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 static 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
 158 static void l2tp_session_free(struct l2tp_session *session)
 159 {
 160         struct l2tp_tunnel *tunnel = session->tunnel;
 161
 162         if (tunnel) {
 163                 if (WARN_ON(tunnel->magic != L2TP_TUNNEL_MAGIC))
 164                         goto out;
 165                 l2tp_tunnel_dec_refcount(tunnel);
 166         }
 167
 168 out:
 169         kfree(session);
 170 }
 171
 172 void l2tp_tunnel_inc_refcount(struct l2tp_tunnel *tunnel)
 173 {
 174         refcount_inc(&tunnel->ref_count);
 175 }
 176 EXPORT_SYMBOL_GPL(l2tp_tunnel_inc_refcount);
 177
 178 void l2tp_tunnel_dec_refcount(struct l2tp_tunnel *tunnel)
 179 {
 180         if (refcount_dec_and_test(&tunnel->ref_count))
 181                 l2tp_tunnel_free(tunnel);
 182 }
 183 EXPORT_SYMBOL_GPL(l2tp_tunnel_dec_refcount);
 184
 185 void l2tp_session_inc_refcount(struct l2tp_session *session)
 186 {
 187         refcount_inc(&session->ref_count);
 188 }
 189 EXPORT_SYMBOL_GPL(l2tp_session_inc_refcount);
 190
 191 void l2tp_session_dec_refcount(struct l2tp_session *session)
 192 {
 193         if (refcount_dec_and_test(&session->ref_count))
 194                 l2tp_session_free(session);
 195 }
 196 EXPORT_SYMBOL_GPL(l2tp_session_dec_refcount);
 197
 198 /* Lookup a tunnel. A new reference is held on the returned tunnel. */
 199 struct l2tp_tunnel *l2tp_tunnel_get(const struct net *net, u32 tunnel_id)
 200 {
 201         const struct l2tp_net *pn = l2tp_pernet(net);
 202         struct l2tp_tunnel *tunnel;
 203
 204         rcu_read_lock_bh();
 205         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 206                 if (tunnel->tunnel_id == tunnel_id &&
 207                     refcount_inc_not_zero(&tunnel->ref_count)) {
 208                         rcu_read_unlock_bh();
 209
 210                         return tunnel;
 211                 }
 212         }
 213         rcu_read_unlock_bh();
 214
 215         return NULL;
 216 }
 217 EXPORT_SYMBOL_GPL(l2tp_tunnel_get);
 218
 219 struct l2tp_tunnel *l2tp_tunnel_get_nth(const struct net *net, int nth)
 220 {
 221         const struct l2tp_net *pn = l2tp_pernet(net);
 222         struct l2tp_tunnel *tunnel;
 223         int count = 0;
 224
 225         rcu_read_lock_bh();
 226         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
 227                 if (++count > nth &&
 228                     refcount_inc_not_zero(&tunnel->ref_count)) {
 229                         rcu_read_unlock_bh();
 230                         return tunnel;
 231                 }
 232         }
 233         rcu_read_unlock_bh();
 234
 235         return NULL;
 236 }
 237 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_nth);
 238
 239 struct l2tp_session *l2tp_tunnel_get_session(struct l2tp_tunnel *tunnel,
 240                                              u32 session_id)
 241 {
 242         struct hlist_head *session_list;
 243         struct l2tp_session *session;
 244
 245         session_list = l2tp_session_id_hash(tunnel, session_id);
 246
 247         read_lock_bh(&tunnel->hlist_lock);
 248         hlist_for_each_entry(session, session_list, hlist)
 249                 if (session->session_id == session_id) {
 250                         l2tp_session_inc_refcount(session);
 251                         read_unlock_bh(&tunnel->hlist_lock);
 252
 253                         return session;
 254                 }
 255         read_unlock_bh(&tunnel->hlist_lock);
 256
 257         return NULL;
 258 }
 259 EXPORT_SYMBOL_GPL(l2tp_tunnel_get_session);
 260
 261 struct l2tp_session *l2tp_session_get(const struct net *net, u32 session_id)
 262 {
 263         struct hlist_head *session_list;
 264         struct l2tp_session *session;
 265
 266         session_list = l2tp_session_id_hash_2(l2tp_pernet(net), session_id);
 267
 268         rcu_read_lock_bh();
 269         hlist_for_each_entry_rcu(session, session_list, global_hlist)
 270                 if (session->session_id == session_id) {
 271                         l2tp_session_inc_refcount(session);
 272                         rcu_read_unlock_bh();
 273
 274                         return session;
 275                 }
 276         rcu_read_unlock_bh();
 277
 278         return NULL;
 279 }
 280 EXPORT_SYMBOL_GPL(l2tp_session_get);
 281
 282 struct l2tp_session *l2tp_session_get_nth(struct l2tp_tunnel *tunnel, int nth)
 283 {
 284         int hash;
 285         struct l2tp_session *session;
 286         int count = 0;
 287
 288         read_lock_bh(&tunnel->hlist_lock);
 289         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
 290                 hlist_for_each_entry(session, &tunnel->session_hlist[hash], hlist) {
 291                         if (++count > nth) {
 292                                 l2tp_session_inc_refcount(session);
 293                                 read_unlock_bh(&tunnel->hlist_lock);
 294                                 return session;
 295                         }
 296                 }
 297         }
 298
 299         read_unlock_bh(&tunnel->hlist_lock);
 300
 301         return NULL;
 302 }
 303 EXPORT_SYMBOL_GPL(l2tp_session_get_nth);
 304
 305 /* Lookup a session by interface name.
 306  * This is very inefficient but is only used by management interfaces.
 307  */
 308 struct l2tp_session *l2tp_session_get_by_ifname(const struct net *net,
 309                                                 const char *ifname)
 310 {
 311         struct l2tp_net *pn = l2tp_pernet(net);
 312         int hash;
 313         struct l2tp_session *session;
 314
 315         rcu_read_lock_bh();
 316         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++) {
 317                 hlist_for_each_entry_rcu(session, &pn->l2tp_session_hlist[hash], global_hlist) {
 318                         if (!strcmp(session->ifname, ifname)) {
 319                                 l2tp_session_inc_refcount(session);
 320                                 rcu_read_unlock_bh();
 321
 322                                 return session;
 323                         }
 324                 }
 325         }
 326
 327         rcu_read_unlock_bh();
 328
 329         return NULL;
 330 }
 331 EXPORT_SYMBOL_GPL(l2tp_session_get_by_ifname);
 332
 333 int l2tp_session_register(struct l2tp_session *session,
 334                           struct l2tp_tunnel *tunnel)
 335 {
 336         struct l2tp_session *session_walk;
 337         struct hlist_head *g_head;
 338         struct hlist_head *head;
 339         struct l2tp_net *pn;
 340         int err;
 341
 342         head = l2tp_session_id_hash(tunnel, session->session_id);
 343
 344         write_lock_bh(&tunnel->hlist_lock);
 345         if (!tunnel->acpt_newsess) {
 346                 err = -ENODEV;
 347                 goto err_tlock;
 348         }
 349
 350         hlist_for_each_entry(session_walk, head, hlist)
 351                 if (session_walk->session_id == session->session_id) {
 352                         err = -EEXIST;
 353                         goto err_tlock;
 354                 }
 355
 356         if (tunnel->version == L2TP_HDR_VER_3) {
 357                 pn = l2tp_pernet(tunnel->l2tp_net);
 358                 g_head = l2tp_session_id_hash_2(pn, session->session_id);
 359
 360                 spin_lock_bh(&pn->l2tp_session_hlist_lock);
 361
 362                 /* IP encap expects session IDs to be globally unique, while
 363                  * UDP encap doesn't.
 364                  */
 365                 hlist_for_each_entry(session_walk, g_head, global_hlist)
 366                         if (session_walk->session_id == session->session_id &&
 367                             (session_walk->tunnel->encap == L2TP_ENCAPTYPE_IP ||
 368                              tunnel->encap == L2TP_ENCAPTYPE_IP)) {
 369                                 err = -EEXIST;
 370                                 goto err_tlock_pnlock;
 371                         }
 372
 373                 l2tp_tunnel_inc_refcount(tunnel);
 374                 hlist_add_head_rcu(&session->global_hlist, g_head);
 375
 376                 spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 377         } else {
 378                 l2tp_tunnel_inc_refcount(tunnel);
 379         }
 380
 381         hlist_add_head(&session->hlist, head);
 382         write_unlock_bh(&tunnel->hlist_lock);
 383
 384         return 0;
 385
 386 err_tlock_pnlock:
 387         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
 388 err_tlock:
 389         write_unlock_bh(&tunnel->hlist_lock);
 390
 391         return err;
 392 }
 393 EXPORT_SYMBOL_GPL(l2tp_session_register);
 394
 395 /*****************************************************************************
 396  * Receive data handling
 397  *****************************************************************************/
 398
 399 /* Queue a skb in order. We come here only if the skb has an L2TP sequence
 400  * number.
 401  */
 402 static void l2tp_recv_queue_skb(struct l2tp_session *session, struct sk_buff *skb)
 403 {
 404         struct sk_buff *skbp;
 405         struct sk_buff *tmp;
 406         u32 ns = L2TP_SKB_CB(skb)->ns;
 407
 408         spin_lock_bh(&session->reorder_q.lock);
 409         skb_queue_walk_safe(&session->reorder_q, skbp, tmp) {
 410                 if (L2TP_SKB_CB(skbp)->ns > ns) {
 411                         __skb_queue_before(&session->reorder_q, skbp, skb);
 412                         l2tp_dbg(session, L2TP_MSG_SEQ,
 413                                  "%s: pkt %hu, inserted before %hu, reorder_q len=%d\n",
 414                                  session->name, ns, L2TP_SKB_CB(skbp)->ns,
 415                                  skb_queue_len(&session->reorder_q));
 416                         atomic_long_inc(&session->stats.rx_oos_packets);
 417                         goto out;
 418                 }
 419         }
 420
 421         __skb_queue_tail(&session->reorder_q, skb);
 422
 423 out:
 424         spin_unlock_bh(&session->reorder_q.lock);
 425 }
 426
 427 /* Dequeue a single skb.
 428  */
 429 static void l2tp_recv_dequeue_skb(struct l2tp_session *session, struct sk_buff *skb)
 430 {
 431         struct l2tp_tunnel *tunnel = session->tunnel;
 432         int length = L2TP_SKB_CB(skb)->length;
 433
 434         /* We're about to requeue the skb, so return resources
 435          * to its current owner (a socket receive buffer).
 436          */
 437         skb_orphan(skb);
 438
 439         atomic_long_inc(&tunnel->stats.rx_packets);
 440         atomic_long_add(length, &tunnel->stats.rx_bytes);
 441         atomic_long_inc(&session->stats.rx_packets);
 442         atomic_long_add(length, &session->stats.rx_bytes);
 443
 444         if (L2TP_SKB_CB(skb)->has_seq) {
 445                 /* Bump our Nr */
 446                 session->nr++;
 447                 session->nr &= session->nr_max;
 448
 449                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated nr to %hu\n",
 450                          session->name, session->nr);
 451         }
 452
 453         /* call private receive handler */
 454         if (session->recv_skb)
 455                 (*session->recv_skb)(session, skb, L2TP_SKB_CB(skb)->length);
 456         else
 457                 kfree_skb(skb);
 458 }
 459
 460 /* Dequeue skbs from the session's reorder_q, subject to packet order.
 461  * Skbs that have been in the queue for too long are simply discarded.
 462  */
 463 static void l2tp_recv_dequeue(struct l2tp_session *session)
 464 {
 465         struct sk_buff *skb;
 466         struct sk_buff *tmp;
 467
 468         /* If the pkt at the head of the queue has the nr that we
 469          * expect to send up next, dequeue it and any other
 470          * in-sequence packets behind it.
 471          */
 472 start:
 473         spin_lock_bh(&session->reorder_q.lock);
 474         skb_queue_walk_safe(&session->reorder_q, skb, tmp) {
 475                 if (time_after(jiffies, L2TP_SKB_CB(skb)->expires)) {
 476                         atomic_long_inc(&session->stats.rx_seq_discards);
 477                         atomic_long_inc(&session->stats.rx_errors);
 478                         l2tp_dbg(session, L2TP_MSG_SEQ,
 479                                  "%s: oos pkt %u len %d discarded (too old), waiting for %u, reorder_q_len=%d\n",
 480                                  session->name, L2TP_SKB_CB(skb)->ns,
 481                                  L2TP_SKB_CB(skb)->length, session->nr,
 482                                  skb_queue_len(&session->reorder_q));
 483                         session->reorder_skip = 1;
 484                         __skb_unlink(skb, &session->reorder_q);
 485                         kfree_skb(skb);
 486                         continue;
 487                 }
 488
 489                 if (L2TP_SKB_CB(skb)->has_seq) {
 490                         if (session->reorder_skip) {
 491                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 492                                          "%s: advancing nr to next pkt: %u -> %u",
 493                                          session->name, session->nr,
 494                                          L2TP_SKB_CB(skb)->ns);
 495                                 session->reorder_skip = 0;
 496                                 session->nr = L2TP_SKB_CB(skb)->ns;
 497                         }
 498                         if (L2TP_SKB_CB(skb)->ns != session->nr) {
 499                                 l2tp_dbg(session, L2TP_MSG_SEQ,
 500                                          "%s: holding oos pkt %u len %d, waiting for %u, reorder_q_len=%d\n",
 501                                          session->name, L2TP_SKB_CB(skb)->ns,
 502                                          L2TP_SKB_CB(skb)->length, session->nr,
 503                                          skb_queue_len(&session->reorder_q));
 504                                 goto out;
 505                         }
 506                 }
 507                 __skb_unlink(skb, &session->reorder_q);
 508
 509                 /* Process the skb. We release the queue lock while we
 510                  * do so to let other contexts process the queue.
 511                  */
 512                 spin_unlock_bh(&session->reorder_q.lock);
 513                 l2tp_recv_dequeue_skb(session, skb);
 514                 goto start;
 515         }
 516
 517 out:
 518         spin_unlock_bh(&session->reorder_q.lock);
 519 }
 520
 521 static int l2tp_seq_check_rx_window(struct l2tp_session *session, u32 nr)
 522 {
 523         u32 nws;
 524
 525         if (nr >= session->nr)
 526                 nws = nr - session->nr;
 527         else
 528                 nws = (session->nr_max + 1) - (session->nr - nr);
 529
 530         return nws < session->nr_window_size;
 531 }
 532
 533 /* If packet has sequence numbers, queue it if acceptable. Returns 0 if
 534  * acceptable, else non-zero.
 535  */
 536 static int l2tp_recv_data_seq(struct l2tp_session *session, struct sk_buff *skb)
 537 {
 538         if (!l2tp_seq_check_rx_window(session, L2TP_SKB_CB(skb)->ns)) {
 539                 /* Packet sequence number is outside allowed window.
 540                  * Discard it.
 541                  */
 542                 l2tp_dbg(session, L2TP_MSG_SEQ,
 543                          "%s: pkt %u len %d discarded, outside window, nr=%u\n",
 544                          session->name, L2TP_SKB_CB(skb)->ns,
 545                          L2TP_SKB_CB(skb)->length, session->nr);
 546                 goto discard;
 547         }
 548
 549         if (session->reorder_timeout != 0) {
 550                 /* Packet reordering enabled. Add skb to session's
 551                  * reorder queue, in order of ns.
 552                  */
 553                 l2tp_recv_queue_skb(session, skb);
 554                 goto out;
 555         }
 556
 557         /* Packet reordering disabled. Discard out-of-sequence packets, while
 558          * tracking the number if in-sequence packets after the first OOS packet
 559          * is seen. After nr_oos_count_max in-sequence packets, reset the
 560          * sequence number to re-enable packet reception.
 561          */
 562         if (L2TP_SKB_CB(skb)->ns == session->nr) {
 563                 skb_queue_tail(&session->reorder_q, skb);
 564         } else {
 565                 u32 nr_oos = L2TP_SKB_CB(skb)->ns;
 566                 u32 nr_next = (session->nr_oos + 1) & session->nr_max;
 567
 568                 if (nr_oos == nr_next)
 569                         session->nr_oos_count++;
 570                 else
 571                         session->nr_oos_count = 0;
 572
 573                 session->nr_oos = nr_oos;
 574                 if (session->nr_oos_count > session->nr_oos_count_max) {
 575                         session->reorder_skip = 1;
 576                         l2tp_dbg(session, L2TP_MSG_SEQ,
 577                                  "%s: %d oos packets received. Resetting sequence numbers\n",
 578                                  session->name, session->nr_oos_count);
 579                 }
 580                 if (!session->reorder_skip) {
 581                         atomic_long_inc(&session->stats.rx_seq_discards);
 582                         l2tp_dbg(session, L2TP_MSG_SEQ,
 583                                  "%s: oos pkt %u len %d discarded, waiting for %u, reorder_q_len=%d\n",
 584                                  session->name, L2TP_SKB_CB(skb)->ns,
 585                                  L2TP_SKB_CB(skb)->length, session->nr,
 586                                  skb_queue_len(&session->reorder_q));
 587                         goto discard;
 588                 }
 589                 skb_queue_tail(&session->reorder_q, skb);
 590         }
 591
 592 out:
 593         return 0;
 594
 595 discard:
 596         return 1;
 597 }
 598
 599 /* Do receive processing of L2TP data frames. We handle both L2TPv2
 600  * and L2TPv3 data frames here.
 601  *
 602  * L2TPv2 Data Message Header
 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  * |T|L|x|x|S|x|O|P|x|x|x|x|  Ver  |          Length (opt)         |
 608  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 609  * |           Tunnel ID           |           Session ID          |
 610  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 611  * |             Ns (opt)          |             Nr (opt)          |
 612  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 613  * |      Offset Size (opt)        |    Offset pad... (opt)
 614  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 615  *
 616  * Data frames are marked by T=0. All other fields are the same as
 617  * those in L2TP control frames.
 618  *
 619  * L2TPv3 Data Message Header
 620  *
 621  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 622  * |                      L2TP Session Header                      |
 623  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 624  * |                      L2-Specific Sublayer                     |
 625  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 626  * |                        Tunnel Payload                      ...
 627  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 628  *
 629  * L2TPv3 Session Header Over IP
 630  *
 631  *  0                   1                   2                   3
 632  *  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
 633  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 634  * |                           Session ID                          |
 635  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 636  * |               Cookie (optional, maximum 64 bits)...
 637  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 638  *                                                                 |
 639  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 640  *
 641  * L2TPv3 L2-Specific Sublayer Format
 642  *
 643  *  0                   1                   2                   3
 644  *  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
 645  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 646  * |x|S|x|x|x|x|x|x|              Sequence Number                  |
 647  * +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
 648  *
 649  * Cookie value and sublayer format are negotiated with the peer when
 650  * the session is set up. Unlike L2TPv2, we do not need to parse the
 651  * packet header to determine if optional fields are present.
 652  *
 653  * Caller must already have parsed the frame and determined that it is
 654  * a data (not control) frame before coming here. Fields up to the
 655  * session-id have already been parsed and ptr points to the data
 656  * after the session-id.
 657  */
 658 void l2tp_recv_common(struct l2tp_session *session, struct sk_buff *skb,
 659                       unsigned char *ptr, unsigned char *optr, u16 hdrflags,
 660                       int length)
 661 {
 662         struct l2tp_tunnel *tunnel = session->tunnel;
 663         int offset;
 664
 665         /* Parse and check optional cookie */
 666         if (session->peer_cookie_len > 0) {
 667                 if (memcmp(ptr, &session->peer_cookie[0], session->peer_cookie_len)) {
 668                         pr_warn_ratelimited("%s: cookie mismatch (%u/%u). Discarding.\n",
 669                                             tunnel->name, tunnel->tunnel_id,
 670                                             session->session_id);
 671                         atomic_long_inc(&session->stats.rx_cookie_discards);
 672                         goto discard;
 673                 }
 674                 ptr += session->peer_cookie_len;
 675         }
 676
 677         /* Handle the optional sequence numbers. Sequence numbers are
 678          * in different places for L2TPv2 and L2TPv3.
 679          *
 680          * If we are the LAC, enable/disable sequence numbers under
 681          * the control of the LNS.  If no sequence numbers present but
 682          * we were expecting them, discard frame.
 683          */
 684         L2TP_SKB_CB(skb)->has_seq = 0;
 685         if (tunnel->version == L2TP_HDR_VER_2) {
 686                 if (hdrflags & L2TP_HDRFLAG_S) {
 687                         /* Store L2TP info in the skb */
 688                         L2TP_SKB_CB(skb)->ns = ntohs(*(__be16 *)ptr);
 689                         L2TP_SKB_CB(skb)->has_seq = 1;
 690                         ptr += 2;
 691                         /* Skip past nr in the header */
 692                         ptr += 2;
 693
 694                 }
 695         } else if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 696                 u32 l2h = ntohl(*(__be32 *)ptr);
 697
 698                 if (l2h & 0x40000000) {
 699                         /* Store L2TP info in the skb */
 700                         L2TP_SKB_CB(skb)->ns = l2h & 0x00ffffff;
 701                         L2TP_SKB_CB(skb)->has_seq = 1;
 702                 }
 703                 ptr += 4;
 704         }
 705
 706         if (L2TP_SKB_CB(skb)->has_seq) {
 707                 /* Received a packet with sequence numbers. If we're the LAC,
 708                  * check if we sre sending sequence numbers and if not,
 709                  * configure it so.
 710                  */
 711                 if (!session->lns_mode && !session->send_seq) {
 712                         l2tp_info(session, L2TP_MSG_SEQ,
 713                                   "%s: requested to enable seq numbers by LNS\n",
 714                                   session->name);
 715                         session->send_seq = 1;
 716                         l2tp_session_set_header_len(session, tunnel->version);
 717                 }
 718         } else {
 719                 /* No sequence numbers.
 720                  * If user has configured mandatory sequence numbers, discard.
 721                  */
 722                 if (session->recv_seq) {
 723                         pr_warn_ratelimited("%s: recv data has no seq numbers when required. Discarding.\n",
 724                                             session->name);
 725                         atomic_long_inc(&session->stats.rx_seq_discards);
 726                         goto discard;
 727                 }
 728
 729                 /* If we're the LAC and we're sending sequence numbers, the
 730                  * LNS has requested that we no longer send sequence numbers.
 731                  * If we're the LNS and we're sending sequence numbers, the
 732                  * LAC is broken. Discard the frame.
 733                  */
 734                 if (!session->lns_mode && session->send_seq) {
 735                         l2tp_info(session, L2TP_MSG_SEQ,
 736                                   "%s: requested to disable seq numbers by LNS\n",
 737                                   session->name);
 738                         session->send_seq = 0;
 739                         l2tp_session_set_header_len(session, tunnel->version);
 740                 } else if (session->send_seq) {
 741                         pr_warn_ratelimited("%s: recv data has no seq numbers when required. Discarding.\n",
 742                                             session->name);
 743                         atomic_long_inc(&session->stats.rx_seq_discards);
 744                         goto discard;
 745                 }
 746         }
 747
 748         /* Session data offset is defined only for L2TPv2 and is
 749          * indicated by an optional 16-bit value in the header.
 750          */
 751         if (tunnel->version == L2TP_HDR_VER_2) {
 752                 /* If offset bit set, skip it. */
 753                 if (hdrflags & L2TP_HDRFLAG_O) {
 754                         offset = ntohs(*(__be16 *)ptr);
 755                         ptr += 2 + offset;
 756                 }
 757         }
 758
 759         offset = ptr - optr;
 760         if (!pskb_may_pull(skb, offset))
 761                 goto discard;
 762
 763         __skb_pull(skb, offset);
 764
 765         /* Prepare skb for adding to the session's reorder_q.  Hold
 766          * packets for max reorder_timeout or 1 second if not
 767          * reordering.
 768          */
 769         L2TP_SKB_CB(skb)->length = length;
 770         L2TP_SKB_CB(skb)->expires = jiffies +
 771                 (session->reorder_timeout ? session->reorder_timeout : HZ);
 772
 773         /* Add packet to the session's receive queue. Reordering is done here, if
 774          * enabled. Saved L2TP protocol info is stored in skb->sb[].
 775          */
 776         if (L2TP_SKB_CB(skb)->has_seq) {
 777                 if (l2tp_recv_data_seq(session, skb))
 778                         goto discard;
 779         } else {
 780                 /* No sequence numbers. Add the skb to the tail of the
 781                  * reorder queue. This ensures that it will be
 782                  * delivered after all previous sequenced skbs.
 783                  */
 784                 skb_queue_tail(&session->reorder_q, skb);
 785         }
 786
 787         /* Try to dequeue as many skbs from reorder_q as we can. */
 788         l2tp_recv_dequeue(session);
 789
 790         return;
 791
 792 discard:
 793         atomic_long_inc(&session->stats.rx_errors);
 794         kfree_skb(skb);
 795 }
 796 EXPORT_SYMBOL_GPL(l2tp_recv_common);
 797
 798 /* Drop skbs from the session's reorder_q
 799  */
 800 static void l2tp_session_queue_purge(struct l2tp_session *session)
 801 {
 802         struct sk_buff *skb = NULL;
 803
 804         if (WARN_ON(session->magic != L2TP_SESSION_MAGIC))
 805                 return;
 806
 807         while ((skb = skb_dequeue(&session->reorder_q))) {
 808                 atomic_long_inc(&session->stats.rx_errors);
 809                 kfree_skb(skb);
 810         }
 811 }
 812
 813 /* Internal UDP receive frame. Do the real work of receiving an L2TP data frame
 814  * here. The skb is not on a list when we get here.
 815  * Returns 0 if the packet was a data packet and was successfully passed on.
 816  * Returns 1 if the packet was not a good data packet and could not be
 817  * forwarded.  All such packets are passed up to userspace to deal with.
 818  */
 819 static int l2tp_udp_recv_core(struct l2tp_tunnel *tunnel, struct sk_buff *skb)
 820 {
 821         struct l2tp_session *session = NULL;
 822         unsigned char *ptr, *optr;
 823         u16 hdrflags;
 824         u32 tunnel_id, session_id;
 825         u16 version;
 826         int length;
 827
 828         /* UDP has verifed checksum */
 829
 830         /* UDP always verifies the packet length. */
 831         __skb_pull(skb, sizeof(struct udphdr));
 832
 833         /* Short packet? */
 834         if (!pskb_may_pull(skb, L2TP_HDR_SIZE_MAX)) {
 835                 pr_warn_ratelimited("%s: recv short packet (len=%d)\n",
 836                                     tunnel->name, skb->len);
 837                 goto error;
 838         }
 839
 840         /* Point to L2TP header */
 841         optr = skb->data;
 842         ptr = skb->data;
 843
 844         /* Get L2TP header flags */
 845         hdrflags = ntohs(*(__be16 *)ptr);
 846
 847         /* Check protocol version */
 848         version = hdrflags & L2TP_HDR_VER_MASK;
 849         if (version != tunnel->version) {
 850                 pr_warn_ratelimited("%s: recv protocol version mismatch: got %d expected %d\n",
 851                                     tunnel->name, version, tunnel->version);
 852                 goto error;
 853         }
 854
 855         /* Get length of L2TP packet */
 856         length = skb->len;
 857
 858         /* If type is control packet, it is handled by userspace. */
 859         if (hdrflags & L2TP_HDRFLAG_T)
 860                 goto error;
 861
 862         /* Skip flags */
 863         ptr += 2;
 864
 865         if (tunnel->version == L2TP_HDR_VER_2) {
 866                 /* If length is present, skip it */
 867                 if (hdrflags & L2TP_HDRFLAG_L)
 868                         ptr += 2;
 869
 870                 /* Extract tunnel and session ID */
 871                 tunnel_id = ntohs(*(__be16 *)ptr);
 872                 ptr += 2;
 873                 session_id = ntohs(*(__be16 *)ptr);
 874                 ptr += 2;
 875         } else {
 876                 ptr += 2;       /* skip reserved bits */
 877                 tunnel_id = tunnel->tunnel_id;
 878                 session_id = ntohl(*(__be32 *)ptr);
 879                 ptr += 4;
 880         }
 881
 882         /* Find the session context */
 883         session = l2tp_tunnel_get_session(tunnel, session_id);
 884         if (!session || !session->recv_skb) {
 885                 if (session)
 886                         l2tp_session_dec_refcount(session);
 887
 888                 /* Not found? Pass to userspace to deal with */
 889                 pr_warn_ratelimited("%s: no session found (%u/%u). Passing up.\n",
 890                                     tunnel->name, tunnel_id, session_id);
 891                 goto error;
 892         }
 893
 894         if (tunnel->version == L2TP_HDR_VER_3 &&
 895             l2tp_v3_ensure_opt_in_linear(session, skb, &ptr, &optr))
 896                 goto error;
 897
 898         l2tp_recv_common(session, skb, ptr, optr, hdrflags, length);
 899         l2tp_session_dec_refcount(session);
 900
 901         return 0;
 902
 903 error:
 904         /* Put UDP header back */
 905         __skb_push(skb, sizeof(struct udphdr));
 906
 907         return 1;
 908 }
 909
 910 /* UDP encapsulation receive handler. See net/ipv4/udp.c.
 911  * Return codes:
 912  * 0 : success.
 913  * <0: error
 914  * >0: skb should be passed up to userspace as UDP.
 915  */
 916 int l2tp_udp_encap_recv(struct sock *sk, struct sk_buff *skb)
 917 {
 918         struct l2tp_tunnel *tunnel;
 919
 920         tunnel = rcu_dereference_sk_user_data(sk);
 921         if (!tunnel)
 922                 goto pass_up;
 923
 924         if (l2tp_udp_recv_core(tunnel, skb))
 925                 goto pass_up;
 926
 927         return 0;
 928
 929 pass_up:
 930         return 1;
 931 }
 932 EXPORT_SYMBOL_GPL(l2tp_udp_encap_recv);
 933
 934 /************************************************************************
 935  * Transmit handling
 936  ***********************************************************************/
 937
 938 /* Build an L2TP header for the session into the buffer provided.
 939  */
 940 static int l2tp_build_l2tpv2_header(struct l2tp_session *session, void *buf)
 941 {
 942         struct l2tp_tunnel *tunnel = session->tunnel;
 943         __be16 *bufp = buf;
 944         __be16 *optr = buf;
 945         u16 flags = L2TP_HDR_VER_2;
 946         u32 tunnel_id = tunnel->peer_tunnel_id;
 947         u32 session_id = session->peer_session_id;
 948
 949         if (session->send_seq)
 950                 flags |= L2TP_HDRFLAG_S;
 951
 952         /* Setup L2TP header. */
 953         *bufp++ = htons(flags);
 954         *bufp++ = htons(tunnel_id);
 955         *bufp++ = htons(session_id);
 956         if (session->send_seq) {
 957                 *bufp++ = htons(session->ns);
 958                 *bufp++ = 0;
 959                 session->ns++;
 960                 session->ns &= 0xffff;
 961                 l2tp_dbg(session, L2TP_MSG_SEQ, "%s: updated ns to %u\n",
 962                          session->name, session->ns);
 963         }
 964
 965         return bufp - optr;
 966 }
 967
 968 static int l2tp_build_l2tpv3_header(struct l2tp_session *session, void *buf)
 969 {
 970         struct l2tp_tunnel *tunnel = session->tunnel;
 971         char *bufp = buf;
 972         char *optr = bufp;
 973
 974         /* Setup L2TP header. The header differs slightly for UDP and
 975          * IP encapsulations. For UDP, there is 4 bytes of flags.
 976          */
 977         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
 978                 u16 flags = L2TP_HDR_VER_3;
 979                 *((__be16 *)bufp) = htons(flags);
 980                 bufp += 2;
 981                 *((__be16 *)bufp) = 0;
 982                 bufp += 2;
 983         }
 984
 985         *((__be32 *)bufp) = htonl(session->peer_session_id);
 986         bufp += 4;
 987         if (session->cookie_len) {
 988                 memcpy(bufp, &session->cookie[0], session->cookie_len);
 989                 bufp += session->cookie_len;
 990         }
 991         if (session->l2specific_type == L2TP_L2SPECTYPE_DEFAULT) {
 992                 u32 l2h = 0;
 993
 994                 if (session->send_seq) {
 995                         l2h = 0x40000000 | session->ns;
 996                         session->ns++;
 997                         session->ns &= 0xffffff;
 998                         l2tp_dbg(session, L2TP_MSG_SEQ,
 999                                  "%s: updated ns to %u\n",
1000                                  session->name, session->ns);
1001                 }
1002
1003                 *((__be32 *)bufp) = htonl(l2h);
1004                 bufp += 4;
1005         }
1006
1007         return bufp - optr;
1008 }
1009
1010 static void l2tp_xmit_core(struct l2tp_session *session, struct sk_buff *skb,
1011                            struct flowi *fl, size_t data_len)
1012 {
1013         struct l2tp_tunnel *tunnel = session->tunnel;
1014         unsigned int len = skb->len;
1015         int error;
1016
1017         /* Queue the packet to IP for output */
1018         skb->ignore_df = 1;
1019         skb_dst_drop(skb);
1020 #if IS_ENABLED(CONFIG_IPV6)
1021         if (l2tp_sk_is_v6(tunnel->sock))
1022                 error = inet6_csk_xmit(tunnel->sock, skb, NULL);
1023         else
1024 #endif
1025                 error = ip_queue_xmit(tunnel->sock, skb, fl);
1026
1027         /* Update stats */
1028         if (error >= 0) {
1029                 atomic_long_inc(&tunnel->stats.tx_packets);
1030                 atomic_long_add(len, &tunnel->stats.tx_bytes);
1031                 atomic_long_inc(&session->stats.tx_packets);
1032                 atomic_long_add(len, &session->stats.tx_bytes);
1033         } else {
1034                 atomic_long_inc(&tunnel->stats.tx_errors);
1035                 atomic_long_inc(&session->stats.tx_errors);
1036         }
1037 }
1038
1039 /* If caller requires the skb to have a ppp header, the header must be
1040  * inserted in the skb data before calling this function.
1041  */
1042 int l2tp_xmit_skb(struct l2tp_session *session, struct sk_buff *skb, int hdr_len)
1043 {
1044         int data_len = skb->len;
1045         struct l2tp_tunnel *tunnel = session->tunnel;
1046         struct sock *sk = tunnel->sock;
1047         struct flowi *fl;
1048         struct udphdr *uh;
1049         struct inet_sock *inet;
1050         int headroom;
1051         int uhlen = (tunnel->encap == L2TP_ENCAPTYPE_UDP) ? sizeof(struct udphdr) : 0;
1052         int udp_len;
1053         int ret = NET_XMIT_SUCCESS;
1054
1055         /* Check that there's enough headroom in the skb to insert IP,
1056          * UDP and L2TP headers. If not enough, expand it to
1057          * make room. Adjust truesize.
1058          */
1059         headroom = NET_SKB_PAD + sizeof(struct iphdr) +
1060                 uhlen + hdr_len;
1061         if (skb_cow_head(skb, headroom)) {
1062                 kfree_skb(skb);
1063                 return NET_XMIT_DROP;
1064         }
1065
1066         /* Setup L2TP header */
1067         if (tunnel->version == L2TP_HDR_VER_2)
1068                 l2tp_build_l2tpv2_header(session, __skb_push(skb, hdr_len));
1069         else
1070                 l2tp_build_l2tpv3_header(session, __skb_push(skb, hdr_len));
1071
1072         /* Reset skb netfilter state */
1073         memset(&(IPCB(skb)->opt), 0, sizeof(IPCB(skb)->opt));
1074         IPCB(skb)->flags &= ~(IPSKB_XFRM_TUNNEL_SIZE | IPSKB_XFRM_TRANSFORMED |
1075                               IPSKB_REROUTED);
1076         nf_reset_ct(skb);
1077
1078         bh_lock_sock(sk);
1079         if (sock_owned_by_user(sk)) {
1080                 kfree_skb(skb);
1081                 ret = NET_XMIT_DROP;
1082                 goto out_unlock;
1083         }
1084
1085         /* The user-space may change the connection status for the user-space
1086          * provided socket at run time: we must check it under the socket lock
1087          */
1088         if (tunnel->fd >= 0 && sk->sk_state != TCP_ESTABLISHED) {
1089                 kfree_skb(skb);
1090                 ret = NET_XMIT_DROP;
1091                 goto out_unlock;
1092         }
1093
1094         inet = inet_sk(sk);
1095         fl = &inet->cork.fl;
1096         switch (tunnel->encap) {
1097         case L2TP_ENCAPTYPE_UDP:
1098                 /* Setup UDP header */
1099                 __skb_push(skb, sizeof(*uh));
1100                 skb_reset_transport_header(skb);
1101                 uh = udp_hdr(skb);
1102                 uh->source = inet->inet_sport;
1103                 uh->dest = inet->inet_dport;
1104                 udp_len = uhlen + hdr_len + data_len;
1105                 uh->len = htons(udp_len);
1106
1107                 /* Calculate UDP checksum if configured to do so */
1108 #if IS_ENABLED(CONFIG_IPV6)
1109                 if (l2tp_sk_is_v6(sk))
1110                         udp6_set_csum(udp_get_no_check6_tx(sk),
1111                                       skb, &inet6_sk(sk)->saddr,
1112                                       &sk->sk_v6_daddr, udp_len);
1113                 else
1114 #endif
1115                         udp_set_csum(sk->sk_no_check_tx, skb, inet->inet_saddr,
1116                                      inet->inet_daddr, udp_len);
1117                 break;
1118
1119         case L2TP_ENCAPTYPE_IP:
1120                 break;
1121         }
1122
1123         l2tp_xmit_core(session, skb, fl, data_len);
1124 out_unlock:
1125         bh_unlock_sock(sk);
1126
1127         return ret;
1128 }
1129 EXPORT_SYMBOL_GPL(l2tp_xmit_skb);
1130
1131 /*****************************************************************************
1132  * Tinnel and session create/destroy.
1133  *****************************************************************************/
1134
1135 /* Tunnel socket destruct hook.
1136  * The tunnel context is deleted only when all session sockets have been
1137  * closed.
1138  */
1139 static void l2tp_tunnel_destruct(struct sock *sk)
1140 {
1141         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1142
1143         if (!tunnel)
1144                 goto end;
1145
1146         /* Disable udp encapsulation */
1147         switch (tunnel->encap) {
1148         case L2TP_ENCAPTYPE_UDP:
1149                 /* No longer an encapsulation socket. See net/ipv4/udp.c */
1150                 (udp_sk(sk))->encap_type = 0;
1151                 (udp_sk(sk))->encap_rcv = NULL;
1152                 (udp_sk(sk))->encap_destroy = NULL;
1153                 break;
1154         case L2TP_ENCAPTYPE_IP:
1155                 break;
1156         }
1157
1158         /* Remove hooks into tunnel socket */
1159         sk->sk_destruct = tunnel->old_sk_destruct;
1160         sk->sk_user_data = NULL;
1161
1162         /* Call the original destructor */
1163         if (sk->sk_destruct)
1164                 (*sk->sk_destruct)(sk);
1165
1166         kfree_rcu(tunnel, rcu);
1167 end:
1168         return;
1169 }
1170
1171 /* Remove an l2tp session from l2tp_core's hash lists. */
1172 static void l2tp_session_unhash(struct l2tp_session *session)
1173 {
1174         struct l2tp_tunnel *tunnel = session->tunnel;
1175
1176         /* Remove the session from core hashes */
1177         if (tunnel) {
1178                 /* Remove from the per-tunnel hash */
1179                 write_lock_bh(&tunnel->hlist_lock);
1180                 hlist_del_init(&session->hlist);
1181                 write_unlock_bh(&tunnel->hlist_lock);
1182
1183                 /* For L2TPv3 we have a per-net hash: remove from there, too */
1184                 if (tunnel->version != L2TP_HDR_VER_2) {
1185                         struct l2tp_net *pn = l2tp_pernet(tunnel->l2tp_net);
1186
1187                         spin_lock_bh(&pn->l2tp_session_hlist_lock);
1188                         hlist_del_init_rcu(&session->global_hlist);
1189                         spin_unlock_bh(&pn->l2tp_session_hlist_lock);
1190                         synchronize_rcu();
1191                 }
1192         }
1193 }
1194
1195 /* When the tunnel is closed, all the attached sessions need to go too.
1196  */
1197 static void l2tp_tunnel_closeall(struct l2tp_tunnel *tunnel)
1198 {
1199         int hash;
1200         struct hlist_node *walk;
1201         struct hlist_node *tmp;
1202         struct l2tp_session *session;
1203
1204         write_lock_bh(&tunnel->hlist_lock);
1205         tunnel->acpt_newsess = false;
1206         for (hash = 0; hash < L2TP_HASH_SIZE; hash++) {
1207 again:
1208                 hlist_for_each_safe(walk, tmp, &tunnel->session_hlist[hash]) {
1209                         session = hlist_entry(walk, struct l2tp_session, hlist);
1210
1211                         hlist_del_init(&session->hlist);
1212
1213                         if (test_and_set_bit(0, &session->dead))
1214                                 goto again;
1215
1216                         write_unlock_bh(&tunnel->hlist_lock);
1217
1218                         l2tp_session_unhash(session);
1219                         l2tp_session_queue_purge(session);
1220
1221                         if (session->session_close)
1222                                 (*session->session_close)(session);
1223
1224                         l2tp_session_dec_refcount(session);
1225
1226                         write_lock_bh(&tunnel->hlist_lock);
1227
1228                         /* Now restart from the beginning of this hash
1229                          * chain.  We always remove a session from the
1230                          * list so we are guaranteed to make forward
1231                          * progress.
1232                          */
1233                         goto again;
1234                 }
1235         }
1236         write_unlock_bh(&tunnel->hlist_lock);
1237 }
1238
1239 /* Tunnel socket destroy hook for UDP encapsulation */
1240 static void l2tp_udp_encap_destroy(struct sock *sk)
1241 {
1242         struct l2tp_tunnel *tunnel = l2tp_tunnel(sk);
1243
1244         if (tunnel)
1245                 l2tp_tunnel_delete(tunnel);
1246 }
1247
1248 /* Workqueue tunnel deletion function */
1249 static void l2tp_tunnel_del_work(struct work_struct *work)
1250 {
1251         struct l2tp_tunnel *tunnel = container_of(work, struct l2tp_tunnel,
1252                                                   del_work);
1253         struct sock *sk = tunnel->sock;
1254         struct socket *sock = sk->sk_socket;
1255         struct l2tp_net *pn;
1256
1257         l2tp_tunnel_closeall(tunnel);
1258
1259         /* If the tunnel socket was created within the kernel, use
1260          * the sk API to release it here.
1261          */
1262         if (tunnel->fd < 0) {
1263                 if (sock) {
1264                         kernel_sock_shutdown(sock, SHUT_RDWR);
1265                         sock_release(sock);
1266                 }
1267         }
1268
1269         /* Remove the tunnel struct from the tunnel list */
1270         pn = l2tp_pernet(tunnel->l2tp_net);
1271         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1272         list_del_rcu(&tunnel->list);
1273         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1274
1275         /* drop initial ref */
1276         l2tp_tunnel_dec_refcount(tunnel);
1277
1278         /* drop workqueue ref */
1279         l2tp_tunnel_dec_refcount(tunnel);
1280 }
1281
1282 /* Create a socket for the tunnel, if one isn't set up by
1283  * userspace. This is used for static tunnels where there is no
1284  * managing L2TP daemon.
1285  *
1286  * Since we don't want these sockets to keep a namespace alive by
1287  * themselves, we drop the socket's namespace refcount after creation.
1288  * These sockets are freed when the namespace exits using the pernet
1289  * exit hook.
1290  */
1291 static int l2tp_tunnel_sock_create(struct net *net,
1292                                    u32 tunnel_id,
1293                                    u32 peer_tunnel_id,
1294                                    struct l2tp_tunnel_cfg *cfg,
1295                                    struct socket **sockp)
1296 {
1297         int err = -EINVAL;
1298         struct socket *sock = NULL;
1299         struct udp_port_cfg udp_conf;
1300
1301         switch (cfg->encap) {
1302         case L2TP_ENCAPTYPE_UDP:
1303                 memset(&udp_conf, 0, sizeof(udp_conf));
1304
1305 #if IS_ENABLED(CONFIG_IPV6)
1306                 if (cfg->local_ip6 && cfg->peer_ip6) {
1307                         udp_conf.family = AF_INET6;
1308                         memcpy(&udp_conf.local_ip6, cfg->local_ip6,
1309                                sizeof(udp_conf.local_ip6));
1310                         memcpy(&udp_conf.peer_ip6, cfg->peer_ip6,
1311                                sizeof(udp_conf.peer_ip6));
1312                         udp_conf.use_udp6_tx_checksums =
1313                           !cfg->udp6_zero_tx_checksums;
1314                         udp_conf.use_udp6_rx_checksums =
1315                           !cfg->udp6_zero_rx_checksums;
1316                 } else
1317 #endif
1318                 {
1319                         udp_conf.family = AF_INET;
1320                         udp_conf.local_ip = cfg->local_ip;
1321                         udp_conf.peer_ip = cfg->peer_ip;
1322                         udp_conf.use_udp_checksums = cfg->use_udp_checksums;
1323                 }
1324
1325                 udp_conf.local_udp_port = htons(cfg->local_udp_port);
1326                 udp_conf.peer_udp_port = htons(cfg->peer_udp_port);
1327
1328                 err = udp_sock_create(net, &udp_conf, &sock);
1329                 if (err < 0)
1330                         goto out;
1331
1332                 break;
1333
1334         case L2TP_ENCAPTYPE_IP:
1335 #if IS_ENABLED(CONFIG_IPV6)
1336                 if (cfg->local_ip6 && cfg->peer_ip6) {
1337                         struct sockaddr_l2tpip6 ip6_addr = {0};
1338
1339                         err = sock_create_kern(net, AF_INET6, SOCK_DGRAM,
1340                                                IPPROTO_L2TP, &sock);
1341                         if (err < 0)
1342                                 goto out;
1343
1344                         ip6_addr.l2tp_family = AF_INET6;
1345                         memcpy(&ip6_addr.l2tp_addr, cfg->local_ip6,
1346                                sizeof(ip6_addr.l2tp_addr));
1347                         ip6_addr.l2tp_conn_id = tunnel_id;
1348                         err = kernel_bind(sock, (struct sockaddr *)&ip6_addr,
1349                                           sizeof(ip6_addr));
1350                         if (err < 0)
1351                                 goto out;
1352
1353                         ip6_addr.l2tp_family = AF_INET6;
1354                         memcpy(&ip6_addr.l2tp_addr, cfg->peer_ip6,
1355                                sizeof(ip6_addr.l2tp_addr));
1356                         ip6_addr.l2tp_conn_id = peer_tunnel_id;
1357                         err = kernel_connect(sock,
1358                                              (struct sockaddr *)&ip6_addr,
1359                                              sizeof(ip6_addr), 0);
1360                         if (err < 0)
1361                                 goto out;
1362                 } else
1363 #endif
1364                 {
1365                         struct sockaddr_l2tpip ip_addr = {0};
1366
1367                         err = sock_create_kern(net, AF_INET, SOCK_DGRAM,
1368                                                IPPROTO_L2TP, &sock);
1369                         if (err < 0)
1370                                 goto out;
1371
1372                         ip_addr.l2tp_family = AF_INET;
1373                         ip_addr.l2tp_addr = cfg->local_ip;
1374                         ip_addr.l2tp_conn_id = tunnel_id;
1375                         err = kernel_bind(sock, (struct sockaddr *)&ip_addr,
1376                                           sizeof(ip_addr));
1377                         if (err < 0)
1378                                 goto out;
1379
1380                         ip_addr.l2tp_family = AF_INET;
1381                         ip_addr.l2tp_addr = cfg->peer_ip;
1382                         ip_addr.l2tp_conn_id = peer_tunnel_id;
1383                         err = kernel_connect(sock, (struct sockaddr *)&ip_addr,
1384                                              sizeof(ip_addr), 0);
1385                         if (err < 0)
1386                                 goto out;
1387                 }
1388                 break;
1389
1390         default:
1391                 goto out;
1392         }
1393
1394 out:
1395         *sockp = sock;
1396         if (err < 0 && sock) {
1397                 kernel_sock_shutdown(sock, SHUT_RDWR);
1398                 sock_release(sock);
1399                 *sockp = NULL;
1400         }
1401
1402         return err;
1403 }
1404
1405 static struct lock_class_key l2tp_socket_class;
1406
1407 int l2tp_tunnel_create(struct net *net, int fd, int version, u32 tunnel_id, u32 peer_tunnel_id,
1408                        struct l2tp_tunnel_cfg *cfg, struct l2tp_tunnel **tunnelp)
1409 {
1410         struct l2tp_tunnel *tunnel = NULL;
1411         int err;
1412         enum l2tp_encap_type encap = L2TP_ENCAPTYPE_UDP;
1413
1414         if (cfg)
1415                 encap = cfg->encap;
1416
1417         tunnel = kzalloc(sizeof(*tunnel), GFP_KERNEL);
1418         if (!tunnel) {
1419                 err = -ENOMEM;
1420                 goto err;
1421         }
1422
1423         tunnel->version = version;
1424         tunnel->tunnel_id = tunnel_id;
1425         tunnel->peer_tunnel_id = peer_tunnel_id;
1426         tunnel->debug = L2TP_DEFAULT_DEBUG_FLAGS;
1427
1428         tunnel->magic = L2TP_TUNNEL_MAGIC;
1429         sprintf(&tunnel->name[0], "tunl %u", tunnel_id);
1430         rwlock_init(&tunnel->hlist_lock);
1431         tunnel->acpt_newsess = true;
1432
1433         if (cfg)
1434                 tunnel->debug = cfg->debug;
1435
1436         tunnel->encap = encap;
1437
1438         refcount_set(&tunnel->ref_count, 1);
1439         tunnel->fd = fd;
1440
1441         /* Init delete workqueue struct */
1442         INIT_WORK(&tunnel->del_work, l2tp_tunnel_del_work);
1443
1444         INIT_LIST_HEAD(&tunnel->list);
1445
1446         err = 0;
1447 err:
1448         if (tunnelp)
1449                 *tunnelp = tunnel;
1450
1451         return err;
1452 }
1453 EXPORT_SYMBOL_GPL(l2tp_tunnel_create);
1454
1455 static int l2tp_validate_socket(const struct sock *sk, const struct net *net,
1456                                 enum l2tp_encap_type encap)
1457 {
1458         if (!net_eq(sock_net(sk), net))
1459                 return -EINVAL;
1460
1461         if (sk->sk_type != SOCK_DGRAM)
1462                 return -EPROTONOSUPPORT;
1463
1464         if (sk->sk_family != PF_INET && sk->sk_family != PF_INET6)
1465                 return -EPROTONOSUPPORT;
1466
1467         if ((encap == L2TP_ENCAPTYPE_UDP && sk->sk_protocol != IPPROTO_UDP) ||
1468             (encap == L2TP_ENCAPTYPE_IP && sk->sk_protocol != IPPROTO_L2TP))
1469                 return -EPROTONOSUPPORT;
1470
1471         if (sk->sk_user_data)
1472                 return -EBUSY;
1473
1474         return 0;
1475 }
1476
1477 int l2tp_tunnel_register(struct l2tp_tunnel *tunnel, struct net *net,
1478                          struct l2tp_tunnel_cfg *cfg)
1479 {
1480         struct l2tp_tunnel *tunnel_walk;
1481         struct l2tp_net *pn;
1482         struct socket *sock;
1483         struct sock *sk;
1484         int ret;
1485
1486         if (tunnel->fd < 0) {
1487                 ret = l2tp_tunnel_sock_create(net, tunnel->tunnel_id,
1488                                               tunnel->peer_tunnel_id, cfg,
1489                                               &sock);
1490                 if (ret < 0)
1491                         goto err;
1492         } else {
1493                 sock = sockfd_lookup(tunnel->fd, &ret);
1494                 if (!sock)
1495                         goto err;
1496
1497                 ret = l2tp_validate_socket(sock->sk, net, tunnel->encap);
1498                 if (ret < 0)
1499                         goto err_sock;
1500         }
1501
1502         tunnel->l2tp_net = net;
1503         pn = l2tp_pernet(net);
1504
1505         spin_lock_bh(&pn->l2tp_tunnel_list_lock);
1506         list_for_each_entry(tunnel_walk, &pn->l2tp_tunnel_list, list) {
1507                 if (tunnel_walk->tunnel_id == tunnel->tunnel_id) {
1508                         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1509
1510                         ret = -EEXIST;
1511                         goto err_sock;
1512                 }
1513         }
1514         list_add_rcu(&tunnel->list, &pn->l2tp_tunnel_list);
1515         spin_unlock_bh(&pn->l2tp_tunnel_list_lock);
1516
1517         sk = sock->sk;
1518         sock_hold(sk);
1519         tunnel->sock = sk;
1520
1521         if (tunnel->encap == L2TP_ENCAPTYPE_UDP) {
1522                 struct udp_tunnel_sock_cfg udp_cfg = {
1523                         .sk_user_data = tunnel,
1524                         .encap_type = UDP_ENCAP_L2TPINUDP,
1525                         .encap_rcv = l2tp_udp_encap_recv,
1526                         .encap_destroy = l2tp_udp_encap_destroy,
1527                 };
1528
1529                 setup_udp_tunnel_sock(net, sock, &udp_cfg);
1530         } else {
1531                 sk->sk_user_data = tunnel;
1532         }
1533
1534         tunnel->old_sk_destruct = sk->sk_destruct;
1535         sk->sk_destruct = &l2tp_tunnel_destruct;
1536         lockdep_set_class_and_name(&sk->sk_lock.slock, &l2tp_socket_class,
1537                                    "l2tp_sock");
1538         sk->sk_allocation = GFP_ATOMIC;
1539
1540         if (tunnel->fd >= 0)
1541                 sockfd_put(sock);
1542
1543         return 0;
1544
1545 err_sock:
1546         if (tunnel->fd < 0)
1547                 sock_release(sock);
1548         else
1549                 sockfd_put(sock);
1550 err:
1551         return ret;
1552 }
1553 EXPORT_SYMBOL_GPL(l2tp_tunnel_register);
1554
1555 /* This function is used by the netlink TUNNEL_DELETE command.
1556  */
1557 void l2tp_tunnel_delete(struct l2tp_tunnel *tunnel)
1558 {
1559         if (!test_and_set_bit(0, &tunnel->dead)) {
1560                 l2tp_tunnel_inc_refcount(tunnel);
1561                 queue_work(l2tp_wq, &tunnel->del_work);
1562         }
1563 }
1564 EXPORT_SYMBOL_GPL(l2tp_tunnel_delete);
1565
1566 void l2tp_session_delete(struct l2tp_session *session)
1567 {
1568         if (test_and_set_bit(0, &session->dead))
1569                 return;
1570
1571         l2tp_session_unhash(session);
1572         l2tp_session_queue_purge(session);
1573         if (session->session_close)
1574                 (*session->session_close)(session);
1575
1576         l2tp_session_dec_refcount(session);
1577 }
1578 EXPORT_SYMBOL_GPL(l2tp_session_delete);
1579
1580 /* We come here whenever a session's send_seq, cookie_len or
1581  * l2specific_type parameters are set.
1582  */
1583 void l2tp_session_set_header_len(struct l2tp_session *session, int version)
1584 {
1585         if (version == L2TP_HDR_VER_2) {
1586                 session->hdr_len = 6;
1587                 if (session->send_seq)
1588                         session->hdr_len += 4;
1589         } else {
1590                 session->hdr_len = 4 + session->cookie_len;
1591                 session->hdr_len += l2tp_get_l2specific_len(session);
1592                 if (session->tunnel->encap == L2TP_ENCAPTYPE_UDP)
1593                         session->hdr_len += 4;
1594         }
1595 }
1596 EXPORT_SYMBOL_GPL(l2tp_session_set_header_len);
1597
1598 struct l2tp_session *l2tp_session_create(int priv_size, struct l2tp_tunnel *tunnel, u32 session_id,
1599                                          u32 peer_session_id, struct l2tp_session_cfg *cfg)
1600 {
1601         struct l2tp_session *session;
1602
1603         session = kzalloc(sizeof(*session) + priv_size, GFP_KERNEL);
1604         if (session) {
1605                 session->magic = L2TP_SESSION_MAGIC;
1606                 session->tunnel = tunnel;
1607
1608                 session->session_id = session_id;
1609                 session->peer_session_id = peer_session_id;
1610                 session->nr = 0;
1611                 if (tunnel->version == L2TP_HDR_VER_2)
1612                         session->nr_max = 0xffff;
1613                 else
1614                         session->nr_max = 0xffffff;
1615                 session->nr_window_size = session->nr_max / 2;
1616                 session->nr_oos_count_max = 4;
1617
1618                 /* Use NR of first received packet */
1619                 session->reorder_skip = 1;
1620
1621                 sprintf(&session->name[0], "sess %u/%u",
1622                         tunnel->tunnel_id, session->session_id);
1623
1624                 skb_queue_head_init(&session->reorder_q);
1625
1626                 INIT_HLIST_NODE(&session->hlist);
1627                 INIT_HLIST_NODE(&session->global_hlist);
1628
1629                 /* Inherit debug options from tunnel */
1630                 session->debug = tunnel->debug;
1631
1632                 if (cfg) {
1633                         session->pwtype = cfg->pw_type;
1634                         session->debug = cfg->debug;
1635                         session->send_seq = cfg->send_seq;
1636                         session->recv_seq = cfg->recv_seq;
1637                         session->lns_mode = cfg->lns_mode;
1638                         session->reorder_timeout = cfg->reorder_timeout;
1639                         session->l2specific_type = cfg->l2specific_type;
1640                         session->cookie_len = cfg->cookie_len;
1641                         memcpy(&session->cookie[0], &cfg->cookie[0], cfg->cookie_len);
1642                         session->peer_cookie_len = cfg->peer_cookie_len;
1643                         memcpy(&session->peer_cookie[0], &cfg->peer_cookie[0], cfg->peer_cookie_len);
1644                 }
1645
1646                 l2tp_session_set_header_len(session, tunnel->version);
1647
1648                 refcount_set(&session->ref_count, 1);
1649
1650                 return session;
1651         }
1652
1653         return ERR_PTR(-ENOMEM);
1654 }
1655 EXPORT_SYMBOL_GPL(l2tp_session_create);
1656
1657 /*****************************************************************************
1658  * Init and cleanup
1659  *****************************************************************************/
1660
1661 static __net_init int l2tp_init_net(struct net *net)
1662 {
1663         struct l2tp_net *pn = net_generic(net, l2tp_net_id);
1664         int hash;
1665
1666         INIT_LIST_HEAD(&pn->l2tp_tunnel_list);
1667         spin_lock_init(&pn->l2tp_tunnel_list_lock);
1668
1669         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1670                 INIT_HLIST_HEAD(&pn->l2tp_session_hlist[hash]);
1671
1672         spin_lock_init(&pn->l2tp_session_hlist_lock);
1673
1674         return 0;
1675 }
1676
1677 static __net_exit void l2tp_exit_net(struct net *net)
1678 {
1679         struct l2tp_net *pn = l2tp_pernet(net);
1680         struct l2tp_tunnel *tunnel = NULL;
1681         int hash;
1682
1683         rcu_read_lock_bh();
1684         list_for_each_entry_rcu(tunnel, &pn->l2tp_tunnel_list, list) {
1685                 l2tp_tunnel_delete(tunnel);
1686         }
1687         rcu_read_unlock_bh();
1688
1689         if (l2tp_wq)
1690                 flush_workqueue(l2tp_wq);
1691         rcu_barrier();
1692
1693         for (hash = 0; hash < L2TP_HASH_SIZE_2; hash++)
1694                 WARN_ON_ONCE(!hlist_empty(&pn->l2tp_session_hlist[hash]));
1695 }
1696
1697 static struct pernet_operations l2tp_net_ops = {
1698         .init = l2tp_init_net,
1699         .exit = l2tp_exit_net,
1700         .id   = &l2tp_net_id,
1701         .size = sizeof(struct l2tp_net),
1702 };
1703
1704 static int __init l2tp_init(void)
1705 {
1706         int rc = 0;
1707
1708         rc = register_pernet_device(&l2tp_net_ops);
1709         if (rc)
1710                 goto out;
1711
1712         l2tp_wq = alloc_workqueue("l2tp", WQ_UNBOUND, 0);
1713         if (!l2tp_wq) {
1714                 pr_err("alloc_workqueue failed\n");
1715                 unregister_pernet_device(&l2tp_net_ops);
1716                 rc = -ENOMEM;
1717                 goto out;
1718         }
1719
1720         pr_info("L2TP core driver, %s\n", L2TP_DRV_VERSION);
1721
1722 out:
1723         return rc;
1724 }
1725
1726 static void __exit l2tp_exit(void)
1727 {
1728         unregister_pernet_device(&l2tp_net_ops);
1729         if (l2tp_wq) {
1730                 destroy_workqueue(l2tp_wq);
1731                 l2tp_wq = NULL;
1732         }
1733 }
1734
1735 module_init(l2tp_init);
1736 module_exit(l2tp_exit);
1737
1738 MODULE_AUTHOR("James Chapman <jchapman@katalix.com>");
1739 MODULE_DESCRIPTION("L2TP core");
1740 MODULE_LICENSE("GPL");
1741 MODULE_VERSION(L2TP_DRV_VERSION);