tunnel's socket was created by the kernel, it doesn't
1707          *  have a file.
1708          */
1709         if (sock && sock->file)
1710                 sockfd_put(sock);
1711
1712         return err;
1713 }
1714 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1715
1716 /* This function is used by the netlink TUNNEL_DELETE command.
1717  */
1718 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1719 {
1720         l2tp_tunnel_closeall(tunnel);
1721         return (false == queue_work(l2tp_wq, &tunnel->del_work));
1722 }
1723 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1724
1725 /* Really kill the session.
1726  */
1727 void l2tp_session_free(struct l2tp_session *session)
1728 {
1729         struct l2tp_tunnel *tunnel = session->tunnel;
1730
1731         BUG_ON(atomic_read(&session->ref_count) != 0);
1732
1733         if (tunnel) {
1734                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1735                 if (session->session_id != 0)
1736                         atomic_dec(&l2tp_session_count);
1737                 sock_put(tunnel->sock);
1738                 session->tunnel = NULL;
1739                 l2tp_tunnel_dec_refcount(tunnel);
1740         }
1741
1742         kfree(session);
1743 }
1744 EXPORT_SYMBOL_GPL(l2tp_session_free);
1745
1746 /* Remove an l2tp session from l2tp_core's hash lists.
1747  * Provides a tidyup interface for pseudowire code which can't just route all
1748  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1749  * callback.
1750  */
1751 void __l2tp_session_unhash(struct l2tp_session *session)
1752 {
1753         struct l2tp_tunnel *tunnel = session->tunnel;
1754
1755         /* Remove the session from core hashes */
1756         if (tunnel) {
1757                 /* Remove from the per-tunnel hash */
1758                 write_lock_bh(&tunnel->hlist_lock);
1759                 hlist_del_init(&session->hlist);
1760                 write_unlock_bh(&tunnel->hlist_lock);
1761
1762                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1763                 if (tunnel->version != L2TP_HDR_VER_2) {
1764                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1765                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1766                         hlist_del_init_rcu(&session->global_hlist);
1767                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1768                         synchronize_rcu();
1769                 }
1770         }
1771 }
1772 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1773
1774 /* This function is used by the netlink SESSION_DELETE command and by
1775    pseudowire modules.
1776  */
1777 int l2tp_session_delete(struct l2tp_session *session)
1778 {
1779         if (session->ref)
1780                 (*session->ref)(session);
1781         __l2tp_session_unhash(session);
1782         l2tp_session_queue_purge(session);
1783         if (session->session_close != NULL)
1784                 (*session->session_close)(session);
1785         if (session->deref)
1786                 (*session->deref)(session);
1787         l2tp_session_dec_refcount(session);
1788         return 0;
1789 }
1790 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1791
1792 /* We come here whenever a session's send_seq, cookie_len or
1793  * l2specific_len parameters are set.
1794  */
1795 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1796 {
1797         if (version == L2TP_HDR_VER_2) {
1798                 session->hdr_len = 6;
1799                 if (session->send_seq)
1800                         session->hdr_len += 4;
1801         } else {
1802                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1803                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1804                         session->hdr_len += 4;
1805         }
1806
1807 }
1808 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1809
1810 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)
1811 {
1812         struct l2tp_session *session;
1813
1814         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1815         if (session != NULL) {
1816                 session->magic = L2TP_SESSION_MAGIC;
1817                 session->tunnel = tunnel;
1818
1819                 session->session_id = session_id;
1820                 session->peer_session_id = peer_session_id;
1821                 session->nr = 0;
1822                 if (tunnel->version == L2TP_HDR_VER_2)
1823                         session->nr_max = 0xffff;
1824                 else
1825                         session->nr_max = 0xffffff;
1826                 session->nr_window_size = session->nr_max / 2;
1827                 session->nr_oos_count_max = 4;
1828
1829                 /* Use NR of first received packet */
1830                 session->reorder_skip = 1;
1831
1832                 sprintf(&session->name[0], "sess %u/%u",
1833                         tunnel->tunnel_id, session->session_id);
1834
1835                 skb_queue_head_init(&session->reorder_q);
1836
1837                 INIT_HLIST_NODE(&session->hlist);
1838                 INIT_HLIST_NODE(&session->global_hlist);
1839
1840                 /* Inherit debug options from tunnel */
1841                 session->debug = tunnel->debug;
1842
1843                 if (cfg) {
1844                         session->pwtype = cfg->pw_type;
1845                         session->debug = cfg->debug;
1846                         session->mtu = cfg->mtu;
1847                         session->mru = cfg->mru;
1848                         session->send_seq = cfg->send_seq;
1849                         session->recv_seq = cfg->recv_seq;
1850                         session->lns_mode = cfg->lns_mode;
1851                         session->reorder_timeout = cfg->reorder_timeout;
1852                         session->offset = cfg->offset;
1853                         session->l2specific_type = cfg->l2specific_type;
1854                         session->l2specific_len = cfg->l2specific_len;
1855                         session->cookie_len = cfg->cookie_len;
1856                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1857                         session->peer_cookie_len = cfg->peer_cookie_len;
1858                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1859                 }
1860
1861                 if (tunnel->version == L2TP_HDR_VER_2)
1862                         session->build_header = l2tp_build_l2tpv2_header;
1863                 else
1864                         session->build_header = l2tp_build_l2tpv3_header;
1865
1866                 l2tp_session_set_header_len(session, tunnel->version);
1867
1868                 /* Bump the reference count. The session context is deleted
1869                  * only when this drops to zero.
1870                  */
1871                 l2tp_session_inc_refcount(session);
1872                 l2tp_tunnel_inc_refcount(tunnel);
1873
1874                 /* Ensure tunnel socket isn't deleted */
1875                 sock_hold(tunnel->sock);
1876
1877                 /* Add session to the tunnel's hash list */
1878                 write_lock_bh(&tunnel->hlist_lock);
1879                 hlist_add_head(&session->hlist,
1880                                l2tp_session_id_hash(tunnel, session_id));
1881                 write_unlock_bh(&tunnel->hlist_lock);
1882
1883                 /* And to the global session list if L2TPv3 */
1884                 if (tunnel->version != L2TP_HDR_VER_2) {
1885                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1886
1887                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1888                         hlist_add_head_rcu(&session->global_hlist,
1889                                            l2tp_session_id_hash_2(pn, session_id));
1890                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1891                 }
1892
1893                 /* Ignore management session in session count value */
1894                 if (session->session_id != 0)
1895                         atomic_inc(&l2tp_session_count);
1896         }
1897
1898         return session;
1899 }
1900 EXPORT_SYMBOL_GPL(l2tp_session_create);
1901
1902 /*****************************************************************************
1903  * Init and cleanup
1904  *****************************************************************************/
1905
1906 static __net_init int l2tp_init_net(struct net *net)
1907 {
1908         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1909         int hash;
1910
1911         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1912         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1913
1914         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1915                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1916
1917         spin_lock_init(&pn->l2tp_session_hlist_lock);
1918
1919         return 0;
1920 }
1921
1922 static __net_exit void l2tp_exit_net(struct net *net)
1923 {
1924         struct l2tp_net *pn = l2tp_pernet(net);
1925         struct l2tp_tunnel *tunnel = NULL;
1926
1927         rcu_read_lock_bh();
1928         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1929                 (void)l2tp_tunnel_delete(tunnel);
1930         }
1931         rcu_read_unlock_bh();
1932 }
1933
1934 static struct pernet_operations l2tp_net_ops = {
1935         .init = l2tp_init_net,
1936         .exit = l2tp_exit_net,
1937         .id   = &l2tp_net_id,
1938         .size = sizeof(struct l2tp_net),
1939 };
1940
1941 static int __init l2tp_init(void)
1942 {
1943         int rc = 0;
1944
1945         rc = register_pernet_device(&l2tp_net_ops);
1946         if (rc)
1947                 goto out;
1948
1949         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1950         if (!l2tp_wq) {
1951                 pr_err("alloc_workqueue failed\n");
1952                 rc = -ENOMEM;
1953                 goto out;
1954         }
1955
1956         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1957
1958 out:
1959         return rc;
1960 }
1961
1962 static void __exit l2tp_exit(void)
1963 {
1964         unregister_pernet_device(&l2tp_net_ops);
1965         if (l2tp_wq) {
1966                 destroy_workqueue(l2tp_wq);
1967                 l2tp_wq = NULL;
1968         }
1969 }
1970
1971 module_init(l2tp_init);
1972 module_exit(l2tp_exit);
1973
1974 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1975 MODULE_DESCRIPTION("L2TP core");
1976 MODULE_LICENSE("GPL");
1977 MODULE_VERSION(L2TP_DRV_VERSION);
1978