net/l2tp/l2tp_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * L2TP core.
   4  *
   5  * Copyright (c) 2008,2009,2010 Katalix Systems Ltd
   6  *
   7  * This file contains some code of the original L2TPv2 pppol2tp
   8  * driver, which has the following copyright:
   9  *
  10  * Authors:     Martijn van Oosterhout <kleptog@svana.org>
  11  *              James Chapman (jchapman@katalix.com)
  12  * Contributors:
  13  *              Michal Ostrowski <mostrows@speakeasy.net>
  14  *              Arnaldo Carvalho de Melo <acme@xconectiva.com.br>
  15  *              David S. Miller (davem@redhat.com)
  16  */
  17
  18 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  19
  20 #include <linux/module.h>
  21 #include <linux/string.h>
  22 #include <linux/list.h>
  23 #include <linux/rculist.h>
  24 #include <linux/uaccess.h>
  25
  26 #include <linux/kernel.h>
  27 #include <linux/spinlock.h>
  28 #include <linux/kthread.h>
  29 #include <linux/sched.h>
  30 #include <linux/slab.h>
  31 #include <linux/errno.h>
  32 #include <linux/jiffies.h>
  33
  34 #include <linux/netdevice.h>
  35 #include <linux/net.h>
  36 #include <linux/inetdevice.h>
  37 #include <linux/skbuff.h>
  38 #include <linux/init.h>
  39 #include <linux/in.h>
  40 #include <linux/ip.h>
  41 #include <linux/udp.h>
  42 #include <linux/l2tp.h>
  43 #include <linux/hash.h>
  44 #include <linux/sort.h>
  45 #include <linux/file.h>
  46 #include <linux/nsproxy.h>
  47 #include <net/net_namespace.h>
  48 #include <net/netns/generic.h>
  49 #include <net/dst.h>
  50 #include <net/ip.h>
  51 #include <net/udp.h>
  52 #include <net/udp_tunnel.h>
  53 #include <net/inet_common.h>
  54 #include <net/xfrm.h>
  55 #include <net/protocol.h>
  56 #include <net/inet6_connection_sock.h>
  57 #include <net/inet_ecn.h>
  58 #include <net/ip6_route.h>
  59 #include <net/ip6_checksum.h>
  60
  61 #include <asm/byteorder.h>
  62 #include <linux/atomic.h>
  63
  64 #include "l2tp_core.h"
  65
  66 #define L2TP_DRV_VERSION        "V2.0"
  67
  68 /* L2TP header constants */
  69 #define L2TP_HDRFLAG_T     0x8000
  70 #define L2TP_HDRFLAG_L     0x4000
  71 #define L2TP_HDRFLAG_S     0x0800
  72 #define L2TP_HDRFLAG_O     0x0200
  73 #define L2TP_HDRFLAG_P     0x0100
  74
  75 #define L2TP_HDR_VER_MASK  0x000F
  76 #define L2TP_HDR_VER_2     0x0002
  77 #define L2TP_HDR_VER_3     0x0003
  78
  79 /* L2TPv3 default L2-specific sublayer */
  80 #define L2TP_SLFLAG_S      0x40000000
  81 #define L2TP_SL_SEQ_MASK   0x00ffffff
  82
  83 #define L2TP_HDR_SIZE_MAX               14
  84
  85 /* Default trace flags */
  86 #define L2TP_DEFAULT_DEBUG_FLAGS        0
  87
  88 /* Private data stored for received packets in the skb.
  89  */
  90 struct l2tp_skb_cb {
  91         u32                     ns;
  92         u16                     has_seq;
  93         u16                     length;
  94         unsigned long           expires;
  95 };
  96
  97 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *) &skb->cb[sizeof(struct inet_skb_parm)])
  98
  99 static struct workqueue_struct *l2tp_wq;
 100
 101 /* per-net private data for this module */
 102 static unsigned int l2tp_net_id;
 103 struct l2tp_net {
 104         struct list_head l2tp_tunnel_list;
 105         spinlock_t l2tp_tunnel_list_lock;
 106         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
 107         spinlock_t l2tp_session_hlist_lock;
 108 };
 109
 110 #if IS_ENABLED(CONFIG_IPV6)
 111 static bool l2tp_sk_is_v6(struct sock *sk)
 112 {
 113         return sk->sk_family == PF_INET6 &&
 114                !ipv6_addr_v4mapped(&sk->sk_v6_daddr);
 115 }
 116 #endif
 117
 118 static inline struct l2tp_tunnel *l2tp_tunnel(struct sock *sk)
 119 {
 120         return sk->sk_user_data;
 121 }
 122
 123 static inline struct l2tp_net *l2tp_pernet(const struct net *net)
 124 {
 125         return net_generic(net, l2tp_net_id);
 126 }
 127
 128 /* Session hash global list for L2TPv3.
 129  * The session_id SHOULD be random according to RFC3931, but several
 130  * L2TP implementations use incrementing session_ids.  So we do a real
 131  * hash on the session_id, rather than a simple bitmask.
 132  */
 133 static inline struct hlist_head *
 134 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 135 {
 136         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 137
 138 }
 139
 140 /* Session hash list.
 141  * The session_id SHOULD be random according to RFC2661, but several
 142  * L2TP implementations (Cisco and Microsoft) use incrementing
 143  * session_ids.  So we do a real hash on the session_id, rather than a
 144  * simple bitmask.
 145  */
 146 static inline struct hlist_head *
 147 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 148 {
 149         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 150 }
 151
 152 void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
 153 {
 154         sock_put(tunnel->sock);
 155         /* the tunnel is freed in the socket destructor */
 156 }
 157 EXPORT_SYMBOL(l2tp_tunnel_free);
 158
 159 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
 160 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
 161 {
 162         const struct l2tp_net *pn = l2tp_pernet(net);
 163         struct l2tp_tunnel *tunnel;
 164
 165         rcu_read_lock_bh();
 166         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 167                 if (tunnel->tunnel_id == tunnel_id &&
 168                     refcount_inc_not_zero(&tunnel->ref_count)) {
 169                         rcu_read_unlock_bh();
 170
 171                         return tunnel;
 172                 }
 173         }
 174         rcu_read_unlock_bh();
 175
 176         return NULL;
 177 }
 178 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
 179
 180 struct l2tp_tunnel *l2tp_tunnel_get_nth(const struct net *net, int nth)
 181 {
 182         const struct l2tp_net *pn = l2tp_pernet(net);
 183         struct l2tp_tunnel *tunnel;
 184         int count = 0;
 185
 186         rcu_read_lock_bh();
 187         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 188                 if (++count > nth &&
 189                     refcount_inc_not_zero(&tunnel->ref_count)) {
 190                         rcu_read_unlock_bh();
 191                         return tunnel;
 192                 }
 193         }
 194         rcu_read_unlock_bh();
 195
 196         return NULL;
 197 }
 198 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_nth);
 199
 200 struct l2tp_session *l2tp_tunnel_get_session(struct l2tp_tunnel *tunnel,
 201                                              u32 session_id)
 202 {
 203         struct hlist_head *session_list;
 204         struct l2tp_session *session;
 205
 206         session_list = l2tp_session_id_hash(tunnel, session_id);
 207
 208         read_lock_bh(&tunnel->hlist_lock);
 209         hlist_for_each_entry(session, session_list, hlist)
 210                 if (session->session_id == session_id) {
 211                         l2tp_session_inc_refcount(session);
 212                         read_unlock_bh(&tunnel->hlist_lock);
 213
 214                         return session;
 215                 }
 216         read_unlock_bh(&tunnel->hlist_lock);
 217
 218         return NULL;
 219 }
 220 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_session);
 221
 222 struct l2tp_session *l2tp_session_get(const struct net *net, u32 session_id)
 223 {
 224         struct hlist_head *session_list;
 225         struct l2tp_session *session;
 226
 227         session_list = l2tp_session_id_hash_2(l2tp_pernet(net), session_id);
 228
 229         rcu_read_lock_bh();
 230         hlist_for_each_entry_rcu(session, session_list, global_hlist)
 231                 if (session->session_id == session_id) {
 232                         l2tp_session_inc_refcount(session);
 233                         rcu_read_unlock_bh();
 234
 235                         return session;
 236                 }
 237         rcu_read_unlock_bh();
 238
 239         return NULL;
 240 }
 241 EXPORT_SYMBOL_GPL(l2tp_session_get);
 242
 243 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth)
 244 {
 245         int hash;
 246         struct l2tp_session *session;
 247         int count = 0;
 248
 249         read_lock_bh(&tunnel->hlist_lock);
 250         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 251                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 252                         if (++count > nth) {
 253                                 l2tp_session_inc_refcount(session);
 254                                 read_unlock_bh(&tunnel->hlist_lock);
 255                                 return session;
 256                         }
 257                 }
 258         }
 259
 260         read_unlock_bh(&tunnel->hlist_lock);
 261
 262         return NULL;
 263 }
 264 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
 265
 266 /* Lookup a session by interface name.
 267  * This is very inefficient but is only used by management interfaces.
 268  */
 269 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
 270                                                 const char *ifname)
 271 {
 272         struct l2tp_net *pn = l2tp_pernet(net);
 273         int hash;
 274         struct l2tp_session *session;
 275
 276         rcu_read_lock_bh();
 277         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 278                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 279                         if (!strcmp(session->ifname, ifname)) {
 280                                 l2tp_session_inc_refcount(session);
 281                                 rcu_read_unlock_bh();
 282
 283                                 return session;
 284                         }
 285                 }
 286         }
 287
 288         rcu_read_unlock_bh();
 289
 290         return NULL;
 291 }
 292 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
 293
 294 int l2tp_session_register(struct l2tp_session *session,
 295                           struct l2tp_tunnel *tunnel)
 296 {
 297         struct l2tp_session *session_walk;
 298         struct hlist_head *g_head;
 299         struct hlist_head *head;
 300         struct l2tp_net *pn;
 301         int err;
 302
 303         head = l2tp_session_id_hash(tunnel, session->session_id);
 304
 305         write_lock_bh(&tunnel->hlist_lock);
 306         if (!tunnel->acpt_newsess) {
 307                 err = -ENODEV;
 308                 goto err_tlock;
 309         }
 310
 311         hlist_for_each_entry(session_walk, head, hlist)
 312                 if (session_walk->session_id == session->session_id) {
 313                         err = -EEXIST;
 314                         goto err_tlock;
 315                 }
 316
 317         if (tunnel->version == L2TP_HDR_VER_3) {
 318                 pn = l2tp_pernet(tunnel->l2tp_net);
 319                 g_head = l2tp_session_id_hash_2(pn, session->session_id);
 320
 321                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
 322
 323                 hlist_for_each_entry(session_walk, g_head, global_hlist)
 324                         if (session_walk->session_id == session->session_id) {
 325                                 err = -EEXIST;
 326                                 goto err_tlock_pnlock;
 327                         }
 328
 329                 l2tp_tunnel_inc_refcount(tunnel);
 330                 hlist_add_head_rcu(&session->global_hlist, g_head);
 331
 332                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 333         } else {
 334                 l2tp_tunnel_inc_refcount(tunnel);
 335         }
 336
 337         hlist_add_head(&session->hlist, head);
 338         write_unlock_bh(&tunnel->hlist_lock);
 339
 340         return 0;
 341
 342 err_tlock_pnlock:
 343         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 344 err_tlock:
 345         write_unlock_bh(&tunnel->hlist_lock);
 346
 347         return err;
 348 }
 349 EXPORT_SYMBOL_GPL(l2tp_session_register);
 350
 351 /*****************************************************************************
 352  * Receive data handling
 353  *****************************************************************************/
 354
 355 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 356  * number.
 357  */
 358 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 359 {
 360         struct sk_buff *skbp;
 361         struct sk_buff *tmp;
 362         u32 ns = L2TP_SKB_CB(skb)->ns;
 363
 364         spin_lock_bh(&session->reorder_q.lock);
 365         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 366                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 367                         __skb_queue_before(&session->reorder_q, skbp, skb);
 368                         l2tp_dbg(session, L2TP_MSG_SEQ,
 369                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 370                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 371                                  skb_queue_len(&session->reorder_q));
 372                         atomic_long_inc(&session->stats.rx_oos_packets);
 373                         goto out;
 374                 }
 375         }
 376
 377         __skb_queue_tail(&session->reorder_q, skb);
 378
 379 out:
 380         spin_unlock_bh(&session->reorder_q.lock);
 381 }
 382
 383 /* Dequeue a single skb.
 384  */
 385 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 386 {
 387         struct l2tp_tunnel *tunnel = session->tunnel;
 388         int length = L2TP_SKB_CB(skb)->length;
 389
 390         /* We're about to requeue the skb, so return resources
 391          * to its current owner (a socket receive buffer).
 392          */
 393         skb_orphan(skb);
 394
 395         atomic_long_inc(&tunnel->stats.rx_packets);
 396         atomic_long_add(length, &tunnel->stats.rx_bytes);
 397         atomic_long_inc(&session->stats.rx_packets);
 398         atomic_long_add(length, &session->stats.rx_bytes);
 399
 400         if (L2TP_SKB_CB(skb)->has_seq) {
 401                 /* Bump our Nr */
 402                 session->nr++;
 403                 session->nr &= session->nr_max;
 404
 405                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 406                          session->name, session->nr);
 407         }
 408
 409         /* call private receive handler */
 410         if (session->recv_skb != NULL)
 411                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 412         else
 413                 kfree_skb(skb);
 414 }
 415
 416 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 417  * Skbs that have been in the queue for too long are simply discarded.
 418  */
 419 static void l2tp_recv_dequeue(struct l2tp_session *session)
 420 {
 421         struct sk_buff *skb;
 422         struct sk_buff *tmp;
 423
 424         /* If the pkt at the head of the queue has the nr that we
 425          * expect to send up next, dequeue it and any other
 426          * in-sequence packets behind it.
 427          */
 428 start:
 429         spin_lock_bh(&session->reorder_q.lock);
 430         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 431                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 432                         atomic_long_inc(&session->stats.rx_seq_discards);
 433                         atomic_long_inc(&session->stats.rx_errors);
 434                         l2tp_dbg(session, L2TP_MSG_SEQ,
 435                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 436                                  session->name, L2TP_SKB_CB(skb)->ns,
 437                                  L2TP_SKB_CB(skb)->length, session->nr,
 438                                  skb_queue_len(&session->reorder_q));
 439                         session->reorder_skip = 1;
 440                         __skb_unlink(skb, &session->reorder_q);
 441                         kfree_skb(skb);
 442                         continue;
 443                 }
 444
 445                 if (L2TP_SKB_CB(skb)->has_seq) {
 446                         if (session->reorder_skip) {
 447                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 448                                          "%s: advancing nr to next pkt: %u -> %u",
 449                                          session->name, session->nr,
 450                                          L2TP_SKB_CB(skb)->ns);
 451                                 session->reorder_skip = 0;
 452                                 session->nr = L2TP_SKB_CB(skb)->ns;
 453                         }
 454                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 455                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 456                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 457                                          session->name, L2TP_SKB_CB(skb)->ns,
 458                                          L2TP_SKB_CB(skb)->length, session->nr,
 459                                          skb_queue_len(&session->reorder_q));
 460                                 goto out;
 461                         }
 462                 }
 463                 __skb_unlink(skb, &session->reorder_q);
 464
 465                 /* Process the skb. We release the queue lock while we
 466                  * do so to let other contexts process the queue.
 467                  */
 468                 spin_unlock_bh(&session->reorder_q.lock);
 469                 l2tp_recv_dequeue_skb(session, skb);
 470                 goto start;
 471         }
 472
 473 out:
 474         spin_unlock_bh(&session->reorder_q.lock);
 475 }
 476
 477 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 478 {
 479         u32 nws;
 480
 481         if (nr >= session->nr)
 482                 nws = nr - session->nr;
 483         else
 484                 nws = (session->nr_max + 1) - (session->nr - nr);
 485
 486         return nws < session->nr_window_size;
 487 }
 488
 489 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 490  * acceptable, else non-zero.
 491  */
 492 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 493 {
 494         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
 495                 /* Packet sequence number is outside allowed window.
 496                  * Discard it.
 497                  */
 498                 l2tp_dbg(session, L2TP_MSG_SEQ,
 499                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
 500                          session->name, L2TP_SKB_CB(skb)->ns,
 501                          L2TP_SKB_CB(skb)->length, session->nr);
 502                 goto discard;
 503         }
 504
 505         if (session->reorder_timeout != 0) {
 506                 /* Packet reordering enabled. Add skb to session's
 507                  * reorder queue, in order of ns.
 508                  */
 509                 l2tp_recv_queue_skb(session, skb);
 510                 goto out;
 511         }
 512
 513         /* Packet reordering disabled. Discard out-of-sequence packets, while
 514          * tracking the number if in-sequence packets after the first OOS packet
 515          * is seen. After nr_oos_count_max in-sequence packets, reset the
 516          * sequence number to re-enable packet reception.
 517          */
 518         if (L2TP_SKB_CB(skb)->ns == session->nr) {
 519                 skb_queue_tail(&session->reorder_q, skb);
 520         } else {
 521                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
 522                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 523
 524                 if (nr_oos == nr_next)
 525                         session->nr_oos_count++;
 526                 else
 527                         session->nr_oos_count = 0;
 528
 529                 session->nr_oos = nr_oos;
 530                 if (session->nr_oos_count > session->nr_oos_count_max) {
 531                         session->reorder_skip = 1;
 532                         l2tp_dbg(session, L2TP_MSG_SEQ,
 533                                  "%s: %d oos packets received. Resetting sequence numbers\n",
 534                                  session->name, session->nr_oos_count);
 535                 }
 536                 if (!session->reorder_skip) {
 537                         atomic_long_inc(&session->stats.rx_seq_discards);
 538                         l2tp_dbg(session, L2TP_MSG_SEQ,
 539                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 540                                  session->name, L2TP_SKB_CB(skb)->ns,
 541                                  L2TP_SKB_CB(skb)->length, session->nr,
 542                                  skb_queue_len(&session->reorder_q));
 543                         goto discard;
 544                 }
 545                 skb_queue_tail(&session->reorder_q, skb);
 546         }
 547
 548 out:
 549         return 0;
 550
 551 discard:
 552         return 1;
 553 }
 554
 555 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 556  * and L2TPv3 data frames here.
 557  *
 558  * L2TPv2 Data Message Header
 559  *
 560  *  0                   1                   2                   3
 561  *  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
 562  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 563  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 564  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 565  * |           Tunnel ID           |           Session ID          |
 566  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 567  * |             Ns (opt)          |             Nr (opt)          |
 568  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 569  * |      Offset Size (opt)        |    Offset pad... (opt)
 570  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 571  *
 572  * Data frames are marked by T=0. All other fields are the same as
 573  * those in L2TP control frames.
 574  *
 575  * L2TPv3 Data Message Header
 576  *
 577  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 578  * |                      L2TP Session Header                      |
 579  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 580  * |                      L2-Specific Sublayer                     |
 581  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 582  * |                        Tunnel Payload                      ...
 583  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 584  *
 585  * L2TPv3 Session Header Over IP
 586  *
 587  *  0                   1                   2                   3
 588  *  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
 589  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 590  * |                           Session ID                          |
 591  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 592  * |               Cookie (optional, maximum 64 bits)...
 593  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 594  *                                                                 |
 595  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 596  *
 597  * L2TPv3 L2-Specific Sublayer Format
 598  *
 599  *  0                   1                   2                   3
 600  *  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
 601  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 602  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 603  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 604  *
 605  * Cookie value and sublayer format are negotiated with the peer when
 606  * the session is set up. Unlike L2TPv2, we do not need to parse the
 607  * packet header to determine if optional fields are present.
 608  *
 609  * Caller must already have parsed the frame and determined that it is
 610  * a data (not control) frame before coming here. Fields up to the
 611  * session-id have already been parsed and ptr points to the data
 612  * after the session-id.
 613  */
 614 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 615                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 616                       int length)
 617 {
 618         struct l2tp_tunnel *tunnel = session->tunnel;
 619         int offset;
 620         u32 ns, nr;
 621
 622         /* Parse and check optional cookie */
 623         if (session->peer_cookie_len > 0) {
 624                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 625                         l2tp_info(tunnel, L2TP_MSG_DATA,
 626                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 627                                   tunnel->name, tunnel->tunnel_id,
 628                                   session->session_id);
 629                         atomic_long_inc(&session->stats.rx_cookie_discards);
 630                         goto discard;
 631                 }
 632                 ptr += session->peer_cookie_len;
 633         }
 634
 635         /* Handle the optional sequence numbers. Sequence numbers are
 636          * in different places for L2TPv2 and L2TPv3.
 637          *
 638          * If we are the LAC, enable/disable sequence numbers under
 639          * the control of the LNS.  If no sequence numbers present but
 640          * we were expecting them, discard frame.
 641          */
 642         ns = nr = 0;
 643         L2TP_SKB_CB(skb)->has_seq = 0;
 644         if (tunnel->version == L2TP_HDR_VER_2) {
 645                 if (hdrflags & L2TP_HDRFLAG_S) {
 646                         ns = ntohs(*(__be16 *) ptr);
 647                         ptr += 2;
 648                         nr = ntohs(*(__be16 *) ptr);
 649                         ptr += 2;
 650
 651                         /* Store L2TP info in the skb */
 652                         L2TP_SKB_CB(skb)->ns = ns;
 653                         L2TP_SKB_CB(skb)->has_seq = 1;
 654
 655                         l2tp_dbg(session, L2TP_MSG_SEQ,
 656                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 657                                  session->name, ns, nr, session->nr);
 658                 }
 659         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 660                 u32 l2h = ntohl(*(__be32 *) ptr);
 661
 662                 if (l2h & 0x40000000) {
 663                         ns = l2h & 0x00ffffff;
 664
 665                         /* Store L2TP info in the skb */
 666                         L2TP_SKB_CB(skb)->ns = ns;
 667                         L2TP_SKB_CB(skb)->has_seq = 1;
 668
 669                         l2tp_dbg(session, L2TP_MSG_SEQ,
 670                                  "%s: recv data ns=%u, session nr=%u\n",
 671                                  session->name, ns, session->nr);
 672                 }
 673                 ptr += 4;
 674         }
 675
 676         if (L2TP_SKB_CB(skb)->has_seq) {
 677                 /* Received a packet with sequence numbers. If we're the LNS,
 678                  * check if we sre sending sequence numbers and if not,
 679                  * configure it so.
 680                  */
 681                 if ((!session->lns_mode) && (!session->send_seq)) {
 682                         l2tp_info(session, L2TP_MSG_SEQ,
 683                                   "%s: requested to enable seq numbers by LNS\n",
 684                                   session->name);
 685                         session->send_seq = 1;
 686                         l2tp_session_set_header_len(session, tunnel->version);
 687                 }
 688         } else {
 689                 /* No sequence numbers.
 690                  * If user has configured mandatory sequence numbers, discard.
 691                  */
 692                 if (session->recv_seq) {
 693                         l2tp_warn(session, L2TP_MSG_SEQ,
 694                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 695                                   session->name);
 696                         atomic_long_inc(&session->stats.rx_seq_discards);
 697                         goto discard;
 698                 }
 699
 700                 /* If we're the LAC and we're sending sequence numbers, the
 701                  * LNS has requested that we no longer send sequence numbers.
 702                  * If we're the LNS and we're sending sequence numbers, the
 703                  * LAC is broken. Discard the frame.
 704                  */
 705                 if ((!session->lns_mode) && (session->send_seq)) {
 706                         l2tp_info(session, L2TP_MSG_SEQ,
 707                                   "%s: requested to disable seq numbers by LNS\n",
 708                                   session->name);
 709                         session->send_seq = 0;
 710                         l2tp_session_set_header_len(session, tunnel->version);
 711                 } else if (session->send_seq) {
 712                         l2tp_warn(session, L2TP_MSG_SEQ,
 713                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 714                                   session->name);
 715                         atomic_long_inc(&session->stats.rx_seq_discards);
 716                         goto discard;
 717                 }
 718         }
 719
 720         /* Session data offset is defined only for L2TPv2 and is
 721          * indicated by an optional 16-bit value in the header.
 722          */
 723         if (tunnel->version == L2TP_HDR_VER_2) {
 724                 /* If offset bit set, skip it. */
 725                 if (hdrflags & L2TP_HDRFLAG_O) {
 726                         offset = ntohs(*(__be16 *)ptr);
 727                         ptr += 2 + offset;
 728                 }
 729         }
 730
 731         offset = ptr - optr;
 732         if (!pskb_may_pull(skb, offset))
 733                 goto discard;
 734
 735         __skb_pull(skb, offset);
 736
 737         /* Prepare skb for adding to the session's reorder_q.  Hold
 738          * packets for max reorder_timeout or 1 second if not
 739          * reordering.
 740          */
 741         L2TP_SKB_CB(skb)->length = length;
 742         L2TP_SKB_CB(skb)->expires = jiffies +
 743                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 744
 745         /* Add packet to the session's receive queue. Reordering is done here, if
 746          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 747          */
 748         if (L2TP_SKB_CB(skb)->has_seq) {
 749                 if (l2tp_recv_data_seq(session, skb))
 750                         goto discard;
 751         } else {
 752                 /* No sequence numbers. Add the skb to the tail of the
 753                  * reorder queue. This ensures that it will be
 754                  * delivered after all previous sequenced skbs.
 755                  */
 756                 skb_queue_tail(&session->reorder_q, skb);
 757         }
 758
 759         /* Try to dequeue as many skbs from reorder_q as we can. */
 760         l2tp_recv_dequeue(session);
 761
 762         return;
 763
 764 discard:
 765         atomic_long_inc(&session->stats.rx_errors);
 766         kfree_skb(skb);
 767 }
 768 EXPORT_SYMBOL(l2tp_recv_common);
 769
 770 /* Drop skbs from the session's reorder_q
 771  */
 772 static int l2tp_session_queue_purge(struct l2tp_session *session)
 773 {
 774         struct sk_buff *skb = NULL;
 775         BUG_ON(!session);
 776         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
 777         while ((skb = skb_dequeue(&session->reorder_q))) {
 778                 atomic_long_inc(&session->stats.rx_errors);
 779                 kfree_skb(skb);
 780         }
 781         return 0;
 782 }
 783
 784 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 785  * here. The skb is not on a list when we get here.
 786  * Returns 0 if the packet was a data packet and was successfully passed on.
 787  * Returns 1 if the packet was not a good data packet and could not be
 788  * forwarded.  All such packets are passed up to userspace to deal with.
 789  */
 790 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb)
 791 {
 792         struct l2tp_session *session = NULL;
 793         unsigned char *ptr, *optr;
 794         u16 hdrflags;
 795         u32 tunnel_id, session_id;
 796         u16 version;
 797         int length;
 798
 799         /* UDP has verifed checksum */
 800
 801         /* UDP always verifies the packet length. */
 802         __skb_pull(skb, sizeof(struct udphdr));
 803
 804         /* Short packet? */
 805         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_MAX)) {
 806                 l2tp_info(tunnel, L2TP_MSG_DATA,
 807                           "%s: recv short packet (len=%d)\n",
 808                           tunnel->name, skb->len);
 809                 goto error;
 810         }
 811
 812         /* Trace packet contents, if enabled */
 813         if (tunnel->debug & L2TP_MSG_DATA) {
 814                 length = min(32u, skb->len);
 815                 if (!pskb_may_pull(skb, length))
 816                         goto error;
 817
 818                 pr_debug("%s: recv\n", tunnel->name);
 819                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 820         }
 821
 822         /* Point to L2TP header */
 823         optr = ptr = skb->data;
 824
 825         /* Get L2TP header flags */
 826         hdrflags = ntohs(*(__be16 *) ptr);
 827
 828         /* Check protocol version */
 829         version = hdrflags & L2TP_HDR_VER_MASK;
 830         if (version != tunnel->version) {
 831                 l2tp_info(tunnel, L2TP_MSG_DATA,
 832                           "%s: recv protocol version mismatch: got %d expected %d\n",
 833                           tunnel->name, version, tunnel->version);
 834                 goto error;
 835         }
 836
 837         /* Get length of L2TP packet */
 838         length = skb->len;
 839
 840         /* If type is control packet, it is handled by userspace. */
 841         if (hdrflags & L2TP_HDRFLAG_T) {
 842                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 843                          "%s: recv control packet, len=%d\n",
 844                          tunnel->name, length);
 845                 goto error;
 846         }
 847
 848         /* Skip flags */
 849         ptr += 2;
 850
 851         if (tunnel->version == L2TP_HDR_VER_2) {
 852                 /* If length is present, skip it */
 853                 if (hdrflags & L2TP_HDRFLAG_L)
 854                         ptr += 2;
 855
 856                 /* Extract tunnel and session ID */
 857                 tunnel_id = ntohs(*(__be16 *) ptr);
 858                 ptr += 2;
 859                 session_id = ntohs(*(__be16 *) ptr);
 860                 ptr += 2;
 861         } else {
 862                 ptr += 2;       /* skip reserved bits */
 863                 tunnel_id = tunnel->tunnel_id;
 864                 session_id = ntohl(*(__be32 *) ptr);
 865                 ptr += 4;
 866         }
 867
 868         /* Find the session context */
 869         session = l2tp_tunnel_get_session(tunnel, session_id);
 870         if (!session || !session->recv_skb) {
 871                 if (session)
 872                         l2tp_session_dec_refcount(session);
 873
 874                 /* Not found? Pass to userspace to deal with */
 875                 l2tp_info(tunnel, L2TP_MSG_DATA,
 876                           "%s: no session found (%u/%u). Passing up.\n",
 877                           tunnel->name, tunnel_id, session_id);
 878                 goto error;
 879         }
 880
 881         if (tunnel->version == L2TP_HDR_VER_3 &&
 882             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr))
 883                 goto error;
 884
 885         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 886         l2tp_session_dec_refcount(session);
 887
 888         return 0;
 889
 890 error:
 891         /* Put UDP header back */
 892         __skb_push(skb, sizeof(struct udphdr));
 893
 894         return 1;
 895 }
 896
 897 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 898  * Return codes:
 899  * 0 : success.
 900  * <0: error
 901  * >0: skb should be passed up to userspace as UDP.
 902  */
 903 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 904 {
 905         struct l2tp_tunnel *tunnel;
 906
 907         tunnel = rcu_dereference_sk_user_data(sk);
 908         if (tunnel == NULL)
 909                 goto pass_up;
 910
 911         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 912                  tunnel->name, skb->len);
 913
 914         if (l2tp_udp_recv_core(tunnel, skb))
 915                 goto pass_up;
 916
 917         return 0;
 918
 919 pass_up:
 920         return 1;
 921 }
 922 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 923
 924 /************************************************************************
 925  * Transmit handling
 926  ***********************************************************************/
 927
 928 /* Build an L2TP header for the session into the buffer provided.
 929  */
 930 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 931 {
 932         struct l2tp_tunnel *tunnel = session->tunnel;
 933         __be16 *bufp = buf;
 934         __be16 *optr = buf;
 935         u16 flags = L2TP_HDR_VER_2;
 936         u32 tunnel_id = tunnel->peer_tunnel_id;
 937         u32 session_id = session->peer_session_id;
 938
 939         if (session->send_seq)
 940                 flags |= L2TP_HDRFLAG_S;
 941
 942         /* Setup L2TP header. */
 943         *bufp++ = htons(flags);
 944         *bufp++ = htons(tunnel_id);
 945         *bufp++ = htons(session_id);
 946         if (session->send_seq) {
 947                 *bufp++ = htons(session->ns);
 948                 *bufp++ = 0;
 949                 session->ns++;
 950                 session->ns &= 0xffff;
 951                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
 952                          session->name, session->ns);
 953         }
 954
 955         return bufp - optr;
 956 }
 957
 958 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 959 {
 960         struct l2tp_tunnel *tunnel = session->tunnel;
 961         char *bufp = buf;
 962         char *optr = bufp;
 963
 964         /* Setup L2TP header. The header differs slightly for UDP and
 965          * IP encapsulations. For UDP, there is 4 bytes of flags.
 966          */
 967         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 968                 u16 flags = L2TP_HDR_VER_3;
 969                 *((__be16 *) bufp) = htons(flags);
 970                 bufp += 2;
 971                 *((__be16 *) bufp) = 0;
 972                 bufp += 2;
 973         }
 974
 975         *((__be32 *) bufp) = htonl(session->peer_session_id);
 976         bufp += 4;
 977         if (session->cookie_len) {
 978                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 979                 bufp += session->cookie_len;
 980         }
 981         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 982                 u32 l2h = 0;
 983
 984                 if (session->send_seq) {
 985                         l2h = 0x40000000 | session->ns;
 986                         session->ns++;
 987                         session->ns &= 0xffffff;
 988                         l2tp_dbg(session, L2TP_MSG_SEQ,
 989                                  "%s: updated ns to %u\n",
 990                                  session->name, session->ns);
 991                 }
 992
 993                 *((__be32 *)bufp) = htonl(l2h);
 994                 bufp += 4;
 995         }
 996
 997         return bufp - optr;
 998 }
 999
1000 static void l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1001                            struct flowi *fl, size_t data_len)
1002 {
1003         struct l2tp_tunnel *tunnel = session->tunnel;
1004         unsigned int len = skb->len;
1005         int error;
1006
1007         /* Debug */
1008         if (session->send_seq)
1009                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes, ns=%u\n",
1010                          session->name, data_len, session->ns - 1);
1011         else
1012                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes\n",
1013                          session->name, data_len);
1014
1015         if (session->debug & L2TP_MSG_DATA) {
1016                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1017                 unsigned char *datap = skb->data + uhlen;
1018
1019                 pr_debug("%s: xmit\n", session->name);
1020                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1021                                      datap, min_t(size_t, 32, len - uhlen));
1022         }
1023
1024         /* Queue the packet to IP for output */
1025         skb->ignore_df = 1;
1026 #if IS_ENABLED(CONFIG_IPV6)
1027         if (l2tp_sk_is_v6(tunnel->sock))
1028                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1029         else
1030 #endif
1031                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1032
1033         /* Update stats */
1034         if (error >= 0) {
1035                 atomic_long_inc(&tunnel->stats.tx_packets);
1036                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1037                 atomic_long_inc(&session->stats.tx_packets);
1038                 atomic_long_add(len, &session->stats.tx_bytes);
1039         } else {
1040                 atomic_long_inc(&tunnel->stats.tx_errors);
1041                 atomic_long_inc(&session->stats.tx_errors);
1042         }
1043 }
1044
1045 /* If caller requires the skb to have a ppp header, the header must be
1046  * inserted in the skb data before calling this function.
1047  */
1048 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1049 {
1050         int data_len = skb->len;
1051         struct l2tp_tunnel *tunnel = session->tunnel;
1052         struct sock *sk = tunnel->sock;
1053         struct flowi *fl;
1054         struct udphdr *uh;
1055         struct inet_sock *inet;
1056         int headroom;
1057         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1058         int udp_len;
1059         int ret = NET_XMIT_SUCCESS;
1060
1061         /* Check that there's enough headroom in the skb to insert IP,
1062          * UDP and L2TP headers. If not enough, expand it to
1063          * make room. Adjust truesize.
1064          */
1065         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1066                 uhlen + hdr_len;
1067         if (skb_cow_head(skb, headroom)) {
1068                 kfree_skb(skb);
1069                 return NET_XMIT_DROP;
1070         }
1071
1072         /* Setup L2TP header */
1073         session->build_header(session, __skb_push(skb, hdr_len));
1074
1075         /* Reset skb netfilter state */
1076         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1077         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1078                               IPSKB_REROUTED);
1079         nf_reset_ct(skb);
1080
1081         bh_lock_sock(sk);
1082         if (sock_owned_by_user(sk)) {
1083                 kfree_skb(skb);
1084                 ret = NET_XMIT_DROP;
1085                 goto out_unlock;
1086         }
1087
1088         /* The user-space may change the connection status for the user-space
1089          * provided socket at run time: we must check it under the socket lock
1090          */
1091         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1092                 kfree_skb(skb);
1093                 ret = NET_XMIT_DROP;
1094                 goto out_unlock;
1095         }
1096
1097         /* Get routing info from the tunnel socket */
1098         skb_dst_drop(skb);
1099         skb_dst_set(skb, sk_dst_check(sk, 0));
1100
1101         inet = inet_sk(sk);
1102         fl = &inet->cork.fl;
1103         switch (tunnel->encap) {
1104         case L2TP_ENCAPTYPE_UDP:
1105                 /* Setup UDP header */
1106                 __skb_push(skb, sizeof(*uh));
1107                 skb_reset_transport_header(skb);
1108                 uh = udp_hdr(skb);
1109                 uh->source = inet->inet_sport;
1110                 uh->dest = inet->inet_dport;
1111                 udp_len = uhlen + hdr_len + data_len;
1112                 uh->len = htons(udp_len);
1113
1114                 /* Calculate UDP checksum if configured to do so */
1115 #if IS_ENABLED(CONFIG_IPV6)
1116                 if (l2tp_sk_is_v6(sk))
1117                         udp6_set_csum(udp_get_no_check6_tx(sk),
1118                                       skb, &inet6_sk(sk)->saddr,
1119                                       &sk->sk_v6_daddr, udp_len);
1120                 else
1121 #endif
1122                 udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1123                              inet->inet_daddr, udp_len);
1124                 break;
1125
1126         case L2TP_ENCAPTYPE_IP:
1127                 break;
1128         }
1129
1130         l2tp_xmit_core(session, skb, fl, data_len);
1131 out_unlock:
1132         bh_unlock_sock(sk);
1133
1134         return ret;
1135 }
1136 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1137
1138 /*****************************************************************************
1139  * Tinnel and session create/destroy.
1140  *****************************************************************************/
1141
1142 /* Tunnel socket destruct hook.
1143  * The tunnel context is deleted only when all session sockets have been
1144  * closed.
1145  */
1146 static void l2tp_tunnel_destruct(struct sock *sk)
1147 {
1148         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1149
1150         if (tunnel == NULL)
1151                 goto end;
1152
1153         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1154
1155         /* Disable udp encapsulation */
1156         switch (tunnel->encap) {
1157         case L2TP_ENCAPTYPE_UDP:
1158                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1159                 (udp_sk(sk))->encap_type = 0;
1160                 (udp_sk(sk))->encap_rcv = NULL;
1161                 (udp_sk(sk))->encap_destroy = NULL;
1162                 break;
1163         case L2TP_ENCAPTYPE_IP:
1164                 break;
1165         }
1166
1167         /* Remove hooks into tunnel socket */
1168         sk->sk_destruct = tunnel->old_sk_destruct;
1169         sk->sk_user_data = NULL;
1170
1171         /* Call the original destructor */
1172         if (sk->sk_destruct)
1173                 (*sk->sk_destruct)(sk);
1174
1175         kfree_rcu(tunnel, rcu);
1176 end:
1177         return;
1178 }
1179
1180 /* When the tunnel is closed, all the attached sessions need to go too.
1181  */
1182 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1183 {
1184         int hash;
1185         struct hlist_node *walk;
1186         struct hlist_node *tmp;
1187         struct l2tp_session *session;
1188
1189         BUG_ON(tunnel == NULL);
1190
1191         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1192                   tunnel->name);
1193
1194         write_lock_bh(&tunnel->hlist_lock);
1195         tunnel->acpt_newsess = false;
1196         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1197 again:
1198                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1199                         session = hlist_entry(walk, struct l2tp_session, hlist);
1200
1201                         l2tp_info(session, L2TP_MSG_CONTROL,
1202                                   "%s: closing session\n", session->name);
1203
1204                         hlist_del_init(&session->hlist);
1205
1206                         if (test_and_set_bit(0, &session->dead))
1207                                 goto again;
1208
1209                         write_unlock_bh(&tunnel->hlist_lock);
1210
1211                         __l2tp_session_unhash(session);
1212                         l2tp_session_queue_purge(session);
1213
1214                         if (session->session_close != NULL)
1215                                 (*session->session_close)(session);
1216
1217                         l2tp_session_dec_refcount(session);
1218
1219                         write_lock_bh(&tunnel->hlist_lock);
1220
1221                         /* Now restart from the beginning of this hash
1222                          * chain.  We always remove a session from the
1223                          * list so we are guaranteed to make forward
1224                          * progress.
1225                          */
1226                         goto again;
1227                 }
1228         }
1229         write_unlock_bh(&tunnel->hlist_lock);
1230 }
1231
1232 /* Tunnel socket destroy hook for UDP encapsulation */
1233 static void l2tp_udp_encap_destroy(struct sock *sk)
1234 {
1235         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1236
1237         if (tunnel)
1238                 l2tp_tunnel_delete(tunnel);
1239 }
1240
1241 /* Workqueue tunnel deletion function */
1242 static void l2tp_tunnel_del_work(struct work_struct *work)
1243 {
1244         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1245                                                   del_work);
1246         struct sock *sk = tunnel->sock;
1247         struct socket *sock = sk->sk_socket;
1248         struct l2tp_net *pn;
1249
1250         l2tp_tunnel_closeall(tunnel);
1251
1252         /* If the tunnel socket was created within the kernel, use
1253          * the sk API to release it here.
1254          */
1255         if (tunnel->fd < 0) {
1256                 if (sock) {
1257                         kernel_sock_shutdown(sock, SHUT_RDWR);
1258                         sock_release(sock);
1259                 }
1260         }
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
1268         /* drop initial ref */
1269         l2tp_tunnel_dec_refcount(tunnel);
1270
1271         /* drop workqueue ref */
1272         l2tp_tunnel_dec_refcount(tunnel);
1273 }
1274
1275 /* Create a socket for the tunnel, if one isn't set up by
1276  * userspace. This is used for static tunnels where there is no
1277  * managing L2TP daemon.
1278  *
1279  * Since we don't want these sockets to keep a namespace alive by
1280  * themselves, we drop the socket's namespace refcount after creation.
1281  * These sockets are freed when the namespace exits using the pernet
1282  * exit hook.
1283  */
1284 static int l2tp_tunnel_sock_create(struct net *net,
1285                                 u32 tunnel_id,
1286                                 u32 peer_tunnel_id,
1287                                 struct l2tp_tunnel_cfg *cfg,
1288                                 struct socket **sockp)
1289 {
1290         int err = -EINVAL;
1291         struct socket *sock = NULL;
1292         struct udp_port_cfg udp_conf;
1293
1294         switch (cfg->encap) {
1295         case L2TP_ENCAPTYPE_UDP:
1296                 memset(&udp_conf, 0, sizeof(udp_conf));
1297
1298 #if IS_ENABLED(CONFIG_IPV6)
1299                 if (cfg->local_ip6 && cfg->peer_ip6) {
1300                         udp_conf.family = AF_INET6;
1301                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1302                                sizeof(udp_conf.local_ip6));
1303                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1304                                sizeof(udp_conf.peer_ip6));
1305                         udp_conf.use_udp6_tx_checksums =
1306                           ! cfg->udp6_zero_tx_checksums;
1307                         udp_conf.use_udp6_rx_checksums =
1308                           ! cfg->udp6_zero_rx_checksums;
1309                 } else
1310 #endif
1311                 {
1312                         udp_conf.family = AF_INET;
1313                         udp_conf.local_ip = cfg->local_ip;
1314                         udp_conf.peer_ip = cfg->peer_ip;
1315                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1316                 }
1317
1318                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1319                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1320
1321                 err = udp_sock_create(net, &udp_conf, &sock);
1322                 if (err < 0)
1323                         goto out;
1324
1325                 break;
1326
1327         case L2TP_ENCAPTYPE_IP:
1328 #if IS_ENABLED(CONFIG_IPV6)
1329                 if (cfg->local_ip6 && cfg->peer_ip6) {
1330                         struct sockaddr_l2tpip6 ip6_addr = {0};
1331
1332                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1333                                           IPPROTO_L2TP, &sock);
1334                         if (err < 0)
1335                                 goto out;
1336
1337                         ip6_addr.l2tp_family = AF_INET6;
1338                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1339                                sizeof(ip6_addr.l2tp_addr));
1340                         ip6_addr.l2tp_conn_id = tunnel_id;
1341                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1342                                           sizeof(ip6_addr));
1343                         if (err < 0)
1344                                 goto out;
1345
1346                         ip6_addr.l2tp_family = AF_INET6;
1347                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1348                                sizeof(ip6_addr.l2tp_addr));
1349                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1350                         err = kernel_connect(sock,
1351                                              (struct sockaddr *) &ip6_addr,
1352                                              sizeof(ip6_addr), 0);
1353                         if (err < 0)
1354                                 goto out;
1355                 } else
1356 #endif
1357                 {
1358                         struct sockaddr_l2tpip ip_addr = {0};
1359
1360                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1361                                           IPPROTO_L2TP, &sock);
1362                         if (err < 0)
1363                                 goto out;
1364
1365                         ip_addr.l2tp_family = AF_INET;
1366                         ip_addr.l2tp_addr = cfg->local_ip;
1367                         ip_addr.l2tp_conn_id = tunnel_id;
1368                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1369                                           sizeof(ip_addr));
1370                         if (err < 0)
1371                                 goto out;
1372
1373                         ip_addr.l2tp_family = AF_INET;
1374                         ip_addr.l2tp_addr = cfg->peer_ip;
1375                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1376                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1377                                              sizeof(ip_addr), 0);
1378                         if (err < 0)
1379                                 goto out;
1380                 }
1381                 break;
1382
1383         default:
1384                 goto out;
1385         }
1386
1387 out:
1388         *sockp = sock;
1389         if ((err < 0) && sock) {
1390                 kernel_sock_shutdown(sock, SHUT_RDWR);
1391                 sock_release(sock);
1392                 *sockp = NULL;
1393         }
1394
1395         return err;
1396 }
1397
1398 static struct lock_class_key l2tp_socket_class;
1399
1400 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)
1401 {
1402         struct l2tp_tunnel *tunnel = NULL;
1403         int err;
1404         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1405
1406         if (cfg != NULL)
1407                 encap = cfg->encap;
1408
1409         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1410         if (tunnel == NULL) {
1411                 err = -ENOMEM;
1412                 goto err;
1413         }
1414
1415         tunnel->version = version;
1416         tunnel->tunnel_id = tunnel_id;
1417         tunnel->peer_tunnel_id = peer_tunnel_id;
1418         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1419
1420         tunnel->magic = L2TP_TUNNEL_MAGIC;
1421         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1422         rwlock_init(&tunnel->hlist_lock);
1423         tunnel->acpt_newsess = true;
1424
1425         if (cfg != NULL)
1426                 tunnel->debug = cfg->debug;
1427
1428         tunnel->encap = encap;
1429
1430         refcount_set(&tunnel->ref_count, 1);
1431         tunnel->fd = fd;
1432
1433         /* Init delete workqueue struct */
1434         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1435
1436         INIT_LIST_HEAD(&tunnel->list);
1437
1438         err = 0;
1439 err:
1440         if (tunnelp)
1441                 *tunnelp = tunnel;
1442
1443         return err;
1444 }
1445 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1446
1447 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1448                                 enum l2tp_encap_type encap)
1449 {
1450         if (!net_eq(sock_net(sk), net))
1451                 return -EINVAL;
1452
1453         if (sk->sk_type != SOCK_DGRAM)
1454                 return -EPROTONOSUPPORT;
1455
1456         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1457             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1458                 return -EPROTONOSUPPORT;
1459
1460         if (sk->sk_user_data)
1461                 return -EBUSY;
1462
1463         return 0;
1464 }
1465
1466 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1467                          struct l2tp_tunnel_cfg *cfg)
1468 {
1469         struct l2tp_tunnel *tunnel_walk;
1470         struct l2tp_net *pn;
1471         struct socket *sock;
1472         struct sock *sk;
1473         int ret;
1474
1475         if (tunnel->fd < 0) {
1476                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1477                                               tunnel->peer_tunnel_id, cfg,
1478                                               &sock);
1479                 if (ret < 0)
1480                         goto err;
1481         } else {
1482                 sock = sockfd_lookup(tunnel->fd, &ret);
1483                 if (!sock)
1484                         goto err;
1485
1486                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1487                 if (ret < 0)
1488                         goto err_sock;
1489         }
1490
1491         tunnel->l2tp_net = net;
1492         pn = l2tp_pernet(net);
1493
1494         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1495         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1496                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1497                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1498
1499                         ret = -EEXIST;
1500                         goto err_sock;
1501                 }
1502         }
1503         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1504         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1505
1506         sk = sock->sk;
1507         sock_hold(sk);
1508         tunnel->sock = sk;
1509
1510         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1511                 struct udp_tunnel_sock_cfg udp_cfg = {
1512                         .sk_user_data = tunnel,
1513                         .encap_type = UDP_ENCAP_L2TPINUDP,
1514                         .encap_rcv = l2tp_udp_encap_recv,
1515                         .encap_destroy = l2tp_udp_encap_destroy,
1516                 };
1517
1518                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1519         } else {
1520                 sk->sk_user_data = tunnel;
1521         }
1522
1523         tunnel->old_sk_destruct = sk->sk_destruct;
1524         sk->sk_destruct = &l2tp_tunnel_destruct;
1525         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1526                                    "l2tp_sock");
1527         sk->sk_allocation = GFP_ATOMIC;
1528
1529         if (tunnel->fd >= 0)
1530                 sockfd_put(sock);
1531
1532         return 0;
1533
1534 err_sock:
1535         if (tunnel->fd < 0)
1536                 sock_release(sock);
1537         else
1538                 sockfd_put(sock);
1539 err:
1540         return ret;
1541 }
1542 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1543
1544 /* This function is used by the netlink TUNNEL_DELETE command.
1545  */
1546 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1547 {
1548         if (!test_and_set_bit(0, &tunnel->dead)) {
1549                 l2tp_tunnel_inc_refcount(tunnel);
1550                 queue_work(l2tp_wq, &tunnel->del_work);
1551         }
1552 }
1553 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1554
1555 /* Really kill the session.
1556  */
1557 void l2tp_session_free(struct l2tp_session *session)
1558 {
1559         struct l2tp_tunnel *tunnel = session->tunnel;
1560
1561         BUG_ON(refcount_read(&session->ref_count) != 0);
1562
1563         if (tunnel) {
1564                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1565                 l2tp_tunnel_dec_refcount(tunnel);
1566         }
1567
1568         kfree(session);
1569 }
1570 EXPORT_SYMBOL_GPL(l2tp_session_free);
1571
1572 /* Remove an l2tp session from l2tp_core's hash lists.
1573  * Provides a tidyup interface for pseudowire code which can't just route all
1574  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1575  * callback.
1576  */
1577 void __l2tp_session_unhash(struct l2tp_session *session)
1578 {
1579         struct l2tp_tunnel *tunnel = session->tunnel;
1580
1581         /* Remove the session from core hashes */
1582         if (tunnel) {
1583                 /* Remove from the per-tunnel hash */
1584                 write_lock_bh(&tunnel->hlist_lock);
1585                 hlist_del_init(&session->hlist);
1586                 write_unlock_bh(&tunnel->hlist_lock);
1587
1588                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1589                 if (tunnel->version != L2TP_HDR_VER_2) {
1590                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1591                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1592                         hlist_del_init_rcu(&session->global_hlist);
1593                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1594                         synchronize_rcu();
1595                 }
1596         }
1597 }
1598 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1599
1600 /* This function is used by the netlink SESSION_DELETE command and by
1601    pseudowire modules.
1602  */
1603 int l2tp_session_delete(struct l2tp_session *session)
1604 {
1605         if (test_and_set_bit(0, &session->dead))
1606                 return 0;
1607
1608         __l2tp_session_unhash(session);
1609         l2tp_session_queue_purge(session);
1610         if (session->session_close != NULL)
1611                 (*session->session_close)(session);
1612
1613         l2tp_session_dec_refcount(session);
1614
1615         return 0;
1616 }
1617 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1618
1619 /* We come here whenever a session's send_seq, cookie_len or
1620  * l2specific_type parameters are set.
1621  */
1622 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1623 {
1624         if (version == L2TP_HDR_VER_2) {
1625                 session->hdr_len = 6;
1626                 if (session->send_seq)
1627                         session->hdr_len += 4;
1628         } else {
1629                 session->hdr_len = 4 + session->cookie_len;
1630                 session->hdr_len += l2tp_get_l2specific_len(session);
1631                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1632                         session->hdr_len += 4;
1633         }
1634
1635 }
1636 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1637
1638 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)
1639 {
1640         struct l2tp_session *session;
1641
1642         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1643         if (session != NULL) {
1644                 session->magic = L2TP_SESSION_MAGIC;
1645                 session->tunnel = tunnel;
1646
1647                 session->session_id = session_id;
1648                 session->peer_session_id = peer_session_id;
1649                 session->nr = 0;
1650                 if (tunnel->version == L2TP_HDR_VER_2)
1651                         session->nr_max = 0xffff;
1652                 else
1653                         session->nr_max = 0xffffff;
1654                 session->nr_window_size = session->nr_max / 2;
1655                 session->nr_oos_count_max = 4;
1656
1657                 /* Use NR of first received packet */
1658                 session->reorder_skip = 1;
1659
1660                 sprintf(&session->name[0], "sess %u/%u",
1661                         tunnel->tunnel_id, session->session_id);
1662
1663                 skb_queue_head_init(&session->reorder_q);
1664
1665                 INIT_HLIST_NODE(&session->hlist);
1666                 INIT_HLIST_NODE(&session->global_hlist);
1667
1668                 /* Inherit debug options from tunnel */
1669                 session->debug = tunnel->debug;
1670
1671                 if (cfg) {
1672                         session->pwtype = cfg->pw_type;
1673                         session->debug = cfg->debug;
1674                         session->send_seq = cfg->send_seq;
1675                         session->recv_seq = cfg->recv_seq;
1676                         session->lns_mode = cfg->lns_mode;
1677                         session->reorder_timeout = cfg->reorder_timeout;
1678                         session->l2specific_type = cfg->l2specific_type;
1679                         session->cookie_len = cfg->cookie_len;
1680                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1681                         session->peer_cookie_len = cfg->peer_cookie_len;
1682                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1683                 }
1684
1685                 if (tunnel->version == L2TP_HDR_VER_2)
1686                         session->build_header = l2tp_build_l2tpv2_header;
1687                 else
1688                         session->build_header = l2tp_build_l2tpv3_header;
1689
1690                 l2tp_session_set_header_len(session, tunnel->version);
1691
1692                 refcount_set(&session->ref_count, 1);
1693
1694                 return session;
1695         }
1696
1697         return ERR_PTR(-ENOMEM);
1698 }
1699 EXPORT_SYMBOL_GPL(l2tp_session_create);
1700
1701 /*****************************************************************************
1702  * Init and cleanup
1703  *****************************************************************************/
1704
1705 static __net_init int l2tp_init_net(struct net *net)
1706 {
1707         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1708         int hash;
1709
1710         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1711         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1712
1713         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1714                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1715
1716         spin_lock_init(&pn->l2tp_session_hlist_lock);
1717
1718         return 0;
1719 }
1720
1721 static __net_exit void l2tp_exit_net(struct net *net)
1722 {
1723         struct l2tp_net *pn = l2tp_pernet(net);
1724         struct l2tp_tunnel *tunnel = NULL;
1725         int hash;
1726
1727         rcu_read_lock_bh();
1728         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1729                 l2tp_tunnel_delete(tunnel);
1730         }
1731         rcu_read_unlock_bh();
1732
1733         if (l2tp_wq)
1734                 flush_workqueue(l2tp_wq);
1735         rcu_barrier();
1736
1737         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1738                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1739 }
1740
1741 static struct pernet_operations l2tp_net_ops = {
1742         .init = l2tp_init_net,
1743         .exit = l2tp_exit_net,
1744         .id   = &l2tp_net_id,
1745         .size = sizeof(struct l2tp_net),
1746 };
1747
1748 static int __init l2tp_init(void)
1749 {
1750         int rc = 0;
1751
1752         rc = register_pernet_device(&l2tp_net_ops);
1753         if (rc)
1754                 goto out;
1755
1756         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1757         if (!l2tp_wq) {
1758                 pr_err("alloc_workqueue failed\n");
1759                 unregister_pernet_device(&l2tp_net_ops);
1760                 rc = -ENOMEM;
1761                 goto out;
1762         }
1763
1764         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1765
1766 out:
1767         return rc;
1768 }
1769
1770 static void __exit l2tp_exit(void)
1771 {
1772         unregister_pernet_device(&l2tp_net_ops);
1773         if (l2tp_wq) {
1774                 destroy_workqueue(l2tp_wq);
1775                 l2tp_wq = NULL;
1776         }
1777 }
1778
1779 module_init(l2tp_init);
1780 module_exit(l2tp_exit);
1781
1782 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1783 MODULE_DESCRIPTION("L2TP core");
1784 MODULE_LICENSE("GPL");
1785 MODULE_VERSION(L2TP_DRV_VERSION);