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         BUG_ON(!net);
 126
 127         return net_generic(net, l2tp_net_id);
 128 }
 129
 130 /* Session hash global list for L2TPv3.
 131  * The session_id SHOULD be random according to RFC3931, but several
 132  * L2TP implementations use incrementing session_ids.  So we do a real
 133  * hash on the session_id, rather than a simple bitmask.
 134  */
 135 static inline struct hlist_head *
 136 l2tp_session_id_hash_2(struct l2tp_net *pn, u32 session_id)
 137 {
 138         return &pn->l2tp_session_hlist[hash_32(session_id, L2TP_HASH_BITS_2)];
 139
 140 }
 141
 142 /* Session hash list.
 143  * The session_id SHOULD be random according to RFC2661, but several
 144  * L2TP implementations (Cisco and Microsoft) use incrementing
 145  * session_ids.  So we do a real hash on the session_id, rather than a
 146  * simple bitmask.
 147  */
 148 static inline struct hlist_head *
 149 l2tp_session_id_hash(struct l2tp_tunnel *tunnel, u32 session_id)
 150 {
 151         return &tunnel->session_hlist[hash_32(session_id, L2TP_HASH_BITS)];
 152 }
 153
 154 void l2tp_tunnel_free(struct l2tp_tunnel *tunnel)
 155 {
 156         sock_put(tunnel->sock);
 157         /* the tunnel is freed in the socket destructor */
 158 }
 159 EXPORT_SYMBOL(l2tp_tunnel_free);
 160
 161 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
 162 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
 163 {
 164         const struct l2tp_net *pn = l2tp_pernet(net);
 165         struct l2tp_tunnel *tunnel;
 166
 167         rcu_read_lock_bh();
 168         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 169                 if (tunnel->tunnel_id == tunnel_id &&
 170                     refcount_inc_not_zero(&tunnel->ref_count)) {
 171                         rcu_read_unlock_bh();
 172
 173                         return tunnel;
 174                 }
 175         }
 176         rcu_read_unlock_bh();
 177
 178         return NULL;
 179 }
 180 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
 181
 182 struct l2tp_tunnel *l2tp_tunnel_get_nth(const struct net *net, int nth)
 183 {
 184         const struct l2tp_net *pn = l2tp_pernet(net);
 185         struct l2tp_tunnel *tunnel;
 186         int count = 0;
 187
 188         rcu_read_lock_bh();
 189         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 190                 if (++count > nth &&
 191                     refcount_inc_not_zero(&tunnel->ref_count)) {
 192                         rcu_read_unlock_bh();
 193                         return tunnel;
 194                 }
 195         }
 196         rcu_read_unlock_bh();
 197
 198         return NULL;
 199 }
 200 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_nth);
 201
 202 struct l2tp_session *l2tp_tunnel_get_session(struct l2tp_tunnel *tunnel,
 203                                              u32 session_id)
 204 {
 205         struct hlist_head *session_list;
 206         struct l2tp_session *session;
 207
 208         session_list = l2tp_session_id_hash(tunnel, session_id);
 209
 210         read_lock_bh(&tunnel->hlist_lock);
 211         hlist_for_each_entry(session, session_list, hlist)
 212                 if (session->session_id == session_id) {
 213                         l2tp_session_inc_refcount(session);
 214                         read_unlock_bh(&tunnel->hlist_lock);
 215
 216                         return session;
 217                 }
 218         read_unlock_bh(&tunnel->hlist_lock);
 219
 220         return NULL;
 221 }
 222 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_session);
 223
 224 struct l2tp_session *l2tp_session_get(const struct net *net, u32 session_id)
 225 {
 226         struct hlist_head *session_list;
 227         struct l2tp_session *session;
 228
 229         session_list = l2tp_session_id_hash_2(l2tp_pernet(net), session_id);
 230
 231         rcu_read_lock_bh();
 232         hlist_for_each_entry_rcu(session, session_list, global_hlist)
 233                 if (session->session_id == session_id) {
 234                         l2tp_session_inc_refcount(session);
 235                         rcu_read_unlock_bh();
 236
 237                         return session;
 238                 }
 239         rcu_read_unlock_bh();
 240
 241         return NULL;
 242 }
 243 EXPORT_SYMBOL_GPL(l2tp_session_get);
 244
 245 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth)
 246 {
 247         int hash;
 248         struct l2tp_session *session;
 249         int count = 0;
 250
 251         read_lock_bh(&tunnel->hlist_lock);
 252         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 253                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 254                         if (++count > nth) {
 255                                 l2tp_session_inc_refcount(session);
 256                                 read_unlock_bh(&tunnel->hlist_lock);
 257                                 return session;
 258                         }
 259                 }
 260         }
 261
 262         read_unlock_bh(&tunnel->hlist_lock);
 263
 264         return NULL;
 265 }
 266 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
 267
 268 /* Lookup a session by interface name.
 269  * This is very inefficient but is only used by management interfaces.
 270  */
 271 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
 272                                                 const char *ifname)
 273 {
 274         struct l2tp_net *pn = l2tp_pernet(net);
 275         int hash;
 276         struct l2tp_session *session;
 277
 278         rcu_read_lock_bh();
 279         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 280                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 281                         if (!strcmp(session->ifname, ifname)) {
 282                                 l2tp_session_inc_refcount(session);
 283                                 rcu_read_unlock_bh();
 284
 285                                 return session;
 286                         }
 287                 }
 288         }
 289
 290         rcu_read_unlock_bh();
 291
 292         return NULL;
 293 }
 294 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
 295
 296 int l2tp_session_register(struct l2tp_session *session,
 297                           struct l2tp_tunnel *tunnel)
 298 {
 299         struct l2tp_session *session_walk;
 300         struct hlist_head *g_head;
 301         struct hlist_head *head;
 302         struct l2tp_net *pn;
 303         int err;
 304
 305         head = l2tp_session_id_hash(tunnel, session->session_id);
 306
 307         write_lock_bh(&tunnel->hlist_lock);
 308         if (!tunnel->acpt_newsess) {
 309                 err = -ENODEV;
 310                 goto err_tlock;
 311         }
 312
 313         hlist_for_each_entry(session_walk, head, hlist)
 314                 if (session_walk->session_id == session->session_id) {
 315                         err = -EEXIST;
 316                         goto err_tlock;
 317                 }
 318
 319         if (tunnel->version == L2TP_HDR_VER_3) {
 320                 pn = l2tp_pernet(tunnel->l2tp_net);
 321                 g_head = l2tp_session_id_hash_2(pn, session->session_id);
 322
 323                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
 324
 325                 /* IP encap expects session IDs to be globally unique, while
 326                  * UDP encap doesn't.
 327                  */
 328                 hlist_for_each_entry(session_walk, g_head, global_hlist)
 329                         if (session_walk->session_id == session->session_id &&
 330                             (session_walk->tunnel->encap == L2TP_ENCAPTYPE_IP ||
 331                              tunnel->encap == L2TP_ENCAPTYPE_IP)) {
 332                                 err = -EEXIST;
 333                                 goto err_tlock_pnlock;
 334                         }
 335
 336                 l2tp_tunnel_inc_refcount(tunnel);
 337                 hlist_add_head_rcu(&session->global_hlist, g_head);
 338
 339                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 340         } else {
 341                 l2tp_tunnel_inc_refcount(tunnel);
 342         }
 343
 344         hlist_add_head(&session->hlist, head);
 345         write_unlock_bh(&tunnel->hlist_lock);
 346
 347         return 0;
 348
 349 err_tlock_pnlock:
 350         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 351 err_tlock:
 352         write_unlock_bh(&tunnel->hlist_lock);
 353
 354         return err;
 355 }
 356 EXPORT_SYMBOL_GPL(l2tp_session_register);
 357
 358 /*****************************************************************************
 359  * Receive data handling
 360  *****************************************************************************/
 361
 362 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 363  * number.
 364  */
 365 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 366 {
 367         struct sk_buff *skbp;
 368         struct sk_buff *tmp;
 369         u32 ns = L2TP_SKB_CB(skb)->ns;
 370
 371         spin_lock_bh(&session->reorder_q.lock);
 372         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 373                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 374                         __skb_queue_before(&session->reorder_q, skbp, skb);
 375                         l2tp_dbg(session, L2TP_MSG_SEQ,
 376                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 377                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 378                                  skb_queue_len(&session->reorder_q));
 379                         atomic_long_inc(&session->stats.rx_oos_packets);
 380                         goto out;
 381                 }
 382         }
 383
 384         __skb_queue_tail(&session->reorder_q, skb);
 385
 386 out:
 387         spin_unlock_bh(&session->reorder_q.lock);
 388 }
 389
 390 /* Dequeue a single skb.
 391  */
 392 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 393 {
 394         struct l2tp_tunnel *tunnel = session->tunnel;
 395         int length = L2TP_SKB_CB(skb)->length;
 396
 397         /* We're about to requeue the skb, so return resources
 398          * to its current owner (a socket receive buffer).
 399          */
 400         skb_orphan(skb);
 401
 402         atomic_long_inc(&tunnel->stats.rx_packets);
 403         atomic_long_add(length, &tunnel->stats.rx_bytes);
 404         atomic_long_inc(&session->stats.rx_packets);
 405         atomic_long_add(length, &session->stats.rx_bytes);
 406
 407         if (L2TP_SKB_CB(skb)->has_seq) {
 408                 /* Bump our Nr */
 409                 session->nr++;
 410                 session->nr &= session->nr_max;
 411
 412                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 413                          session->name, session->nr);
 414         }
 415
 416         /* call private receive handler */
 417         if (session->recv_skb != NULL)
 418                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 419         else
 420                 kfree_skb(skb);
 421 }
 422
 423 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 424  * Skbs that have been in the queue for too long are simply discarded.
 425  */
 426 static void l2tp_recv_dequeue(struct l2tp_session *session)
 427 {
 428         struct sk_buff *skb;
 429         struct sk_buff *tmp;
 430
 431         /* If the pkt at the head of the queue has the nr that we
 432          * expect to send up next, dequeue it and any other
 433          * in-sequence packets behind it.
 434          */
 435 start:
 436         spin_lock_bh(&session->reorder_q.lock);
 437         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 438                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 439                         atomic_long_inc(&session->stats.rx_seq_discards);
 440                         atomic_long_inc(&session->stats.rx_errors);
 441                         l2tp_dbg(session, L2TP_MSG_SEQ,
 442                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 443                                  session->name, L2TP_SKB_CB(skb)->ns,
 444                                  L2TP_SKB_CB(skb)->length, session->nr,
 445                                  skb_queue_len(&session->reorder_q));
 446                         session->reorder_skip = 1;
 447                         __skb_unlink(skb, &session->reorder_q);
 448                         kfree_skb(skb);
 449                         continue;
 450                 }
 451
 452                 if (L2TP_SKB_CB(skb)->has_seq) {
 453                         if (session->reorder_skip) {
 454                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 455                                          "%s: advancing nr to next pkt: %u -> %u",
 456                                          session->name, session->nr,
 457                                          L2TP_SKB_CB(skb)->ns);
 458                                 session->reorder_skip = 0;
 459                                 session->nr = L2TP_SKB_CB(skb)->ns;
 460                         }
 461                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 462                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 463                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 464                                          session->name, L2TP_SKB_CB(skb)->ns,
 465                                          L2TP_SKB_CB(skb)->length, session->nr,
 466                                          skb_queue_len(&session->reorder_q));
 467                                 goto out;
 468                         }
 469                 }
 470                 __skb_unlink(skb, &session->reorder_q);
 471
 472                 /* Process the skb. We release the queue lock while we
 473                  * do so to let other contexts process the queue.
 474                  */
 475                 spin_unlock_bh(&session->reorder_q.lock);
 476                 l2tp_recv_dequeue_skb(session, skb);
 477                 goto start;
 478         }
 479
 480 out:
 481         spin_unlock_bh(&session->reorder_q.lock);
 482 }
 483
 484 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 485 {
 486         u32 nws;
 487
 488         if (nr >= session->nr)
 489                 nws = nr - session->nr;
 490         else
 491                 nws = (session->nr_max + 1) - (session->nr - nr);
 492
 493         return nws < session->nr_window_size;
 494 }
 495
 496 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 497  * acceptable, else non-zero.
 498  */
 499 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 500 {
 501         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
 502                 /* Packet sequence number is outside allowed window.
 503                  * Discard it.
 504                  */
 505                 l2tp_dbg(session, L2TP_MSG_SEQ,
 506                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
 507                          session->name, L2TP_SKB_CB(skb)->ns,
 508                          L2TP_SKB_CB(skb)->length, session->nr);
 509                 goto discard;
 510         }
 511
 512         if (session->reorder_timeout != 0) {
 513                 /* Packet reordering enabled. Add skb to session's
 514                  * reorder queue, in order of ns.
 515                  */
 516                 l2tp_recv_queue_skb(session, skb);
 517                 goto out;
 518         }
 519
 520         /* Packet reordering disabled. Discard out-of-sequence packets, while
 521          * tracking the number if in-sequence packets after the first OOS packet
 522          * is seen. After nr_oos_count_max in-sequence packets, reset the
 523          * sequence number to re-enable packet reception.
 524          */
 525         if (L2TP_SKB_CB(skb)->ns == session->nr) {
 526                 skb_queue_tail(&session->reorder_q, skb);
 527         } else {
 528                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
 529                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 530
 531                 if (nr_oos == nr_next)
 532                         session->nr_oos_count++;
 533                 else
 534                         session->nr_oos_count = 0;
 535
 536                 session->nr_oos = nr_oos;
 537                 if (session->nr_oos_count > session->nr_oos_count_max) {
 538                         session->reorder_skip = 1;
 539                         l2tp_dbg(session, L2TP_MSG_SEQ,
 540                                  "%s: %d oos packets received. Resetting sequence numbers\n",
 541                                  session->name, session->nr_oos_count);
 542                 }
 543                 if (!session->reorder_skip) {
 544                         atomic_long_inc(&session->stats.rx_seq_discards);
 545                         l2tp_dbg(session, L2TP_MSG_SEQ,
 546                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 547                                  session->name, L2TP_SKB_CB(skb)->ns,
 548                                  L2TP_SKB_CB(skb)->length, session->nr,
 549                                  skb_queue_len(&session->reorder_q));
 550                         goto discard;
 551                 }
 552                 skb_queue_tail(&session->reorder_q, skb);
 553         }
 554
 555 out:
 556         return 0;
 557
 558 discard:
 559         return 1;
 560 }
 561
 562 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 563  * and L2TPv3 data frames here.
 564  *
 565  * L2TPv2 Data Message Header
 566  *
 567  *  0                   1                   2                   3
 568  *  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
 569  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 570  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 571  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 572  * |           Tunnel ID           |           Session ID          |
 573  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 574  * |             Ns (opt)          |             Nr (opt)          |
 575  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 576  * |      Offset Size (opt)        |    Offset pad... (opt)
 577  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 578  *
 579  * Data frames are marked by T=0. All other fields are the same as
 580  * those in L2TP control frames.
 581  *
 582  * L2TPv3 Data Message Header
 583  *
 584  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 585  * |                      L2TP Session Header                      |
 586  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 587  * |                      L2-Specific Sublayer                     |
 588  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 589  * |                        Tunnel Payload                      ...
 590  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 591  *
 592  * L2TPv3 Session Header Over IP
 593  *
 594  *  0                   1                   2                   3
 595  *  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
 596  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 597  * |                           Session ID                          |
 598  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 599  * |               Cookie (optional, maximum 64 bits)...
 600  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 601  *                                                                 |
 602  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 603  *
 604  * L2TPv3 L2-Specific Sublayer Format
 605  *
 606  *  0                   1                   2                   3
 607  *  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
 608  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 609  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 610  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 611  *
 612  * Cookie value and sublayer format are negotiated with the peer when
 613  * the session is set up. Unlike L2TPv2, we do not need to parse the
 614  * packet header to determine if optional fields are present.
 615  *
 616  * Caller must already have parsed the frame and determined that it is
 617  * a data (not control) frame before coming here. Fields up to the
 618  * session-id have already been parsed and ptr points to the data
 619  * after the session-id.
 620  */
 621 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 622                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 623                       int length)
 624 {
 625         struct l2tp_tunnel *tunnel = session->tunnel;
 626         int offset;
 627         u32 ns, nr;
 628
 629         /* Parse and check optional cookie */
 630         if (session->peer_cookie_len > 0) {
 631                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 632                         l2tp_info(tunnel, L2TP_MSG_DATA,
 633                                   "%s: cookie mismatch (%u/%u). Discarding.\n",
 634                                   tunnel->name, tunnel->tunnel_id,
 635                                   session->session_id);
 636                         atomic_long_inc(&session->stats.rx_cookie_discards);
 637                         goto discard;
 638                 }
 639                 ptr += session->peer_cookie_len;
 640         }
 641
 642         /* Handle the optional sequence numbers. Sequence numbers are
 643          * in different places for L2TPv2 and L2TPv3.
 644          *
 645          * If we are the LAC, enable/disable sequence numbers under
 646          * the control of the LNS.  If no sequence numbers present but
 647          * we were expecting them, discard frame.
 648          */
 649         ns = nr = 0;
 650         L2TP_SKB_CB(skb)->has_seq = 0;
 651         if (tunnel->version == L2TP_HDR_VER_2) {
 652                 if (hdrflags & L2TP_HDRFLAG_S) {
 653                         ns = ntohs(*(__be16 *) ptr);
 654                         ptr += 2;
 655                         nr = ntohs(*(__be16 *) ptr);
 656                         ptr += 2;
 657
 658                         /* Store L2TP info in the skb */
 659                         L2TP_SKB_CB(skb)->ns = ns;
 660                         L2TP_SKB_CB(skb)->has_seq = 1;
 661
 662                         l2tp_dbg(session, L2TP_MSG_SEQ,
 663                                  "%s: recv data ns=%u, nr=%u, session nr=%u\n",
 664                                  session->name, ns, nr, session->nr);
 665                 }
 666         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 667                 u32 l2h = ntohl(*(__be32 *) ptr);
 668
 669                 if (l2h & 0x40000000) {
 670                         ns = l2h & 0x00ffffff;
 671
 672                         /* Store L2TP info in the skb */
 673                         L2TP_SKB_CB(skb)->ns = ns;
 674                         L2TP_SKB_CB(skb)->has_seq = 1;
 675
 676                         l2tp_dbg(session, L2TP_MSG_SEQ,
 677                                  "%s: recv data ns=%u, session nr=%u\n",
 678                                  session->name, ns, session->nr);
 679                 }
 680                 ptr += 4;
 681         }
 682
 683         if (L2TP_SKB_CB(skb)->has_seq) {
 684                 /* Received a packet with sequence numbers. If we're the LNS,
 685                  * check if we sre sending sequence numbers and if not,
 686                  * configure it so.
 687                  */
 688                 if ((!session->lns_mode) && (!session->send_seq)) {
 689                         l2tp_info(session, L2TP_MSG_SEQ,
 690                                   "%s: requested to enable seq numbers by LNS\n",
 691                                   session->name);
 692                         session->send_seq = 1;
 693                         l2tp_session_set_header_len(session, tunnel->version);
 694                 }
 695         } else {
 696                 /* No sequence numbers.
 697                  * If user has configured mandatory sequence numbers, discard.
 698                  */
 699                 if (session->recv_seq) {
 700                         l2tp_warn(session, L2TP_MSG_SEQ,
 701                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 702                                   session->name);
 703                         atomic_long_inc(&session->stats.rx_seq_discards);
 704                         goto discard;
 705                 }
 706
 707                 /* If we're the LAC and we're sending sequence numbers, the
 708                  * LNS has requested that we no longer send sequence numbers.
 709                  * If we're the LNS and we're sending sequence numbers, the
 710                  * LAC is broken. Discard the frame.
 711                  */
 712                 if ((!session->lns_mode) && (session->send_seq)) {
 713                         l2tp_info(session, L2TP_MSG_SEQ,
 714                                   "%s: requested to disable seq numbers by LNS\n",
 715                                   session->name);
 716                         session->send_seq = 0;
 717                         l2tp_session_set_header_len(session, tunnel->version);
 718                 } else if (session->send_seq) {
 719                         l2tp_warn(session, L2TP_MSG_SEQ,
 720                                   "%s: recv data has no seq numbers when required. Discarding.\n",
 721                                   session->name);
 722                         atomic_long_inc(&session->stats.rx_seq_discards);
 723                         goto discard;
 724                 }
 725         }
 726
 727         /* Session data offset is defined only for L2TPv2 and is
 728          * indicated by an optional 16-bit value in the header.
 729          */
 730         if (tunnel->version == L2TP_HDR_VER_2) {
 731                 /* If offset bit set, skip it. */
 732                 if (hdrflags & L2TP_HDRFLAG_O) {
 733                         offset = ntohs(*(__be16 *)ptr);
 734                         ptr += 2 + offset;
 735                 }
 736         }
 737
 738         offset = ptr - optr;
 739         if (!pskb_may_pull(skb, offset))
 740                 goto discard;
 741
 742         __skb_pull(skb, offset);
 743
 744         /* Prepare skb for adding to the session's reorder_q.  Hold
 745          * packets for max reorder_timeout or 1 second if not
 746          * reordering.
 747          */
 748         L2TP_SKB_CB(skb)->length = length;
 749         L2TP_SKB_CB(skb)->expires = jiffies +
 750                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 751
 752         /* Add packet to the session's receive queue. Reordering is done here, if
 753          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 754          */
 755         if (L2TP_SKB_CB(skb)->has_seq) {
 756                 if (l2tp_recv_data_seq(session, skb))
 757                         goto discard;
 758         } else {
 759                 /* No sequence numbers. Add the skb to the tail of the
 760                  * reorder queue. This ensures that it will be
 761                  * delivered after all previous sequenced skbs.
 762                  */
 763                 skb_queue_tail(&session->reorder_q, skb);
 764         }
 765
 766         /* Try to dequeue as many skbs from reorder_q as we can. */
 767         l2tp_recv_dequeue(session);
 768
 769         return;
 770
 771 discard:
 772         atomic_long_inc(&session->stats.rx_errors);
 773         kfree_skb(skb);
 774 }
 775 EXPORT_SYMBOL(l2tp_recv_common);
 776
 777 /* Drop skbs from the session's reorder_q
 778  */
 779 static int l2tp_session_queue_purge(struct l2tp_session *session)
 780 {
 781         struct sk_buff *skb = NULL;
 782         BUG_ON(!session);
 783         BUG_ON(session->magic != L2TP_SESSION_MAGIC);
 784         while ((skb = skb_dequeue(&session->reorder_q))) {
 785                 atomic_long_inc(&session->stats.rx_errors);
 786                 kfree_skb(skb);
 787         }
 788         return 0;
 789 }
 790
 791 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 792  * here. The skb is not on a list when we get here.
 793  * Returns 0 if the packet was a data packet and was successfully passed on.
 794  * Returns 1 if the packet was not a good data packet and could not be
 795  * forwarded.  All such packets are passed up to userspace to deal with.
 796  */
 797 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb)
 798 {
 799         struct l2tp_session *session = NULL;
 800         unsigned char *ptr, *optr;
 801         u16 hdrflags;
 802         u32 tunnel_id, session_id;
 803         u16 version;
 804         int length;
 805
 806         /* UDP has verifed checksum */
 807
 808         /* UDP always verifies the packet length. */
 809         __skb_pull(skb, sizeof(struct udphdr));
 810
 811         /* Short packet? */
 812         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_MAX)) {
 813                 l2tp_info(tunnel, L2TP_MSG_DATA,
 814                           "%s: recv short packet (len=%d)\n",
 815                           tunnel->name, skb->len);
 816                 goto error;
 817         }
 818
 819         /* Trace packet contents, if enabled */
 820         if (tunnel->debug & L2TP_MSG_DATA) {
 821                 length = min(32u, skb->len);
 822                 if (!pskb_may_pull(skb, length))
 823                         goto error;
 824
 825                 pr_debug("%s: recv\n", tunnel->name);
 826                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET, skb->data, length);
 827         }
 828
 829         /* Point to L2TP header */
 830         optr = ptr = skb->data;
 831
 832         /* Get L2TP header flags */
 833         hdrflags = ntohs(*(__be16 *) ptr);
 834
 835         /* Check protocol version */
 836         version = hdrflags & L2TP_HDR_VER_MASK;
 837         if (version != tunnel->version) {
 838                 l2tp_info(tunnel, L2TP_MSG_DATA,
 839                           "%s: recv protocol version mismatch: got %d expected %d\n",
 840                           tunnel->name, version, tunnel->version);
 841                 goto error;
 842         }
 843
 844         /* Get length of L2TP packet */
 845         length = skb->len;
 846
 847         /* If type is control packet, it is handled by userspace. */
 848         if (hdrflags & L2TP_HDRFLAG_T) {
 849                 l2tp_dbg(tunnel, L2TP_MSG_DATA,
 850                          "%s: recv control packet, len=%d\n",
 851                          tunnel->name, length);
 852                 goto error;
 853         }
 854
 855         /* Skip flags */
 856         ptr += 2;
 857
 858         if (tunnel->version == L2TP_HDR_VER_2) {
 859                 /* If length is present, skip it */
 860                 if (hdrflags & L2TP_HDRFLAG_L)
 861                         ptr += 2;
 862
 863                 /* Extract tunnel and session ID */
 864                 tunnel_id = ntohs(*(__be16 *) ptr);
 865                 ptr += 2;
 866                 session_id = ntohs(*(__be16 *) ptr);
 867                 ptr += 2;
 868         } else {
 869                 ptr += 2;       /* skip reserved bits */
 870                 tunnel_id = tunnel->tunnel_id;
 871                 session_id = ntohl(*(__be32 *) ptr);
 872                 ptr += 4;
 873         }
 874
 875         /* Find the session context */
 876         session = l2tp_tunnel_get_session(tunnel, session_id);
 877         if (!session || !session->recv_skb) {
 878                 if (session)
 879                         l2tp_session_dec_refcount(session);
 880
 881                 /* Not found? Pass to userspace to deal with */
 882                 l2tp_info(tunnel, L2TP_MSG_DATA,
 883                           "%s: no session found (%u/%u). Passing up.\n",
 884                           tunnel->name, tunnel_id, session_id);
 885                 goto error;
 886         }
 887
 888         if (tunnel->version == L2TP_HDR_VER_3 &&
 889             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr)) {
 890                 l2tp_session_dec_refcount(session);
 891                 goto error;
 892         }
 893
 894         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 895         l2tp_session_dec_refcount(session);
 896
 897         return 0;
 898
 899 error:
 900         /* Put UDP header back */
 901         __skb_push(skb, sizeof(struct udphdr));
 902
 903         return 1;
 904 }
 905
 906 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 907  * Return codes:
 908  * 0 : success.
 909  * <0: error
 910  * >0: skb should be passed up to userspace as UDP.
 911  */
 912 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 913 {
 914         struct l2tp_tunnel *tunnel;
 915
 916         tunnel = rcu_dereference_sk_user_data(sk);
 917         if (tunnel == NULL)
 918                 goto pass_up;
 919
 920         l2tp_dbg(tunnel, L2TP_MSG_DATA, "%s: received %d bytes\n",
 921                  tunnel->name, skb->len);
 922
 923         if (l2tp_udp_recv_core(tunnel, skb))
 924                 goto pass_up;
 925
 926         return 0;
 927
 928 pass_up:
 929         return 1;
 930 }
 931 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 932
 933 /************************************************************************
 934  * Transmit handling
 935  ***********************************************************************/
 936
 937 /* Build an L2TP header for the session into the buffer provided.
 938  */
 939 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 940 {
 941         struct l2tp_tunnel *tunnel = session->tunnel;
 942         __be16 *bufp = buf;
 943         __be16 *optr = buf;
 944         u16 flags = L2TP_HDR_VER_2;
 945         u32 tunnel_id = tunnel->peer_tunnel_id;
 946         u32 session_id = session->peer_session_id;
 947
 948         if (session->send_seq)
 949                 flags |= L2TP_HDRFLAG_S;
 950
 951         /* Setup L2TP header. */
 952         *bufp++ = htons(flags);
 953         *bufp++ = htons(tunnel_id);
 954         *bufp++ = htons(session_id);
 955         if (session->send_seq) {
 956                 *bufp++ = htons(session->ns);
 957                 *bufp++ = 0;
 958                 session->ns++;
 959                 session->ns &= 0xffff;
 960                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
 961                          session->name, session->ns);
 962         }
 963
 964         return bufp - optr;
 965 }
 966
 967 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 968 {
 969         struct l2tp_tunnel *tunnel = session->tunnel;
 970         char *bufp = buf;
 971         char *optr = bufp;
 972
 973         /* Setup L2TP header. The header differs slightly for UDP and
 974          * IP encapsulations. For UDP, there is 4 bytes of flags.
 975          */
 976         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 977                 u16 flags = L2TP_HDR_VER_3;
 978                 *((__be16 *) bufp) = htons(flags);
 979                 bufp += 2;
 980                 *((__be16 *) bufp) = 0;
 981                 bufp += 2;
 982         }
 983
 984         *((__be32 *) bufp) = htonl(session->peer_session_id);
 985         bufp += 4;
 986         if (session->cookie_len) {
 987                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 988                 bufp += session->cookie_len;
 989         }
 990         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 991                 u32 l2h = 0;
 992
 993                 if (session->send_seq) {
 994                         l2h = 0x40000000 | session->ns;
 995                         session->ns++;
 996                         session->ns &= 0xffffff;
 997                         l2tp_dbg(session, L2TP_MSG_SEQ,
 998                                  "%s: updated ns to %u\n",
 999                                  session->name, session->ns);
1000                 }
1001
1002                 *((__be32 *)bufp) = htonl(l2h);
1003                 bufp += 4;
1004         }
1005
1006         return bufp - optr;
1007 }
1008
1009 static void l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1010                            struct flowi *fl, size_t data_len)
1011 {
1012         struct l2tp_tunnel *tunnel = session->tunnel;
1013         unsigned int len = skb->len;
1014         int error;
1015
1016         /* Debug */
1017         if (session->send_seq)
1018                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes, ns=%u\n",
1019                          session->name, data_len, session->ns - 1);
1020         else
1021                 l2tp_dbg(session, L2TP_MSG_DATA, "%s: send %zd bytes\n",
1022                          session->name, data_len);
1023
1024         if (session->debug & L2TP_MSG_DATA) {
1025                 int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1026                 unsigned char *datap = skb->data + uhlen;
1027
1028                 pr_debug("%s: xmit\n", session->name);
1029                 print_hex_dump_bytes("", DUMP_PREFIX_OFFSET,
1030                                      datap, min_t(size_t, 32, len - uhlen));
1031         }
1032
1033         /* Queue the packet to IP for output */
1034         skb->ignore_df = 1;
1035         skb_dst_drop(skb);
1036 #if IS_ENABLED(CONFIG_IPV6)
1037         if (l2tp_sk_is_v6(tunnel->sock))
1038                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1039         else
1040 #endif
1041                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1042
1043         /* Update stats */
1044         if (error >= 0) {
1045                 atomic_long_inc(&tunnel->stats.tx_packets);
1046                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1047                 atomic_long_inc(&session->stats.tx_packets);
1048                 atomic_long_add(len, &session->stats.tx_bytes);
1049         } else {
1050                 atomic_long_inc(&tunnel->stats.tx_errors);
1051                 atomic_long_inc(&session->stats.tx_errors);
1052         }
1053 }
1054
1055 /* If caller requires the skb to have a ppp header, the header must be
1056  * inserted in the skb data before calling this function.
1057  */
1058 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1059 {
1060         int data_len = skb->len;
1061         struct l2tp_tunnel *tunnel = session->tunnel;
1062         struct sock *sk = tunnel->sock;
1063         struct flowi *fl;
1064         struct udphdr *uh;
1065         struct inet_sock *inet;
1066         int headroom;
1067         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1068         int udp_len;
1069         int ret = NET_XMIT_SUCCESS;
1070
1071         /* Check that there's enough headroom in the skb to insert IP,
1072          * UDP and L2TP headers. If not enough, expand it to
1073          * make room. Adjust truesize.
1074          */
1075         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1076                 uhlen + hdr_len;
1077         if (skb_cow_head(skb, headroom)) {
1078                 kfree_skb(skb);
1079                 return NET_XMIT_DROP;
1080         }
1081
1082         /* Setup L2TP header */
1083         session->build_header(session, __skb_push(skb, hdr_len));
1084
1085         /* Reset skb netfilter state */
1086         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1087         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1088                               IPSKB_REROUTED);
1089         nf_reset_ct(skb);
1090
1091         bh_lock_sock(sk);
1092         if (sock_owned_by_user(sk)) {
1093                 kfree_skb(skb);
1094                 ret = NET_XMIT_DROP;
1095                 goto out_unlock;
1096         }
1097
1098         /* The user-space may change the connection status for the user-space
1099          * provided socket at run time: we must check it under the socket lock
1100          */
1101         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1102                 kfree_skb(skb);
1103                 ret = NET_XMIT_DROP;
1104                 goto out_unlock;
1105         }
1106
1107         inet = inet_sk(sk);
1108         fl = &inet->cork.fl;
1109         switch (tunnel->encap) {
1110         case L2TP_ENCAPTYPE_UDP:
1111                 /* Setup UDP header */
1112                 __skb_push(skb, sizeof(*uh));
1113                 skb_reset_transport_header(skb);
1114                 uh = udp_hdr(skb);
1115                 uh->source = inet->inet_sport;
1116                 uh->dest = inet->inet_dport;
1117                 udp_len = uhlen + hdr_len + data_len;
1118                 uh->len = htons(udp_len);
1119
1120                 /* Calculate UDP checksum if configured to do so */
1121 #if IS_ENABLED(CONFIG_IPV6)
1122                 if (l2tp_sk_is_v6(sk))
1123                         udp6_set_csum(udp_get_no_check6_tx(sk),
1124                                       skb, &inet6_sk(sk)->saddr,
1125                                       &sk->sk_v6_daddr, udp_len);
1126                 else
1127 #endif
1128                 udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1129                              inet->inet_daddr, udp_len);
1130                 break;
1131
1132         case L2TP_ENCAPTYPE_IP:
1133                 break;
1134         }
1135
1136         l2tp_xmit_core(session, skb, fl, data_len);
1137 out_unlock:
1138         bh_unlock_sock(sk);
1139
1140         return ret;
1141 }
1142 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1143
1144 /*****************************************************************************
1145  * Tinnel and session create/destroy.
1146  *****************************************************************************/
1147
1148 /* Tunnel socket destruct hook.
1149  * The tunnel context is deleted only when all session sockets have been
1150  * closed.
1151  */
1152 static void l2tp_tunnel_destruct(struct sock *sk)
1153 {
1154         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1155
1156         if (tunnel == NULL)
1157                 goto end;
1158
1159         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing...\n", tunnel->name);
1160
1161         /* Disable udp encapsulation */
1162         switch (tunnel->encap) {
1163         case L2TP_ENCAPTYPE_UDP:
1164                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1165                 (udp_sk(sk))->encap_type = 0;
1166                 (udp_sk(sk))->encap_rcv = NULL;
1167                 (udp_sk(sk))->encap_destroy = NULL;
1168                 break;
1169         case L2TP_ENCAPTYPE_IP:
1170                 break;
1171         }
1172
1173         /* Remove hooks into tunnel socket */
1174         sk->sk_destruct = tunnel->old_sk_destruct;
1175         sk->sk_user_data = NULL;
1176
1177         /* Call the original destructor */
1178         if (sk->sk_destruct)
1179                 (*sk->sk_destruct)(sk);
1180
1181         kfree_rcu(tunnel, rcu);
1182 end:
1183         return;
1184 }
1185
1186 /* When the tunnel is closed, all the attached sessions need to go too.
1187  */
1188 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1189 {
1190         int hash;
1191         struct hlist_node *walk;
1192         struct hlist_node *tmp;
1193         struct l2tp_session *session;
1194
1195         BUG_ON(tunnel == NULL);
1196
1197         l2tp_info(tunnel, L2TP_MSG_CONTROL, "%s: closing all sessions...\n",
1198                   tunnel->name);
1199
1200         write_lock_bh(&tunnel->hlist_lock);
1201         tunnel->acpt_newsess = false;
1202         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1203 again:
1204                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1205                         session = hlist_entry(walk, struct l2tp_session, hlist);
1206
1207                         l2tp_info(session, L2TP_MSG_CONTROL,
1208                                   "%s: closing session\n", session->name);
1209
1210                         hlist_del_init(&session->hlist);
1211
1212                         if (test_and_set_bit(0, &session->dead))
1213                                 goto again;
1214
1215                         write_unlock_bh(&tunnel->hlist_lock);
1216
1217                         __l2tp_session_unhash(session);
1218                         l2tp_session_queue_purge(session);
1219
1220                         if (session->session_close != NULL)
1221                                 (*session->session_close)(session);
1222
1223                         l2tp_session_dec_refcount(session);
1224
1225                         write_lock_bh(&tunnel->hlist_lock);
1226
1227                         /* Now restart from the beginning of this hash
1228                          * chain.  We always remove a session from the
1229                          * list so we are guaranteed to make forward
1230                          * progress.
1231                          */
1232                         goto again;
1233                 }
1234         }
1235         write_unlock_bh(&tunnel->hlist_lock);
1236 }
1237
1238 /* Tunnel socket destroy hook for UDP encapsulation */
1239 static void l2tp_udp_encap_destroy(struct sock *sk)
1240 {
1241         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1242
1243         if (tunnel)
1244                 l2tp_tunnel_delete(tunnel);
1245 }
1246
1247 /* Workqueue tunnel deletion function */
1248 static void l2tp_tunnel_del_work(struct work_struct *work)
1249 {
1250         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1251                                                   del_work);
1252         struct sock *sk = tunnel->sock;
1253         struct socket *sock = sk->sk_socket;
1254         struct l2tp_net *pn;
1255
1256         l2tp_tunnel_closeall(tunnel);
1257
1258         /* If the tunnel socket was created within the kernel, use
1259          * the sk API to release it here.
1260          */
1261         if (tunnel->fd < 0) {
1262                 if (sock) {
1263                         kernel_sock_shutdown(sock, SHUT_RDWR);
1264                         sock_release(sock);
1265                 }
1266         }
1267
1268         /* Remove the tunnel struct from the tunnel list */
1269         pn = l2tp_pernet(tunnel->l2tp_net);
1270         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1271         list_del_rcu(&tunnel->list);
1272         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1273
1274         /* drop initial ref */
1275         l2tp_tunnel_dec_refcount(tunnel);
1276
1277         /* drop workqueue ref */
1278         l2tp_tunnel_dec_refcount(tunnel);
1279 }
1280
1281 /* Create a socket for the tunnel, if one isn't set up by
1282  * userspace. This is used for static tunnels where there is no
1283  * managing L2TP daemon.
1284  *
1285  * Since we don't want these sockets to keep a namespace alive by
1286  * themselves, we drop the socket's namespace refcount after creation.
1287  * These sockets are freed when the namespace exits using the pernet
1288  * exit hook.
1289  */
1290 static int l2tp_tunnel_sock_create(struct net *net,
1291                                 u32 tunnel_id,
1292                                 u32 peer_tunnel_id,
1293                                 struct l2tp_tunnel_cfg *cfg,
1294                                 struct socket **sockp)
1295 {
1296         int err = -EINVAL;
1297         struct socket *sock = NULL;
1298         struct udp_port_cfg udp_conf;
1299
1300         switch (cfg->encap) {
1301         case L2TP_ENCAPTYPE_UDP:
1302                 memset(&udp_conf, 0, sizeof(udp_conf));
1303
1304 #if IS_ENABLED(CONFIG_IPV6)
1305                 if (cfg->local_ip6 && cfg->peer_ip6) {
1306                         udp_conf.family = AF_INET6;
1307                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1308                                sizeof(udp_conf.local_ip6));
1309                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1310                                sizeof(udp_conf.peer_ip6));
1311                         udp_conf.use_udp6_tx_checksums =
1312                           ! cfg->udp6_zero_tx_checksums;
1313                         udp_conf.use_udp6_rx_checksums =
1314                           ! cfg->udp6_zero_rx_checksums;
1315                 } else
1316 #endif
1317                 {
1318                         udp_conf.family = AF_INET;
1319                         udp_conf.local_ip = cfg->local_ip;
1320                         udp_conf.peer_ip = cfg->peer_ip;
1321                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1322                 }
1323
1324                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1325                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1326
1327                 err = udp_sock_create(net, &udp_conf, &sock);
1328                 if (err < 0)
1329                         goto out;
1330
1331                 break;
1332
1333         case L2TP_ENCAPTYPE_IP:
1334 #if IS_ENABLED(CONFIG_IPV6)
1335                 if (cfg->local_ip6 && cfg->peer_ip6) {
1336                         struct sockaddr_l2tpip6 ip6_addr = {0};
1337
1338                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1339                                           IPPROTO_L2TP, &sock);
1340                         if (err < 0)
1341                                 goto out;
1342
1343                         ip6_addr.l2tp_family = AF_INET6;
1344                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1345                                sizeof(ip6_addr.l2tp_addr));
1346                         ip6_addr.l2tp_conn_id = tunnel_id;
1347                         err = kernel_bind(sock, (struct sockaddr *) &ip6_addr,
1348                                           sizeof(ip6_addr));
1349                         if (err < 0)
1350                                 goto out;
1351
1352                         ip6_addr.l2tp_family = AF_INET6;
1353                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1354                                sizeof(ip6_addr.l2tp_addr));
1355                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1356                         err = kernel_connect(sock,
1357                                              (struct sockaddr *) &ip6_addr,
1358                                              sizeof(ip6_addr), 0);
1359                         if (err < 0)
1360                                 goto out;
1361                 } else
1362 #endif
1363                 {
1364                         struct sockaddr_l2tpip ip_addr = {0};
1365
1366                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1367                                           IPPROTO_L2TP, &sock);
1368                         if (err < 0)
1369                                 goto out;
1370
1371                         ip_addr.l2tp_family = AF_INET;
1372                         ip_addr.l2tp_addr = cfg->local_ip;
1373                         ip_addr.l2tp_conn_id = tunnel_id;
1374                         err = kernel_bind(sock, (struct sockaddr *) &ip_addr,
1375                                           sizeof(ip_addr));
1376                         if (err < 0)
1377                                 goto out;
1378
1379                         ip_addr.l2tp_family = AF_INET;
1380                         ip_addr.l2tp_addr = cfg->peer_ip;
1381                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1382                         err = kernel_connect(sock, (struct sockaddr *) &ip_addr,
1383                                              sizeof(ip_addr), 0);
1384                         if (err < 0)
1385                                 goto out;
1386                 }
1387                 break;
1388
1389         default:
1390                 goto out;
1391         }
1392
1393 out:
1394         *sockp = sock;
1395         if ((err < 0) && sock) {
1396                 kernel_sock_shutdown(sock, SHUT_RDWR);
1397                 sock_release(sock);
1398                 *sockp = NULL;
1399         }
1400
1401         return err;
1402 }
1403
1404 static struct lock_class_key l2tp_socket_class;
1405
1406 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)
1407 {
1408         struct l2tp_tunnel *tunnel = NULL;
1409         int err;
1410         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1411
1412         if (cfg != NULL)
1413                 encap = cfg->encap;
1414
1415         tunnel = kzalloc(sizeof(struct l2tp_tunnel), GFP_KERNEL);
1416         if (tunnel == NULL) {
1417                 err = -ENOMEM;
1418                 goto err;
1419         }
1420
1421         tunnel->version = version;
1422         tunnel->tunnel_id = tunnel_id;
1423         tunnel->peer_tunnel_id = peer_tunnel_id;
1424         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1425
1426         tunnel->magic = L2TP_TUNNEL_MAGIC;
1427         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1428         rwlock_init(&tunnel->hlist_lock);
1429         tunnel->acpt_newsess = true;
1430
1431         if (cfg != NULL)
1432                 tunnel->debug = cfg->debug;
1433
1434         tunnel->encap = encap;
1435
1436         refcount_set(&tunnel->ref_count, 1);
1437         tunnel->fd = fd;
1438
1439         /* Init delete workqueue struct */
1440         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1441
1442         INIT_LIST_HEAD(&tunnel->list);
1443
1444         err = 0;
1445 err:
1446         if (tunnelp)
1447                 *tunnelp = tunnel;
1448
1449         return err;
1450 }
1451 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1452
1453 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1454                                 enum l2tp_encap_type encap)
1455 {
1456         if (!net_eq(sock_net(sk), net))
1457                 return -EINVAL;
1458
1459         if (sk->sk_type != SOCK_DGRAM)
1460                 return -EPROTONOSUPPORT;
1461
1462         if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1463                 return -EPROTONOSUPPORT;
1464
1465         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1466             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1467                 return -EPROTONOSUPPORT;
1468
1469         if (sk->sk_user_data)
1470                 return -EBUSY;
1471
1472         return 0;
1473 }
1474
1475 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1476                          struct l2tp_tunnel_cfg *cfg)
1477 {
1478         struct l2tp_tunnel *tunnel_walk;
1479         struct l2tp_net *pn;
1480         struct socket *sock;
1481         struct sock *sk;
1482         int ret;
1483
1484         if (tunnel->fd < 0) {
1485                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1486                                               tunnel->peer_tunnel_id, cfg,
1487                                               &sock);
1488                 if (ret < 0)
1489                         goto err;
1490         } else {
1491                 sock = sockfd_lookup(tunnel->fd, &ret);
1492                 if (!sock)
1493                         goto err;
1494
1495                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1496                 if (ret < 0)
1497                         goto err_sock;
1498         }
1499
1500         tunnel->l2tp_net = net;
1501         pn = l2tp_pernet(net);
1502
1503         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1504         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1505                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1506                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1507
1508                         ret = -EEXIST;
1509                         goto err_sock;
1510                 }
1511         }
1512         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1513         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1514
1515         sk = sock->sk;
1516         sock_hold(sk);
1517         tunnel->sock = sk;
1518
1519         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1520                 struct udp_tunnel_sock_cfg udp_cfg = {
1521                         .sk_user_data = tunnel,
1522                         .encap_type = UDP_ENCAP_L2TPINUDP,
1523                         .encap_rcv = l2tp_udp_encap_recv,
1524                         .encap_destroy = l2tp_udp_encap_destroy,
1525                 };
1526
1527                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1528         } else {
1529                 sk->sk_user_data = tunnel;
1530         }
1531
1532         tunnel->old_sk_destruct = sk->sk_destruct;
1533         sk->sk_destruct = &l2tp_tunnel_destruct;
1534         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1535                                    "l2tp_sock");
1536         sk->sk_allocation = GFP_ATOMIC;
1537
1538         if (tunnel->fd >= 0)
1539                 sockfd_put(sock);
1540
1541         return 0;
1542
1543 err_sock:
1544         if (tunnel->fd < 0)
1545                 sock_release(sock);
1546         else
1547                 sockfd_put(sock);
1548 err:
1549         return ret;
1550 }
1551 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1552
1553 /* This function is used by the netlink TUNNEL_DELETE command.
1554  */
1555 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1556 {
1557         if (!test_and_set_bit(0, &tunnel->dead)) {
1558                 l2tp_tunnel_inc_refcount(tunnel);
1559                 queue_work(l2tp_wq, &tunnel->del_work);
1560         }
1561 }
1562 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1563
1564 /* Really kill the session.
1565  */
1566 void l2tp_session_free(struct l2tp_session *session)
1567 {
1568         struct l2tp_tunnel *tunnel = session->tunnel;
1569
1570         BUG_ON(refcount_read(&session->ref_count) != 0);
1571
1572         if (tunnel) {
1573                 BUG_ON(tunnel->magic != L2TP_TUNNEL_MAGIC);
1574                 l2tp_tunnel_dec_refcount(tunnel);
1575         }
1576
1577         kfree(session);
1578 }
1579 EXPORT_SYMBOL_GPL(l2tp_session_free);
1580
1581 /* Remove an l2tp session from l2tp_core's hash lists.
1582  * Provides a tidyup interface for pseudowire code which can't just route all
1583  * shutdown via. l2tp_session_delete and a pseudowire-specific session_close
1584  * callback.
1585  */
1586 void __l2tp_session_unhash(struct l2tp_session *session)
1587 {
1588         struct l2tp_tunnel *tunnel = session->tunnel;
1589
1590         /* Remove the session from core hashes */
1591         if (tunnel) {
1592                 /* Remove from the per-tunnel hash */
1593                 write_lock_bh(&tunnel->hlist_lock);
1594                 hlist_del_init(&session->hlist);
1595                 write_unlock_bh(&tunnel->hlist_lock);
1596
1597                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1598                 if (tunnel->version != L2TP_HDR_VER_2) {
1599                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1600                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1601                         hlist_del_init_rcu(&session->global_hlist);
1602                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1603                         synchronize_rcu();
1604                 }
1605         }
1606 }
1607 EXPORT_SYMBOL_GPL(__l2tp_session_unhash);
1608
1609 /* This function is used by the netlink SESSION_DELETE command and by
1610    pseudowire modules.
1611  */
1612 int l2tp_session_delete(struct l2tp_session *session)
1613 {
1614         if (test_and_set_bit(0, &session->dead))
1615                 return 0;
1616
1617         __l2tp_session_unhash(session);
1618         l2tp_session_queue_purge(session);
1619         if (session->session_close != NULL)
1620                 (*session->session_close)(session);
1621
1622         l2tp_session_dec_refcount(session);
1623
1624         return 0;
1625 }
1626 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1627
1628 /* We come here whenever a session's send_seq, cookie_len or
1629  * l2specific_type parameters are set.
1630  */
1631 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1632 {
1633         if (version == L2TP_HDR_VER_2) {
1634                 session->hdr_len = 6;
1635                 if (session->send_seq)
1636                         session->hdr_len += 4;
1637         } else {
1638                 session->hdr_len = 4 + session->cookie_len;
1639                 session->hdr_len += l2tp_get_l2specific_len(session);
1640                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1641                         session->hdr_len += 4;
1642         }
1643
1644 }
1645 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1646
1647 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)
1648 {
1649         struct l2tp_session *session;
1650
1651         session = kzalloc(sizeof(struct l2tp_session) + priv_size, GFP_KERNEL);
1652         if (session != NULL) {
1653                 session->magic = L2TP_SESSION_MAGIC;
1654                 session->tunnel = tunnel;
1655
1656                 session->session_id = session_id;
1657                 session->peer_session_id = peer_session_id;
1658                 session->nr = 0;
1659                 if (tunnel->version == L2TP_HDR_VER_2)
1660                         session->nr_max = 0xffff;
1661                 else
1662                         session->nr_max = 0xffffff;
1663                 session->nr_window_size = session->nr_max / 2;
1664                 session->nr_oos_count_max = 4;
1665
1666                 /* Use NR of first received packet */
1667                 session->reorder_skip = 1;
1668
1669                 sprintf(&session->name[0], "sess %u/%u",
1670                         tunnel->tunnel_id, session->session_id);
1671
1672                 skb_queue_head_init(&session->reorder_q);
1673
1674                 INIT_HLIST_NODE(&session->hlist);
1675                 INIT_HLIST_NODE(&session->global_hlist);
1676
1677                 /* Inherit debug options from tunnel */
1678                 session->debug = tunnel->debug;
1679
1680                 if (cfg) {
1681                         session->pwtype = cfg->pw_type;
1682                         session->debug = cfg->debug;
1683                         session->send_seq = cfg->send_seq;
1684                         session->recv_seq = cfg->recv_seq;
1685                         session->lns_mode = cfg->lns_mode;
1686                         session->reorder_timeout = cfg->reorder_timeout;
1687                         session->l2specific_type = cfg->l2specific_type;
1688                         session->cookie_len = cfg->cookie_len;
1689                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1690                         session->peer_cookie_len = cfg->peer_cookie_len;
1691                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1692                 }
1693
1694                 if (tunnel->version == L2TP_HDR_VER_2)
1695                         session->build_header = l2tp_build_l2tpv2_header;
1696                 else
1697                         session->build_header = l2tp_build_l2tpv3_header;
1698
1699                 l2tp_session_set_header_len(session, tunnel->version);
1700
1701                 refcount_set(&session->ref_count, 1);
1702
1703                 return session;
1704         }
1705
1706         return ERR_PTR(-ENOMEM);
1707 }
1708 EXPORT_SYMBOL_GPL(l2tp_session_create);
1709
1710 /*****************************************************************************
1711  * Init and cleanup
1712  *****************************************************************************/
1713
1714 static __net_init int l2tp_init_net(struct net *net)
1715 {
1716         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1717         int hash;
1718
1719         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1720         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1721
1722         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1723                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1724
1725         spin_lock_init(&pn->l2tp_session_hlist_lock);
1726
1727         return 0;
1728 }
1729
1730 static __net_exit void l2tp_exit_net(struct net *net)
1731 {
1732         struct l2tp_net *pn = l2tp_pernet(net);
1733         struct l2tp_tunnel *tunnel = NULL;
1734         int hash;
1735
1736         rcu_read_lock_bh();
1737         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1738                 l2tp_tunnel_delete(tunnel);
1739         }
1740         rcu_read_unlock_bh();
1741
1742         if (l2tp_wq)
1743                 flush_workqueue(l2tp_wq);
1744         rcu_barrier();
1745
1746         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1747                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1748 }
1749
1750 static struct pernet_operations l2tp_net_ops = {
1751         .init = l2tp_init_net,
1752         .exit = l2tp_exit_net,
1753         .id   = &l2tp_net_id,
1754         .size = sizeof(struct l2tp_net),
1755 };
1756
1757 static int __init l2tp_init(void)
1758 {
1759         int rc = 0;
1760
1761         rc = register_pernet_device(&l2tp_net_ops);
1762         if (rc)
1763                 goto out;
1764
1765         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1766         if (!l2tp_wq) {
1767                 pr_err("alloc_workqueue failed\n");
1768                 unregister_pernet_device(&l2tp_net_ops);
1769                 rc = -ENOMEM;
1770                 goto out;
1771         }
1772
1773         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1774
1775 out:
1776         return rc;
1777 }
1778
1779 static void __exit l2tp_exit(void)
1780 {
1781         unregister_pernet_device(&l2tp_net_ops);
1782         if (l2tp_wq) {
1783                 destroy_workqueue(l2tp_wq);
1784                 l2tp_wq = NULL;
1785         }
1786 }
1787
1788 module_init(l2tp_init);
1789 module_exit(l2tp_exit);
1790
1791 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1792 MODULE_DESCRIPTION("L2TP core");
1793 MODULE_LICENSE("GPL");
1794 MODULE_VERSION(L2TP_DRV_VERSION);