net/l2tp/l2tp_core.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   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
  17 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  18
  19 #include <linux/module.h>
  20 #include <linux/string.h>
  21 #include <linux/list.h>
  22 #include <linux/rculist.h>
  23 #include <linux/uaccess.h>
  24
  25 #include <linux/kernel.h>
  26 #include <linux/spinlock.h>
  27 #include <linux/kthread.h>
  28 #include <linux/sched.h>
  29 #include <linux/slab.h>
  30 #include <linux/errno.h>
  31 #include <linux/jiffies.h>
  32
  33 #include <linux/netdevice.h>
  34 #include <linux/net.h>
  35 #include <linux/inetdevice.h>
  36 #include <linux/skbuff.h>
  37 #include <linux/init.h>
  38 #include <linux/in.h>
  39 #include <linux/ip.h>
  40 #include <linux/udp.h>
  41 #include <linux/l2tp.h>
  42 #include <linux/hash.h>
  43 #include <linux/sort.h>
  44 #include <linux/file.h>
  45 #include <linux/nsproxy.h>
  46 #include <net/net_namespace.h>
  47 #include <net/netns/generic.h>
  48 #include <net/dst.h>
  49 #include <net/ip.h>
  50 #include <net/udp.h>
  51 #include <net/udp_tunnel.h>
  52 #include <net/inet_common.h>
  53 #include <net/xfrm.h>
  54 #include <net/protocol.h>
  55 #include <net/inet6_connection_sock.h>
  56 #include <net/inet_ecn.h>
  57 #include <net/ip6_route.h>
  58 #include <net/ip6_checksum.h>
  59
  60 #include <asm/byteorder.h>
  61 #include <linux/atomic.h>
  62
  63 #include "l2tp_core.h"
  64
  65 #define L2TP_DRV_VERSION        "V2.0"
  66
  67 /* L2TP header constants */
  68 #define L2TP_HDRFLAG_T     0x8000
  69 #define L2TP_HDRFLAG_L     0x4000
  70 #define L2TP_HDRFLAG_S     0x0800
  71 #define L2TP_HDRFLAG_O     0x0200
  72 #define L2TP_HDRFLAG_P     0x0100
  73
  74 #define L2TP_HDR_VER_MASK  0x000F
  75 #define L2TP_HDR_VER_2     0x0002
  76 #define L2TP_HDR_VER_3     0x0003
  77
  78 /* L2TPv3 default L2-specific sublayer */
  79 #define L2TP_SLFLAG_S      0x40000000
  80 #define L2TP_SL_SEQ_MASK   0x00ffffff
  81
  82 #define L2TP_HDR_SIZE_MAX               14
  83
  84 /* Default trace flags */
  85 #define L2TP_DEFAULT_DEBUG_FLAGS        0
  86
  87 /* Private data stored for received packets in the skb.
  88  */
  89 struct l2tp_skb_cb {
  90         u32                     ns;
  91         u16                     has_seq;
  92         u16                     length;
  93         unsigned long           expires;
  94 };
  95
  96 #define L2TP_SKB_CB(skb)        ((struct l2tp_skb_cb *)&(skb)->cb[sizeof(struct inet_skb_parm)])
  97
  98 static struct workqueue_struct *l2tp_wq;
  99
 100 /* per-net private data for this module */
 101 static unsigned int l2tp_net_id;
 102 struct l2tp_net {
 103         struct list_head l2tp_tunnel_list;
 104         /* Lock for write access to l2tp_tunnel_list */
 105         spinlock_t l2tp_tunnel_list_lock;
 106         struct hlist_head l2tp_session_hlist[L2TP_HASH_SIZE_2];
 107         /* Lock for write access to l2tp_session_hlist */
 108         spinlock_t l2tp_session_hlist_lock;
 109 };
 110
 111 #if IS_ENABLED(CONFIG_IPV6)
 112 static bool l2tp_sk_is_v6(struct sock *sk)
 113 {
 114         return sk->sk_family == PF_INET6 &&
 115                !ipv6_addr_v4mapped(&sk->sk_v6_daddr);
 116 }
 117 #endif
 118
 119 static inline struct l2tp_tunnel *l2tp_tunnel(struct sock *sk)
 120 {
 121         return sk->sk_user_data;
 122 }
 123
 124 static inline struct l2tp_net *l2tp_pernet(const struct net *net)
 125 {
 126         return net_generic(net, l2tp_net_id);
 127 }
 128
 129 /* Session hash global list for L2TPv3.
 130  * The session_id SHOULD be random according to RFC3931, but several
 131  * L2TP implementations use incrementing session_ids.  So we do a real
 132  * hash on the session_id, rather than a simple bitmask.
 133  */
 134 static inline struct hlist_head *
 135 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 136 {
 137         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 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                 /* IP encap expects session IDs to be globally unique, while
 324                  * UDP encap doesn't.
 325                  */
 326                 hlist_for_each_entry(session_walk, g_head, global_hlist)
 327                         if (session_walk->session_id == session->session_id &&
 328                             (session_walk->tunnel->encap == L2TP_ENCAPTYPE_IP ||
 329                              tunnel->encap == L2TP_ENCAPTYPE_IP)) {
 330                                 err = -EEXIST;
 331                                 goto err_tlock_pnlock;
 332                         }
 333
 334                 l2tp_tunnel_inc_refcount(tunnel);
 335                 hlist_add_head_rcu(&session->global_hlist, g_head);
 336
 337                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 338         } else {
 339                 l2tp_tunnel_inc_refcount(tunnel);
 340         }
 341
 342         hlist_add_head(&session->hlist, head);
 343         write_unlock_bh(&tunnel->hlist_lock);
 344
 345         return 0;
 346
 347 err_tlock_pnlock:
 348         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 349 err_tlock:
 350         write_unlock_bh(&tunnel->hlist_lock);
 351
 352         return err;
 353 }
 354 EXPORT_SYMBOL_GPL(l2tp_session_register);
 355
 356 /*****************************************************************************
 357  * Receive data handling
 358  *****************************************************************************/
 359
 360 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 361  * number.
 362  */
 363 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 364 {
 365         struct sk_buff *skbp;
 366         struct sk_buff *tmp;
 367         u32 ns = L2TP_SKB_CB(skb)->ns;
 368
 369         spin_lock_bh(&session->reorder_q.lock);
 370         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 371                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 372                         __skb_queue_before(&session->reorder_q, skbp, skb);
 373                         l2tp_dbg(session, L2TP_MSG_SEQ,
 374                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 375                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 376                                  skb_queue_len(&session->reorder_q));
 377                         atomic_long_inc(&session->stats.rx_oos_packets);
 378                         goto out;
 379                 }
 380         }
 381
 382         __skb_queue_tail(&session->reorder_q, skb);
 383
 384 out:
 385         spin_unlock_bh(&session->reorder_q.lock);
 386 }
 387
 388 /* Dequeue a single skb.
 389  */
 390 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 391 {
 392         struct l2tp_tunnel *tunnel = session->tunnel;
 393         int length = L2TP_SKB_CB(skb)->length;
 394
 395         /* We're about to requeue the skb, so return resources
 396          * to its current owner (a socket receive buffer).
 397          */
 398         skb_orphan(skb);
 399
 400         atomic_long_inc(&tunnel->stats.rx_packets);
 401         atomic_long_add(length, &tunnel->stats.rx_bytes);
 402         atomic_long_inc(&session->stats.rx_packets);
 403         atomic_long_add(length, &session->stats.rx_bytes);
 404
 405         if (L2TP_SKB_CB(skb)->has_seq) {
 406                 /* Bump our Nr */
 407                 session->nr++;
 408                 session->nr &= session->nr_max;
 409
 410                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 411                          session->name, session->nr);
 412         }
 413
 414         /* call private receive handler */
 415         if (session->recv_skb)
 416                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 417         else
 418                 kfree_skb(skb);
 419 }
 420
 421 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 422  * Skbs that have been in the queue for too long are simply discarded.
 423  */
 424 static void l2tp_recv_dequeue(struct l2tp_session *session)
 425 {
 426         struct sk_buff *skb;
 427         struct sk_buff *tmp;
 428
 429         /* If the pkt at the head of the queue has the nr that we
 430          * expect to send up next, dequeue it and any other
 431          * in-sequence packets behind it.
 432          */
 433 start:
 434         spin_lock_bh(&session->reorder_q.lock);
 435         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 436                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 437                         atomic_long_inc(&session->stats.rx_seq_discards);
 438                         atomic_long_inc(&session->stats.rx_errors);
 439                         l2tp_dbg(session, L2TP_MSG_SEQ,
 440                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 441                                  session->name, L2TP_SKB_CB(skb)->ns,
 442                                  L2TP_SKB_CB(skb)->length, session->nr,
 443                                  skb_queue_len(&session->reorder_q));
 444                         session->reorder_skip = 1;
 445                         __skb_unlink(skb, &session->reorder_q);
 446                         kfree_skb(skb);
 447                         continue;
 448                 }
 449
 450                 if (L2TP_SKB_CB(skb)->has_seq) {
 451                         if (session->reorder_skip) {
 452                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 453                                          "%s: advancing nr to next pkt: %u -> %u",
 454                                          session->name, session->nr,
 455                                          L2TP_SKB_CB(skb)->ns);
 456                                 session->reorder_skip = 0;
 457                                 session->nr = L2TP_SKB_CB(skb)->ns;
 458                         }
 459                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 460                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 461                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 462                                          session->name, L2TP_SKB_CB(skb)->ns,
 463                                          L2TP_SKB_CB(skb)->length, session->nr,
 464                                          skb_queue_len(&session->reorder_q));
 465                                 goto out;
 466                         }
 467                 }
 468                 __skb_unlink(skb, &session->reorder_q);
 469
 470                 /* Process the skb. We release the queue lock while we
 471                  * do so to let other contexts process the queue.
 472                  */
 473                 spin_unlock_bh(&session->reorder_q.lock);
 474                 l2tp_recv_dequeue_skb(session, skb);
 475                 goto start;
 476         }
 477
 478 out:
 479         spin_unlock_bh(&session->reorder_q.lock);
 480 }
 481
 482 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 483 {
 484         u32 nws;
 485
 486         if (nr >= session->nr)
 487                 nws = nr - session->nr;
 488         else
 489                 nws = (session->nr_max + 1) - (session->nr - nr);
 490
 491         return nws < session->nr_window_size;
 492 }
 493
 494 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 495  * acceptable, else non-zero.
 496  */
 497 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 498 {
 499         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
 500                 /* Packet sequence number is outside allowed window.
 501                  * Discard it.
 502                  */
 503                 l2tp_dbg(session, L2TP_MSG_SEQ,
 504                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
 505                          session->name, L2TP_SKB_CB(skb)->ns,
 506                          L2TP_SKB_CB(skb)->length, session->nr);
 507                 goto discard;
 508         }
 509
 510         if (session->reorder_timeout != 0) {
 511                 /* Packet reordering enabled. Add skb to session's
 512                  * reorder queue, in order of ns.
 513                  */
 514                 l2tp_recv_queue_skb(session, skb);
 515                 goto out;
 516         }
 517
 518         /* Packet reordering disabled. Discard out-of-sequence packets, while
 519          * tracking the number if in-sequence packets after the first OOS packet
 520          * is seen. After nr_oos_count_max in-sequence packets, reset the
 521          * sequence number to re-enable packet reception.
 522          */
 523         if (L2TP_SKB_CB(skb)->ns == session->nr) {
 524                 skb_queue_tail(&session->reorder_q, skb);
 525         } else {
 526                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
 527                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 528
 529                 if (nr_oos == nr_next)
 530                         session->nr_oos_count++;
 531                 else
 532                         session->nr_oos_count = 0;
 533
 534                 session->nr_oos = nr_oos;
 535                 if (session->nr_oos_count > session->nr_oos_count_max) {
 536                         session->reorder_skip = 1;
 537                         l2tp_dbg(session, L2TP_MSG_SEQ,
 538                                  "%s: %d oos packets received. Resetting sequence numbers\n",
 539                                  session->name, session->nr_oos_count);
 540                 }
 541                 if (!session->reorder_skip) {
 542                         atomic_long_inc(&session->stats.rx_seq_discards);
 543                         l2tp_dbg(session, L2TP_MSG_SEQ,
 544                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 545                                  session->name, L2TP_SKB_CB(skb)->ns,
 546                                  L2TP_SKB_CB(skb)->length, session->nr,
 547                                  skb_queue_len(&session->reorder_q));
 548                         goto discard;
 549                 }
 550                 skb_queue_tail(&session->reorder_q, skb);
 551         }
 552
 553 out:
 554         return 0;
 555
 556 discard:
 557         return 1;
 558 }
 559
 560 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 561  * and L2TPv3 data frames here.
 562  *
 563  * L2TPv2 Data Message Header
 564  *
 565  *  0                   1                   2                   3
 566  *  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
 567  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 568  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 569  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 570  * |           Tunnel ID           |           Session ID          |
 571  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 572  * |             Ns (opt)          |             Nr (opt)          |
 573  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 574  * |      Offset Size (opt)        |    Offset pad... (opt)
 575  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 576  *
 577  * Data frames are marked by T=0. All other fields are the same as
 578  * those in L2TP control frames.
 579  *
 580  * L2TPv3 Data Message Header
 581  *
 582  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 583  * |                      L2TP Session Header                      |
 584  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 585  * |                      L2-Specific Sublayer                     |
 586  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 587  * |                        Tunnel Payload                      ...
 588  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 589  *
 590  * L2TPv3 Session Header Over IP
 591  *
 592  *  0                   1                   2                   3
 593  *  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
 594  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 595  * |                           Session ID                          |
 596  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 597  * |               Cookie (optional, maximum 64 bits)...
 598  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 599  *                                                                 |
 600  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 601  *
 602  * L2TPv3 L2-Specific Sublayer Format
 603  *
 604  *  0                   1                   2                   3
 605  *  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
 606  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 607  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 608  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 609  *
 610  * Cookie value and sublayer format are negotiated with the peer when
 611  * the session is set up. Unlike L2TPv2, we do not need to parse the
 612  * packet header to determine if optional fields are 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)
 622 {
 623         struct l2tp_tunnel *tunnel = session->tunnel;
 624         u32 ns = 0, nr = 0;
 625         int offset;
 626
 627         /* Parse and check optional cookie */
 628         if (session->peer_cookie_len > 0) {
 629                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 630                         l2tp_info(tunnel, L2TP_MSG_DATA,
 631                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 632                                   tunnel->name, tunnel->tunnel_id,
 633                                   session->session_id);
 634                         atomic_long_inc(&session->stats.rx_cookie_discards);
 635                         goto discard;
 636                 }
 637                 ptr += session->peer_cookie_len;
 638         }
 639
 640         /* Handle the optional sequence numbers. Sequence numbers are
 641          * in different places for L2TPv2 and L2TPv3.
 642          *
 643          * If we are the LAC, enable/disable sequence numbers under
 644          * the control of the LNS.  If no sequence numbers present but
 645          * we were expecting them, discard frame.
 646          */
 647         L2TP_SKB_CB(skb)->has_seq = 0;
 648         if (tunnel->version == L2TP_HDR_VER_2) {
 649                 if (hdrflags & L2TP_HDRFLAG_S) {
 650                         ns = ntohs(*(__be16 *)ptr);
 651                         ptr += 2;
 652                         nr = ntohs(*(__be16 *)ptr);
 653                         ptr += 2;
 654
 655                         /* Store L2TP info in the skb */
 656                         L2TP_SKB_CB(skb)->ns = ns;
 657                         L2TP_SKB_CB(skb)->has_seq = 1;
 658
 659                         l2tp_dbg(session, L2TP_MSG_SEQ,
 660                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 661                                  session->name, ns, nr, session->nr);
 662                 }
 663         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 664                 u32 l2h = ntohl(*(__be32 *)ptr);
 665
 666                 if (l2h & 0x40000000) {
 667                         ns = l2h & 0x00ffffff;
 668
 669                         /* Store L2TP info in the skb */
 670                         L2TP_SKB_CB(skb)->ns = ns;
 671                         L2TP_SKB_CB(skb)->has_seq = 1;
 672
 673                         l2tp_dbg(session, L2TP_MSG_SEQ,
 674                                  "%s: recv data ns=%u, session nr=%u\n",
 675                                  session->name, ns, session->nr);
 676                 }
 677                 ptr += 4;
 678         }
 679
 680         if (L2TP_SKB_CB(skb)->has_seq) {
 681                 /* Received a packet with sequence numbers. If we're the LAC,
 682                  * check if we sre sending sequence numbers and if not,
 683                  * configure it so.
 684                  */
 685                 if (!session->lns_mode && !session->send_seq) {
 686                         l2tp_info(session, L2TP_MSG_SEQ,
 687                                   "%s: requested to enable seq numbers by LNS\n",
 688                                   session->name);
 689                         session->send_seq = 1;
 690                         l2tp_session_set_header_len(session, tunnel->version);
 691                 }
 692         } else {
 693                 /* No sequence numbers.
 694                  * If user has configured mandatory sequence numbers, discard.
 695                  */
 696                 if (session->recv_seq) {
 697                         l2tp_warn(session, L2TP_MSG_SEQ,
 698                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 699                                   session->name);
 700                         atomic_long_inc(&session->stats.rx_seq_discards);
 701                         goto discard;
 702                 }
 703
 704                 /* If we're the LAC and we're sending sequence numbers, the
 705                  * LNS has requested that we no longer send sequence numbers.
 706                  * If we're the LNS and we're sending sequence numbers, the
 707                  * LAC is broken. Discard the frame.
 708                  */
 709                 if (!session->lns_mode && session->send_seq) {
 710                         l2tp_info(session, L2TP_MSG_SEQ,
 711                                   "%s: requested to disable seq numbers by LNS\n",
 712                                   session->name);
 713                         session->send_seq = 0;
 714                         l2tp_session_set_header_len(session, tunnel->version);
 715                 } else if (session->send_seq) {
 716                         l2tp_warn(session, L2TP_MSG_SEQ,
 717                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 718                                   session->name);
 719                         atomic_long_inc(&session->stats.rx_seq_discards);
 720                         goto discard;
 721                 }
 722         }
 723
 724         /* Session data offset is defined only for L2TPv2 and is
 725          * indicated by an optional 16-bit value in the header.
 726          */
 727         if (tunnel->version == L2TP_HDR_VER_2) {
 728                 /* If offset bit set, skip it. */
 729                 if (hdrflags & L2TP_HDRFLAG_O) {
 730                         offset = ntohs(*(__be16 *)ptr);
 731                         ptr += 2 + offset;
 732                 }
 733         }
 734
 735         offset = ptr - optr;
 736         if (!pskb_may_pull(skb, offset))
 737                 goto discard;
 738
 739         __skb_pull(skb, offset);
 740
 741         /* Prepare skb for adding to the session's reorder_q.  Hold
 742          * packets for max reorder_timeout or 1 second if not
 743          * reordering.
 744          */
 745         L2TP_SKB_CB(skb)->length = length;
 746         L2TP_SKB_CB(skb)->expires = jiffies +
 747                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 748
 749         /* Add packet to the session's receive queue. Reordering is done here, if
 750          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 751          */
 752         if (L2TP_SKB_CB(skb)->has_seq) {
 753                 if (l2tp_recv_data_seq(session, skb))
 754                         goto discard;
 755         } else {
 756                 /* No sequence numbers. Add the skb to the tail of the
 757                  * reorder queue. This ensures that it will be
 758                  * delivered after all previous sequenced skbs.
 759                  */
 760                 skb_queue_tail(&session->reorder_q, skb);
 761         }
 762
 763         /* Try to dequeue as many skbs from reorder_q as we can. */
 764         l2tp_recv_dequeue(session);
 765
 766         return;
 767
 768 discard:
 769         atomic_long_inc(&session->stats.rx_errors);
 770         kfree_skb(skb);
 771 }
 772 EXPORT_SYMBOL(l2tp_recv_common);
 773
 774 /* Drop skbs from the session's reorder_q
 775  */
 776 static void l2tp_session_queue_purge(struct l2tp_session *session)
 777 {
 778         struct sk_buff *skb = NULL;
 779
 780         if (WARN_ON(session->magic != L2TP_SESSION_MAGIC))
 781                 return;
 782
 783         while ((skb = skb_dequeue(&session->reorder_q))) {
 784                 atomic_long_inc(&session->stats.rx_errors);
 785                 kfree_skb(skb);
 786         }
 787 }
 788
 789 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 790  * here. The skb is not on a list when we get here.
 791  * Returns 0 if the packet was a data packet and was successfully passed on.
 792  * Returns 1 if the packet was not a good data packet and could not be
 793  * forwarded.  All such packets are passed up to userspace to deal with.
 794  */
 795 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb)
 796 {
 797         struct l2tp_session *session = NULL;
 798         unsigned char *ptr, *optr;
 799         u16 hdrflags;
 800         u32 tunnel_id, session_id;
 801         u16 version;
 802         int length;
 803
 804         /* UDP has verifed checksum */
 805
 806         /* UDP always verifies the packet length. */
 807         __skb_pull(skb, sizeof(struct udphdr));
 808
 809         /* Short packet? */
 810         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_MAX)) {
 811                 l2tp_info(tunnel, L2TP_MSG_DATA,
 812                           "%s: recv short packet (len=%d)\n",
 813                           tunnel->name, skb->len);
 814                 goto error;
 815         }
 816
 817         /* Trace packet contents, if enabled */
 818         if (tunnel->debug & L2TP_MSG_DATA) {
 819                 length = min(32u, skb->len);
 820                 if (!pskb_may_pull(skb, length))
 821                         goto error;
 822
 823                 pr_debug("%s: recv\n", tunnel->name);
 824                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 825         }
 826
 827         /* Point to L2TP header */
 828         optr = skb->data;
 829         ptr = skb->data;
 830
 831         /* Get L2TP header flags */
 832         hdrflags = ntohs(*(__be16 *)ptr);
 833
 834         /* Check protocol version */
 835         version = hdrflags & L2TP_HDR_VER_MASK;
 836         if (version != tunnel->version) {
 837                 l2tp_info(tunnel, L2TP_MSG_DATA,
 838                           "%s: recv protocol version mismatch: got %d expected %d\n",
 839                           tunnel->name, version, tunnel->version);
 840                 goto error;
 841         }
 842
 843         /* Get length of L2TP packet */
 844         length = skb->len;
 845
 846         /* If type is control packet, it is handled by userspace. */
 847         if (hdrflags & L2TP_HDRFLAG_T) {
 848                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 849                          "%s: recv control packet, len=%d\n",
 850                          tunnel->name, length);
 851                 goto error;
 852         }
 853
 854         /* Skip flags */
 855         ptr += 2;
 856
 857         if (tunnel->version == L2TP_HDR_VER_2) {
 858                 /* If length is present, skip it */
 859                 if (hdrflags & L2TP_HDRFLAG_L)
 860                         ptr += 2;
 861
 862                 /* Extract tunnel and session ID */
 863                 tunnel_id = ntohs(*(__be16 *)ptr);
 864                 ptr += 2;
 865                 session_id = ntohs(*(__be16 *)ptr);
 866                 ptr += 2;
 867         } else {
 868                 ptr += 2;       /* skip reserved bits */
 869                 tunnel_id = tunnel->tunnel_id;
 870                 session_id = ntohl(*(__be32 *)ptr);
 871                 ptr += 4;
 872         }
 873
 874         /* Find the session context */
 875         session = l2tp_tunnel_get_session(tunnel, session_id);
 876         if (!session || !session->recv_skb) {
 877                 if (session)
 878                         l2tp_session_dec_refcount(session);
 879
 880                 /* Not found? Pass to userspace to deal with */
 881                 l2tp_info(tunnel, L2TP_MSG_DATA,
 882                           "%s: no session found (%u/%u). Passing up.\n",
 883                           tunnel->name, tunnel_id, session_id);
 884                 goto error;
 885         }
 886
 887         if (tunnel->version == L2TP_HDR_VER_3 &&
 888             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr))
 889                 goto error;
 890
 891         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 892         l2tp_session_dec_refcount(session);
 893
 894         return 0;
 895
 896 error:
 897         /* Put UDP header back */
 898         __skb_push(skb, sizeof(struct udphdr));
 899
 900         return 1;
 901 }
 902
 903 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 904  * Return codes:
 905  * 0 : success.
 906  * <0: error
 907  * >0: skb should be passed up to userspace as UDP.
 908  */
 909 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 910 {
 911         struct l2tp_tunnel *tunnel;
 912
 913         tunnel = rcu_dereference_sk_user_data(sk);
 914         if (!tunnel)
 915                 goto pass_up;
 916
 917         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 918                  tunnel->name, skb->len);
 919
 920         if (l2tp_udp_recv_core(tunnel, skb))
 921                 goto pass_up;
 922
 923         return 0;
 924
 925 pass_up:
 926         return 1;
 927 }
 928 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 929
 930 /************************************************************************
 931  * Transmit handling
 932  ***********************************************************************/
 933
 934 /* Build an L2TP header for the session into the buffer provided.
 935  */
 936 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 937 {
 938         struct l2tp_tunnel *tunnel = session->tunnel;
 939         __be16 *bufp = buf;
 940         __be16 *optr = buf;
 941         u16 flags = L2TP_HDR_VER_2;
 942         u32 tunnel_id = tunnel->peer_tunnel_id;
 943         u32 session_id = session->peer_session_id;
 944
 945         if (session->send_seq)
 946                 flags |= L2TP_HDRFLAG_S;
 947
 948         /* Setup L2TP header. */
 949         *bufp++ = htons(flags);
 950         *bufp++ = htons(tunnel_id);
 951         *bufp++ = htons(session_id);
 952         if (session->send_seq) {
 953                 *bufp++ = htons(session->ns);
 954                 *bufp++ = 0;
 955                 session->ns++;
 956                 session->ns &= 0xffff;
 957                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
 958                          session->name, session->ns);
 959         }
 960
 961         return bufp - optr;
 962 }
 963
 964 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 965 {
 966         struct l2tp_tunnel *tunnel = session->tunnel;
 967         char *bufp = buf;
 968         char *optr = bufp;
 969
 970         /* Setup L2TP header. The header differs slightly for UDP and
 971          * IP encapsulations. For UDP, there is 4 bytes of flags.
 972          */
 973         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 974                 u16 flags = L2TP_HDR_VER_3;
 975                 *((__be16 *)bufp) = htons(flags);
 976                 bufp += 2;
 977                 *((__be16 *)bufp) = 0;
 978                 bufp += 2;
 979         }
 980
 981         *((__be32 *)bufp) = htonl(session->peer_session_id);
 982         bufp += 4;
 983         if (session->cookie_len) {
 984                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 985                 bufp += session->cookie_len;
 986         }
 987         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 988                 u32 l2h = 0;
 989
 990                 if (session->send_seq) {
 991                         l2h = 0x40000000 | session->ns;
 992                         session->ns++;
 993                         session->ns &= 0xffffff;
 994                         l2tp_dbg(session, L2TP_MSG_SEQ,
 995                                  "%s: updated ns to %u\n",
 996                                  session->name, session->ns);
 997                 }
 998
 999                 *((__be32 *)bufp) = htonl(l2h);
1000                 bufp += 4;
1001         }
1002
1003         return bufp - optr;
1004 }
1005
1006 static void l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1007                            struct flowi *fl, size_t data_len)
1008 {
1009         struct l2tp_tunnel *tunnel = session->tunnel;
1010         unsigned int len = skb->len;
1011         int error;
1012
1013         /* Debug */
1014         if (session->send_seq)
1015                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes, ns=%u\n",
1016                          session->name, data_len, session->ns - 1);
1017         else
1018                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes\n",
1019                          session->name, data_len);
1020
1021         if (session->debug & L2TP_MSG_DATA) {
1022                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1023                 unsigned char *datap = skb->data + uhlen;
1024
1025                 pr_debug("%s: xmit\n", session->name);
1026                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1027                                      datap, min_t(size_t, 32, len - uhlen));
1028         }
1029
1030         /* Queue the packet to IP for output */
1031         skb->ignore_df = 1;
1032         skb_dst_drop(skb);
1033 #if IS_ENABLED(CONFIG_IPV6)
1034         if (l2tp_sk_is_v6(tunnel->sock))
1035                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1036         else
1037 #endif
1038                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1039
1040         /* Update stats */
1041         if (error >= 0) {
1042                 atomic_long_inc(&tunnel->stats.tx_packets);
1043                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1044                 atomic_long_inc(&session->stats.tx_packets);
1045                 atomic_long_add(len, &session->stats.tx_bytes);
1046         } else {
1047                 atomic_long_inc(&tunnel->stats.tx_errors);
1048                 atomic_long_inc(&session->stats.tx_errors);
1049         }
1050 }
1051
1052 /* If caller requires the skb to have a ppp header, the header must be
1053  * inserted in the skb data before calling this function.
1054  */
1055 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1056 {
1057         int data_len = skb->len;
1058         struct l2tp_tunnel *tunnel = session->tunnel;
1059         struct sock *sk = tunnel->sock;
1060         struct flowi *fl;
1061         struct udphdr *uh;
1062         struct inet_sock *inet;
1063         int headroom;
1064         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1065         int udp_len;
1066         int ret = NET_XMIT_SUCCESS;
1067
1068         /* Check that there's enough headroom in the skb to insert IP,
1069          * UDP and L2TP headers. If not enough, expand it to
1070          * make room. Adjust truesize.
1071          */
1072         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1073                 uhlen + hdr_len;
1074         if (skb_cow_head(skb, headroom)) {
1075                 kfree_skb(skb);
1076                 return NET_XMIT_DROP;
1077         }
1078
1079         /* Setup L2TP header */
1080         session->build_header(session, __skb_push(skb, hdr_len));
1081
1082         /* Reset skb netfilter state */
1083         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1084         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1085                               IPSKB_REROUTED);
1086         nf_reset_ct(skb);
1087
1088         bh_lock_sock(sk);
1089         if (sock_owned_by_user(sk)) {
1090                 kfree_skb(skb);
1091                 ret = NET_XMIT_DROP;
1092                 goto out_unlock;
1093         }
1094
1095         /* The user-space may change the connection status for the user-space
1096          * provided socket at run time: we must check it under the socket lock
1097          */
1098         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1099                 kfree_skb(skb);
1100                 ret = NET_XMIT_DROP;
1101                 goto out_unlock;
1102         }
1103
1104         inet = inet_sk(sk);
1105         fl = &inet->cork.fl;
1106         switch (tunnel->encap) {
1107         case L2TP_ENCAPTYPE_UDP:
1108                 /* Setup UDP header */
1109                 __skb_push(skb, sizeof(*uh));
1110                 skb_reset_transport_header(skb);
1111                 uh = udp_hdr(skb);
1112                 uh->source = inet->inet_sport;
1113                 uh->dest = inet->inet_dport;
1114                 udp_len = uhlen + hdr_len + data_len;
1115                 uh->len = htons(udp_len);
1116
1117                 /* Calculate UDP checksum if configured to do so */
1118 #if IS_ENABLED(CONFIG_IPV6)
1119                 if (l2tp_sk_is_v6(sk))
1120                         udp6_set_csum(udp_get_no_check6_tx(sk),
1121                                       skb, &inet6_sk(sk)->saddr,
1122                                       &sk->sk_v6_daddr, udp_len);
1123                 else
1124 #endif
1125                         udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1126                                      inet->inet_daddr, udp_len);
1127                 break;
1128
1129         case L2TP_ENCAPTYPE_IP:
1130                 break;
1131         }
1132
1133         l2tp_xmit_core(session, skb, fl, data_len);
1134 out_unlock:
1135         bh_unlock_sock(sk);
1136
1137         return ret;
1138 }
1139 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1140
1141 /*****************************************************************************
1142  * Tinnel and session create/destroy.
1143  *****************************************************************************/
1144
1145 /* Tunnel socket destruct hook.
1146  * The tunnel context is deleted only when all session sockets have been
1147  * closed.
1148  */
1149 static void l2tp_tunnel_destruct(struct sock *sk)
1150 {
1151         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1152
1153         if (!tunnel)
1154                 goto end;
1155
1156         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1157
1158         /* Disable udp encapsulation */
1159         switch (tunnel->encap) {
1160         case L2TP_ENCAPTYPE_UDP:
1161                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1162                 (udp_sk(sk))->encap_type = 0;
1163                 (udp_sk(sk))->encap_rcv = NULL;
1164                 (udp_sk(sk))->encap_destroy = NULL;
1165                 break;
1166         case L2TP_ENCAPTYPE_IP:
1167                 break;
1168         }
1169
1170         /* Remove hooks into tunnel socket */
1171         sk->sk_destruct = tunnel->old_sk_destruct;
1172         sk->sk_user_data = NULL;
1173
1174         /* Call the original destructor */
1175         if (sk->sk_destruct)
1176                 (*sk->sk_destruct)(sk);
1177
1178         kfree_rcu(tunnel, rcu);
1179 end:
1180         return;
1181 }
1182
1183 /* When the tunnel is closed, all the attached sessions need to go too.
1184  */
1185 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1186 {
1187         int hash;
1188         struct hlist_node *walk;
1189         struct hlist_node *tmp;
1190         struct l2tp_session *session;
1191
1192         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1193                   tunnel->name);
1194
1195         write_lock_bh(&tunnel->hlist_lock);
1196         tunnel->acpt_newsess = false;
1197         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1198 again:
1199                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1200                         session = hlist_entry(walk, struct l2tp_session, hlist);
1201
1202                         l2tp_info(session, L2TP_MSG_CONTROL,
1203                                   "%s: closing session\n", session->name);
1204
1205                         hlist_del_init(&session->hlist);
1206
1207                         if (test_and_set_bit(0, &session->dead))
1208                                 goto again;
1209
1210                         write_unlock_bh(&tunnel->hlist_lock);
1211
1212                         __l2tp_session_unhash(session);
1213                         l2tp_session_queue_purge(session);
1214
1215                         if (session->session_close)
1216                                 (*session->session_close)(session);
1217
1218                         l2tp_session_dec_refcount(session);
1219
1220                         write_lock_bh(&tunnel->hlist_lock);
1221
1222                         /* Now restart from the beginning of this hash
1223                          * chain.  We always remove a session from the
1224                          * list so we are guaranteed to make forward
1225                          * progress.
1226                          */
1227                         goto again;
1228                 }
1229         }
1230         write_unlock_bh(&tunnel->hlist_lock);
1231 }
1232
1233 /* Tunnel socket destroy hook for UDP encapsulation */
1234 static void l2tp_udp_encap_destroy(struct sock *sk)
1235 {
1236         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1237
1238         if (tunnel)
1239                 l2tp_tunnel_delete(tunnel);
1240 }
1241
1242 /* Workqueue tunnel deletion function */
1243 static void l2tp_tunnel_del_work(struct work_struct *work)
1244 {
1245         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1246                                                   del_work);
1247         struct sock *sk = tunnel->sock;
1248         struct socket *sock = sk->sk_socket;
1249         struct l2tp_net *pn;
1250
1251         l2tp_tunnel_closeall(tunnel);
1252
1253         /* If the tunnel socket was created within the kernel, use
1254          * the sk API to release it here.
1255          */
1256         if (tunnel->fd < 0) {
1257                 if (sock) {
1258                         kernel_sock_shutdown(sock, SHUT_RDWR);
1259                         sock_release(sock);
1260                 }
1261         }
1262
1263         /* Remove the tunnel struct from the tunnel list */
1264         pn = l2tp_pernet(tunnel->l2tp_net);
1265         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1266         list_del_rcu(&tunnel->list);
1267         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1268
1269         /* drop initial ref */
1270         l2tp_tunnel_dec_refcount(tunnel);
1271
1272         /* drop workqueue ref */
1273         l2tp_tunnel_dec_refcount(tunnel);
1274 }
1275
1276 /* Create a socket for the tunnel, if one isn't set up by
1277  * userspace. This is used for static tunnels where there is no
1278  * managing L2TP daemon.
1279  *
1280  * Since we don't want these sockets to keep a namespace alive by
1281  * themselves, we drop the socket's namespace refcount after creation.
1282  * These sockets are freed when the namespace exits using the pernet
1283  * exit hook.
1284  */
1285 static int l2tp_tunnel_sock_create(struct net *net,
1286                                    u32 tunnel_id,
1287                                    u32 peer_tunnel_id,
1288                                    struct l2tp_tunnel_cfg *cfg,
1289                                    struct socket **sockp)
1290 {
1291         int err = -EINVAL;
1292         struct socket *sock = NULL;
1293         struct udp_port_cfg udp_conf;
1294
1295         switch (cfg->encap) {
1296         case L2TP_ENCAPTYPE_UDP:
1297                 memset(&udp_conf, 0, sizeof(udp_conf));
1298
1299 #if IS_ENABLED(CONFIG_IPV6)
1300                 if (cfg->local_ip6 && cfg->peer_ip6) {
1301                         udp_conf.family = AF_INET6;
1302                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1303                                sizeof(udp_conf.local_ip6));
1304                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1305                                sizeof(udp_conf.peer_ip6));
1306                         udp_conf.use_udp6_tx_checksums =
1307                           !cfg->udp6_zero_tx_checksums;
1308                         udp_conf.use_udp6_rx_checksums =
1309                           !cfg->udp6_zero_rx_checksums;
1310                 } else
1311 #endif
1312                 {
1313                         udp_conf.family = AF_INET;
1314                         udp_conf.local_ip = cfg->local_ip;
1315                         udp_conf.peer_ip = cfg->peer_ip;
1316                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1317                 }
1318
1319                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1320                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1321
1322                 err = udp_sock_create(net, &udp_conf, &sock);
1323                 if (err < 0)
1324                         goto out;
1325
1326                 break;
1327
1328         case L2TP_ENCAPTYPE_IP:
1329 #if IS_ENABLED(CONFIG_IPV6)
1330                 if (cfg->local_ip6 && cfg->peer_ip6) {
1331                         struct sockaddr_l2tpip6 ip6_addr = {0};
1332
1333                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1334                                                IPPROTO_L2TP, &sock);
1335                         if (err < 0)
1336                                 goto out;
1337
1338                         ip6_addr.l2tp_family = AF_INET6;
1339                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1340                                sizeof(ip6_addr.l2tp_addr));
1341                         ip6_addr.l2tp_conn_id = tunnel_id;
1342                         err = kernel_bind(sock, (struct sockaddr *)&ip6_addr,
1343                                           sizeof(ip6_addr));
1344                         if (err < 0)
1345                                 goto out;
1346
1347                         ip6_addr.l2tp_family = AF_INET6;
1348                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1349                                sizeof(ip6_addr.l2tp_addr));
1350                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1351                         err = kernel_connect(sock,
1352                                              (struct sockaddr *)&ip6_addr,
1353                                              sizeof(ip6_addr), 0);
1354                         if (err < 0)
1355                                 goto out;
1356                 } else
1357 #endif
1358                 {
1359                         struct sockaddr_l2tpip ip_addr = {0};
1360
1361                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1362                                                IPPROTO_L2TP, &sock);
1363                         if (err < 0)
1364                                 goto out;
1365
1366                         ip_addr.l2tp_family = AF_INET;
1367                         ip_addr.l2tp_addr = cfg->local_ip;
1368                         ip_addr.l2tp_conn_id = tunnel_id;
1369                         err = kernel_bind(sock, (struct sockaddr *)&ip_addr,
1370                                           sizeof(ip_addr));
1371                         if (err < 0)
1372                                 goto out;
1373
1374                         ip_addr.l2tp_family = AF_INET;
1375                         ip_addr.l2tp_addr = cfg->peer_ip;
1376                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1377                         err = kernel_connect(sock, (struct sockaddr *)&ip_addr,
1378                                              sizeof(ip_addr), 0);
1379                         if (err < 0)
1380                                 goto out;
1381                 }
1382                 break;
1383
1384         default:
1385                 goto out;
1386         }
1387
1388 out:
1389         *sockp = sock;
1390         if (err < 0 && sock) {
1391                 kernel_sock_shutdown(sock, SHUT_RDWR);
1392                 sock_release(sock);
1393                 *sockp = NULL;
1394         }
1395
1396         return err;
1397 }
1398
1399 static struct lock_class_key l2tp_socket_class;
1400
1401 int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32 peer_tunnel_id,
1402                        struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1403 {
1404         struct l2tp_tunnel *tunnel = NULL;
1405         int err;
1406         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1407
1408         if (cfg)
1409                 encap = cfg->encap;
1410
1411         tunnel = kzalloc(sizeof(*tunnel), GFP_KERNEL);
1412         if (!tunnel) {
1413                 err = -ENOMEM;
1414                 goto err;
1415         }
1416
1417         tunnel->version = version;
1418         tunnel->tunnel_id = tunnel_id;
1419         tunnel->peer_tunnel_id = peer_tunnel_id;
1420         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1421
1422         tunnel->magic = L2TP_TUNNEL_MAGIC;
1423         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1424         rwlock_init(&tunnel->hlist_lock);
1425         tunnel->acpt_newsess = true;
1426
1427         if (cfg)
1428                 tunnel->debug = cfg->debug;
1429
1430         tunnel->encap = encap;
1431
1432         refcount_set(&tunnel->ref_count, 1);
1433         tunnel->fd = fd;
1434
1435         /* Init delete workqueue struct */
1436         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1437
1438         INIT_LIST_HEAD(&tunnel->list);
1439
1440         err = 0;
1441 err:
1442         if (tunnelp)
1443                 *tunnelp = tunnel;
1444
1445         return err;
1446 }
1447 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1448
1449 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1450                                 enum l2tp_encap_type encap)
1451 {
1452         if (!net_eq(sock_net(sk), net))
1453                 return -EINVAL;
1454
1455         if (sk->sk_type != SOCK_DGRAM)
1456                 return -EPROTONOSUPPORT;
1457
1458         if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1459                 return -EPROTONOSUPPORT;
1460
1461         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1462             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1463                 return -EPROTONOSUPPORT;
1464
1465         if (sk->sk_user_data)
1466                 return -EBUSY;
1467
1468         return 0;
1469 }
1470
1471 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1472                          struct l2tp_tunnel_cfg *cfg)
1473 {
1474         struct l2tp_tunnel *tunnel_walk;
1475         struct l2tp_net *pn;
1476         struct socket *sock;
1477         struct sock *sk;
1478         int ret;
1479
1480         if (tunnel->fd < 0) {
1481                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1482                                               tunnel->peer_tunnel_id, cfg,
1483                                               &sock);
1484                 if (ret < 0)
1485                         goto err;
1486         } else {
1487                 sock = sockfd_lookup(tunnel->fd, &ret);
1488                 if (!sock)
1489                         goto err;
1490
1491                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1492                 if (ret < 0)
1493                         goto err_sock;
1494         }
1495
1496         tunnel->l2tp_net = net;
1497         pn = l2tp_pernet(net);
1498
1499         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1500         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1501                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1502                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1503
1504                         ret = -EEXIST;
1505                         goto err_sock;
1506                 }
1507         }
1508         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1509         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1510
1511         sk = sock->sk;
1512         sock_hold(sk);
1513         tunnel->sock = sk;
1514
1515         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1516                 struct udp_tunnel_sock_cfg udp_cfg = {
1517                         .sk_user_data = tunnel,
1518                         .encap_type = UDP_ENCAP_L2TPINUDP,
1519                         .encap_rcv = l2tp_udp_encap_recv,
1520                         .encap_destroy = l2tp_udp_encap_destroy,
1521                 };
1522
1523                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1524         } else {
1525                 sk->sk_user_data = tunnel;
1526         }
1527
1528         tunnel->old_sk_destruct = sk->sk_destruct;
1529         sk->sk_destruct = &l2tp_tunnel_destruct;
1530         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1531                                    "l2tp_sock");
1532         sk->sk_allocation = GFP_ATOMIC;
1533
1534         if (tunnel->fd >= 0)
1535                 sockfd_put(sock);
1536
1537         return 0;
1538
1539 err_sock:
1540         if (tunnel->fd < 0)
1541                 sock_release(sock);
1542         else
1543                 sockfd_put(sock);
1544 err:
1545         return ret;
1546 }
1547 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1548
1549 /* This function is used by the netlink TUNNEL_DELETE command.
1550  */
1551 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1552 {
1553         if (!test_and_set_bit(0, &tunnel->dead)) {
1554                 l2tp_tunnel_inc_refcount(tunnel);
1555                 queue_work(l2tp_wq, &tunnel->del_work);
1556         }
1557 }
1558 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1559
1560 /* Really kill the session.
1561  */
1562 void l2tp_session_free(struct l2tp_session *session)
1563 {
1564         struct l2tp_tunnel *tunnel = session->tunnel;
1565
1566         if (tunnel) {
1567                 if (WARN_ON(tunnel->magic != L2TP_TUNNEL_MAGIC))
1568                         goto out;
1569                 l2tp_tunnel_dec_refcount(tunnel);
1570         }
1571
1572 out:
1573         kfree(session);
1574 }
1575 EXPORT_SYMBOL_GPL(l2tp_session_free);
1576
1577 /* Remove an l2tp session from l2tp_core's hash lists.
1578  * Provides a tidyup interface for pseudowire code which can't just route all
1579  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1580  * callback.
1581  */
1582 void __l2tp_session_unhash(struct l2tp_session *session)
1583 {
1584         struct l2tp_tunnel *tunnel = session->tunnel;
1585
1586         /* Remove the session from core hashes */
1587         if (tunnel) {
1588                 /* Remove from the per-tunnel hash */
1589                 write_lock_bh(&tunnel->hlist_lock);
1590                 hlist_del_init(&session->hlist);
1591                 write_unlock_bh(&tunnel->hlist_lock);
1592
1593                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1594                 if (tunnel->version != L2TP_HDR_VER_2) {
1595                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1596
1597                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1598                         hlist_del_init_rcu(&session->global_hlist);
1599                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1600                         synchronize_rcu();
1601                 }
1602         }
1603 }
1604 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1605
1606 /* This function is used by the netlink SESSION_DELETE command and by
1607  * pseudowire modules.
1608  */
1609 int l2tp_session_delete(struct l2tp_session *session)
1610 {
1611         if (test_and_set_bit(0, &session->dead))
1612                 return 0;
1613
1614         __l2tp_session_unhash(session);
1615         l2tp_session_queue_purge(session);
1616         if (session->session_close)
1617                 (*session->session_close)(session);
1618
1619         l2tp_session_dec_refcount(session);
1620
1621         return 0;
1622 }
1623 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1624
1625 /* We come here whenever a session's send_seq, cookie_len or
1626  * l2specific_type parameters are set.
1627  */
1628 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1629 {
1630         if (version == L2TP_HDR_VER_2) {
1631                 session->hdr_len = 6;
1632                 if (session->send_seq)
1633                         session->hdr_len += 4;
1634         } else {
1635                 session->hdr_len = 4 + session->cookie_len;
1636                 session->hdr_len += l2tp_get_l2specific_len(session);
1637                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1638                         session->hdr_len += 4;
1639         }
1640 }
1641 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1642
1643 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id,
1644                                          u32 peer_session_id, struct l2tp_session_cfg *cfg)
1645 {
1646         struct l2tp_session *session;
1647
1648         session = kzalloc(sizeof(*session) + priv_size, GFP_KERNEL);
1649         if (session) {
1650                 session->magic = L2TP_SESSION_MAGIC;
1651                 session->tunnel = tunnel;
1652
1653                 session->session_id = session_id;
1654                 session->peer_session_id = peer_session_id;
1655                 session->nr = 0;
1656                 if (tunnel->version == L2TP_HDR_VER_2)
1657                         session->nr_max = 0xffff;
1658                 else
1659                         session->nr_max = 0xffffff;
1660                 session->nr_window_size = session->nr_max / 2;
1661                 session->nr_oos_count_max = 4;
1662
1663                 /* Use NR of first received packet */
1664                 session->reorder_skip = 1;
1665
1666                 sprintf(&session->name[0], "sess %u/%u",
1667                         tunnel->tunnel_id, session->session_id);
1668
1669                 skb_queue_head_init(&session->reorder_q);
1670
1671                 INIT_HLIST_NODE(&session->hlist);
1672                 INIT_HLIST_NODE(&session->global_hlist);
1673
1674                 /* Inherit debug options from tunnel */
1675                 session->debug = tunnel->debug;
1676
1677                 if (cfg) {
1678                         session->pwtype = cfg->pw_type;
1679                         session->debug = cfg->debug;
1680                         session->send_seq = cfg->send_seq;
1681                         session->recv_seq = cfg->recv_seq;
1682                         session->lns_mode = cfg->lns_mode;
1683                         session->reorder_timeout = cfg->reorder_timeout;
1684                         session->l2specific_type = cfg->l2specific_type;
1685                         session->cookie_len = cfg->cookie_len;
1686                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1687                         session->peer_cookie_len = cfg->peer_cookie_len;
1688                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1689                 }
1690
1691                 if (tunnel->version == L2TP_HDR_VER_2)
1692                         session->build_header = l2tp_build_l2tpv2_header;
1693                 else
1694                         session->build_header = l2tp_build_l2tpv3_header;
1695
1696                 l2tp_session_set_header_len(session, tunnel->version);
1697
1698                 refcount_set(&session->ref_count, 1);
1699
1700                 return session;
1701         }
1702
1703         return ERR_PTR(-ENOMEM);
1704 }
1705 EXPORT_SYMBOL_GPL(l2tp_session_create);
1706
1707 /*****************************************************************************
1708  * Init and cleanup
1709  *****************************************************************************/
1710
1711 static __net_init int l2tp_init_net(struct net *net)
1712 {
1713         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1714         int hash;
1715
1716         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1717         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1718
1719         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1720                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1721
1722         spin_lock_init(&pn->l2tp_session_hlist_lock);
1723
1724         return 0;
1725 }
1726
1727 static __net_exit void l2tp_exit_net(struct net *net)
1728 {
1729         struct l2tp_net *pn = l2tp_pernet(net);
1730         struct l2tp_tunnel *tunnel = NULL;
1731         int hash;
1732
1733         rcu_read_lock_bh();
1734         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1735                 l2tp_tunnel_delete(tunnel);
1736         }
1737         rcu_read_unlock_bh();
1738
1739         if (l2tp_wq)
1740                 flush_workqueue(l2tp_wq);
1741         rcu_barrier();
1742
1743         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1744                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1745 }
1746
1747 static struct pernet_operations l2tp_net_ops = {
1748         .init = l2tp_init_net,
1749         .exit = l2tp_exit_net,
1750         .id   = &l2tp_net_id,
1751         .size = sizeof(struct l2tp_net),
1752 };
1753
1754 static int __init l2tp_init(void)
1755 {
1756         int rc = 0;
1757
1758         rc = register_pernet_device(&l2tp_net_ops);
1759         if (rc)
1760                 goto out;
1761
1762         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1763         if (!l2tp_wq) {
1764                 pr_err("alloc_workqueue failed\n");
1765                 unregister_pernet_device(&l2tp_net_ops);
1766                 rc = -ENOMEM;
1767                 goto out;
1768         }
1769
1770         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1771
1772 out:
1773         return rc;
1774 }
1775
1776 static void __exit l2tp_exit(void)
1777 {
1778         unregister_pernet_device(&l2tp_net_ops);
1779         if (l2tp_wq) {
1780                 destroy_workqueue(l2tp_wq);
1781                 l2tp_wq = NULL;
1782         }
1783 }
1784
1785 module_init(l2tp_init);
1786 module_exit(l2tp_exit);
1787
1788 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1789 MODULE_DESCRIPTION("L2TP core");
1790 MODULE_LICENSE("GPL");
1791 MODULE_VERSION(L2TP_DRV_VERSION);