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_session_set_header_len(struct l2tp_session *session, int version);
 116 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel);
 117 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel);
 118
 119 static inline struct l2tp_net *l2tp_pernet(struct net *net)
 120 {
 121         BUG_ON(!net);
 122
 123         return net_generic(net, l2tp_net_id);
 124 }
 125
 126 /* Tunnel reference counts. Incremented per session that is added to
 127  * the tunnel.
 128  */
 129 static inline void l2tp_tunnel_inc_refcount_1(struct l2tp_tunnel *tunnel)
 130 {
 131         atomic_inc(&tunnel->ref_count);
 132 }
 133
 134 static inline void l2tp_tunnel_dec_refcount_1(struct l2tp_tunnel *tunnel)
 135 {
 136         if (atomic_dec_and_test(&tunnel->ref_count))
 137                 l2tp_tunnel_free(tunnel);
 138 }
 139 #ifdef L2TP_REFCNT_DEBUG
 140 #define l2tp_tunnel_inc_refcount(_t)                                    \
 141 do {                                                                    \
 142         pr_debug("l2tp_tunnel_inc_refcount: %s:%d %s: cnt=%d\n",        \
 143                  __func__, __LINE__, (_t)->name,                        \
 144                  atomic_read(&_t->ref_count));                          \
 145         l2tp_tunnel_inc_refcount_1(_t);                                 \
 146 } while (0)
 147 #define l2tp_tunnel_dec_refcount(_t)
 148 do {                                                                    \
 149         pr_debug("l2tp_tunnel_dec_refcount: %s:%d %s: cnt=%d\n",        \
 150                  __func__, __LINE__, (_t)->name,                        \
 151                  atomic_read(&_t->ref_count));                          \
 152         l2tp_tunnel_dec_refcount_1(_t);                                 \
 153 } while (0)
 154 #else
 155 #define l2tp_tunnel_inc_refcount(t) l2tp_tunnel_inc_refcount_1(t)
 156 #define l2tp_tunnel_dec_refcount(t) l2tp_tunnel_dec_refcount_1(t)
 157 #endif
 158
 159 /* Session hash global list for L2TPv3.
 160  * The session_id SHOULD be random according to RFC3931, but several
 161  * L2TP implementations use incrementing session_ids.  So we do a real
 162  * hash on the session_id, rather than a simple bitmask.
 163  */
 164 static inline struct hlist_head *
 165 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 166 {
 167         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 168
 169 }
 170
 171 /* Lookup the tunnel socket, possibly involving the fs code if the socket is
 172  * owned by userspace.  A struct sock returned from this function must be
 173  * released using l2tp_tunnel_sock_put once you're done with it.
 174  */
 175 struct sock *l2tp_tunnel_sock_lookup(struct l2tp_tunnel *tunnel)
 176 {
 177         int err = 0;
 178         struct socket *sock = NULL;
 179         struct sock *sk = NULL;
 180
 181         if (!tunnel)
 182                 goto out;
 183
 184         if (tunnel->fd >= 0) {
 185                 /* Socket is owned by userspace, who might be in the process
 186                  * of closing it.  Look the socket up using the fd to ensure
 187                  * consistency.
 188                  */
 189                 sock = sockfd_lookup(tunnel->fd, &err);
 190                 if (sock)
 191                         sk = sock->sk;
 192         } else {
 193                 /* Socket is owned by kernelspace */
 194                 sk = tunnel->sock;
 195         }
 196
 197 out:
 198         return sk;
 199 }
 200 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_lookup);
 201
 202 /* Drop a reference to a tunnel socket obtained via. l2tp_tunnel_sock_put */
 203 void l2tp_tunnel_sock_put(struct sock *sk)
 204 {
 205         struct l2tp_tunnel *tunnel = l2tp_sock_to_tunnel(sk);
 206         if (tunnel) {
 207                 if (tunnel->fd >= 0) {
 208                         /* Socket is owned by userspace */
 209                         sockfd_put(sk->sk_socket);
 210                 }
 211                 sock_put(sk);
 212         }
 213 }
 214 EXPORT_SYMBOL_GPL(l2tp_tunnel_sock_put);
 215
 216 /* Lookup a session by id in the global session list
 217  */
 218 static struct l2tp_session *l2tp_session_find_2(struct net *net, u32 session_id)
 219 {
 220         struct l2tp_net *pn = l2tp_pernet(net);
 221         struct hlist_head *session_list =
 222                 l2tp_session_id_hash_2(pn, session_id);
 223         struct l2tp_session *session;
 224
 225         rcu_read_lock_bh();
 226         hlist_for_each_entry_rcu(session, session_list, global_hlist) {
 227                 if (session->session_id == session_id) {
 228                         rcu_read_unlock_bh();
 229                         return session;
 230                 }
 231         }
 232         rcu_read_unlock_bh();
 233
 234         return NULL;
 235 }
 236
 237 /* Session hash list.
 238  * The session_id SHOULD be random according to RFC2661, but several
 239  * L2TP implementations (Cisco and Microsoft) use incrementing
 240  * session_ids.  So we do a real hash on the session_id, rather than a
 241  * simple bitmask.
 242  */
 243 static inline struct hlist_head *
 244 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 245 {
 246         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 247 }
 248
 249 /* Lookup a session by id
 250  */
 251 struct l2tp_session *l2tp_session_find(struct net *net, struct l2tp_tunnel *tunnel, u32 session_id)
 252 {
 253         struct hlist_head *session_list;
 254         struct l2tp_session *session;
 255
 256         /* In L2TPv3, session_ids are unique over all tunnels and we
 257          * sometimes need to look them up before we know the
 258          * tunnel.
 259          */
 260         if (tunnel == NULL)
 261                 return l2tp_session_find_2(net, session_id);
 262
 263         session_list = l2tp_session_id_hash(tunnel, session_id);
 264         read_lock_bh(&tunnel->hlist_lock);
 265         hlist_for_each_entry(session, session_list, hlist) {
 266                 if (session->session_id == session_id) {
 267                         read_unlock_bh(&tunnel->hlist_lock);
 268                         return session;
 269                 }
 270         }
 271         read_unlock_bh(&tunnel->hlist_lock);
 272
 273         return NULL;
 274 }
 275 EXPORT_SYMBOL_GPL(l2tp_session_find);
 276
 277 struct l2tp_session *l2tp_session_find_nth(struct l2tp_tunnel *tunnel, int nth)
 278 {
 279         int hash;
 280         struct l2tp_session *session;
 281         int count = 0;
 282
 283         read_lock_bh(&tunnel->hlist_lock);
 284         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 285                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 286                         if (++count > nth) {
 287                                 read_unlock_bh(&tunnel->hlist_lock);
 288                                 return session;
 289                         }
 290                 }
 291         }
 292
 293         read_unlock_bh(&tunnel->hlist_lock);
 294
 295         return NULL;
 296 }
 297 EXPORT_SYMBOL_GPL(l2tp_session_find_nth);
 298
 299 /* Lookup a session by interface name.
 300  * This is very inefficient but is only used by management interfaces.
 301  */
 302 struct l2tp_session *l2tp_session_find_by_ifname(struct net *net, char *ifname)
 303 {
 304         struct l2tp_net *pn = l2tp_pernet(net);
 305         int hash;
 306         struct l2tp_session *session;
 307
 308         rcu_read_lock_bh();
 309         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 310                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 311                         if (!strcmp(session->ifname, ifname)) {
 312                                 rcu_read_unlock_bh();
 313                                 return session;
 314                         }
 315                 }
 316         }
 317
 318         rcu_read_unlock_bh();
 319
 320         return NULL;
 321 }
 322 EXPORT_SYMBOL_GPL(l2tp_session_find_by_ifname);
 323
 324 /* Lookup a tunnel by id
 325  */
 326 struct l2tp_tunnel *l2tp_tunnel_find(struct net *net, u32 tunnel_id)
 327 {
 328         struct l2tp_tunnel *tunnel;
 329         struct l2tp_net *pn = l2tp_pernet(net);
 330
 331         rcu_read_lock_bh();
 332         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 333                 if (tunnel->tunnel_id == tunnel_id) {
 334                         rcu_read_unlock_bh();
 335                         return tunnel;
 336                 }
 337         }
 338         rcu_read_unlock_bh();
 339
 340         return NULL;
 341 }
 342 EXPORT_SYMBOL_GPL(l2tp_tunnel_find);
 343
 344 struct l2tp_tunnel *l2tp_tunnel_find_nth(struct net *net, int nth)
 345 {
 346         struct l2tp_net *pn = l2tp_pernet(net);
 347         struct l2tp_tunnel *tunnel;
 348         int count = 0;
 349
 350         rcu_read_lock_bh();
 351         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 352                 if (++count > nth) {
 353                         rcu_read_unlock_bh();
 354                         return tunnel;
 355                 }
 356         }
 357
 358         rcu_read_unlock_bh();
 359
 360         return NULL;
 361 }
 362 EXPORT_SYMBOL_GPL(l2tp_tunnel_find_nth);
 363
 364 /*****************************************************************************
 365  * Receive data handling
 366  *****************************************************************************/
 367
 368 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 369  * number.
 370  */
 371 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 372 {
 373         struct sk_buff *skbp;
 374         struct sk_buff *tmp;
 375         u32 ns = L2TP_SKB_CB(skb)->ns;
 376         struct l2tp_stats *sstats;
 377
 378         spin_lock_bh(&session->reorder_q.lock);
 379         sstats = &session->stats;
 380         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 381                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 382                         __skb_queue_before(&session->reorder_q, skbp, skb);
 383                         l2tp_dbg(session, L2TP_MSG_SEQ,
 384                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 385                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 386                                  skb_queue_len(&session->reorder_q));
 387                         u64_stats_update_begin(&sstats->syncp);
 388                         sstats->rx_oos_packets++;
 389                         u64_stats_update_end(&sstats->syncp);
 390                         goto out;
 391                 }
 392         }
 393
 394         __skb_queue_tail(&session->reorder_q, skb);
 395
 396 out:
 397         spin_unlock_bh(&session->reorder_q.lock);
 398 }
 399
 400 /* Dequeue a single skb.
 401  */
 402 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 403 {
 404         struct l2tp_tunnel *tunnel = session->tunnel;
 405         int length = L2TP_SKB_CB(skb)->length;
 406         struct l2tp_stats *tstats, *sstats;
 407
 408         /* We're about to requeue the skb, so return resources
 409          * to its current owner (a socket receive buffer).
 410          */
 411         skb_orphan(skb);
 412
 413         tstats = &tunnel->stats;
 414         u64_stats_update_begin(&tstats->syncp);
 415         sstats = &session->stats;
 416         u64_stats_update_begin(&sstats->syncp);
 417         tstats->rx_packets++;
 418         tstats->rx_bytes += length;
 419         sstats->rx_packets++;
 420         sstats->rx_bytes += length;
 421         u64_stats_update_end(&tstats->syncp);
 422         u64_stats_update_end(&sstats->syncp);
 423
 424         if (L2TP_SKB_CB(skb)->has_seq) {
 425                 /* Bump our Nr */
 426                 session->nr++;
 427                 if (tunnel->version == L2TP_HDR_VER_2)
 428                         session->nr &= 0xffff;
 429                 else
 430                         session->nr &= 0xffffff;
 431
 432                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 433                          session->name, session->nr);
 434         }
 435
 436         /* call private receive handler */
 437         if (session->recv_skb != NULL)
 438                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 439         else
 440                 kfree_skb(skb);
 441
 442         if (session->deref)
 443                 (*session->deref)(session);
 444 }
 445
 446 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 447  * Skbs that have been in the queue for too long are simply discarded.
 448  */
 449 static void l2tp_recv_dequeue(struct l2tp_session *session)
 450 {
 451         struct sk_buff *skb;
 452         struct sk_buff *tmp;
 453         struct l2tp_stats *sstats;
 454
 455         /* If the pkt at the head of the queue has the nr that we
 456          * expect to send up next, dequeue it and any other
 457          * in-sequence packets behind it.
 458          */
 459 start:
 460         spin_lock_bh(&session->reorder_q.lock);
 461         sstats = &session->stats;
 462         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 463                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 464                         u64_stats_update_begin(&sstats->syncp);
 465                         sstats->rx_seq_discards++;
 466                         sstats->rx_errors++;
 467                         u64_stats_update_end(&sstats->syncp);
 468                         l2tp_dbg(session, L2TP_MSG_SEQ,
 469                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 470                                  session->name, L2TP_SKB_CB(skb)->ns,
 471                                  L2TP_SKB_CB(skb)->length, session->nr,
 472                                  skb_queue_len(&session->reorder_q));
 473                         session->reorder_skip = 1;
 474                         __skb_unlink(skb, &session->reorder_q);
 475                         kfree_skb(skb);
 476                         if (session->deref)
 477                                 (*session->deref)(session);
 478                         continue;
 479                 }
 480
 481                 if (L2TP_SKB_CB(skb)->has_seq) {
 482                         if (session->reorder_skip) {
 483                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 484                                          "%s: advancing nr to next pkt: %u -> %u",
 485                                          session->name, session->nr,
 486                                          L2TP_SKB_CB(skb)->ns);
 487                                 session->reorder_skip = 0;
 488                                 session->nr = L2TP_SKB_CB(skb)->ns;
 489                         }
 490                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 491                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 492                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 493                                          session->name, L2TP_SKB_CB(skb)->ns,
 494                                          L2TP_SKB_CB(skb)->length, session->nr,
 495                                          skb_queue_len(&session->reorder_q));
 496                                 goto out;
 497                         }
 498                 }
 499                 __skb_unlink(skb, &session->reorder_q);
 500
 501                 /* Process the skb. We release the queue lock while we
 502                  * do so to let other contexts process the queue.
 503                  */
 504                 spin_unlock_bh(&session->reorder_q.lock);
 505                 l2tp_recv_dequeue_skb(session, skb);
 506                 goto start;
 507         }
 508
 509 out:
 510         spin_unlock_bh(&session->reorder_q.lock);
 511 }
 512
 513 static inline int l2tp_verify_udp_checksum(struct sock *sk,
 514                                            struct sk_buff *skb)
 515 {
 516         struct udphdr *uh = udp_hdr(skb);
 517         u16 ulen = ntohs(uh->len);
 518         __wsum psum;
 519
 520         if (sk->sk_no_check || skb_csum_unnecessary(skb))
 521                 return 0;
 522
 523 #if IS_ENABLED(CONFIG_IPV6)
 524         if (sk->sk_family == PF_INET6) {
 525                 if (!uh->check) {
 526                         LIMIT_NETDEBUG(KERN_INFO "L2TP: IPv6: checksum is 0\n");
 527                         return 1;
 528                 }
 529                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 530                     !csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 531                                      &ipv6_hdr(skb)->daddr, ulen,
 532                                      IPPROTO_UDP, skb->csum)) {
 533                         skb->ip_summed = CHECKSUM_UNNECESSARY;
 534                         return 0;
 535                 }
 536                 skb->csum = ~csum_unfold(csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
 537                                                          &ipv6_hdr(skb)->daddr,
 538                                                          skb->len, IPPROTO_UDP,
 539                                                          0));
 540         } else
 541 #endif
 542         {
 543                 struct inet_sock *inet;
 544                 if (!uh->check)
 545                         return 0;
 546                 inet = inet_sk(sk);
 547                 psum = csum_tcpudp_nofold(inet->inet_saddr, inet->inet_daddr,
 548                                           ulen, IPPROTO_UDP, 0);
 549
 550                 if ((skb->ip_summed == CHECKSUM_COMPLETE) &&
 551                     !csum_fold(csum_add(psum, skb->csum)))
 552                         return 0;
 553                 skb->csum = psum;
 554         }
 555
 556         return __skb_checksum_complete(skb);
 557 }
 558
 559 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 560  * and L2TPv3 data frames here.
 561  *
 562  * L2TPv2 Data Message Header
 563  *
 564  *  0                   1                   2                   3
 565  *  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
 566  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 567  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 568  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 569  * |           Tunnel ID           |           Session ID          |
 570  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 571  * |             Ns (opt)          |             Nr (opt)          |
 572  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 573  * |      Offset Size (opt)        |    Offset pad... (opt)
 574  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 575  *
 576  * Data frames are marked by T=0. All other fields are the same as
 577  * those in L2TP control frames.
 578  *
 579  * L2TPv3 Data Message Header
 580  *
 581  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 582  * |                      L2TP Session Header                      |
 583  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 584  * |                      L2-Specific Sublayer                     |
 585  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 586  * |                        Tunnel Payload                      ...
 587  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 588  *
 589  * L2TPv3 Session Header Over IP
 590  *
 591  *  0                   1                   2                   3
 592  *  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
 593  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 594  * |                           Session ID                          |
 595  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 596  * |               Cookie (optional, maximum 64 bits)...
 597  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 598  *                                                                 |
 599  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 600  *
 601  * L2TPv3 L2-Specific Sublayer Format
 602  *
 603  *  0                   1                   2                   3
 604  *  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
 605  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 606  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 607  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 608  *
 609  * Cookie value, sublayer format and offset (pad) are negotiated with
 610  * the peer when the session is set up. Unlike L2TPv2, we do not need
 611  * to parse the packet header to determine if optional fields are
 612  * present.
 613  *
 614  * Caller must already have parsed the frame and determined that it is
 615  * a data (not control) frame before coming here. Fields up to the
 616  * session-id have already been parsed and ptr points to the data
 617  * after the session-id.
 618  */
 619 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 620                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 621                       int length, int (*payload_hook)(struct sk_buff *skb))
 622 {
 623         struct l2tp_tunnel *tunnel = session->tunnel;
 624         int offset;
 625         u32 ns, nr;
 626         struct l2tp_stats *sstats = &session->stats;
 627
 628         /* The ref count is increased since we now hold a pointer to
 629          * the session. Take care to decrement the refcnt when exiting
 630          * this function from now on...
 631          */
 632         l2tp_session_inc_refcount(session);
 633         if (session->ref)
 634                 (*session->ref)(session);
 635
 636         /* Parse and check optional cookie */
 637         if (session->peer_cookie_len > 0) {
 638                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 639                         l2tp_info(tunnel, L2TP_MSG_DATA,
 640                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 641                                   tunnel->name, tunnel->tunnel_id,
 642                                   session->session_id);
 643                         u64_stats_update_begin(&sstats->syncp);
 644                         sstats->rx_cookie_discards++;
 645                         u64_stats_update_end(&sstats->syncp);
 646                         goto discard;
 647                 }
 648                 ptr += session->peer_cookie_len;
 649         }
 650
 651         /* Handle the optional sequence numbers. Sequence numbers are
 652          * in different places for L2TPv2 and L2TPv3.
 653          *
 654          * If we are the LAC, enable/disable sequence numbers under
 655          * the control of the LNS.  If no sequence numbers present but
 656          * we were expecting them, discard frame.
 657          */
 658         ns = nr = 0;
 659         L2TP_SKB_CB(skb)->has_seq = 0;
 660         if (tunnel->version == L2TP_HDR_VER_2) {
 661                 if (hdrflags & L2TP_HDRFLAG_S) {
 662                         ns = ntohs(*(__be16 *) ptr);
 663                         ptr += 2;
 664                         nr = ntohs(*(__be16 *) ptr);
 665                         ptr += 2;
 666
 667                         /* Store L2TP info in the skb */
 668                         L2TP_SKB_CB(skb)->ns = ns;
 669                         L2TP_SKB_CB(skb)->has_seq = 1;
 670
 671                         l2tp_dbg(session, L2TP_MSG_SEQ,
 672                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 673                                  session->name, ns, nr, session->nr);
 674                 }
 675         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 676                 u32 l2h = ntohl(*(__be32 *) ptr);
 677
 678                 if (l2h & 0x40000000) {
 679                         ns = l2h & 0x00ffffff;
 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, session nr=%u\n",
 687                                  session->name, ns, session->nr);
 688                 }
 689         }
 690
 691         /* Advance past L2-specific header, if present */
 692         ptr += session->l2specific_len;
 693
 694         if (L2TP_SKB_CB(skb)->has_seq) {
 695                 /* Received a packet with sequence numbers. If we're the LNS,
 696                  * check if we sre sending sequence numbers and if not,
 697                  * configure it so.
 698                  */
 699                 if ((!session->lns_mode) && (!session->send_seq)) {
 700                         l2tp_info(session, L2TP_MSG_SEQ,
 701                                   "%s: requested to enable seq numbers by LNS\n",
 702                                   session->name);
 703                         session->send_seq = -1;
 704                         l2tp_session_set_header_len(session, tunnel->version);
 705                 }
 706         } else {
 707                 /* No sequence numbers.
 708                  * If user has configured mandatory sequence numbers, discard.
 709                  */
 710                 if (session->recv_seq) {
 711                         l2tp_warn(session, L2TP_MSG_SEQ,
 712                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 713                                   session->name);
 714                         u64_stats_update_begin(&sstats->syncp);
 715                         sstats->rx_seq_discards++;
 716                         u64_stats_update_end(&sstats->syncp);
 717                         goto discard;
 718                 }
 719
 720                 /* If we're the LAC and we're sending sequence numbers, the
 721                  * LNS has requested that we no longer send sequence numbers.
 722                  * If we're the LNS and we're sending sequence numbers, the
 723                  * LAC is broken. Discard the frame.
 724                  */
 725                 if ((!session->lns_mode) && (session->send_seq)) {
 726                         l2tp_info(session, L2TP_MSG_SEQ,
 727                                   "%s: requested to disable seq numbers by LNS\n",
 728                                   session->name);
 729                         session->send_seq = 0;
 730                         l2tp_session_set_header_len(session, tunnel->version);
 731                 } else if (session->send_seq) {
 732                         l2tp_warn(session, L2TP_MSG_SEQ,
 733                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 734                                   session->name);
 735                         u64_stats_update_begin(&sstats->syncp);
 736                         sstats->rx_seq_discards++;
 737                         u64_stats_update_end(&sstats->syncp);
 738                         goto discard;
 739                 }
 740         }
 741
 742         /* Session data offset is handled differently for L2TPv2 and
 743          * L2TPv3. For L2TPv2, there is an optional 16-bit value in
 744          * the header. For L2TPv3, the offset is negotiated using AVPs
 745          * in the session setup control protocol.
 746          */
 747         if (tunnel->version == L2TP_HDR_VER_2) {
 748                 /* If offset bit set, skip it. */
 749                 if (hdrflags & L2TP_HDRFLAG_O) {
 750                         offset = ntohs(*(__be16 *)ptr);
 751                         ptr += 2 + offset;
 752                 }
 753         } else
 754                 ptr += session->offset;
 755
 756         offset = ptr - optr;
 757         if (!pskb_may_pull(skb, offset))
 758                 goto discard;
 759
 760         __skb_pull(skb, offset);
 761
 762         /* If caller wants to process the payload before we queue the
 763          * packet, do so now.
 764          */
 765         if (payload_hook)
 766                 if ((*payload_hook)(skb))
 767                         goto discard;
 768
 769         /* Prepare skb for adding to the session's reorder_q.  Hold
 770          * packets for max reorder_timeout or 1 second if not
 771          * reordering.
 772          */
 773         L2TP_SKB_CB(skb)->length = length;
 774         L2TP_SKB_CB(skb)->expires = jiffies +
 775                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 776
 777         /* Add packet to the session's receive queue. Reordering is done here, if
 778          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 779          */
 780         if (L2TP_SKB_CB(skb)->has_seq) {
 781                 if (session->reorder_timeout != 0) {
 782                         /* Packet reordering enabled. Add skb to session's
 783                          * reorder queue, in order of ns.
 784                          */
 785                         l2tp_recv_queue_skb(session, skb);
 786                 } else {
 787                         /* Packet reordering disabled. Discard out-of-sequence
 788                          * packets
 789                          */
 790                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 791                                 u64_stats_update_begin(&sstats->syncp);
 792                                 sstats->rx_seq_discards++;
 793                                 u64_stats_update_end(&sstats->syncp);
 794                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 795                                          "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 796                                          session->name, L2TP_SKB_CB(skb)->ns,
 797                                          L2TP_SKB_CB(skb)->length, session->nr,
 798                                          skb_queue_len(&session->reorder_q));
 799                                 goto discard;
 800                         }
 801                         skb_queue_tail(&session->reorder_q, skb);
 802                 }
 803         } else {
 804                 /* No sequence numbers. Add the skb to the tail of the
 805                  * reorder queue. This ensures that it will be
 806                  * delivered after all previous sequenced skbs.
 807                  */
 808                 skb_queue_tail(&session->reorder_q, skb);
 809         }
 810
 811         /* Try to dequeue as many skbs from reorder_q as we can. */
 812         l2tp_recv_dequeue(session);
 813
 814         l2tp_session_dec_refcount(session);
 815
 816         return;
 817
 818 discard:
 819         u64_stats_update_begin(&sstats->syncp);
 820         sstats->rx_errors++;
 821         u64_stats_update_end(&sstats->syncp);
 822         kfree_skb(skb);
 823
 824         if (session->deref)
 825                 (*session->deref)(session);
 826
 827         l2tp_session_dec_refcount(session);
 828 }
 829 EXPORT_SYMBOL(l2tp_recv_common);
 830
 831 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 832  * here. The skb is not on a list when we get here.
 833  * Returns 0 if the packet was a data packet and was successfully passed on.
 834  * Returns 1 if the packet was not a good data packet and could not be
 835  * forwarded.  All such packets are passed up to userspace to deal with.
 836  */
 837 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb,
 838                               int (*payload_hook)(struct sk_buff *skb))
 839 {
 840         struct l2tp_session *session = NULL;
 841         unsigned char *ptr, *optr;
 842         u16 hdrflags;
 843         u32 tunnel_id, session_id;
 844         u16 version;
 845         int length;
 846         struct l2tp_stats *tstats;
 847
 848         if (tunnel->sock && l2tp_verify_udp_checksum(tunnel->sock, skb))
 849                 goto discard_bad_csum;
 850
 851         /* UDP always verifies the packet length. */
 852         __skb_pull(skb, sizeof(struct udphdr));
 853
 854         /* Short packet? */
 855         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_SEQ)) {
 856                 l2tp_info(tunnel, L2TP_MSG_DATA,
 857                           "%s: recv short packet (len=%d)\n",
 858                           tunnel->name, skb->len);
 859                 goto error;
 860         }
 861
 862         /* Trace packet contents, if enabled */
 863         if (tunnel->debug & L2TP_MSG_DATA) {
 864                 length = min(32u, skb->len);
 865                 if (!pskb_may_pull(skb, length))
 866                         goto error;
 867
 868                 pr_debug("%s: recv\n", tunnel->name);
 869                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 870         }
 871
 872         /* Point to L2TP header */
 873         optr = ptr = skb->data;
 874
 875         /* Get L2TP header flags */
 876         hdrflags = ntohs(*(__be16 *) ptr);
 877
 878         /* Check protocol version */
 879         version = hdrflags & L2TP_HDR_VER_MASK;
 880         if (version != tunnel->version) {
 881                 l2tp_info(tunnel, L2TP_MSG_DATA,
 882                           "%s: recv protocol version mismatch: got %d expected %d\n",
 883                           tunnel->name, version, tunnel->version);
 884                 goto error;
 885         }
 886
 887         /* Get length of L2TP packet */
 888         length = skb->len;
 889
 890         /* If type is control packet, it is handled by userspace. */
 891         if (hdrflags & L2TP_HDRFLAG_T) {
 892                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 893                          "%s: recv control packet, len=%d\n",
 894                          tunnel->name, length);
 895                 goto error;
 896         }
 897
 898         /* Skip flags */
 899         ptr += 2;
 900
 901         if (tunnel->version == L2TP_HDR_VER_2) {
 902                 /* If length is present, skip it */
 903                 if (hdrflags & L2TP_HDRFLAG_L)
 904                         ptr += 2;
 905
 906                 /* Extract tunnel and session ID */
 907                 tunnel_id = ntohs(*(__be16 *) ptr);
 908                 ptr += 2;
 909                 session_id = ntohs(*(__be16 *) ptr);
 910                 ptr += 2;
 911         } else {
 912                 ptr += 2;       /* skip reserved bits */
 913                 tunnel_id = tunnel->tunnel_id;
 914                 session_id = ntohl(*(__be32 *) ptr);
 915                 ptr += 4;
 916         }
 917
 918         /* Find the session context */
 919         session = l2tp_session_find(tunnel->l2tp_net, tunnel, session_id);
 920         if (!session || !session->recv_skb) {
 921                 /* Not found? Pass to userspace to deal with */
 922                 l2tp_info(tunnel, L2TP_MSG_DATA,
 923                           "%s: no session found (%u/%u). Passing up.\n",
 924                           tunnel->name, tunnel_id, session_id);
 925                 goto error;
 926         }
 927
 928         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length, payload_hook);
 929
 930         return 0;
 931
 932 discard_bad_csum:
 933         LIMIT_NETDEBUG("%s: UDP: bad checksum\n", tunnel->name);
 934         UDP_INC_STATS_USER(tunnel->l2tp_net, UDP_MIB_INERRORS, 0);
 935         tstats = &tunnel->stats;
 936         u64_stats_update_begin(&tstats->syncp);
 937         tstats->rx_errors++;
 938         u64_stats_update_end(&tstats->syncp);
 939         kfree_skb(skb);
 940
 941         return 0;
 942
 943 error:
 944         /* Put UDP header back */
 945         __skb_push(skb, sizeof(struct udphdr));
 946
 947         return 1;
 948 }
 949
 950 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 951  * Return codes:
 952  * 0 : success.
 953  * <0: error
 954  * >0: skb should be passed up to userspace as UDP.
 955  */
 956 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 957 {
 958         struct l2tp_tunnel *tunnel;
 959
 960         tunnel = l2tp_sock_to_tunnel(sk);
 961         if (tunnel == NULL)
 962                 goto pass_up;
 963
 964         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 965                  tunnel->name, skb->len);
 966
 967         if (l2tp_udp_recv_core(tunnel, skb, tunnel->recv_payload_hook))
 968                 goto pass_up_put;
 969
 970         sock_put(sk);
 971         return 0;
 972
 973 pass_up_put:
 974         sock_put(sk);
 975 pass_up:
 976         return 1;
 977 }
 978 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 979
 980 /************************************************************************
 981  * Transmit handling
 982  ***********************************************************************/
 983
 984 /* Build an L2TP header for the session into the buffer provided.
 985  */
 986 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 987 {
 988         struct l2tp_tunnel *tunnel = session->tunnel;
 989         __be16 *bufp = buf;
 990         __be16 *optr = buf;
 991         u16 flags = L2TP_HDR_VER_2;
 992         u32 tunnel_id = tunnel->peer_tunnel_id;
 993         u32 session_id = session->peer_session_id;
 994
 995         if (session->send_seq)
 996                 flags |= L2TP_HDRFLAG_S;
 997
 998         /* Setup L2TP header. */
 999         *bufp++ = htons(flags);
1000         *bufp++ = htons(tunnel_id);
1001         *bufp++ = htons(session_id);
1002         if (session->send_seq) {
1003                 *bufp++ = htons(session->ns);
1004                 *bufp++ = 0;
1005                 session->ns++;
1006                 session->ns &= 0xffff;
1007                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
1008                          session->name, session->ns);
1009         }
1010
1011         return bufp - optr;
1012 }
1013
1014 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
1015 {
1016         struct l2tp_tunnel *tunnel = session->tunnel;
1017         char *bufp = buf;
1018         char *optr = bufp;
1019
1020         /* Setup L2TP header. The header differs slightly for UDP and
1021          * IP encapsulations. For UDP, there is 4 bytes of flags.
1022          */
1023         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1024                 u16 flags = L2TP_HDR_VER_3;
1025                 *((__be16 *) bufp) = htons(flags);
1026                 bufp += 2;
1027                 *((__be16 *) bufp) = 0;
1028                 bufp += 2;
1029         }
1030
1031         *((__be32 *) bufp) = htonl(session->peer_session_id);
1032         bufp += 4;
1033         if (session->cookie_len) {
1034                 memcpy(bufp, &session->cookie[0], session->cookie_len);
1035                 bufp += session->cookie_len;
1036         }
1037         if (session->l2specific_len) {
1038                 if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
1039                         u32 l2h = 0;
1040                         if (session->send_seq) {
1041                                 l2h = 0x40000000 | session->ns;
1042                                 session->ns++;
1043                                 session->ns &= 0xffffff;
1044                                 l2tp_dbg(session, L2TP_MSG_SEQ,
1045                                          "%s: updated ns to %u\n",
1046                                          session->name, session->ns);
1047                         }
1048
1049                         *((__be32 *) bufp) = htonl(l2h);
1050                 }
1051                 bufp += session->l2specific_len;
1052         }
1053         if (session->offset)
1054                 bufp += session->offset;
1055
1056         return bufp - optr;
1057 }
1058
1059 static int l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1060                           struct flowi *fl, size_t data_len)
1061 {
1062         struct l2tp_tunnel *tunnel = session->tunnel;
1063         unsigned int len = skb->len;
1064         int error;
1065         struct l2tp_stats *tstats, *sstats;
1066
1067         /* Debug */
1068         if (session->send_seq)
1069                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes, ns=%u\n",
1070                          session->name, data_len, session->ns - 1);
1071         else
1072                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %Zd bytes\n",
1073                          session->name, data_len);
1074
1075         if (session->debug & L2TP_MSG_DATA) {
1076                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1077                 unsigned char *datap = skb->data + uhlen;
1078
1079                 pr_debug("%s: xmit\n", session->name);
1080                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1081                                      datap, min_t(size_t, 32, len - uhlen));
1082         }
1083
1084         /* Queue the packet to IP for output */
1085         skb->local_df = 1;
1086 #if IS_ENABLED(CONFIG_IPV6)
1087         if (skb->sk->sk_family == PF_INET6)
1088                 error = inet6_csk_xmit(skb, NULL);
1089         else
1090 #endif
1091                 error = ip_queue_xmit(skb, fl);
1092
1093         /* Update stats */
1094         tstats = &tunnel->stats;
1095         u64_stats_update_begin(&tstats->syncp);
1096         sstats = &session->stats;
1097         u64_stats_update_begin(&sstats->syncp);
1098         if (error >= 0) {
1099                 tstats->tx_packets++;
1100                 tstats->tx_bytes += len;
1101                 sstats->tx_packets++;
1102                 sstats->tx_bytes += len;
1103         } else {
1104                 tstats->tx_errors++;
1105                 sstats->tx_errors++;
1106         }
1107         u64_stats_update_end(&tstats->syncp);
1108         u64_stats_update_end(&sstats->syncp);
1109
1110         return 0;
1111 }
1112
1113 /* Automatically called when the skb is freed.
1114  */
1115 static void l2tp_sock_wfree(struct sk_buff *skb)
1116 {
1117         sock_put(skb->sk);
1118 }
1119
1120 /* For data skbs that we transmit, we associate with the tunnel socket
1121  * but don't do accounting.
1122  */
1123 static inline void l2tp_skb_set_owner_w(struct sk_buff *skb, struct sock *sk)
1124 {
1125         sock_hold(sk);
1126         skb->sk = sk;
1127         skb->destructor = l2tp_sock_wfree;
1128 }
1129
1130 #if IS_ENABLED(CONFIG_IPV6)
1131 static void l2tp_xmit_ipv6_csum(struct sock *sk, struct sk_buff *skb,
1132                                 int udp_len)
1133 {
1134         struct ipv6_pinfo *np = inet6_sk(sk);
1135         struct udphdr *uh = udp_hdr(skb);
1136
1137         if (!skb_dst(skb) || !skb_dst(skb)->dev ||
1138             !(skb_dst(skb)->dev->features & NETIF_F_IPV6_CSUM)) {
1139                 __wsum csum = skb_checksum(skb, 0, udp_len, 0);
1140                 skb->ip_summed = CHECKSUM_UNNECESSARY;
1141                 uh->check = csum_ipv6_magic(&np->saddr, &np->daddr, udp_len,
1142                                             IPPROTO_UDP, csum);
1143                 if (uh->check == 0)
1144                         uh->check = CSUM_MANGLED_0;
1145         } else {
1146                 skb->ip_summed = CHECKSUM_PARTIAL;
1147                 skb->csum_start = skb_transport_header(skb) - skb->head;
1148                 skb->csum_offset = offsetof(struct udphdr, check);
1149                 uh->check = ~csum_ipv6_magic(&np->saddr, &np->daddr,
1150                                              udp_len, IPPROTO_UDP, 0);
1151         }
1152 }
1153 #endif
1154
1155 /* If caller requires the skb to have a ppp header, the header must be
1156  * inserted in the skb data before calling this function.
1157  */
1158 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1159 {
1160         int data_len = skb->len;
1161         struct l2tp_tunnel *tunnel = session->tunnel;
1162         struct sock *sk = tunnel->sock;
1163         struct flowi *fl;
1164         struct udphdr *uh;
1165         struct inet_sock *inet;
1166         __wsum csum;
1167         int headroom;
1168         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1169         int udp_len;
1170         int ret = NET_XMIT_SUCCESS;
1171
1172         /* Check that there's enough headroom in the skb to insert IP,
1173          * UDP and L2TP headers. If not enough, expand it to
1174          * make room. Adjust truesize.
1175          */
1176         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1177                 uhlen + hdr_len;
1178         if (skb_cow_head(skb, headroom)) {
1179                 kfree_skb(skb);
1180                 return NET_XMIT_DROP;
1181         }
1182
1183         skb_orphan(skb);
1184         /* Setup L2TP header */
1185         session->build_header(session, __skb_push(skb, hdr_len));
1186
1187         /* Reset skb netfilter state */
1188         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1189         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1190                               IPSKB_REROUTED);
1191         nf_reset(skb);
1192
1193         bh_lock_sock(sk);
1194         if (sock_owned_by_user(sk)) {
1195                 kfree_skb(skb);
1196                 ret = NET_XMIT_DROP;
1197                 goto out_unlock;
1198         }
1199
1200         /* Get routing info from the tunnel socket */
1201         skb_dst_drop(skb);
1202         skb_dst_set(skb, dst_clone(__sk_dst_check(sk, 0)));
1203
1204         inet = inet_sk(sk);
1205         fl = &inet->cork.fl;
1206         switch (tunnel->encap) {
1207         case L2TP_ENCAPTYPE_UDP:
1208                 /* Setup UDP header */
1209                 __skb_push(skb, sizeof(*uh));
1210                 skb_reset_transport_header(skb);
1211                 uh = udp_hdr(skb);
1212                 uh->source = inet->inet_sport;
1213                 uh->dest = inet->inet_dport;
1214                 udp_len = uhlen + hdr_len + data_len;
1215                 uh->len = htons(udp_len);
1216                 uh->check = 0;
1217
1218                 /* Calculate UDP checksum if configured to do so */
1219 #if IS_ENABLED(CONFIG_IPV6)
1220                 if (sk->sk_family == PF_INET6)
1221                         l2tp_xmit_ipv6_csum(sk, skb, udp_len);
1222                 else
1223 #endif
1224                 if (sk->sk_no_check == UDP_CSUM_NOXMIT)
1225                         skb->ip_summed = CHECKSUM_NONE;
1226                 else if ((skb_dst(skb) && skb_dst(skb)->dev) &&
1227                          (!(skb_dst(skb)->dev->features & NETIF_F_V4_CSUM))) {
1228                         skb->ip_summed = CHECKSUM_COMPLETE;
1229                         csum = skb_checksum(skb, 0, udp_len, 0);
1230                         uh->check = csum_tcpudp_magic(inet->inet_saddr,
1231                                                       inet->inet_daddr,
1232                                                       udp_len, IPPROTO_UDP, csum);
1233                         if (uh->check == 0)
1234                                 uh->check = CSUM_MANGLED_0;
1235                 } else {
1236                         skb->ip_summed = CHECKSUM_PARTIAL;
1237                         skb->csum_start = skb_transport_header(skb) - skb->head;
1238                         skb->csum_offset = offsetof(struct udphdr, check);
1239                         uh->check = ~csum_tcpudp_magic(inet->inet_saddr,
1240                                                        inet->inet_daddr,
1241                                                        udp_len, IPPROTO_UDP, 0);
1242                 }
1243                 break;
1244
1245         case L2TP_ENCAPTYPE_IP:
1246                 break;
1247         }
1248
1249         l2tp_skb_set_owner_w(skb, sk);
1250
1251         l2tp_xmit_core(session, skb, fl, data_len);
1252 out_unlock:
1253         bh_unlock_sock(sk);
1254
1255         return ret;
1256 }
1257 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1258
1259 /*****************************************************************************
1260  * Tinnel and session create/destroy.
1261  *****************************************************************************/
1262
1263 /* Tunnel socket destruct hook.
1264  * The tunnel context is deleted only when all session sockets have been
1265  * closed.
1266  */
1267 static void l2tp_tunnel_destruct(struct sock *sk)
1268 {
1269         struct l2tp_tunnel *tunnel;
1270         struct l2tp_net *pn;
1271
1272         tunnel = sk->sk_user_data;
1273         if (tunnel == NULL)
1274                 goto end;
1275
1276         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1277
1278
1279         /* Disable udp encapsulation */
1280         switch (tunnel->encap) {
1281         case L2TP_ENCAPTYPE_UDP:
1282                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1283                 (udp_sk(sk))->encap_type = 0;
1284                 (udp_sk(sk))->encap_rcv = NULL;
1285                 break;
1286         case L2TP_ENCAPTYPE_IP:
1287                 break;
1288         }
1289
1290         /* Remove hooks into tunnel socket */
1291         sk->sk_destruct = tunnel->old_sk_destruct;
1292         sk->sk_user_data = NULL;
1293         tunnel->sock = NULL;
1294
1295         /* Remove the tunnel struct from the tunnel list */
1296         pn = l2tp_pernet(tunnel->l2tp_net);
1297         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1298         list_del_rcu(&tunnel->list);
1299         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1300         atomic_dec(&l2tp_tunnel_count);
1301
1302         l2tp_tunnel_closeall(tunnel);
1303         l2tp_tunnel_dec_refcount(tunnel);
1304
1305         /* Call the original destructor */
1306         if (sk->sk_destruct)
1307                 (*sk->sk_destruct)(sk);
1308 end:
1309         return;
1310 }
1311
1312 /* When the tunnel is closed, all the attached sessions need to go too.
1313  */
1314 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1315 {
1316         int hash;
1317         struct hlist_node *walk;
1318         struct hlist_node *tmp;
1319         struct l2tp_session *session;
1320
1321         BUG_ON(tunnel == NULL);
1322
1323         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1324                   tunnel->name);
1325
1326         write_lock_bh(&tunnel->hlist_lock);
1327         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1328 again:
1329                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1330                         session = hlist_entry(walk, struct l2tp_session, hlist);
1331
1332                         l2tp_info(session, L2TP_MSG_CONTROL,
1333                                   "%s: closing session\n", session->name);
1334
1335                         hlist_del_init(&session->hlist);
1336
1337                         /* Since we should hold the sock lock while
1338                          * doing any unbinding, we need to release the
1339                          * lock we're holding before taking that lock.
1340                          * Hold a reference to the sock so it doesn't
1341                          * disappear as we're jumping between locks.
1342                          */
1343                         if (session->ref != NULL)
1344                                 (*session->ref)(session);
1345
1346                         write_unlock_bh(&tunnel->hlist_lock);
1347
1348                         if (tunnel->version != L2TP_HDR_VER_2) {
1349                                 struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1350
1351                                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
1352                                 hlist_del_init_rcu(&session->global_hlist);
1353                                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1354                                 synchronize_rcu();
1355                         }
1356
1357                         if (session->session_close != NULL)
1358                                 (*session->session_close)(session);
1359
1360                         if (session->deref != NULL)
1361                                 (*session->deref)(session);
1362
1363                         write_lock_bh(&tunnel->hlist_lock);
1364
1365                         /* Now restart from the beginning of this hash
1366                          * chain.  We always remove a session from the
1367                          * list so we are guaranteed to make forward
1368                          * progress.
1369                          */
1370                         goto again;
1371                 }
1372         }
1373         write_unlock_bh(&tunnel->hlist_lock);
1374 }
1375
1376 /* Really kill the tunnel.
1377  * Come here only when all sessions have been cleared from the tunnel.
1378  */
1379 static void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
1380 {
1381         BUG_ON(atomic_read(&tunnel->ref_count) != 0);
1382         BUG_ON(tunnel->sock != NULL);
1383         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: free...\n", tunnel->name);
1384         kfree_rcu(tunnel, rcu);
1385 }
1386
1387 /* Workqueue tunnel deletion function */
1388 static void l2tp_tunnel_del_work(struct work_struct *work)
1389 {
1390         struct l2tp_tunnel *tunnel = NULL;
1391         struct socket *sock = NULL;
1392         struct sock *sk = NULL;
1393
1394         tunnel = container_of(work, struct l2tp_tunnel, del_work);
1395         sk = l2tp_tunnel_sock_lookup(tunnel);
1396         if (!sk)
1397                 return;
1398
1399         sock = sk->sk_socket;
1400         BUG_ON(!sock);
1401
1402         /* If the tunnel socket was created directly by the kernel, use the
1403          * sk_* API to release the socket now.  Otherwise go through the
1404          * inet_* layer to shut the socket down, and let userspace close it.
1405          * In either case the tunnel resources are freed in the socket
1406          * destructor when the tunnel socket goes away.
1407          */
1408         if (sock->file == NULL) {
1409                 kernel_sock_shutdown(sock, SHUT_RDWR);
1410                 sk_release_kernel(sk);
1411         } else {
1412                 inet_shutdown(sock, 2);
1413         }
1414
1415         l2tp_tunnel_sock_put(sk);
1416 }
1417
1418 /* Create a socket for the tunnel, if one isn't set up by
1419  * userspace. This is used for static tunnels where there is no
1420  * managing L2TP daemon.
1421  *
1422  * Since we don't want these sockets to keep a namespace alive by
1423  * themselves, we drop the socket's namespace refcount after creation.
1424  * These sockets are freed when the namespace exits using the pernet
1425  * exit hook.
1426  */
1427 static int l2tp_tunnel_sock_create(struct net *net,
1428                                 u32 tunnel_id,
1429                                 u32 peer_tunnel_id,
1430                                 struct l2tp_tunnel_cfg *cfg,
1431                                 struct socket **sockp)
1432 {
1433         int err = -EINVAL;
1434         struct socket *sock = NULL;
1435         struct sockaddr_in udp_addr = {0};
1436         struct sockaddr_l2tpip ip_addr = {0};
1437 #if IS_ENABLED(CONFIG_IPV6)
1438         struct sockaddr_in6 udp6_addr = {0};
1439         struct sockaddr_l2tpip6 ip6_addr = {0};
1440 #endif
1441
1442         switch (cfg->encap) {
1443         case L2TP_ENCAPTYPE_UDP:
1444 #if IS_ENABLED(CONFIG_IPV6)
1445                 if (cfg->local_ip6 && cfg->peer_ip6) {
1446                         err = sock_create_kern(AF_INET6, SOCK_DGRAM, 0, &sock);
1447                         if (err < 0)
1448                                 goto out;
1449
1450                         sk_change_net(sock->sk, net);
1451
1452                         udp6_addr.sin6_family = AF_INET6;
1453                         memcpy(&udp6_addr.sin6_addr, cfg->local_ip6,
1454                                sizeof(udp6_addr.sin6_addr));
1455                         udp6_addr.sin6_port = htons(cfg->local_udp_port);
1456                         err = kernel_bind(sock, (struct sockaddr *) &udp6_addr,
1457                                           sizeof(udp6_addr));
1458                         if (err < 0)
1459                                 goto out;
1460
1461                         udp6_addr.sin6_family = AF_INET6;
1462                         memcpy(&udp6_addr.sin6_addr, cfg->peer_ip6,
1463                                sizeof(udp6_addr.sin6_addr));
1464                         udp6_addr.sin6_port = htons(cfg->peer_udp_port);
1465                         err = kernel_connect(sock,
1466                                              (struct sockaddr *) &udp6_addr,
1467                                              sizeof(udp6_addr), 0);
1468                         if (err < 0)
1469                                 goto out;
1470                 } else
1471 #endif
1472                 {
1473                         err = sock_create_kern(AF_INET, SOCK_DGRAM, 0, &sock);
1474                         if (err < 0)
1475                                 goto out;
1476
1477                         sk_change_net(sock->sk, net);
1478
1479                         udp_addr.sin_family = AF_INET;
1480                         udp_addr.sin_addr = cfg->local_ip;
1481                         udp_addr.sin_port = htons(cfg->local_udp_port);
1482                         err = kernel_bind(sock, (struct sockaddr *) &udp_addr,
1483                                           sizeof(udp_addr));
1484                         if (err < 0)
1485                                 goto out;
1486
1487                         udp_addr.sin_family = AF_INET;
1488                         udp_addr.sin_addr = cfg->peer_ip;
1489                         udp_addr.sin_port = htons(cfg->peer_udp_port);
1490                         err = kernel_connect(sock,
1491                                              (struct sockaddr *) &udp_addr,
1492                                              sizeof(udp_addr), 0);
1493                         if (err < 0)
1494                                 goto out;
1495                 }
1496
1497                 if (!cfg->use_udp_checksums)
1498                         sock->sk->sk_no_check = UDP_CSUM_NOXMIT;
1499
1500                 break;
1501
1502         case L2TP_ENCAPTYPE_IP:
1503 #if IS_ENABLED(CONFIG_IPV6)
1504                 if (cfg->local_ip6 && cfg->peer_ip6) {
1505                         err = sock_create_kern(AF_INET6, SOCK_DGRAM,
1506                                           IPPROTO_L2TP, &sock);
1507                         if (err < 0)
1508                                 goto out;
1509
1510                         sk_change_net(sock->sk, net);
1511
1512                         ip6_addr.l2tp_family = AF_INET6;
1513                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1514                                sizeof(ip6_addr.l2tp_addr));
1515                         ip6_addr.l2tp_conn_id = tunnel_id;
1516                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1517                                           sizeof(ip6_addr));
1518                         if (err < 0)
1519                                 goto out;
1520
1521                         ip6_addr.l2tp_family = AF_INET6;
1522                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1523                                sizeof(ip6_addr.l2tp_addr));
1524                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1525                         err = kernel_connect(sock,
1526                                              (struct sockaddr *) &ip6_addr,
1527                                              sizeof(ip6_addr), 0);
1528                         if (err < 0)
1529                                 goto out;
1530                 } else
1531 #endif
1532                 {
1533                         err = sock_create_kern(AF_INET, SOCK_DGRAM,
1534                                           IPPROTO_L2TP, &sock);
1535                         if (err < 0)
1536                                 goto out;
1537
1538                         sk_change_net(sock->sk, net);
1539
1540                         ip_addr.l2tp_family = AF_INET;
1541                         ip_addr.l2tp_addr = cfg->local_ip;
1542                         ip_addr.l2tp_conn_id = tunnel_id;
1543                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1544                                           sizeof(ip_addr));
1545                         if (err < 0)
1546                                 goto out;
1547
1548                         ip_addr.l2tp_family = AF_INET;
1549                         ip_addr.l2tp_addr = cfg->peer_ip;
1550                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1551                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1552                                              sizeof(ip_addr), 0);
1553                         if (err < 0)
1554                                 goto out;
1555                 }
1556                 break;
1557
1558         default:
1559                 goto out;
1560         }
1561
1562 out:
1563         *sockp = sock;
1564         if ((err < 0) && sock) {
1565                 kernel_sock_shutdown(sock, SHUT_RDWR);
1566                 sk_release_kernel(sock->sk);
1567                 *sockp = NULL;
1568         }
1569
1570         return err;
1571 }
1572
1573 static struct lock_class_key l2tp_socket_class;
1574
1575 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)
1576 {
1577         struct l2tp_tunnel *tunnel = NULL;
1578         int err;
1579         struct socket *sock = NULL;
1580         struct sock *sk = NULL;
1581         struct l2tp_net *pn;
1582         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1583
1584         /* Get the tunnel socket from the fd, which was opened by
1585          * the userspace L2TP daemon. If not specified, create a
1586          * kernel socket.
1587          */
1588         if (fd < 0) {
1589                 err = l2tp_tunnel_sock_create(net, tunnel_id, peer_tunnel_id,
1590                                 cfg, &sock);
1591                 if (err < 0)
1592                         goto err;
1593         } else {
1594                 sock = sockfd_lookup(fd, &err);
1595                 if (!sock) {
1596                         pr_err("tunl %u: sockfd_lookup(fd=%d) returned %d\n",
1597                                tunnel_id, fd, err);
1598                         err = -EBADF;
1599                         goto err;
1600                 }
1601
1602                 /* Reject namespace mismatches */
1603                 if (!net_eq(sock_net(sock->sk), net)) {
1604                         pr_err("tunl %u: netns mismatch\n", tunnel_id);
1605                         err = -EINVAL;
1606                         goto err;
1607                 }
1608         }
1609
1610         sk = sock->sk;
1611
1612         if (cfg != NULL)
1613                 encap = cfg->encap;
1614
1615         /* Quick sanity checks */
1616         switch (encap) {
1617         case L2TP_ENCAPTYPE_UDP:
1618                 err = -EPROTONOSUPPORT;
1619                 if (sk->sk_protocol != IPPROTO_UDP) {
1620                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1621                                tunnel_id, fd, sk->sk_protocol, IPPROTO_UDP);
1622                         goto err;
1623                 }
1624                 break;
1625         case L2TP_ENCAPTYPE_IP:
1626                 err = -EPROTONOSUPPORT;
1627                 if (sk->sk_protocol != IPPROTO_L2TP) {
1628                         pr_err("tunl %hu: fd %d wrong protocol, got %d, expected %d\n",
1629                                tunnel_id, fd, sk->sk_protocol, IPPROTO_L2TP);
1630                         goto err;
1631                 }
1632                 break;
1633         }
1634
1635         /* Check if this socket has already been prepped */
1636         tunnel = (struct l2tp_tunnel *)sk->sk_user_data;
1637         if (tunnel != NULL) {
1638                 /* This socket has already been prepped */
1639                 err = -EBUSY;
1640                 goto err;
1641         }
1642
1643         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1644         if (tunnel == NULL) {
1645                 err = -ENOMEM;
1646                 goto err;
1647         }
1648
1649         tunnel->version = version;
1650         tunnel->tunnel_id = tunnel_id;
1651         tunnel->peer_tunnel_id = peer_tunnel_id;
1652         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1653
1654         tunnel->magic = L2TP_TUNNEL_MAGIC;
1655         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1656         rwlock_init(&tunnel->hlist_lock);
1657
1658         /* The net we belong to */
1659         tunnel->l2tp_net = net;
1660         pn = l2tp_pernet(net);
1661
1662         if (cfg != NULL)
1663                 tunnel->debug = cfg->debug;
1664
1665         /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1666         tunnel->encap = encap;
1667         if (encap == L2TP_ENCAPTYPE_UDP) {
1668                 /* Mark socket as an encapsulation socket. See net/ipv4/udp.c */
1669                 udp_sk(sk)->encap_type = UDP_ENCAP_L2TPINUDP;
1670                 udp_sk(sk)->encap_rcv = l2tp_udp_encap_recv;
1671 #if IS_ENABLED(CONFIG_IPV6)
1672                 if (sk->sk_family == PF_INET6)
1673                         udpv6_encap_enable();
1674                 else
1675 #endif
1676                 udp_encap_enable();
1677         }
1678
1679         sk->sk_user_data = tunnel;
1680
1681         /* Hook on the tunnel socket destructor so that we can cleanup
1682          * if the tunnel socket goes away.
1683          */
1684         tunnel->old_sk_destruct = sk->sk_destruct;
1685         sk->sk_destruct = &l2tp_tunnel_destruct;
1686         tunnel->sock = sk;
1687         tunnel->fd = fd;
1688         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class, "l2tp_sock");
1689
1690         sk->sk_allocation = GFP_ATOMIC;
1691
1692         /* Init delete workqueue struct */
1693         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1694
1695         /* Add tunnel to our list */
1696         INIT_LIST_HEAD(&tunnel->list);
1697         atomic_inc(&l2tp_tunnel_count);
1698
1699         /* Bump the reference count. The tunnel context is deleted
1700          * only when this drops to zero. Must be done before list insertion
1701          */
1702         l2tp_tunnel_inc_refcount(tunnel);
1703         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1704         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1705         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1706
1707         err = 0;
1708 err:
1709         if (tunnelp)
1710                 *tunnelp = tunnel;
1711
1712         /* If tunnel's socket was created by the kernel, it doesn't
1713          *  have a file.
1714          */
1715         if (sock && sock->file)
1716                 sockfd_put(sock);
1717
1718         return err;
1719 }
1720 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1721
1722 /* This function is used by the netlink TUNNEL_DELETE command.
1723  */
1724 int l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1725 {
1726         return (false == queue_work(l2tp_wq, &tunnel->del_work));
1727 }
1728 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1729
1730 /* Really kill the session.
1731  */
1732 void l2tp_session_free(struct l2tp_session *session)
1733 {
1734         struct l2tp_tunnel *tunnel;
1735
1736         BUG_ON(atomic_read(&session->ref_count) != 0);
1737
1738         tunnel = session->tunnel;
1739         if (tunnel != NULL) {
1740                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1741
1742                 /* Delete the session from the hash */
1743                 write_lock_bh(&tunnel->hlist_lock);
1744                 hlist_del_init(&session->hlist);
1745                 write_unlock_bh(&tunnel->hlist_lock);
1746
1747                 /* Unlink from the global hash if not L2TPv2 */
1748                 if (tunnel->version != L2TP_HDR_VER_2) {
1749                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1750
1751                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1752                         hlist_del_init_rcu(&session->global_hlist);
1753                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1754                         synchronize_rcu();
1755                 }
1756
1757                 if (session->session_id != 0)
1758                         atomic_dec(&l2tp_session_count);
1759
1760                 sock_put(tunnel->sock);
1761
1762                 /* This will delete the tunnel context if this
1763                  * is the last session on the tunnel.
1764                  */
1765                 session->tunnel = NULL;
1766                 l2tp_tunnel_dec_refcount(tunnel);
1767         }
1768
1769         kfree(session);
1770
1771         return;
1772 }
1773 EXPORT_SYMBOL_GPL(l2tp_session_free);
1774
1775 /* This function is used by the netlink SESSION_DELETE command and by
1776    pseudowire modules.
1777  */
1778 int l2tp_session_delete(struct l2tp_session *session)
1779 {
1780         if (session->session_close != NULL)
1781                 (*session->session_close)(session);
1782
1783         l2tp_session_dec_refcount(session);
1784
1785         return 0;
1786 }
1787 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1788
1789
1790 /* We come here whenever a session's send_seq, cookie_len or
1791  * l2specific_len parameters are set.
1792  */
1793 static void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1794 {
1795         if (version == L2TP_HDR_VER_2) {
1796                 session->hdr_len = 6;
1797                 if (session->send_seq)
1798                         session->hdr_len += 4;
1799         } else {
1800                 session->hdr_len = 4 + session->cookie_len + session->l2specific_len + session->offset;
1801                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1802                         session->hdr_len += 4;
1803         }
1804
1805 }
1806
1807 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)
1808 {
1809         struct l2tp_session *session;
1810
1811         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1812         if (session != NULL) {
1813                 session->magic = L2TP_SESSION_MAGIC;
1814                 session->tunnel = tunnel;
1815
1816                 session->session_id = session_id;
1817                 session->peer_session_id = peer_session_id;
1818                 session->nr = 0;
1819
1820                 sprintf(&session->name[0], "sess %u/%u",
1821                         tunnel->tunnel_id, session->session_id);
1822
1823                 skb_queue_head_init(&session->reorder_q);
1824
1825                 INIT_HLIST_NODE(&session->hlist);
1826                 INIT_HLIST_NODE(&session->global_hlist);
1827
1828                 /* Inherit debug options from tunnel */
1829                 session->debug = tunnel->debug;
1830
1831                 if (cfg) {
1832                         session->pwtype = cfg->pw_type;
1833                         session->debug = cfg->debug;
1834                         session->mtu = cfg->mtu;
1835                         session->mru = cfg->mru;
1836                         session->send_seq = cfg->send_seq;
1837                         session->recv_seq = cfg->recv_seq;
1838                         session->lns_mode = cfg->lns_mode;
1839                         session->reorder_timeout = cfg->reorder_timeout;
1840                         session->offset = cfg->offset;
1841                         session->l2specific_type = cfg->l2specific_type;
1842                         session->l2specific_len = cfg->l2specific_len;
1843                         session->cookie_len = cfg->cookie_len;
1844                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1845                         session->peer_cookie_len = cfg->peer_cookie_len;
1846                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1847                 }
1848
1849                 if (tunnel->version == L2TP_HDR_VER_2)
1850                         session->build_header = l2tp_build_l2tpv2_header;
1851                 else
1852                         session->build_header = l2tp_build_l2tpv3_header;
1853
1854                 l2tp_session_set_header_len(session, tunnel->version);
1855
1856                 /* Bump the reference count. The session context is deleted
1857                  * only when this drops to zero.
1858                  */
1859                 l2tp_session_inc_refcount(session);
1860                 l2tp_tunnel_inc_refcount(tunnel);
1861
1862                 /* Ensure tunnel socket isn't deleted */
1863                 sock_hold(tunnel->sock);
1864
1865                 /* Add session to the tunnel's hash list */
1866                 write_lock_bh(&tunnel->hlist_lock);
1867                 hlist_add_head(&session->hlist,
1868                                l2tp_session_id_hash(tunnel, session_id));
1869                 write_unlock_bh(&tunnel->hlist_lock);
1870
1871                 /* And to the global session list if L2TPv3 */
1872                 if (tunnel->version != L2TP_HDR_VER_2) {
1873                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1874
1875                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1876                         hlist_add_head_rcu(&session->global_hlist,
1877                                            l2tp_session_id_hash_2(pn, session_id));
1878                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1879                 }
1880
1881                 /* Ignore management session in session count value */
1882                 if (session->session_id != 0)
1883                         atomic_inc(&l2tp_session_count);
1884         }
1885
1886         return session;
1887 }
1888 EXPORT_SYMBOL_GPL(l2tp_session_create);
1889
1890 /*****************************************************************************
1891  * Init and cleanup
1892  *****************************************************************************/
1893
1894 static __net_init int l2tp_init_net(struct net *net)
1895 {
1896         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1897         int hash;
1898
1899         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1900         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1901
1902         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1903                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1904
1905         spin_lock_init(&pn->l2tp_session_hlist_lock);
1906
1907         return 0;
1908 }
1909
1910 static __net_exit void l2tp_exit_net(struct net *net)
1911 {
1912         struct l2tp_net *pn = l2tp_pernet(net);
1913         struct l2tp_tunnel *tunnel = NULL;
1914
1915         rcu_read_lock_bh();
1916         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1917                 (void)l2tp_tunnel_delete(tunnel);
1918         }
1919         rcu_read_unlock_bh();
1920 }
1921
1922 static struct pernet_operations l2tp_net_ops = {
1923         .init = l2tp_init_net,
1924         .exit = l2tp_exit_net,
1925         .id   = &l2tp_net_id,
1926         .size = sizeof(struct l2tp_net),
1927 };
1928
1929 static int __init l2tp_init(void)
1930 {
1931         int rc = 0;
1932
1933         rc = register_pernet_device(&l2tp_net_ops);
1934         if (rc)
1935                 goto out;
1936
1937         l2tp_wq = alloc_workqueue("l2tp", WQ_NON_REENTRANT | WQ_UNBOUND, 0);
1938         if (!l2tp_wq) {
1939                 pr_err("alloc_workqueue failed\n");
1940                 rc = -ENOMEM;
1941                 goto out;
1942         }
1943
1944         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1945
1946 out:
1947         return rc;
1948 }
1949
1950 static void __exit l2tp_exit(void)
1951 {
1952         unregister_pernet_device(&l2tp_net_ops);
1953         if (l2tp_wq) {
1954                 destroy_workqueue(l2tp_wq);
1955                 l2tp_wq = NULL;
1956         }
1957 }
1958
1959 module_init(l2tp_init);
1960 module_exit(l2tp_exit);
1961
1962 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1963 MODULE_DESCRIPTION("L2TP core");
1964 MODULE_LICENSE("GPL");
1965 MODULE_VERSION(L2TP_DRV_VERSION);
1966